xref: /freebsd/sys/vm/vm_mmap.c (revision 77a0943ded95b9e6438f7db70c4a28e4d93946d4)
1 /*
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
39  *
40  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
41  * $FreeBSD$
42  */
43 
44 /*
45  * Mapped file (mmap) interface to VM
46  */
47 
48 #include "opt_compat.h"
49 #include "opt_rlimit.h"
50 
51 #include <sys/param.h>
52 #include <sys/kernel.h>
53 #include <sys/systm.h>
54 #include <sys/sysproto.h>
55 #include <sys/filedesc.h>
56 #include <sys/proc.h>
57 #include <sys/vnode.h>
58 #include <sys/fcntl.h>
59 #include <sys/file.h>
60 #include <sys/mman.h>
61 #include <sys/conf.h>
62 #include <sys/stat.h>
63 #include <sys/vmmeter.h>
64 #include <sys/sysctl.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <sys/lock.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pager.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 
79 #ifndef _SYS_SYSPROTO_H_
80 struct sbrk_args {
81 	int incr;
82 };
83 #endif
84 
85 static int max_proc_mmap;
86 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 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 __P((void *));
98 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
99 
100 static void
101 vmmapentry_rsrc_init(dummy)
102         void *dummy;
103 {
104     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
105     max_proc_mmap /= 100;
106 }
107 
108 /* ARGSUSED */
109 int
110 sbrk(p, uap)
111 	struct proc *p;
112 	struct sbrk_args *uap;
113 {
114 
115 	/* Not yet implemented */
116 	return (EOPNOTSUPP);
117 }
118 
119 #ifndef _SYS_SYSPROTO_H_
120 struct sstk_args {
121 	int incr;
122 };
123 #endif
124 
125 /* ARGSUSED */
126 int
127 sstk(p, uap)
128 	struct proc *p;
129 	struct sstk_args *uap;
130 {
131 
132 	/* Not yet implemented */
133 	return (EOPNOTSUPP);
134 }
135 
136 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
137 #ifndef _SYS_SYSPROTO_H_
138 struct getpagesize_args {
139 	int dummy;
140 };
141 #endif
142 
143 /* ARGSUSED */
144 int
145 ogetpagesize(p, uap)
146 	struct proc *p;
147 	struct getpagesize_args *uap;
148 {
149 
150 	p->p_retval[0] = PAGE_SIZE;
151 	return (0);
152 }
153 #endif				/* COMPAT_43 || COMPAT_SUNOS */
154 
155 
156 /*
157  * Memory Map (mmap) system call.  Note that the file offset
158  * and address are allowed to be NOT page aligned, though if
159  * the MAP_FIXED flag it set, both must have the same remainder
160  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
161  * page-aligned, the actual mapping starts at trunc_page(addr)
162  * and the return value is adjusted up by the page offset.
163  *
164  * Generally speaking, only character devices which are themselves
165  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
166  * there would be no cache coherency between a descriptor and a VM mapping
167  * both to the same character device.
168  *
169  * Block devices can be mmap'd no matter what they represent.  Cache coherency
170  * is maintained as long as you do not write directly to the underlying
171  * character device.
172  */
173 #ifndef _SYS_SYSPROTO_H_
174 struct mmap_args {
175 	void *addr;
176 	size_t len;
177 	int prot;
178 	int flags;
179 	int fd;
180 	long pad;
181 	off_t pos;
182 };
183 #endif
184 
185 int
186 mmap(p, uap)
187 	struct proc *p;
188 	register struct mmap_args *uap;
189 {
190 	register struct filedesc *fdp = p->p_fd;
191 	register struct file *fp = NULL;
192 	struct vnode *vp;
193 	vm_offset_t addr;
194 	vm_size_t size, pageoff;
195 	vm_prot_t prot, maxprot;
196 	void *handle;
197 	int flags, error;
198 	int disablexworkaround;
199 	off_t pos;
200 	struct vmspace *vms = p->p_vmspace;
201 	vm_object_t obj;
202 
203 	addr = (vm_offset_t) uap->addr;
204 	size = uap->len;
205 	prot = uap->prot & VM_PROT_ALL;
206 	flags = uap->flags;
207 	pos = uap->pos;
208 
209 	/* make sure mapping fits into numeric range etc */
210 	if ((ssize_t) uap->len < 0 ||
211 	    ((flags & MAP_ANON) && uap->fd != -1))
212 		return (EINVAL);
213 
214 	if (flags & MAP_STACK) {
215 		if ((uap->fd != -1) ||
216 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
217 			return (EINVAL);
218 		flags |= MAP_ANON;
219 		pos = 0;
220 	}
221 
222 	/*
223 	 * Align the file position to a page boundary,
224 	 * and save its page offset component.
