xref: /freebsd/sys/vm/vm_mmap.c (revision 4f29da19bd44f0e99f021510460a81bf754c21d2)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35  *
36  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
37  */
38 
39 /*
40  * Mapped file (mmap) interface to VM
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48 #include "opt_mac.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/proc.h>
58 #include <sys/resource.h>
59 #include <sys/resourcevar.h>
60 #include <sys/vnode.h>
61 #include <sys/fcntl.h>
62 #include <sys/file.h>
63 #include <sys/mac.h>
64 #include <sys/mman.h>
65 #include <sys/mount.h>
66 #include <sys/conf.h>
67 #include <sys/stat.h>
68 #include <sys/vmmeter.h>
69 #include <sys/sysctl.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_param.h>
73 #include <vm/pmap.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_pager.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_kern.h>
82 
83 #ifdef HWPMC_HOOKS
84 #include <sys/pmckern.h>
85 #endif
86 
87 #ifndef _SYS_SYSPROTO_H_
88 struct sbrk_args {
89 	int incr;
90 };
91 #endif
92 
93 static int max_proc_mmap;
94 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
95 
96 /*
97  * Set the maximum number of vm_map_entry structures per process.  Roughly
98  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
99  * of our KVM malloc space still results in generous limits.  We want a
100  * default that is good enough to prevent the kernel running out of resources
101  * if attacked from compromised user account but generous enough such that
102  * multi-threaded processes are not unduly inconvenienced.
103  */
104 static void vmmapentry_rsrc_init(void *);
105 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
106 
107 static void
108 vmmapentry_rsrc_init(dummy)
109         void *dummy;
110 {
111     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
112     max_proc_mmap /= 100;
113 }
114 
115 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
116     int *, struct vnode *, vm_ooffset_t, vm_object_t *);
117 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
118     int *, struct cdev *, vm_ooffset_t, vm_object_t *);
119 
120 /*
121  * MPSAFE
122  */
123 /* ARGSUSED */
124 int
125 sbrk(td, uap)
126 	struct thread *td;
127 	struct sbrk_args *uap;
128 {
129 	/* Not yet implemented */
130 	return (EOPNOTSUPP);
131 }
132 
133 #ifndef _SYS_SYSPROTO_H_
134 struct sstk_args {
135 	int incr;
136 };
137 #endif
138 
139 /*
140  * MPSAFE
141  */
142 /* ARGSUSED */
143 int
144 sstk(td, uap)
145 	struct thread *td;
146 	struct sstk_args *uap;
147 {
148 	/* Not yet implemented */
149 	return (EOPNOTSUPP);
150 }
151 
152 #if defined(COMPAT_43)
153 #ifndef _SYS_SYSPROTO_H_
154 struct getpagesize_args {
155 	int dummy;
156 };
157 #endif
158 
159 /* ARGSUSED */
160 int
161 ogetpagesize(td, uap)
162 	struct thread *td;
163 	struct getpagesize_args *uap;
164 {
165 	/* MP SAFE */
166 	td->td_retval[0] = PAGE_SIZE;
167 	return (0);
168 }
169 #endif				/* COMPAT_43 */
170 
171 
172 /*
173  * Memory Map (mmap) system call.  Note that the file offset
174  * and address are allowed to be NOT page aligned, though if
175  * the MAP_FIXED flag it set, both must have the same remainder
176  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
177  * page-aligned, the actual mapping starts at trunc_page(addr)
178  * and the return value is adjusted up by the page offset.
179  *
180  * Generally speaking, only character devices which are themselves
181  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
182  * there would be no cache coherency between a descriptor and a VM mapping
183  * both to the same character device.
184  *
185  * Block devices can be mmap'd no matter what they represent.  Cache coherency
186  * is maintained as long as you do not write directly to the underlying
187  * character device.
188  */
189 #ifndef _SYS_SYSPROTO_H_
190 struct mmap_args {
191 	void *addr;
192 	size_t len;
193 	int prot;
194 	int flags;
195 	int fd;
196 	long pad;
197 	off_t pos;
198 };
199 #endif
200 
201 /*
202  * MPSAFE
203  */
204 int
205 mmap(td, uap)
206 	struct thread *td;
207 	struct mmap_args *uap;
208 {
209 #ifdef HWPMC_HOOKS
210 	struct pmckern_map_in pkm;
211 #endif
212 	struct file *fp;
213 	struct vnode *vp;
214 	vm_offset_t addr;
215 	vm_size_t size, pageoff;
216 	vm_prot_t prot, maxprot;
217 	void *handle;
218 	objtype_t handle_type;
219 	int flags, error;
220 	off_t pos;
221 	struct vmspace *vms = td->td_proc->p_vmspace;
222 
223 	addr = (vm_offset_t) uap->addr;
224 	size = uap->len;
225 	prot = uap->prot & VM_PROT_ALL;
226 	flags = uap->flags;
227 	pos = uap->pos;
228 
229 	fp = NULL;
230 	/* make sure mapping fits into numeric range etc */
231 	if ((ssize_t) uap->len < 0 ||
232 	    ((flags & MAP_ANON) && uap->fd != -1))
233 		return (EINVAL);
234 
235 	if (flags & MAP_STACK) {
236 		if ((uap->fd != -1) ||
237 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
238 			return (EINVAL);
239 		flags |= MAP_ANON;
240 		pos = 0;
241 	}
242 
243 	/*
244 	 * Align the file position to a page boundary,
245 	 * and save its page offset component.
