xref: /dflybsd-src/sys/vm/vm_mmap.c (revision 201c8c4447cad562e0a54ebbe0e7ee4e8a0be647)
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 1.39 2007/04/30 07:18:57 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/priv.h>
57 #include <sys/resource.h>
58 #include <sys/resourcevar.h>
59 #include <sys/vnode.h>
60 #include <sys/fcntl.h>
61 #include <sys/file.h>
62 #include <sys/mman.h>
63 #include <sys/conf.h>
64 #include <sys/stat.h>
65 #include <sys/vmmeter.h>
66 #include <sys/sysctl.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_param.h>
70 #include <sys/lock.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 #include <sys/file2.h>
82 #include <sys/thread2.h>
83 
84 static int max_proc_mmap;
85 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
86 int vkernel_enable;
87 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, "");
88 
89 /*
90  * Set the maximum number of vm_map_entry structures per process.  Roughly
91  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
92  * of our KVM malloc space still results in generous limits.  We want a
93  * default that is good enough to prevent the kernel running out of resources
94  * if attacked from compromised user account but generous enough such that
95  * multi-threaded processes are not unduly inconvenienced.
96  */
97 
98 static void vmmapentry_rsrc_init (void *);
99 SYSINIT(vmmersrc, SI_BOOT1_POST, SI_ORDER_ANY, vmmapentry_rsrc_init, NULL)
100 
101 static void
102 vmmapentry_rsrc_init(void *dummy)
103 {
104     max_proc_mmap = KvaSize / 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 | MAP_TRYFIXED)) {
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_MAX_USER_ADDRESS > 0 && addr + size > VM_MAX_USER_ADDRESS)
228 			return (EINVAL);
229 		if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
230 			return (EINVAL);
231 		if (addr + size < addr)
232 			return (EINVAL);
233 	} else {
234 		/*
235 		 * Set a reasonable start point for the hint if it was
236 		 * not specified or if it falls within the heap space.
237 		 * Hinted mmap()s do not allocate out of the heap space.
238 		 */
239 		if (addr == 0 ||
240 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
241 		     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
242 			addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
243 	}
244 
245 	if (flags & MAP_ANON) {
246 		/*
247 		 * Mapping blank space is trivial.
248 		 */
249 		handle = NULL;
250 		maxprot = VM_PROT_ALL;
251 		pos = 0;
252 	} else {
253 		/*
254 		 * Mapping file, get fp for validation. Obtain vnode and make
255 		 * sure it is of appropriate type.
256 		 */
257 		fp = holdfp(p->p_fd, fd, -1);
258 		if (fp == NULL)
259 			return (EBADF);
260 		if (fp->f_type != DTYPE_VNODE) {
261 			error = EINVAL;
262 			goto done;
263 		}
264 		/*
265 		 * POSIX shared-memory objects are defined to have
266 		 * kernel persistence, and are not defined to support
267 		 * read(2)/write(2) -- or even open(2).  Thus, we can
268 		 * use MAP_ASYNC to trade on-disk coherence for speed.
269 		 * The shm_open(3) library routine turns on the FPOSIXSHM
270 		 * flag to request this behavior.
271 		 */
272 		if (fp->f_flag & FPOSIXSHM)
273 			flags |= MAP_NOSYNC;
274 		vp = (struct vnode *) fp->f_data;
275 
276 		/*
277 		 * Validate the vnode for the operation.
278 		 */
279 		switch(vp->v_type) {
280 		case VREG:
281 			/*
282 			 * Get the proper underlying object
283 			 */
284 			if ((obj = vp->v_object) == NULL) {
285 				error = EINVAL;
286 				goto done;
287 			}
288 			KKASSERT((struct vnode *)obj->handle == vp);
289 			break;
290 		case VCHR:
291 			/*
292 			 * Make sure a device has not been revoked.
293 			 * Mappability is handled by the device layer.