225 	 */
226 	pageoff = (pos & PAGE_MASK);
227 	pos -= pageoff;
228 
229 	/* Adjust size for rounding (on both ends). */
230 	size += pageoff;			/* low end... */
231 	size = (vm_size_t) round_page(size);	/* hi end */
232 
233 	/*
234 	 * Check for illegal addresses.  Watch out for address wrap... Note
235 	 * that VM_*_ADDRESS are not constants due to casts (argh).
236 	 */
237 	if (flags & MAP_FIXED) {
238 		/*
239 		 * The specified address must have the same remainder
240 		 * as the file offset taken modulo PAGE_SIZE, so it
241 		 * should be aligned after adjustment by pageoff.
242 		 */
243 		addr -= pageoff;
244 		if (addr & PAGE_MASK)
245 			return (EINVAL);
246 		/* Address range must be all in user VM space. */
247 		if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
248 			return (EINVAL);
249 #ifndef i386
250 		if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
251 			return (EINVAL);
252 #endif
253 		if (addr + size < addr)
254 			return (EINVAL);
255 	}
256 	/*
257 	 * XXX for non-fixed mappings where no hint is provided or
258 	 * the hint would fall in the potential heap space,
259 	 * place it after the end of the largest possible heap.
260 	 *
261 	 * There should really be a pmap call to determine a reasonable
262 	 * location.
263 	 */
264 	else if (addr == 0 ||
265 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
266 	     addr < round_page((vm_offset_t)vms->vm_daddr + MAXDSIZ)))
267 		addr = round_page((vm_offset_t)vms->vm_daddr + MAXDSIZ);
268 
269 	if (flags & MAP_ANON) {
270 		/*
271 		 * Mapping blank space is trivial.
272 		 */
273 		handle = NULL;
274 		maxprot = VM_PROT_ALL;
275 		pos = 0;
276 	} else {
277 		/*
278 		 * Mapping file, get fp for validation. Obtain vnode and make
279 		 * sure it is of appropriate type.
280 		 */
281 		if (((unsigned) uap->fd) >= fdp->fd_nfiles ||
282 		    (fp = fdp->fd_ofiles[uap->fd]) == NULL)
283 			return (EBADF);
284 		if (fp->f_type != DTYPE_VNODE)
285 			return (EINVAL);
286 
287 		/*
288 		 * don't let the descriptor disappear on us if we block
289 		 */
290 		fhold(fp);
291 
292 		/*
293 		 * POSIX shared-memory objects are defined to have
294 		 * kernel persistence, and are not defined to support
295 		 * read(2)/write(2) -- or even open(2).  Thus, we can
296 		 * use MAP_ASYNC to trade on-disk coherence for speed.
297 		 * The shm_open(3) library routine turns on the FPOSIXSHM
298 		 * flag to request this behavior.
299 		 */
300 		if (fp->f_flag & FPOSIXSHM)
301 			flags |= MAP_NOSYNC;
302 		vp = (struct vnode *) fp->f_data;
303 		if (vp->v_type != VREG && vp->v_type != VCHR)
304 			return (EINVAL);
305 		if (vp->v_type == VREG) {
306 			/*
307 			 * Get the proper underlying object
308 			 */
309 			if (VOP_GETVOBJECT(vp, &obj) != 0)
310 				return (EINVAL);
311 			vp = (struct vnode*)obj->handle;
312 		}
313 		/*
314 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
315 		 * SunOS).
316 		 */
317 		if ((vp->v_type == VCHR) &&
318 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
319 			handle = NULL;
320 			maxprot = VM_PROT_ALL;
321 			flags |= MAP_ANON;
322 			pos = 0;
323 		} else {
324 			/*
325 			 * cdevs does not provide private mappings of any kind.
326 			 */
327 			/*
328 			 * However, for XIG X server to continue to work,
329 			 * we should allow the superuser to do it anyway.
330 			 * We only allow it at securelevel < 1.