246 	 */
247 	pageoff = (pos & PAGE_MASK);
248 	pos -= pageoff;
249 
250 	/* Adjust size for rounding (on both ends). */
251 	size += pageoff;			/* low end... */
252 	size = (vm_size_t) round_page(size);	/* hi end */
253 
254 	/*
255 	 * Check for illegal addresses.  Watch out for address wrap... Note
256 	 * that VM_*_ADDRESS are not constants due to casts (argh).
257 	 */
258 	if (flags & MAP_FIXED) {
259 		/*
260 		 * The specified address must have the same remainder
261 		 * as the file offset taken modulo PAGE_SIZE, so it
262 		 * should be aligned after adjustment by pageoff.
263 		 */
264 		addr -= pageoff;
265 		if (addr & PAGE_MASK)
266 			return (EINVAL);
267 		/* Address range must be all in user VM space. */
268 		if (addr < vm_map_min(&vms->vm_map) ||
269 		    addr + size > vm_map_max(&vms->vm_map))
270 			return (EINVAL);
271 		if (addr + size < addr)
272 			return (EINVAL);
273 	} else {
274 	/*
275 	 * XXX for non-fixed mappings where no hint is provided or
276 	 * the hint would fall in the potential heap space,
277 	 * place it after the end of the largest possible heap.
278 	 *
279 	 * There should really be a pmap call to determine a reasonable
280 	 * location.
281 	 */
282 		PROC_LOCK(td->td_proc);
283 		if (addr == 0 ||
284 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
285 		    addr < round_page((vm_offset_t)vms->vm_daddr +
286 		    lim_max(td->td_proc, RLIMIT_DATA))))
287 			addr = round_page((vm_offset_t)vms->vm_daddr +
288 			    lim_max(td->td_proc, RLIMIT_DATA));
289 		PROC_UNLOCK(td->td_proc);
290 	}
291 	if (flags & MAP_ANON) {
292 		/*
293 		 * Mapping blank space is trivial.
294 		 */
295 		handle = NULL;
296 		handle_type = OBJT_DEFAULT;
297 		maxprot = VM_PROT_ALL;
298 		pos = 0;
299 	} else {
300 		/*
301 		 * Mapping file, get fp for validation. Obtain vnode and make
302 		 * sure it is of appropriate type.
303 		 * don't let the descriptor disappear on us if we block
304 		 */
305 		if ((error = fget(td, uap->fd, &fp)) != 0)
306 			goto done;
307 		if (fp->f_type != DTYPE_VNODE) {
308 			error = EINVAL;
309 			goto done;
310 		}
311 		/*
312 		 * POSIX shared-memory objects are defined to have
313 		 * kernel persistence, and are not defined to support
314 		 * read(2)/write(2) -- or even open(2).  Thus, we can
315 		 * use MAP_ASYNC to trade on-disk coherence for speed.
316 		 * The shm_open(3) library routine turns on the FPOSIXSHM
317 		 * flag to request this behavior.
318 		 */
319 		if (fp->f_flag & FPOSIXSHM)
320 			flags |= MAP_NOSYNC;
321 		vp = fp->f_vnode;
322 		/*
323 		 * Ensure that file and memory protections are
324 		 * compatible.  Note that we only worry about
325 		 * writability if mapping is shared; in this case,
326 		 * current and max prot are dictated by the open file.
327 		 * XXX use the vnode instead?  Problem is: what
328 		 * credentials do we use for determination? What if
329 		 * proc does a setuid?
330 		 */
331 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
332 			maxprot = VM_PROT_NONE;
333 		else
334 			maxprot = VM_PROT_EXECUTE;
335 		if (fp->f_flag & FREAD) {
336 			maxprot |= VM_PROT_READ;
337 		} else if (prot & PROT_READ) {
338 			error = EACCES;
339 			goto done;
340 		}
341 		/*
342 		 * If we are sharing potential changes (either via
343 		 * MAP_SHARED or via the implicit sharing of character
344 		 * device mappings), and we are trying to get write
345 		 * permission although we opened it without asking
346 		 * for it, bail out.
347 		 */
348 		if ((flags & MAP_SHARED) != 0) {
349 			if ((fp->f_flag & FWRITE) != 0) {
350 				maxprot |= VM_PROT_WRITE;
351 			} else if ((prot & PROT_WRITE) != 0) {
352 				error = EACCES;
353 				goto done;
354 			}
355 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
356 			maxprot |= VM_PROT_WRITE;
357 		}
358 		handle = (void *)vp;
359 		handle_type = OBJT_VNODE;
360 	}
361 
362 	/*
363 	 * Do not allow more then a certain number of vm_map_entry structures
364 	 * per process.  Scale with the number of rforks sharing the map
365 	 * to make the limit reasonable for threads.