294 			 */
295 			if (vp->v_rdev == NULL) {
296 				error = EBADF;
297 				goto done;
298 			}
299 			break;
300 		default:
301 			/*
302 			 * Nothing else is mappable.
303 			 */
304 			error = EINVAL;
305 			goto done;
306 		}
307 
308 		/*
309 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
310 		 * SunOS).
311 		 */
312 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
313 			handle = NULL;
314 			maxprot = VM_PROT_ALL;
315 			flags |= MAP_ANON;
316 			pos = 0;
317 		} else {
318 			/*
319 			 * cdevs does not provide private mappings of any kind.
320 			 */
321 			/*
322 			 * However, for XIG X server to continue to work,
323 			 * we should allow the superuser to do it anyway.
324 			 * We only allow it at securelevel < 1.
325 			 * (Because the XIG X server writes directly to video
326 			 * memory via /dev/mem, it should never work at any
327 			 * other securelevel.
328 			 * XXX this will have to go
329 			 */
330 			if (securelevel >= 1)
331 				disablexworkaround = 1;
332 			else
333 				disablexworkaround = priv_check(td, PRIV_ROOT);
334 			if (vp->v_type == VCHR && disablexworkaround &&
335 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
336 				error = EINVAL;
337 				goto done;
338 			}
339 			/*
340 			 * Ensure that file and memory protections are
341 			 * compatible.  Note that we only worry about
342 			 * writability if mapping is shared; in this case,
343 			 * current and max prot are dictated by the open file.
344 			 * XXX use the vnode instead?  Problem is: what
345 			 * credentials do we use for determination? What if
346 			 * proc does a setuid?
347 			 */
348 			maxprot = VM_PROT_EXECUTE;	/* ??? */
349 			if (fp->f_flag & FREAD) {
350 				maxprot |= VM_PROT_READ;
351 			} else if (prot & PROT_READ) {
352 				error = EACCES;
353 				goto done;
354 			}
355 			/*
356 			 * If we are sharing potential changes (either via
357 			 * MAP_SHARED or via the implicit sharing of character
358 			 * device mappings), and we are trying to get write
359 			 * permission although we opened it without asking
360 			 * for it, bail out.  Check for superuser, only if
361 			 * we're at securelevel < 1, to allow the XIG X server
362 			 * to continue to work.
363 			 */
364 
365 			if ((flags & MAP_SHARED) != 0 ||
366 			    (vp->v_type == VCHR && disablexworkaround)) {
367 				if ((fp->f_flag & FWRITE) != 0) {
368 					struct vattr va;
369 					if ((error = VOP_GETATTR(vp, &va))) {
370 						goto done;
371 					}
372 					if ((va.va_flags &
373 					    (IMMUTABLE|APPEND)) == 0) {
374 						maxprot |= VM_PROT_WRITE;
375 					} else if (prot & PROT_WRITE) {
376 						error = EPERM;
377 						goto done;
378 					}
379 				} else if ((prot & PROT_WRITE) != 0) {
380 					error = EACCES;
381 					goto done;
382 				}
383 			} else {
384 				maxprot |= VM_PROT_WRITE;
385 			}
386 			handle = (void *)vp;
387 		}
388 	}
389 
390 	/*
391 	 * Do not allow more then a certain number of vm_map_entry structures
392 	 * per process.  Scale with the number of rforks sharing the map
393 	 * to make the limit reasonable for threads.