331 			 * (Because the XIG X server writes directly to video
332 			 * memory via /dev/mem, it should never work at any
333 			 * other securelevel.
334 			 * XXX this will have to go
335 			 */
336 			if (securelevel >= 1)
337 				disablexworkaround = 1;
338 			else
339 				disablexworkaround = suser(p);
340 			if (vp->v_type == VCHR && disablexworkaround &&
341 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
342 				error = EINVAL;
343 				goto done;
344 			}
345 			/*
346 			 * Ensure that file and memory protections are
347 			 * compatible.  Note that we only worry about
348 			 * writability if mapping is shared; in this case,
349 			 * current and max prot are dictated by the open file.
350 			 * XXX use the vnode instead?  Problem is: what
351 			 * credentials do we use for determination? What if
352 			 * proc does a setuid?
353 			 */
354 			maxprot = VM_PROT_EXECUTE;	/* ??? */
355 			if (fp->f_flag & FREAD) {
356 				maxprot |= VM_PROT_READ;
357 			} else if (prot & PROT_READ) {
358 				error = EACCES;
359 				goto done;
360 			}
361 			/*
362 			 * If we are sharing potential changes (either via
363 			 * MAP_SHARED or via the implicit sharing of character
364 			 * device mappings), and we are trying to get write
365 			 * permission although we opened it without asking
366 			 * for it, bail out.  Check for superuser, only if
367 			 * we're at securelevel < 1, to allow the XIG X server
368 			 * to continue to work.
369 			 */
370 
371 			if ((flags & MAP_SHARED) != 0 ||
372 			    (vp->v_type == VCHR && disablexworkaround)) {
373 				if ((fp->f_flag & FWRITE) != 0) {
374 					struct vattr va;
375 					if ((error =
376 					    VOP_GETATTR(vp, &va,
377 						        p->p_ucred, p))) {
378 						goto done;
379 					}
380 					if ((va.va_flags &
381 					   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
382 						maxprot |= VM_PROT_WRITE;
383 					} else if (prot & PROT_WRITE) {
384 						error = EPERM;
385 						goto done;
386 					}
387 				} else if ((prot & PROT_WRITE) != 0) {
388 					error = EACCES;
389 					goto done;
390 				}
391 			} else {
392 				maxprot |= VM_PROT_WRITE;
393 			}
394 
395 			handle = (void *)vp;
396 		}
397 	}
398 
399 	/*
400 	 * Do not allow more then a certain number of vm_map_entry structures
401 	 * per process.  Scale with the number of rforks sharing the map
402 	 * to make the limit reasonable for threads.
403 	 */
404 	if (max_proc_mmap &&
405 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
406 		error = ENOMEM;
407 		goto done;
408 	}
409 
410 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
411 	    flags, handle, pos);
412 	if (error == 0)
413 		p->p_retval[0] = (register_t) (addr + pageoff);
414 done:
415 	if (fp)
416 		fdrop(fp, p);
417 	return (error);
418 }
419 
420 #ifdef COMPAT_43
421 #ifndef _SYS_SYSPROTO_H_
422 struct ommap_args {
423 	caddr_t addr;
424 	int len;
425 	int prot;
426 	int flags;
427 	int fd;
428 	long pos;
429 };
430 #endif
431 int
432 ommap(p, uap)
433 	struct proc *p;
434 	register struct ommap_args *uap;
435 {
436 	struct mmap_args nargs;
437 	static const char cvtbsdprot[8] = {
438 		0,
439 		PROT_EXEC,
440 		PROT_WRITE,
441 		PROT_EXEC | PROT_WRITE,
442 		PROT_READ,
443 		PROT_EXEC | PROT_READ,
444 		PROT_WRITE | PROT_READ,
445 		PROT_EXEC | PROT_WRITE | PROT_READ,
446 	};
447 
448 #define	OMAP_ANON	0x0002
449 #define	OMAP_COPY	0x0020
450 #define	OMAP_SHARED	0x0010
451 #define	OMAP_FIXED	0x0100