366 	 */
367 	if (max_proc_mmap &&
368 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
369 		error = ENOMEM;
370 		goto done;
371 	}
372 
373 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
374 	    flags, handle_type, handle, pos);
375 #ifdef HWPMC_HOOKS
376 	/* inform hwpmc(4) if an executable is being mapped */
377 	if (error == 0 && handle_type == OBJT_VNODE &&
378 	    (prot & PROT_EXEC)) {
379 		pkm.pm_file = handle;
380 		pkm.pm_address = (uintptr_t) addr;
381 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
382 	}
383 #endif
384 	if (error == 0)
385 		td->td_retval[0] = (register_t) (addr + pageoff);
386 done:
387 	if (fp)
388 		fdrop(fp, td);
389 
390 	return (error);
391 }
392 
393 #ifdef COMPAT_43
394 #ifndef _SYS_SYSPROTO_H_
395 struct ommap_args {
396 	caddr_t addr;
397 	int len;
398 	int prot;
399 	int flags;
400 	int fd;
401 	long pos;
402 };
403 #endif
404 int
405 ommap(td, uap)
406 	struct thread *td;
407 	struct ommap_args *uap;
408 {
409 	struct mmap_args nargs;
410 	static const char cvtbsdprot[8] = {
411 		0,
412 		PROT_EXEC,
413 		PROT_WRITE,
414 		PROT_EXEC | PROT_WRITE,
415 		PROT_READ,
416 		PROT_EXEC | PROT_READ,
417 		PROT_WRITE | PROT_READ,
418 		PROT_EXEC | PROT_WRITE | PROT_READ,
419 	};
420 
421 #define	OMAP_ANON	0x0002
422 #define	OMAP_COPY	0x0020
423 #define	OMAP_SHARED	0x0010
424 #define	OMAP_FIXED	0x0100
425 
426 	nargs.addr = uap->addr;
427 	nargs.len = uap->len;
428 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
429 	nargs.flags = 0;
430 	if (uap->flags & OMAP_ANON)
431 		nargs.flags |= MAP_ANON;
432 	if (uap->flags & OMAP_COPY)
433 		nargs.flags |= MAP_COPY;
434 	if (uap->flags & OMAP_SHARED)
435 		nargs.flags |= MAP_SHARED;
436 	else
437 		nargs.flags |= MAP_PRIVATE;
438 	if (uap->flags & OMAP_FIXED)
439 		nargs.flags |= MAP_FIXED;
440 	nargs.fd = uap->fd;
441 	nargs.pos = uap->pos;
442 	return (mmap(td, &nargs));
443 }
444 #endif				/* COMPAT_43 */
445 
446 
447 #ifndef _SYS_SYSPROTO_H_
448 struct msync_args {
449 	void *addr;
450 	int len;
451 	int flags;
452 };
453 #endif
454 /*
455  * MPSAFE
456  */
457 int
458 msync(td, uap)
459 	struct thread *td;
460 	struct msync_args *uap;
461 {
462 	vm_offset_t addr;
463 	vm_size_t size, pageoff;
464 	int flags;
465 	vm_map_t map;
466 	int rv;
467 
468 	addr = (vm_offset_t) uap->addr;
469 	size = uap->len;
470 	flags = uap->flags;
471 
472 	pageoff = (addr & PAGE_MASK);
473 	addr -= pageoff;
474 	size += pageoff;
475 	size = (vm_size_t) round_page(size);
476 	if (addr + size < addr)
477 		return (EINVAL);
478 
479 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
480 		return (EINVAL);
481 
482 	map = &td->td_proc->p_vmspace->vm_map;
483 
484 	/*
485 	 * Clean the pages and interpret the return value.
486 	 */
487 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
488 	    (flags & MS_INVALIDATE) != 0);
489 	switch (rv) {
490 	case KERN_SUCCESS:
491 		return (0);
492 	case KERN_INVALID_ADDRESS:
493 		return (EINVAL);	/* Sun returns ENOMEM? */
494 	case KERN_INVALID_ARGUMENT:
495 		return (EBUSY);
496 	default:
497 		return (EINVAL);
498 	}
499 }
500 
501 #ifndef _SYS_SYSPROTO_H_
502 struct munmap_args {
503 	void *addr;
504 	size_t len;
505 };
506 #endif
507 /*
508  * MPSAFE
509  */
510 int
511 munmap(td, uap)
512 	struct thread *td;
513 	struct munmap_args *uap;
514 {
515 #ifdef HWPMC_HOOKS
516 	struct pmckern_map_out pkm;
517 	vm_map_entry_t entry;
518 #endif
519 	vm_offset_t addr;
520 	vm_size_t size, pageoff;
521 	vm_map_t map;
522 
523 	addr = (vm_offset_t) uap->addr;
524 	size = uap->len;
525 	if (size == 0)
526 		return (EINVAL);
527 
528 	pageoff = (addr & PAGE_MASK);
529 	addr -= pageoff;
530 	size += pageoff;
531 	size = (vm_size_t) round_page(size);
532 	if (addr + size < addr)
533 		return (EINVAL);
534 
535 	/*
536 	 * Check for illegal addresses.  Watch out for address wrap...
537 	 */
538 	map = &td->td_proc->p_vmspace->vm_map;
539 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
540 		return (EINVAL);
541 	vm_map_lock(map);
542 	/*
543 	 * Make sure entire range is allocated.