394 	 */
395 	if (max_proc_mmap &&
396 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) {
397 		error = ENOMEM;
398 		goto done;
399 	}
400 
401 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
402 	    flags, handle, pos);
403 	if (error == 0)
404 		*res = (void *)(addr + pageoff);
405 done:
406 	if (fp)
407 		fdrop(fp);
408 	return (error);
409 }
410 
411 int
412 sys_mmap(struct mmap_args *uap)
413 {
414 	int error;
415 
416 	error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
417 			  uap->prot, uap->flags,
418 			  uap->fd, uap->pos, &uap->sysmsg_resultp);
419 
420 	return (error);
421 }
422 
423 /*
424  * msync_args(void *addr, int len, int flags)
425  */
426 int
427 sys_msync(struct msync_args *uap)
428 {
429 	struct proc *p = curproc;
430 	vm_offset_t addr;
431 	vm_size_t size, pageoff;
432 	int flags;
433 	vm_map_t map;
434 	int rv;
435 
436 	addr = (vm_offset_t) uap->addr;
437 	size = uap->len;
438 	flags = uap->flags;
439 
440 	pageoff = (addr & PAGE_MASK);
441 	addr -= pageoff;
442 	size += pageoff;
443 	size = (vm_size_t) round_page(size);
444 	if (addr + size < addr)
445 		return(EINVAL);
446 
447 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
448 		return (EINVAL);
449 
450 	map = &p->p_vmspace->vm_map;
451 
452 	/*
453 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
454 	 * pages with the region containing addr".  Unfortunately, we don't
455 	 * really keep track of individual mmaps so we approximate by flushing
456 	 * the range of the map entry containing addr. This can be incorrect
457 	 * if the region splits or is coalesced with a neighbor.
458 	 */
459 	if (size == 0) {
460 		vm_map_entry_t entry;
461 
462 		vm_map_lock_read(map);
463 		rv = vm_map_lookup_entry(map, addr, &entry);
464 		vm_map_unlock_read(map);
465 		if (rv == FALSE)
466 			return (EINVAL);
467 		addr = entry->start;
468 		size = entry->end - entry->start;
469 	}
470 
471 	/*
472 	 * Clean the pages and interpret the return value.
473 	 */
474 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
475 	    (flags & MS_INVALIDATE) != 0);
476 
477 	switch (rv) {
478 	case KERN_SUCCESS:
479 		break;
480 	case KERN_INVALID_ADDRESS:
481 		return (EINVAL);	/* Sun returns ENOMEM? */
482 	case KERN_FAILURE:
483 		return (EIO);
484 	default:
485 		return (EINVAL);
486 	}
487 
488 	return (0);
489 }
490 
491 /*
492  * munmap_args(void *addr, size_t len)
493  */
494 int
495 sys_munmap(struct munmap_args *uap)
496 {
497 	struct proc *p = curproc;
498 	vm_offset_t addr;
499 	vm_size_t size, pageoff;
500 	vm_map_t map;
501 
502 	addr = (vm_offset_t) uap->addr;
503 	size = uap->len;
504 
505 	pageoff = (addr & PAGE_MASK);
506 	addr -= pageoff;
507 	size += pageoff;
508 	size = (vm_size_t) round_page(size);
509 	if (addr + size < addr)
510 		return(EINVAL);
511 
512 	if (size == 0)
513 		return (0);
514 
515 	/*
516 	 * Check for illegal addresses.  Watch out for address wrap... Note
517 	 * that VM_*_ADDRESS are not constants due to casts (argh).
518 	 */
519 	if (VM_MAX_USER_ADDRESS > 0 && addr + size > VM_MAX_USER_ADDRESS)
520 		return (EINVAL);
521 	if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
522 		return (EINVAL);
523 	map = &p->p_vmspace->vm_map;
524 	/*
525 	 * Make sure entire range is allocated.