452 #define	OMAP_INHERIT	0x0800
453 
454 	nargs.addr = uap->addr;
455 	nargs.len = uap->len;
456 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
457 	nargs.flags = 0;
458 	if (uap->flags & OMAP_ANON)
459 		nargs.flags |= MAP_ANON;
460 	if (uap->flags & OMAP_COPY)
461 		nargs.flags |= MAP_COPY;
462 	if (uap->flags & OMAP_SHARED)
463 		nargs.flags |= MAP_SHARED;
464 	else
465 		nargs.flags |= MAP_PRIVATE;
466 	if (uap->flags & OMAP_FIXED)
467 		nargs.flags |= MAP_FIXED;
468 	if (uap->flags & OMAP_INHERIT)
469 		nargs.flags |= MAP_INHERIT;
470 	nargs.fd = uap->fd;
471 	nargs.pos = uap->pos;
472 	return (mmap(p, &nargs));
473 }
474 #endif				/* COMPAT_43 */
475 
476 
477 #ifndef _SYS_SYSPROTO_H_
478 struct msync_args {
479 	void *addr;
480 	int len;
481 	int flags;
482 };
483 #endif
484 int
485 msync(p, uap)
486 	struct proc *p;
487 	struct msync_args *uap;
488 {
489 	vm_offset_t addr;
490 	vm_size_t size, pageoff;
491 	int flags;
492 	vm_map_t map;
493 	int rv;
494 
495 	addr = (vm_offset_t) uap->addr;
496 	size = uap->len;
497 	flags = uap->flags;
498 
499 	pageoff = (addr & PAGE_MASK);
500 	addr -= pageoff;
501 	size += pageoff;
502 	size = (vm_size_t) round_page(size);
503 	if (addr + size < addr)
504 		return(EINVAL);
505 
506 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
507 		return (EINVAL);
508 
509 	map = &p->p_vmspace->vm_map;
510 
511 	/*
512 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
513 	 * pages with the region containing addr".  Unfortunately, we don't
514 	 * really keep track of individual mmaps so we approximate by flushing
515 	 * the range of the map entry containing addr. This can be incorrect
516 	 * if the region splits or is coalesced with a neighbor.
517 	 */
518 	if (size == 0) {
519 		vm_map_entry_t entry;
520 
521 		vm_map_lock_read(map);
522 		rv = vm_map_lookup_entry(map, addr, &entry);
523 		vm_map_unlock_read(map);
524 		if (rv == FALSE)
525 			return (EINVAL);
526 		addr = entry->start;
527 		size = entry->end - entry->start;
528 	}
529 
530 	/*
531 	 * Clean the pages and interpret the return value.
532 	 */
533 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
534 	    (flags & MS_INVALIDATE) != 0);
535 
536 	switch (rv) {
537 	case KERN_SUCCESS:
538 		break;
539 	case KERN_INVALID_ADDRESS:
540 		return (EINVAL);	/* Sun returns ENOMEM? */
541 	case KERN_FAILURE:
542 		return (EIO);
543 	default:
544 		return (EINVAL);
545 	}
546 
547 	return (0);
548 }
549 
550 #ifndef _SYS_SYSPROTO_H_
551 struct munmap_args {
552 	void *addr;
553 	size_t len;
554 };
555 #endif
556 int
557 munmap(p, uap)
558 	register struct proc *p;
559 	register struct munmap_args *uap;
560 {
561 	vm_offset_t addr;
562 	vm_size_t size, pageoff;
563 	vm_map_t map;
564 
565 	addr = (vm_offset_t) uap->addr;
566 	size = uap->len;
567 
568 	pageoff = (addr & PAGE_MASK);
569 	addr -= pageoff;
570 	size += pageoff;
571 	size = (vm_size_t) round_page(size);
572 	if (addr + size < addr)
573 		return(EINVAL);
574 
575 	if (size == 0)
576 		return (0);
577 
578 	/*
579 	 * Check for illegal addresses.  Watch out for address wrap... Note
580 	 * that VM_*_ADDRESS are not constants due to casts (argh).
581 	 */
582 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
583 		return (EINVAL);
584 #ifndef i386
585 	if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
586 		return (EINVAL);
587 #endif
588 	map = &p->p_vmspace->vm_map;
589 	/*
590 	 * Make sure entire range is allocated.