544 	 */
545 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) {
546 		vm_map_unlock(map);
547 		return (EINVAL);
548 	}
549 #ifdef HWPMC_HOOKS
550 	/*
551 	 * Inform hwpmc if the address range being unmapped contains
552 	 * an executable region.
553 	 */
554 	if (vm_map_lookup_entry(map, addr, &entry)) {
555 		for (;
556 		     entry != &map->header && entry->start < addr + size;
557 		     entry = entry->next) {
558 			if (vm_map_check_protection(map, entry->start,
559 				entry->end, VM_PROT_EXECUTE) == TRUE) {
560 				pkm.pm_address = (uintptr_t) addr;
561 				pkm.pm_size = (size_t) size;
562 				PMC_CALL_HOOK(td, PMC_FN_MUNMAP,
563 				    (void *) &pkm);
564 				break;
565 			}
566 		}
567 	}
568 #endif
569 	/* returns nothing but KERN_SUCCESS anyway */
570 	vm_map_delete(map, addr, addr + size);
571 	vm_map_unlock(map);
572 	return (0);
573 }
574 
575 #ifndef _SYS_SYSPROTO_H_
576 struct mprotect_args {
577 	const void *addr;
578 	size_t len;
579 	int prot;
580 };
581 #endif
582 /*
583  * MPSAFE
584  */
585 int
586 mprotect(td, uap)
587 	struct thread *td;
588 	struct mprotect_args *uap;
589 {
590 	vm_offset_t addr;
591 	vm_size_t size, pageoff;
592 	vm_prot_t prot;
593 
594 	addr = (vm_offset_t) uap->addr;
595 	size = uap->len;
596 	prot = uap->prot & VM_PROT_ALL;
597 #if defined(VM_PROT_READ_IS_EXEC)
598 	if (prot & VM_PROT_READ)
599 		prot |= VM_PROT_EXECUTE;
600 #endif
601 
602 	pageoff = (addr & PAGE_MASK);
603 	addr -= pageoff;
604 	size += pageoff;
605 	size = (vm_size_t) round_page(size);
606 	if (addr + size < addr)
607 		return (EINVAL);
608 
609 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
610 	    addr + size, prot, FALSE)) {
611 	case KERN_SUCCESS:
612 		return (0);
613 	case KERN_PROTECTION_FAILURE:
614 		return (EACCES);
615 	}
616 	return (EINVAL);
617 }
618 
619 #ifndef _SYS_SYSPROTO_H_
620 struct minherit_args {
621 	void *addr;
622 	size_t len;
623 	int inherit;
624 };
625 #endif
626 /*
627  * MPSAFE
628  */
629 int
630 minherit(td, uap)
631 	struct thread *td;
632 	struct minherit_args *uap;
633 {
634 	vm_offset_t addr;
635 	vm_size_t size, pageoff;
636 	vm_inherit_t inherit;
637 
638 	addr = (vm_offset_t)uap->addr;
639 	size = uap->len;
640 	inherit = uap->inherit;
641 
642 	pageoff = (addr & PAGE_MASK);
643 	addr -= pageoff;
644 	size += pageoff;
645 	size = (vm_size_t) round_page(size);
646 	if (addr + size < addr)
647 		return (EINVAL);
648 
649 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
650 	    addr + size, inherit)) {
651 	case KERN_SUCCESS:
652 		return (0);
653 	case KERN_PROTECTION_FAILURE:
654 		return (EACCES);
655 	}
656 	return (EINVAL);
657 }
658 
659 #ifndef _SYS_SYSPROTO_H_
660 struct madvise_args {
661 	void *addr;
662 	size_t len;
663 	int behav;
664 };
665 #endif
666 
667 /*
668  * MPSAFE
669  */
670 /* ARGSUSED */
671 int
672 madvise(td, uap)
673 	struct thread *td;
674 	struct madvise_args *uap;
675 {
676 	vm_offset_t start, end;
677 	vm_map_t map;
678 	struct proc *p;
679 	int error;
680 
681 	/*
682 	 * Check for our special case, advising the swap pager we are
683 	 * "immortal."
684 	 */
685 	if (uap->behav == MADV_PROTECT) {
686 		error = suser(td);
687 		if (error == 0) {
688 			p = td->td_proc;
689 			PROC_LOCK(p);
690 			p->p_flag |= P_PROTECTED;
691 			PROC_UNLOCK(p);
692 		}
693 		return (error);
694 	}
695 	/*
696 	 * Check for illegal behavior
697 	 */
698 	if (uap->behav < 0 || uap->behav > MADV_CORE)
699 		return (EINVAL);
700 	/*
701 	 * Check for illegal addresses.  Watch out for address wrap... Note
702 	 * that VM_*_ADDRESS are not constants due to casts (argh).
703 	 */
704 	map = &td->td_proc->p_vmspace->vm_map;
705 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
706 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
707 		return (EINVAL);
708 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
709 		return (EINVAL);
710 
711 	/*
712 	 * Since this routine is only advisory, we default to conservative
713 	 * behavior.