526 	 */
527 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
528 		return (EINVAL);
529 	/* returns nothing but KERN_SUCCESS anyway */
530 	vm_map_remove(map, addr, addr + size);
531 	return (0);
532 }
533 
534 /*
535  * mprotect_args(const void *addr, size_t len, int prot)
536  */
537 int
538 sys_mprotect(struct mprotect_args *uap)
539 {
540 	struct proc *p = curproc;
541 	vm_offset_t addr;
542 	vm_size_t size, pageoff;
543 	vm_prot_t prot;
544 
545 	addr = (vm_offset_t) uap->addr;
546 	size = uap->len;
547 	prot = uap->prot & VM_PROT_ALL;
548 #if defined(VM_PROT_READ_IS_EXEC)
549 	if (prot & VM_PROT_READ)
550 		prot |= VM_PROT_EXECUTE;
551 #endif
552 
553 	pageoff = (addr & PAGE_MASK);
554 	addr -= pageoff;
555 	size += pageoff;
556 	size = (vm_size_t) round_page(size);
557 	if (addr + size < addr)
558 		return(EINVAL);
559 
560 	switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
561 		FALSE)) {
562 	case KERN_SUCCESS:
563 		return (0);
564 	case KERN_PROTECTION_FAILURE:
565 		return (EACCES);
566 	}
567 	return (EINVAL);
568 }
569 
570 /*
571  * minherit_args(void *addr, size_t len, int inherit)
572  */
573 int
574 sys_minherit(struct minherit_args *uap)
575 {
576 	struct proc *p = curproc;
577 	vm_offset_t addr;
578 	vm_size_t size, pageoff;
579 	vm_inherit_t inherit;
580 
581 	addr = (vm_offset_t)uap->addr;
582 	size = uap->len;
583 	inherit = uap->inherit;
584 
585 	pageoff = (addr & PAGE_MASK);
586 	addr -= pageoff;
587 	size += pageoff;
588 	size = (vm_size_t) round_page(size);
589 	if (addr + size < addr)
590 		return(EINVAL);
591 
592 	switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
593 	    inherit)) {
594 	case KERN_SUCCESS:
595 		return (0);
596 	case KERN_PROTECTION_FAILURE:
597 		return (EACCES);
598 	}
599 	return (EINVAL);
600 }
601 
602 /*
603  * madvise_args(void *addr, size_t len, int behav)
604  */
605 /* ARGSUSED */
606 int
607 sys_madvise(struct madvise_args *uap)
608 {
609 	struct proc *p = curproc;
610 	vm_offset_t start, end;
611 
612 	/*
613 	 * Check for illegal behavior
614 	 */
615 	if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
616 		return (EINVAL);
617 	/*
618 	 * Check for illegal addresses.  Watch out for address wrap... Note
619 	 * that VM_*_ADDRESS are not constants due to casts (argh).
620 	 */
621 	if (VM_MAX_USER_ADDRESS > 0 &&
622 		((vm_offset_t) uap->addr + uap->len) > VM_MAX_USER_ADDRESS)
623 		return (EINVAL);
624 	if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
625 		return (EINVAL);
626 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
627 		return (EINVAL);
628 
629 	/*
630 	 * Since this routine is only advisory, we default to conservative
631 	 * behavior.
632 	 */
633 	start = trunc_page((vm_offset_t) uap->addr);
634 	end = round_page((vm_offset_t) uap->addr + uap->len);
635 
636 	return (vm_map_madvise(&p->p_vmspace->vm_map, start, end,
637 		uap->behav, 0));
638 }
639 
640 /*
641  * mcontrol_args(void *addr, size_t len, int behav, off_t value)
642  */
643 /* ARGSUSED */
644 int
645 sys_mcontrol(struct mcontrol_args *uap)
646 {
647 	struct proc *p = curproc;
648 	vm_offset_t start, end;
649 
650 	/*
651 	 * Check for illegal behavior
652 	 */
653 	if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
654 		return (EINVAL);
655 	/*
656 	 * Check for illegal addresses.  Watch out for address wrap... Note
657 	 * that VM_*_ADDRESS are not constants due to casts (argh).
658 	 */
659 	if (VM_MAX_USER_ADDRESS > 0 &&
660 		((vm_offset_t) uap->addr + uap->len) > VM_MAX_USER_ADDRESS)
661 		return (EINVAL);
662 	if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
663 		return (EINVAL);
664 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
665 		return (EINVAL);
666 
667 	/*
668 	 * Since this routine is only advisory, we default to conservative
669 	 * behavior.