591 	 */
592 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
593 		return (EINVAL);
594 	/* returns nothing but KERN_SUCCESS anyway */
595 	(void) vm_map_remove(map, addr, addr + size);
596 	return (0);
597 }
598 
599 #if 0
600 void
601 munmapfd(p, fd)
602 	struct proc *p;
603 	int fd;
604 {
605 	/*
606 	 * XXX should unmap any regions mapped to this file
607 	 */
608 	p->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
609 }
610 #endif
611 
612 #ifndef _SYS_SYSPROTO_H_
613 struct mprotect_args {
614 	const void *addr;
615 	size_t len;
616 	int prot;
617 };
618 #endif
619 int
620 mprotect(p, uap)
621 	struct proc *p;
622 	struct mprotect_args *uap;
623 {
624 	vm_offset_t addr;
625 	vm_size_t size, pageoff;
626 	register vm_prot_t prot;
627 
628 	addr = (vm_offset_t) uap->addr;
629 	size = uap->len;
630 	prot = uap->prot & VM_PROT_ALL;
631 #if defined(VM_PROT_READ_IS_EXEC)
632 	if (prot & VM_PROT_READ)
633 		prot |= VM_PROT_EXECUTE;
634 #endif
635 
636 	pageoff = (addr & PAGE_MASK);
637 	addr -= pageoff;
638 	size += pageoff;
639 	size = (vm_size_t) round_page(size);
640 	if (addr + size < addr)
641 		return(EINVAL);
642 
643 	switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
644 		FALSE)) {
645 	case KERN_SUCCESS:
646 		return (0);
647 	case KERN_PROTECTION_FAILURE:
648 		return (EACCES);
649 	}
650 	return (EINVAL);
651 }
652 
653 #ifndef _SYS_SYSPROTO_H_
654 struct minherit_args {
655 	void *addr;
656 	size_t len;
657 	int inherit;
658 };
659 #endif
660 int
661 minherit(p, uap)
662 	struct proc *p;
663 	struct minherit_args *uap;
664 {
665 	vm_offset_t addr;
666 	vm_size_t size, pageoff;
667 	register vm_inherit_t inherit;
668 
669 	addr = (vm_offset_t)uap->addr;
670 	size = uap->len;
671 	inherit = uap->inherit;
672 
673 	pageoff = (addr & PAGE_MASK);
674 	addr -= pageoff;
675 	size += pageoff;
676 	size = (vm_size_t) round_page(size);
677 	if (addr + size < addr)
678 		return(EINVAL);
679 
680 	switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
681 	    inherit)) {
682 	case KERN_SUCCESS:
683 		return (0);
684 	case KERN_PROTECTION_FAILURE:
685 		return (EACCES);
686 	}
687 	return (EINVAL);
688 }
689 
690 #ifndef _SYS_SYSPROTO_H_
691 struct madvise_args {
692 	void *addr;
693 	size_t len;
694 	int behav;
695 };
696 #endif
697 
698 /* ARGSUSED */
699 int
700 madvise(p, uap)
701 	struct proc *p;
702 	struct madvise_args *uap;
703 {
704 	vm_offset_t start, end;
705 
706 	/*
707 	 * Check for illegal behavior
708 	 */
709 	if (uap->behav < 0 || uap->behav > MADV_CORE)
710 		return (EINVAL);
711 	/*
712 	 * Check for illegal addresses.  Watch out for address wrap... Note
713 	 * that VM_*_ADDRESS are not constants due to casts (argh).
714 	 */
715 	if (VM_MAXUSER_ADDRESS > 0 &&
716 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
717 		return (EINVAL);
718 #ifndef i386
719 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
720 		return (EINVAL);
721 #endif
722 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
723 		return (EINVAL);
724 
725 	/*
726 	 * Since this routine is only advisory, we default to conservative
727 	 * behavior.
728 	 */
729 	start = trunc_page((vm_offset_t) uap->addr);
730 	end = round_page((vm_offset_t) uap->addr + uap->len);
731 
732 	if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav))
733 		return (EINVAL);
734 	return (0);
735 }
736 
737 #ifndef _SYS_SYSPROTO_H_
738 struct mincore_args {
739 	const void *addr;
740 	size_t len;
741 	char *vec;
742 };
743 #endif
744 
745 /* ARGSUSED */
746 int
747 mincore(p, uap)
748 	struct proc *p;
749 	struct mincore_args *uap;
750 {
751 	vm_offset_t addr, first_addr;
752 	vm_offset_t end, cend;
753 	pmap_t pmap;
754 	vm_map_t map;
755 	char *vec;
756 	int error;
757 	int vecindex, lastvecindex;
758 	register vm_map_entry_t current;
759 	vm_map_entry_t entry;
760 	int mincoreinfo;
761 	unsigned int timestamp;
762 
763 	/*
764 	 * Make sure that the addresses presented are valid for user
765 	 * mode.