714 	 */
715 	start = trunc_page((vm_offset_t) uap->addr);
716 	end = round_page((vm_offset_t) uap->addr + uap->len);
717 
718 	if (vm_map_madvise(map, start, end, uap->behav))
719 		return (EINVAL);
720 	return (0);
721 }
722 
723 #ifndef _SYS_SYSPROTO_H_
724 struct mincore_args {
725 	const void *addr;
726 	size_t len;
727 	char *vec;
728 };
729 #endif
730 
731 /*
732  * MPSAFE
733  */
734 /* ARGSUSED */
735 int
736 mincore(td, uap)
737 	struct thread *td;
738 	struct mincore_args *uap;
739 {
740 	vm_offset_t addr, first_addr;
741 	vm_offset_t end, cend;
742 	pmap_t pmap;
743 	vm_map_t map;
744 	char *vec;
745 	int error = 0;
746 	int vecindex, lastvecindex;
747 	vm_map_entry_t current;
748 	vm_map_entry_t entry;
749 	int mincoreinfo;
750 	unsigned int timestamp;
751 
752 	/*
753 	 * Make sure that the addresses presented are valid for user
754 	 * mode.
755 	 */
756 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
757 	end = addr + (vm_size_t)round_page(uap->len);
758 	map = &td->td_proc->p_vmspace->vm_map;
759 	if (end > vm_map_max(map) || end < addr)
760 		return (EINVAL);
761 
762 	/*
763 	 * Address of byte vector
764 	 */
765 	vec = uap->vec;
766 
767 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
768 
769 	vm_map_lock_read(map);
770 RestartScan:
771 	timestamp = map->timestamp;
772 
773 	if (!vm_map_lookup_entry(map, addr, &entry))
774 		entry = entry->next;
775 
776 	/*
777 	 * Do this on a map entry basis so that if the pages are not
778 	 * in the current processes address space, we can easily look
779 	 * up the pages elsewhere.
780 	 */
781 	lastvecindex = -1;
782 	for (current = entry;
783 	    (current != &map->header) && (current->start < end);
784 	    current = current->next) {
785 
786 		/*
787 		 * ignore submaps (for now) or null objects
788 		 */
789 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
790 			current->object.vm_object == NULL)
791 			continue;
792 
793 		/*
794 		 * limit this scan to the current map entry and the
795 		 * limits for the mincore call
796 		 */
797 		if (addr < current->start)
798 			addr = current->start;
799 		cend = current->end;
800 		if (cend > end)
801 			cend = end;
802 
803 		/*
804 		 * scan this entry one page at a time
805 		 */
806 		while (addr < cend) {
807 			/*
808 			 * Check pmap first, it is likely faster, also
809 			 * it can provide info as to whether we are the
810 			 * one referencing or modifying the page.
811 			 */
812 			mincoreinfo = pmap_mincore(pmap, addr);
813 			if (!mincoreinfo) {
814 				vm_pindex_t pindex;
815 				vm_ooffset_t offset;
816 				vm_page_t m;
817 				/*
818 				 * calculate the page index into the object
819 				 */
820 				offset = current->offset + (addr - current->start);
821 				pindex = OFF_TO_IDX(offset);
822 				VM_OBJECT_LOCK(current->object.vm_object);
823 				m = vm_page_lookup(current->object.vm_object,
824 					pindex);
825 				/*
826 				 * if the page is resident, then gather information about
827 				 * it.
828 				 */
829 				if (m != NULL && m->valid != 0) {
830 					mincoreinfo = MINCORE_INCORE;
831 					vm_page_lock_queues();
832 					if (m->dirty ||
833 						pmap_is_modified(m))
834 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
835 					if ((m->flags & PG_REFERENCED) ||
836 						pmap_ts_referenced(m)) {
837 						vm_page_flag_set(m, PG_REFERENCED);
838 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
839 					}
840 					vm_page_unlock_queues();
841 				}
842 				VM_OBJECT_UNLOCK(current->object.vm_object);
843 			}
844 
845 			/*
846 			 * subyte may page fault.  In case it needs to modify
847 			 * the map, we release the lock.
848 			 */
849 			vm_map_unlock_read(map);
850 
851 			/*
852 			 * calculate index into user supplied byte vector
853 			 */
854 			vecindex = OFF_TO_IDX(addr - first_addr);
855 
856 			/*
857 			 * If we have skipped map entries, we need to make sure that
858 			 * the byte vector is zeroed for those skipped entries.
859 			 */
860 			while ((lastvecindex + 1) < vecindex) {
861 				error = subyte(vec + lastvecindex, 0);
862 				if (error) {
863 					error = EFAULT;
864 					goto done2;
865 				}
866 				++lastvecindex;
867 			}
868 
869 			/*
870 			 * Pass the page information to the user
871 			 */
872 			error = subyte(vec + vecindex, mincoreinfo);
873 			if (error) {
874 				error = EFAULT;
875 				goto done2;
876 			}
877 
878 			/*
879 			 * If the map has changed, due to the subyte, the previous
880 			 * output may be invalid.
881 			 */
882 			vm_map_lock_read(map);
883 			if (timestamp != map->timestamp)
884 				goto RestartScan;
885 
886 			lastvecindex = vecindex;
887 			addr += PAGE_SIZE;
888 		}
889 	}
890 
891 	/*
892 	 * subyte may page fault.  In case it needs to modify
893 	 * the map, we release the lock.
894 	 */
895 	vm_map_unlock_read(map);
896 
897 	/*
898 	 * Zero the last entries in the byte vector.