670 	 */
671 	start = trunc_page((vm_offset_t) uap->addr);
672 	end = round_page((vm_offset_t) uap->addr + uap->len);
673 
674 	return (vm_map_madvise(&p->p_vmspace->vm_map, start, end,
675 			      uap->behav, uap->value));
676 }
677 
678 
679 /*
680  * mincore_args(const void *addr, size_t len, char *vec)
681  */
682 /* ARGSUSED */
683 int
684 sys_mincore(struct mincore_args *uap)
685 {
686 	struct proc *p = curproc;
687 	vm_offset_t addr, first_addr;
688 	vm_offset_t end, cend;
689 	pmap_t pmap;
690 	vm_map_t map;
691 	char *vec;
692 	int error;
693 	int vecindex, lastvecindex;
694 	vm_map_entry_t current;
695 	vm_map_entry_t entry;
696 	int mincoreinfo;
697 	unsigned int timestamp;
698 
699 	/*
700 	 * Make sure that the addresses presented are valid for user
701 	 * mode.
702 	 */
703 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
704 	end = addr + (vm_size_t)round_page(uap->len);
705 	if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS)
706 		return (EINVAL);
707 	if (end < addr)
708 		return (EINVAL);
709 
710 	/*
711 	 * Address of byte vector
712 	 */
713 	vec = uap->vec;
714 
715 	map = &p->p_vmspace->vm_map;
716 	pmap = vmspace_pmap(p->p_vmspace);
717 
718 	vm_map_lock_read(map);
719 RestartScan:
720 	timestamp = map->timestamp;
721 
722 	if (!vm_map_lookup_entry(map, addr, &entry))
723 		entry = entry->next;
724 
725 	/*
726 	 * Do this on a map entry basis so that if the pages are not
727 	 * in the current processes address space, we can easily look
728 	 * up the pages elsewhere.
729 	 */
730 	lastvecindex = -1;
731 	for(current = entry;
732 		(current != &map->header) && (current->start < end);
733 		current = current->next) {
734 
735 		/*
736 		 * ignore submaps (for now) or null objects
737 		 */
738 		if (current->maptype != VM_MAPTYPE_NORMAL &&
739 		    current->maptype != VM_MAPTYPE_VPAGETABLE) {
740 			continue;
741 		}
742 		if (current->object.vm_object == NULL)
743 			continue;
744 
745 		/*
746 		 * limit this scan to the current map entry and the
747 		 * limits for the mincore call
748 		 */
749 		if (addr < current->start)
750 			addr = current->start;
751 		cend = current->end;
752 		if (cend > end)
753 			cend = end;
754 
755 		/*
756 		 * scan this entry one page at a time
757 		 */
758 		while (addr < cend) {
759 			/*
760 			 * Check pmap first, it is likely faster, also
761 			 * it can provide info as to whether we are the
762 			 * one referencing or modifying the page.
763 			 *
764 			 * If we have to check the VM object, only mess
765 			 * around with normal maps.  Do not mess around
766 			 * with virtual page tables (XXX).
767 			 */
768 			mincoreinfo = pmap_mincore(pmap, addr);
769 			if (mincoreinfo == 0 &&
770 			    current->maptype == VM_MAPTYPE_NORMAL) {
771 				vm_pindex_t pindex;
772 				vm_ooffset_t offset;
773 				vm_page_t m;
774 
775 				/*
776 				 * calculate the page index into the object
777 				 */
778 				offset = current->offset + (addr - current->start);
779 				pindex = OFF_TO_IDX(offset);
780 
781 				/*
782 				 * if the page is resident, then gather
783 				 * information about it.  spl protection is
784 				 * required to maintain the object
785 				 * association.  And XXX what if the page is
786 				 * busy?  What's the deal with that?