766 	 */
767 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
768 	end = addr + (vm_size_t)round_page(uap->len);
769 	if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
770 		return (EINVAL);
771 	if (end < addr)
772 		return (EINVAL);
773 
774 	/*
775 	 * Address of byte vector
776 	 */
777 	vec = uap->vec;
778 
779 	map = &p->p_vmspace->vm_map;
780 	pmap = vmspace_pmap(p->p_vmspace);
781 
782 	vm_map_lock_read(map);
783 RestartScan:
784 	timestamp = map->timestamp;
785 
786 	if (!vm_map_lookup_entry(map, addr, &entry))
787 		entry = entry->next;
788 
789 	/*
790 	 * Do this on a map entry basis so that if the pages are not
791 	 * in the current processes address space, we can easily look
792 	 * up the pages elsewhere.
793 	 */
794 	lastvecindex = -1;
795 	for(current = entry;
796 		(current != &map->header) && (current->start < end);
797 		current = current->next) {
798 
799 		/*
800 		 * ignore submaps (for now) or null objects
801 		 */
802 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
803 			current->object.vm_object == NULL)
804 			continue;
805 
806 		/*
807 		 * limit this scan to the current map entry and the
808 		 * limits for the mincore call
809 		 */
810 		if (addr < current->start)
811 			addr = current->start;
812 		cend = current->end;
813 		if (cend > end)
814 			cend = end;
815 
816 		/*
817 		 * scan this entry one page at a time
818 		 */
819 		while(addr < cend) {
820 			/*
821 			 * Check pmap first, it is likely faster, also
822 			 * it can provide info as to whether we are the
823 			 * one referencing or modifying the page.
824 			 */
825 			mincoreinfo = pmap_mincore(pmap, addr);
826 			if (!mincoreinfo) {
827 				vm_pindex_t pindex;
828 				vm_ooffset_t offset;
829 				vm_page_t m;
830 				/*
831 				 * calculate the page index into the object
832 				 */
833 				offset = current->offset + (addr - current->start);
834 				pindex = OFF_TO_IDX(offset);
835 				m = vm_page_lookup(current->object.vm_object,
836 					pindex);
837 				/*
838 				 * if the page is resident, then gather information about
839 				 * it.
840 				 */
841 				if (m) {
842 					mincoreinfo = MINCORE_INCORE;
843 					if (m->dirty ||
844 						pmap_is_modified(m))
845 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
846 					if ((m->flags & PG_REFERENCED) ||
847 						pmap_ts_referenced(m)) {
848 						vm_page_flag_set(m, PG_REFERENCED);
849 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
850 					}
851 				}
852 			}
853 
854 			/*
855 			 * subyte may page fault.  In case it needs to modify
856 			 * the map, we release the lock.
857 			 */
858 			vm_map_unlock_read(map);
859 
860 			/*
861 			 * calculate index into user supplied byte vector
862 			 */
863 			vecindex = OFF_TO_IDX(addr - first_addr);
864 
865 			/*
866 			 * If we have skipped map entries, we need to make sure that
867 			 * the byte vector is zeroed for those skipped entries.
868 			 */
869 			while((lastvecindex + 1) < vecindex) {
870 				error = subyte( vec + lastvecindex, 0);
871 				if (error) {
872 					return (EFAULT);
873 				}
874 				++lastvecindex;
875 			}
876 
877 			/*
878 			 * Pass the page information to the user
879 			 */
880 			error = subyte( vec + vecindex, mincoreinfo);
881 			if (error) {
882 				return (EFAULT);
883 			}
884 
885 			/*
886 			 * If the map has changed, due to the subyte, the previous
887 			 * output may be invalid.
888 			 */
889 			vm_map_lock_read(map);
890 			if (timestamp != map->timestamp)
891 				goto RestartScan;
892 
893 			lastvecindex = vecindex;
894 			addr += PAGE_SIZE;
895 		}
896 	}
897 
898 	/*
899 	 * subyte may page fault.  In case it needs to modify
900 	 * the map, we release the lock.
901 	 */
902 	vm_map_unlock_read(map);
903 
904 	/*
905 	 * Zero the last entries in the byte vector.