899 	 */
900 	vecindex = OFF_TO_IDX(end - first_addr);
901 	while ((lastvecindex + 1) < vecindex) {
902 		error = subyte(vec + lastvecindex, 0);
903 		if (error) {
904 			error = EFAULT;
905 			goto done2;
906 		}
907 		++lastvecindex;
908 	}
909 
910 	/*
911 	 * If the map has changed, due to the subyte, the previous
912 	 * output may be invalid.
913 	 */
914 	vm_map_lock_read(map);
915 	if (timestamp != map->timestamp)
916 		goto RestartScan;
917 	vm_map_unlock_read(map);
918 done2:
919 	return (error);
920 }
921 
922 #ifndef _SYS_SYSPROTO_H_
923 struct mlock_args {
924 	const void *addr;
925 	size_t len;
926 };
927 #endif
928 /*
929  * MPSAFE
930  */
931 int
932 mlock(td, uap)
933 	struct thread *td;
934 	struct mlock_args *uap;
935 {
936 	struct proc *proc;
937 	vm_offset_t addr, end, last, start;
938 	vm_size_t npages, size;
939 	int error;
940 
941 	error = suser(td);
942 	if (error)
943 		return (error);
944 	addr = (vm_offset_t)uap->addr;
945 	size = uap->len;
946 	last = addr + size;
947 	start = trunc_page(addr);
948 	end = round_page(last);
949 	if (last < addr || end < addr)
950 		return (EINVAL);
951 	npages = atop(end - start);
952 	if (npages > vm_page_max_wired)
953 		return (ENOMEM);
954 	proc = td->td_proc;
955 	PROC_LOCK(proc);
956 	if (ptoa(npages +
957 	    pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
958 	    lim_cur(proc, RLIMIT_MEMLOCK)) {
959 		PROC_UNLOCK(proc);
960 		return (ENOMEM);
961 	}
962 	PROC_UNLOCK(proc);
963 	if (npages + cnt.v_wire_count > vm_page_max_wired)
964 		return (EAGAIN);
965 	error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
966 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
967 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
968 }
969 
970 #ifndef _SYS_SYSPROTO_H_
971 struct mlockall_args {
972 	int	how;
973 };
974 #endif
975 
976 /*
977  * MPSAFE
978  */
979 int
980 mlockall(td, uap)
981 	struct thread *td;
982 	struct mlockall_args *uap;
983 {
984 	vm_map_t map;
985 	int error;
986 
987 	map = &td->td_proc->p_vmspace->vm_map;
988 	error = 0;
989 
990 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
991 		return (EINVAL);
992 
993 #if 0
994 	/*
995 	 * If wiring all pages in the process would cause it to exceed
996 	 * a hard resource limit, return ENOMEM.
997 	 */
998 	PROC_LOCK(td->td_proc);
999 	if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1000 		lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1001 		PROC_UNLOCK(td->td_proc);
1002 		return (ENOMEM);
1003 	}
1004 	PROC_UNLOCK(td->td_proc);
1005 #else
1006 	error = suser(td);
1007 	if (error)
1008 		return (error);
1009 #endif
1010 
1011 	if (uap->how & MCL_FUTURE) {
1012 		vm_map_lock(map);
1013 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1014 		vm_map_unlock(map);
1015 		error = 0;
1016 	}
1017 
1018 	if (uap->how & MCL_CURRENT) {
1019 		/*
1020 		 * P1003.1-2001 mandates that all currently mapped pages
1021 		 * will be memory resident and locked (wired) upon return
1022 		 * from mlockall(). vm_map_wire() will wire pages, by
1023 		 * calling vm_fault_wire() for each page in the region.
1024 		 */
1025 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1026 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1027 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1028 	}
1029 
1030 	return (error);
1031 }
1032 
1033 #ifndef _SYS_SYSPROTO_H_
1034 struct munlockall_args {
1035 	register_t dummy;
1036 };
1037 #endif
1038 
1039 /*
1040  * MPSAFE
1041  */
1042 int
1043 munlockall(td, uap)
1044 	struct thread *td;
1045 	struct munlockall_args *uap;
1046 {
1047 	vm_map_t map;
1048 	int error;
1049 
1050 	map = &td->td_proc->p_vmspace->vm_map;
1051 	error = suser(td);
1052 	if (error)
1053 		return (error);
1054 
1055 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1056 	vm_map_lock(map);
1057 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1058 	vm_map_unlock(map);
1059 
1060 	/* Forcibly unwire all pages. */
1061 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1062 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1063 
1064 	return (error);
1065 }
1066 
1067 #ifndef _SYS_SYSPROTO_H_
1068 struct munlock_args {
1069 	const void *addr;
1070 	size_t len;
1071 };
1072 #endif
1073 /*
1074  * MPSAFE
1075  */
1076 int
1077 munlock(td, uap)
1078 	struct thread *td;
1079 	struct munlock_args *uap;
1080 {
1081 	vm_offset_t addr, end, last, start;
1082 	vm_size_t size;
1083 	int error;
1084 
1085 	error = suser(td);
1086 	if (error)
1087 		return (error);
1088 	addr = (vm_offset_t)uap->addr;
1089 	size = uap->len;
1090 	last = addr + size;
1091 	start = trunc_page(addr);
1092 	end = round_page(last);
1093 	if (last < addr || end < addr)
1094 		return (EINVAL);
1095 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1096 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1097 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1098 }
1099 
1100 /*
1101  * vm_mmap_vnode()
1102  *
1103  * MPSAFE
1104  *
1105  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1106  * operations on vnodes.