787 				 */
788 				crit_enter();
789 				m = vm_page_lookup(current->object.vm_object,
790 						    pindex);
791 				if (m && m->valid) {
792 					mincoreinfo = MINCORE_INCORE;
793 					if (m->dirty ||
794 						pmap_is_modified(m))
795 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
796 					if ((m->flags & PG_REFERENCED) ||
797 						pmap_ts_referenced(m)) {
798 						vm_page_flag_set(m, PG_REFERENCED);
799 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
800 					}
801 				}
802 				crit_exit();
803 			}
804 
805 			/*
806 			 * subyte may page fault.  In case it needs to modify
807 			 * the map, we release the lock.
808 			 */
809 			vm_map_unlock_read(map);
810 
811 			/*
812 			 * calculate index into user supplied byte vector
813 			 */
814 			vecindex = OFF_TO_IDX(addr - first_addr);
815 
816 			/*
817 			 * If we have skipped map entries, we need to make sure that
818 			 * the byte vector is zeroed for those skipped entries.
819 			 */
820 			while((lastvecindex + 1) < vecindex) {
821 				error = subyte( vec + lastvecindex, 0);
822 				if (error) {
823 					return (EFAULT);
824 				}
825 				++lastvecindex;
826 			}
827 
828 			/*
829 			 * Pass the page information to the user
830 			 */
831 			error = subyte( vec + vecindex, mincoreinfo);
832 			if (error) {
833 				return (EFAULT);
834 			}
835 
836 			/*
837 			 * If the map has changed, due to the subyte, the previous
838 			 * output may be invalid.
839 			 */
840 			vm_map_lock_read(map);
841 			if (timestamp != map->timestamp)
842 				goto RestartScan;
843 
844 			lastvecindex = vecindex;
845 			addr += PAGE_SIZE;
846 		}
847 	}
848 
849 	/*
850 	 * subyte may page fault.  In case it needs to modify
851 	 * the map, we release the lock.
852 	 */
853 	vm_map_unlock_read(map);
854 
855 	/*
856 	 * Zero the last entries in the byte vector.
857 	 */
858 	vecindex = OFF_TO_IDX(end - first_addr);
859 	while((lastvecindex + 1) < vecindex) {
860 		error = subyte( vec + lastvecindex, 0);
861 		if (error) {
862 			return (EFAULT);
863 		}
864 		++lastvecindex;
865 	}
866 
867 	/*
868 	 * If the map has changed, due to the subyte, the previous
869 	 * output may be invalid.
870 	 */
871 	vm_map_lock_read(map);
872 	if (timestamp != map->timestamp)
873 		goto RestartScan;
874 	vm_map_unlock_read(map);
875 
876 	return (0);
877 }
878 
879 /*
880  * mlock_args(const void *addr, size_t len)
881  */
882 int
883 sys_mlock(struct mlock_args *uap)
884 {
885 	vm_offset_t addr;
886 	vm_size_t size, pageoff;
887 	int error;
888 	struct proc *p = curproc;
889 
890 	addr = (vm_offset_t) uap->addr;
891 	size = uap->len;
892 
893 	pageoff = (addr & PAGE_MASK);
894 	addr -= pageoff;
895 	size += pageoff;
896 	size = (vm_size_t) round_page(size);
897 
898 	/* disable wrap around */
899 	if (addr + size < addr)
900 		return (EINVAL);
901 
902 	if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
903 		return (EAGAIN);
904 
905 #ifdef pmap_wired_count
906 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
907 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
908 		return (ENOMEM);
909 #else
910 	error = priv_check_cred(p->p_ucred, PRIV_ROOT, 0);
911 	if (error)
912 		return (error);
913 #endif
914 
915 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
916 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
917 }
918 
919 /*
920  * mlockall_args(int how)
921  */
922 int
923 sys_mlockall(struct mlockall_args *uap)
924 {
925 	return 0;
926 }
927 
928 /*
929  * munlockall_args(void)
930  */
931 int
932 sys_munlockall(struct munlockall_args *uap)
933 {
934 	return 0;
935 }
936 
937 /*
938  * munlock_args(const void *addr, size_t len)
939  */
940 int
941 sys_munlock(struct munlock_args *uap)
942 {
943 	struct thread *td = curthread;
944 	struct proc *p = td->td_proc;
945 	vm_offset_t addr;
946 	vm_size_t size, pageoff;
947 	int error;
948 
949 	addr = (vm_offset_t) uap->addr;
950 	size = uap->len;
951 
952 	pageoff = (addr & PAGE_MASK);
953 	addr -= pageoff;
954 	size += pageoff;
955 	size = (vm_size_t) round_page(size);
956 
957 	/* disable wrap around */
958 	if (addr + size < addr)
959 		return (EINVAL);
960 
961 #ifndef pmap_wired_count
962 	error = priv_check(td, PRIV_ROOT);
963 	if (error)
964 		return (error);
965 #endif
966 
967 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
968 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
969 }
970 
971 /*
972  * Internal version of mmap.