906 	 */
907 	vecindex = OFF_TO_IDX(end - first_addr);
908 	while((lastvecindex + 1) < vecindex) {
909 		error = subyte( vec + lastvecindex, 0);
910 		if (error) {
911 			return (EFAULT);
912 		}
913 		++lastvecindex;
914 	}
915 
916 	/*
917 	 * If the map has changed, due to the subyte, the previous
918 	 * output may be invalid.
919 	 */
920 	vm_map_lock_read(map);
921 	if (timestamp != map->timestamp)
922 		goto RestartScan;
923 	vm_map_unlock_read(map);
924 
925 	return (0);
926 }
927 
928 #ifndef _SYS_SYSPROTO_H_
929 struct mlock_args {
930 	const void *addr;
931 	size_t len;
932 };
933 #endif
934 int
935 mlock(p, uap)
936 	struct proc *p;
937 	struct mlock_args *uap;
938 {
939 	vm_offset_t addr;
940 	vm_size_t size, pageoff;
941 	int error;
942 
943 	addr = (vm_offset_t) uap->addr;
944 	size = uap->len;
945 
946 	pageoff = (addr & PAGE_MASK);
947 	addr -= pageoff;
948 	size += pageoff;
949 	size = (vm_size_t) round_page(size);
950 
951 	/* disable wrap around */
952 	if (addr + size < addr)
953 		return (EINVAL);
954 
955 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
956 		return (EAGAIN);
957 
958 #ifdef pmap_wired_count
959 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
960 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
961 		return (ENOMEM);
962 #else
963 	error = suser(p);
964 	if (error)
965 		return (error);
966 #endif
967 
968 	error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
969 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
970 }
971 
972 #ifndef _SYS_SYSPROTO_H_
973 struct mlockall_args {
974 	int	how;
975 };
976 #endif
977 
978 int
979 mlockall(p, uap)
980 	struct proc *p;
981 	struct mlockall_args *uap;
982 {
983 	return 0;
984 }
985 
986 #ifndef _SYS_SYSPROTO_H_
987 struct mlockall_args {
988 	int	how;
989 };
990 #endif
991 
992 int
993 munlockall(p, uap)
994 	struct proc *p;
995 	struct munlockall_args *uap;
996 {
997 	return 0;
998 }
999 
1000 #ifndef _SYS_SYSPROTO_H_
1001 struct munlock_args {
1002 	const void *addr;
1003 	size_t len;
1004 };
1005 #endif
1006 int
1007 munlock(p, uap)
1008 	struct proc *p;
1009 	struct munlock_args *uap;
1010 {
1011 	vm_offset_t addr;
1012 	vm_size_t size, pageoff;
1013 	int error;
1014 
1015 	addr = (vm_offset_t) uap->addr;
1016 	size = uap->len;
1017 
1018 	pageoff = (addr & PAGE_MASK);
1019 	addr -= pageoff;
1020 	size += pageoff;
1021 	size = (vm_size_t) round_page(size);
1022 
1023 	/* disable wrap around */
1024 	if (addr + size < addr)
1025 		return (EINVAL);
1026 
1027 #ifndef pmap_wired_count
1028 	error = suser(p);
1029 	if (error)
1030 		return (error);
1031 #endif
1032 
1033 	error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
1034 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1035 }
1036 
1037 /*
1038  * Internal version of mmap.
1039  * Currently used by mmap, exec, and sys5 shared memory.
1040  * Handle is either a vnode pointer or NULL for MAP_ANON.
1041  */
1042 int
1043 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1044 	vm_prot_t maxprot, int flags,
1045 	void *handle,
1046 	vm_ooffset_t foff)
1047 {
1048 	boolean_t fitit;
1049 	vm_object_t object;
1050 	struct vnode *vp = NULL;
1051 	objtype_t type;
1052 	int rv = KERN_SUCCESS;
1053 	vm_ooffset_t objsize;
1054 	int docow;
1055 	struct proc *p = curproc;
1056 
1057 	if (size == 0)
1058 		return (0);
1059 
1060 	objsize = size = round_page(size);
1061 
1062 	/*
1063 	 * We currently can only deal with page aligned file offsets.
1064 	 * The check is here rather than in the syscall because the
1065 	 * kernel calls this function internally for other mmaping
1066 	 * operations (such as in exec) and non-aligned offsets will
1067 	 * cause pmap inconsistencies...so we want to be sure to
1068 	 * disallow this in all cases.