1107  */
1108 int
1109 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1110     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1111     struct vnode *vp, vm_ooffset_t foff, vm_object_t *objp)
1112 {
1113 	struct vattr va;
1114 	void *handle;
1115 	vm_object_t obj;
1116 	struct mount *mp;
1117 	int error, flags, type;
1118 	int vfslocked;
1119 
1120 	mp = vp->v_mount;
1121 	vfslocked = VFS_LOCK_GIANT(mp);
1122 	if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0) {
1123 		VFS_UNLOCK_GIANT(vfslocked);
1124 		return (error);
1125 	}
1126 	flags = *flagsp;
1127 	obj = vp->v_object;
1128 	if (vp->v_type == VREG) {
1129 		/*
1130 		 * Get the proper underlying object
1131 		 */
1132 		if (obj == NULL) {
1133 			error = EINVAL;
1134 			goto done;
1135 		}
1136 		if (obj->handle != vp) {
1137 			vput(vp);
1138 			vp = (struct vnode*)obj->handle;
1139 			vget(vp, LK_EXCLUSIVE, td);
1140 		}
1141 		type = OBJT_VNODE;
1142 		handle = vp;
1143 	} else if (vp->v_type == VCHR) {
1144 		type = OBJT_DEVICE;
1145 		handle = vp->v_rdev;
1146 
1147 		/* XXX: lack thredref on device */
1148 		if(vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON) {
1149 			*maxprotp = VM_PROT_ALL;
1150 			*flagsp |= MAP_ANON;
1151 			error = 0;
1152 			goto done;
1153 		}
1154 		/*
1155 		 * cdevs does not provide private mappings of any kind.
1156 		 */
1157 		if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1158 		    (prot & PROT_WRITE) != 0) {
1159 			error = EACCES;
1160 			goto done;
1161 		}
1162 		if (flags & (MAP_PRIVATE|MAP_COPY)) {
1163 			error = EINVAL;
1164 			goto done;
1165 		}
1166 		/*
1167 		 * Force device mappings to be shared.
1168 		 */
1169 		flags |= MAP_SHARED;
1170 	} else {
1171 		error = EINVAL;
1172 		goto done;
1173 	}
1174 	if ((error = VOP_GETATTR(vp, &va, td->td_ucred, td))) {
1175 		goto done;
1176 	}
1177 #ifdef MAC
1178 	error = mac_check_vnode_mmap(td->td_ucred, vp, prot, flags);
1179 	if (error != 0)
1180 		goto done;
1181 #endif
1182 	if ((flags & MAP_SHARED) != 0) {
1183 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1184 			if (prot & PROT_WRITE) {
1185 				error = EPERM;
1186 				goto done;
1187 			}
1188 			*maxprotp &= ~VM_PROT_WRITE;
1189 		}
1190 	}
1191 	/*
1192 	 * If it is a regular file without any references
1193 	 * we do not need to sync it.
1194 	 * Adjust object size to be the size of actual file.
1195 	 */
1196 	if (vp->v_type == VREG) {
1197 		objsize = round_page(va.va_size);
1198 		if (va.va_nlink == 0)
1199 			flags |= MAP_NOSYNC;
1200 	}
1201 	obj = vm_pager_allocate(type, handle, objsize, prot, foff);
1202 	if (obj == NULL) {
1203 		error = (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1204 		goto done;
1205 	}
1206 	*objp = obj;
1207 	*flagsp = flags;
1208 	vfs_mark_atime(vp, td);
1209 
1210 done:
1211 	vput(vp);
1212 	VFS_UNLOCK_GIANT(vfslocked);
1213 	return (error);
1214 }
1215 
1216 /*
1217  * vm_mmap_cdev()
1218  *
1219  * MPSAFE
1220  *
1221  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1222  * operations on cdevs.
1223  */
1224 int
1225 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1226     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1227     struct cdev *cdev, vm_ooffset_t foff, vm_object_t *objp)
1228 {
1229 	vm_object_t obj;
1230 	int flags;
1231 
1232 	flags = *flagsp;
1233 
1234 	/* XXX: lack thredref on device */
1235 	if (cdev->si_devsw->d_flags & D_MMAP_ANON) {
1236 		*maxprotp = VM_PROT_ALL;
1237 		*flagsp |= MAP_ANON;
1238 		return (0);
1239 	}
1240 	/*
1241 	 * cdevs does not provide private mappings of any kind.
1242 	 */
1243 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1244 	    (prot & PROT_WRITE) != 0)
1245 		return (EACCES);
1246 	if (flags & (MAP_PRIVATE|MAP_COPY))
1247 		return (EINVAL);
1248 	/*
1249 	 * Force device mappings to be shared.