973  * Currently used by mmap, exec, and sys5 shared memory.
974  * Handle is either a vnode pointer or NULL for MAP_ANON.
975  */
976 int
977 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
978 	vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff)
979 {
980 	boolean_t fitit;
981 	vm_object_t object;
982 	vm_offset_t eaddr;
983 	vm_size_t   esize;
984 	struct vnode *vp;
985 	struct thread *td = curthread;
986 	struct proc *p;
987 	objtype_t type;
988 	int rv = KERN_SUCCESS;
989 	off_t objsize;
990 	int docow;
991 
992 	if (size == 0)
993 		return (0);
994 
995 	objsize = size = round_page(size);
996 
997 	/*
998 	 * XXX messy code, fixme
999 	 *
1000 	 * NOTE: Overflow checks require discrete statements or GCC4
1001 	 * will optimize it out.
1002 	 */
1003 	if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
1004 		esize = map->size + size;
1005 		if (esize < map->size ||
1006 		    esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1007 			return(ENOMEM);
1008 		}
1009 	}
1010 
1011 	/*
1012 	 * We currently can only deal with page aligned file offsets.
1013 	 * The check is here rather than in the syscall because the
1014 	 * kernel calls this function internally for other mmaping
1015 	 * operations (such as in exec) and non-aligned offsets will
1016 	 * cause pmap inconsistencies...so we want to be sure to
1017 	 * disallow this in all cases.
1018 	 *
1019 	 * NOTE: Overflow checks require discrete statements or GCC4
1020 	 * will optimize it out.
1021 	 */
1022 	if (foff & PAGE_MASK)
1023 		return (EINVAL);
1024 
1025 	if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
1026 		fitit = TRUE;
1027 		*addr = round_page(*addr);
1028 	} else {
1029 		if (*addr != trunc_page(*addr))
1030 			return (EINVAL);
1031 		eaddr = *addr + size;
1032 		if (eaddr < *addr)
1033 			return (EINVAL);
1034 		fitit = FALSE;
1035 		if ((flags & MAP_TRYFIXED) == 0)
1036 			vm_map_remove(map, *addr, *addr + size);
1037 	}
1038 
1039 	/*
1040 	 * Lookup/allocate object.
1041 	 */
1042 	if (flags & MAP_ANON) {
1043 		type = OBJT_DEFAULT;
1044 		/*
1045 		 * Unnamed anonymous regions always start at 0.