1069 	 */
1070 	if (foff & PAGE_MASK)
1071 		return (EINVAL);
1072 
1073 	if ((flags & MAP_FIXED) == 0) {
1074 		fitit = TRUE;
1075 		*addr = round_page(*addr);
1076 	} else {
1077 		if (*addr != trunc_page(*addr))
1078 			return (EINVAL);
1079 		fitit = FALSE;
1080 		(void) vm_map_remove(map, *addr, *addr + size);
1081 	}
1082 
1083 	/*
1084 	 * Lookup/allocate object.
1085 	 */
1086 	if (flags & MAP_ANON) {
1087 		type = OBJT_DEFAULT;
1088 		/*
1089 		 * Unnamed anonymous regions always start at 0.
1090 		 */
1091 		if (handle == 0)
1092 			foff = 0;
1093 	} else {
1094 		vp = (struct vnode *) handle;
1095 		if (vp->v_type == VCHR) {
1096 			type = OBJT_DEVICE;
1097 			handle = (void *)(intptr_t)vp->v_rdev;
1098 		} else {
1099 			struct vattr vat;
1100 			int error;
1101 
1102 			error = VOP_GETATTR(vp, &vat, p->p_ucred, p);
1103 			if (error)
1104 				return (error);
1105 			objsize = round_page(vat.va_size);
1106 			type = OBJT_VNODE;
1107 			/*
1108 			 * if it is a regular file without any references
1109 			 * we do not need to sync it.
1110 			 */
1111 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1112 				flags |= MAP_NOSYNC;
1113 			}
1114 		}
1115 	}
1116 
1117 	if (handle == NULL) {
1118 		object = NULL;
1119 		docow = 0;
1120 	} else {
1121 		object = vm_pager_allocate(type,
1122 			handle, objsize, prot, foff);
1123 		if (object == NULL)
1124 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1125 		docow = MAP_PREFAULT_PARTIAL;
1126 	}
1127 
1128 	/*
1129 	 * Force device mappings to be shared.
1130 	 */
1131 	if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1132 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1133 		flags |= MAP_SHARED;
1134 	}
1135 
1136 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1137 		docow |= MAP_COPY_ON_WRITE;
1138 	if (flags & MAP_NOSYNC)
1139 		docow |= MAP_DISABLE_SYNCER;
1140 	if (flags & MAP_NOCORE)
1141 		docow |= MAP_DISABLE_COREDUMP;
1142 
1143 #if defined(VM_PROT_READ_IS_EXEC)
1144 	if (prot & VM_PROT_READ)
1145 		prot |= VM_PROT_EXECUTE;
1146 
1147 	if (maxprot & VM_PROT_READ)
1148 		maxprot |= VM_PROT_EXECUTE;
1149 #endif
1150 
1151 	if (fitit) {
1152 		*addr = pmap_addr_hint(object, *addr, size);
1153 	}
1154 
1155 	if (flags & MAP_STACK)
1156 		rv = vm_map_stack (map, *addr, size, prot,
1157 				   maxprot, docow);
1158 	else
1159 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1160 				 prot, maxprot, docow);
1161 
1162 	if (rv != KERN_SUCCESS) {
1163 		/*
1164 		 * Lose the object reference. Will destroy the
1165 		 * object if it's an unnamed anonymous mapping
1166 		 * or named anonymous without other references.
1167 		 */
1168 		vm_object_deallocate(object);
1169 		goto out;
1170 	}
1171 
1172 	/*
1173 	 * Shared memory is also shared with children.
1174 	 */
1175 	if (flags & (MAP_SHARED|MAP_INHERIT)) {
1176 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1177 		if (rv != KERN_SUCCESS) {
1178 			(void) vm_map_remove(map, *addr, *addr + size);
1179 			goto out;
1180 		}
1181 	}
1182 out:
1183 	switch (rv) {
1184 	case KERN_SUCCESS:
1185 		return (0);
1186 	case KERN_INVALID_ADDRESS:
1187 	case KERN_NO_SPACE:
1188 		return (ENOMEM);
1189 	case KERN_PROTECTION_FAILURE:
1190 		return (EACCES);
1191 	default:
1192 		return (EINVAL);
1193 	}
1194 }
1195