1250 	 */
1251 	flags |= MAP_SHARED;
1252 #ifdef MAC_XXX
1253 	error = mac_check_cdev_mmap(td->td_ucred, cdev, prot);
1254 	if (error != 0)
1255 		return (error);
1256 #endif
1257 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, foff);
1258 	if (obj == NULL)
1259 		return (EINVAL);
1260 	*objp = obj;
1261 	*flagsp = flags;
1262 	return (0);
1263 }
1264 
1265 /*
1266  * vm_mmap()
1267  *
1268  * MPSAFE
1269  *
1270  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1271  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1272  */
1273 int
1274 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1275 	vm_prot_t maxprot, int flags,
1276 	objtype_t handle_type, void *handle,
1277 	vm_ooffset_t foff)
1278 {
1279 	boolean_t fitit;
1280 	vm_object_t object;
1281 	int rv = KERN_SUCCESS;
1282 	int docow, error;
1283 	struct thread *td = curthread;
1284 
1285 	if (size == 0)
1286 		return (0);
1287 
1288 	size = round_page(size);
1289 
1290 	PROC_LOCK(td->td_proc);
1291 	if (td->td_proc->p_vmspace->vm_map.size + size >
1292 	    lim_cur(td->td_proc, RLIMIT_VMEM)) {
1293 		PROC_UNLOCK(td->td_proc);
1294 		return(ENOMEM);
1295 	}
1296 	PROC_UNLOCK(td->td_proc);
1297 
1298 	/*
1299 	 * We currently can only deal with page aligned file offsets.
1300 	 * The check is here rather than in the syscall because the
1301 	 * kernel calls this function internally for other mmaping
1302 	 * operations (such as in exec) and non-aligned offsets will
1303 	 * cause pmap inconsistencies...so we want to be sure to
1304 	 * disallow this in all cases.
1305 	 */
1306 	if (foff & PAGE_MASK)
1307 		return (EINVAL);
1308 
1309 	if ((flags & MAP_FIXED) == 0) {
1310 		fitit = TRUE;
1311 		*addr = round_page(*addr);
1312 	} else {
1313 		if (*addr != trunc_page(*addr))
1314 			return (EINVAL);
1315 		fitit = FALSE;
1316 		(void) vm_map_remove(map, *addr, *addr + size);
1317 	}
1318 	/*
1319 	 * Lookup/allocate object.
1320 	 */
1321 	switch (handle_type) {
1322 	case OBJT_DEVICE:
1323 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1324 		    handle, foff, &object);
1325 		break;
1326 	case OBJT_VNODE:
1327 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1328 		    handle, foff, &object);
1329 		break;
1330 	case OBJT_DEFAULT:
1331 		if (handle == NULL) {
1332 			error = 0;
1333 			break;
1334 		}
1335 		/* FALLTHROUGH */
1336 	default:
1337 		error = EINVAL;
1338 	}
1339 	if (error)
1340 		return (error);
1341 	if (flags & MAP_ANON) {
1342 		object = NULL;
1343 		docow = 0;
1344 		/*
1345 		 * Unnamed anonymous regions always start at 0.
1346 		 */
1347 		if (handle == 0)
1348 			foff = 0;
1349 	} else {
1350 		docow = MAP_PREFAULT_PARTIAL;
1351 	}
1352 
1353 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1354 		docow |= MAP_COPY_ON_WRITE;
1355 	if (flags & MAP_NOSYNC)
1356 		docow |= MAP_DISABLE_SYNCER;
1357 	if (flags & MAP_NOCORE)
1358 		docow |= MAP_DISABLE_COREDUMP;
1359 
1360 #if defined(VM_PROT_READ_IS_EXEC)
1361 	if (prot & VM_PROT_READ)
1362 		prot |= VM_PROT_EXECUTE;
1363 
1364 	if (maxprot & VM_PROT_READ)
1365 		maxprot |= VM_PROT_EXECUTE;
1366 #endif
1367 
1368 	if (fitit)
1369 		*addr = pmap_addr_hint(object, *addr, size);
1370 
1371 	if (flags & MAP_STACK)
1372 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1373 		    docow | MAP_STACK_GROWS_DOWN);
1374 	else
1375 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1376 				 prot, maxprot, docow);
1377 
1378 	if (rv != KERN_SUCCESS) {
1379 		/*
1380 		 * Lose the object reference. Will destroy the
1381 		 * object if it's an unnamed anonymous mapping
1382 		 * or named anonymous without other references.
1383 		 */
1384 		vm_object_deallocate(object);
1385 	} else if (flags & MAP_SHARED) {
1386 		/*
1387 		 * Shared memory is also shared with children.
1388 		 */
1389 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1390 		if (rv != KERN_SUCCESS)
1391 			(void) vm_map_remove(map, *addr, *addr + size);
1392 	}
1393 
1394 	/*
1395 	 * If the process has requested that all future mappings
1396 	 * be wired, then heed this.
1397 	 */
1398 	if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1399 		vm_map_wire(map, *addr, *addr + size,
1400 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1401 
1402 	switch (rv) {
1403 	case KERN_SUCCESS:
1404 		return (0);
1405 	case KERN_INVALID_ADDRESS:
1406 	case KERN_NO_SPACE:
1407 		return (ENOMEM);
1408 	case KERN_PROTECTION_FAILURE:
1409 		return (EACCES);
1410 	default:
1411 		return (EINVAL);
1412 	}
1413 }
1414