1046 		 */
1047 		if (handle == NULL)
1048 			foff = 0;
1049 		vp = NULL;
1050 	} else {
1051 		vp = (struct vnode *)handle;
1052 		if (vp->v_type == VCHR) {
1053 			type = OBJT_DEVICE;
1054 			handle = (void *)(intptr_t)vp->v_rdev;
1055 		} else {
1056 			struct vattr vat;
1057 			int error;
1058 
1059 			error = VOP_GETATTR(vp, &vat);
1060 			if (error)
1061 				return (error);
1062 			objsize = vat.va_size;
1063 			type = OBJT_VNODE;
1064 			/*
1065 			 * if it is a regular file without any references
1066 			 * we do not need to sync it.
1067 			 */
1068 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1069 				flags |= MAP_NOSYNC;
1070 			}
1071 		}
1072 	}
1073 
1074 	if (handle == NULL) {
1075 		object = NULL;
1076 		docow = 0;
1077 	} else {
1078 		object = vm_pager_allocate(type, handle, objsize, prot, foff);
1079 		if (object == NULL)
1080 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1081 		docow = MAP_PREFAULT_PARTIAL;
1082 	}
1083 
1084 	/*
1085 	 * Force device mappings to be shared.
1086 	 */
1087 	if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1088 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1089 		flags |= MAP_SHARED;
1090 	}
1091 
1092 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1093 		docow |= MAP_COPY_ON_WRITE;
1094 	if (flags & MAP_NOSYNC)
1095 		docow |= MAP_DISABLE_SYNCER;
1096 	if (flags & MAP_NOCORE)
1097 		docow |= MAP_DISABLE_COREDUMP;
1098 
1099 #if defined(VM_PROT_READ_IS_EXEC)
1100 	if (prot & VM_PROT_READ)
1101 		prot |= VM_PROT_EXECUTE;
1102 
1103 	if (maxprot & VM_PROT_READ)
1104 		maxprot |= VM_PROT_EXECUTE;
1105 #endif
1106 
1107 	/*
1108 	 * This may place the area in its own page directory if (size) is
1109 	 * large enough, otherwise it typically returns its argument.
1110 	 */
1111 	if (fitit) {
1112 		*addr = pmap_addr_hint(object, *addr, size);
1113 	}
1114 
1115 	/*
1116 	 * Stack mappings need special attention.
1117 	 *
1118 	 * Mappings that use virtual page tables will default to storing
1119 	 * the page table at offset 0.
1120 	 */
1121 	if (flags & MAP_STACK) {
1122 		rv = vm_map_stack(map, *addr, size, flags,
1123 				  prot, maxprot, docow);
1124 	} else if (flags & MAP_VPAGETABLE) {
1125 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1126 				 VM_MAPTYPE_VPAGETABLE, prot, maxprot, docow);
1127 	} else {
1128 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1129 				 VM_MAPTYPE_NORMAL, prot, maxprot, docow);
1130 	}
1131 
1132 	if (rv != KERN_SUCCESS) {
1133 		/*
1134 		 * Lose the object reference. Will destroy the
1135 		 * object if it's an unnamed anonymous mapping
1136 		 * or named anonymous without other references.
1137 		 */
1138 		vm_object_deallocate(object);
1139 		goto out;
1140 	}
1141 
1142 	/*
1143 	 * Shared memory is also shared with children.
1144 	 */
1145 	if (flags & (MAP_SHARED|MAP_INHERIT)) {
1146 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1147 		if (rv != KERN_SUCCESS) {
1148 			vm_map_remove(map, *addr, *addr + size);
1149 			goto out;
1150 		}
1151 	}
1152 
1153 	/*
1154 	 * Set the access time on the vnode
1155 	 */
1156 	if (vp != NULL)
1157 		vn_mark_atime(vp, td);
1158 out:
1159 	switch (rv) {
1160 	case KERN_SUCCESS:
1161 		return (0);
1162 	case KERN_INVALID_ADDRESS:
1163 	case KERN_NO_SPACE:
1164 		return (ENOMEM);
1165 	case KERN_PROTECTION_FAILURE:
1166 		return (EACCES);
1167 	default:
1168 		return (EINVAL);
1169 	}
1170 }
1171