xref: /freebsd/sys/vm/vm_mmap.c (revision ee2ea5ceafed78a5bd9810beb9e3ca927180c226)
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 
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/vnode.h>
59 #include <sys/fcntl.h>
60 #include <sys/file.h>
61 #include <sys/mman.h>
62 #include <sys/conf.h>
63 #include <sys/stat.h>
64 #include <sys/vmmeter.h>
65 #include <sys/sysctl.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_param.h>
69 #include <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 static void vmmapentry_rsrc_init(void *);
97 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
98 
99 static void
100 vmmapentry_rsrc_init(dummy)
101         void *dummy;
102 {
103     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
104     max_proc_mmap /= 100;
105 }
106 
107 /*
108  * MPSAFE
109  */
110 /* ARGSUSED */
111 int
112 sbrk(td, uap)
113 	struct thread *td;
114 	struct sbrk_args *uap;
115 {
116 	/* Not yet implemented */
117 	/* mtx_lock(&Giant); */
118 	/* mtx_unlock(&Giant); */
119 	return (EOPNOTSUPP);
120 }
121 
122 #ifndef _SYS_SYSPROTO_H_
123 struct sstk_args {
124 	int incr;
125 };
126 #endif
127 
128 /*
129  * MPSAFE
130  */
131 /* ARGSUSED */
132 int
133 sstk(td, uap)
134 	struct thread *td;
135 	struct sstk_args *uap;
136 {
137 	/* Not yet implemented */
138 	/* mtx_lock(&Giant); */
139 	/* mtx_unlock(&Giant); */
140 	return (EOPNOTSUPP);
141 }
142 
143 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
144 #ifndef _SYS_SYSPROTO_H_
145 struct getpagesize_args {
146 	int dummy;
147 };
148 #endif
149 
150 /* ARGSUSED */
151 int
152 ogetpagesize(td, uap)
153 	struct thread *td;
154 	struct getpagesize_args *uap;
155 {
156 	/* MP SAFE */
157 	td->td_retval[0] = PAGE_SIZE;
158 	return (0);
159 }
160 #endif				/* COMPAT_43 || COMPAT_SUNOS */
161 
162 
163 /*
164  * Memory Map (mmap) system call.  Note that the file offset
165  * and address are allowed to be NOT page aligned, though if
166  * the MAP_FIXED flag it set, both must have the same remainder
167  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
168  * page-aligned, the actual mapping starts at trunc_page(addr)
169  * and the return value is adjusted up by the page offset.
170  *
171  * Generally speaking, only character devices which are themselves
172  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
173  * there would be no cache coherency between a descriptor and a VM mapping
174  * both to the same character device.
175  *
176  * Block devices can be mmap'd no matter what they represent.  Cache coherency
177  * is maintained as long as you do not write directly to the underlying
178  * character device.
179  */
180 #ifndef _SYS_SYSPROTO_H_
181 struct mmap_args {
182 	void *addr;
183 	size_t len;
184 	int prot;
185 	int flags;
186 	int fd;
187 	long pad;
188 	off_t pos;
189 };
190 #endif
191 
192 /*
193  * MPSAFE
194  */
195 int
196 mmap(td, uap)
197 	struct thread *td;
198 	struct mmap_args *uap;
199 {
200 	struct file *fp = NULL;
201 	struct vnode *vp;
202 	vm_offset_t addr;
203 	vm_size_t size, pageoff;
204 	vm_prot_t prot, maxprot;
205 	void *handle;
206 	int flags, error;
207 	int disablexworkaround;
208 	off_t pos;
209 	struct vmspace *vms = td->td_proc->p_vmspace;
210 	vm_object_t obj;
211 
212 	addr = (vm_offset_t) uap->addr;
213 	size = uap->len;
214 	prot = uap->prot & VM_PROT_ALL;
215 	flags = uap->flags;
216 	pos = uap->pos;
217 
218 	fp = NULL;
219 	/* make sure mapping fits into numeric range etc */
220 	if ((ssize_t) uap->len < 0 ||
221 	    ((flags & MAP_ANON) && uap->fd != -1))
222 		return (EINVAL);
223 
224 	if (flags & MAP_STACK) {
225 		if ((uap->fd != -1) ||
226 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
227 			return (EINVAL);
228 		flags |= MAP_ANON;
229 		pos = 0;
230 	}
231 
232 	/*
233 	 * Align the file position to a page boundary,
234 	 * and save its page offset component.
235 	 */
236 	pageoff = (pos & PAGE_MASK);
237 	pos -= pageoff;
238 
239 	/* Adjust size for rounding (on both ends). */
240 	size += pageoff;			/* low end... */
241 	size = (vm_size_t) round_page(size);	/* hi end */
242 
243 	/*
244 	 * Check for illegal addresses.  Watch out for address wrap... Note
245 	 * that VM_*_ADDRESS are not constants due to casts (argh).
246 	 */
247 	if (flags & MAP_FIXED) {
248 		/*
249 		 * The specified address must have the same remainder
250 		 * as the file offset taken modulo PAGE_SIZE, so it
251 		 * should be aligned after adjustment by pageoff.
252 		 */
253 		addr -= pageoff;
254 		if (addr & PAGE_MASK)
255 			return (EINVAL);
256 		/* Address range must be all in user VM space. */
257 		if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
258 			return (EINVAL);
259 #ifndef i386
260 		if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
261 			return (EINVAL);
262 #endif
263 		if (addr + size < addr)
264 			return (EINVAL);
265 	}
266 	/*
267 	 * XXX for non-fixed mappings where no hint is provided or
268 	 * the hint would fall in the potential heap space,
269 	 * place it after the end of the largest possible heap.
270 	 *
271 	 * There should really be a pmap call to determine a reasonable
272 	 * location.
273 	 */
274 	else if (addr == 0 ||
275 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
276 	     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
277 		addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
278 
279 	mtx_lock(&Giant);	/* syscall marked mp-safe but isn't */
280 	if (flags & MAP_ANON) {
281 		/*
282 		 * Mapping blank space is trivial.
283 		 */
284 		handle = NULL;
285 		maxprot = VM_PROT_ALL;
286 		pos = 0;
287 	} else {
288 		/*
289 		 * Mapping file, get fp for validation. Obtain vnode and make
290 		 * sure it is of appropriate type.
291 		 * don't let the descriptor disappear on us if we block
292 		 */
293 		if ((error = fget(td, uap->fd, &fp)) != 0)
294 			goto done;
295 		if (fp->f_type != DTYPE_VNODE) {
296 			error = EINVAL;
297 			goto done;
298 		}
299 
300 		/*
301 		 * POSIX shared-memory objects are defined to have
302 		 * kernel persistence, and are not defined to support
303 		 * read(2)/write(2) -- or even open(2).  Thus, we can
304 		 * use MAP_ASYNC to trade on-disk coherence for speed.
305 		 * The shm_open(3) library routine turns on the FPOSIXSHM
306 		 * flag to request this behavior.
307 		 */
308 		if (fp->f_flag & FPOSIXSHM)
309 			flags |= MAP_NOSYNC;
310 		vp = (struct vnode *) fp->f_data;
311 		if (vp->v_type != VREG && vp->v_type != VCHR) {
312 			error = EINVAL;
313 			goto done;
314 		}
315 		if (vp->v_type == VREG) {
316 			/*
317 			 * Get the proper underlying object
318 			 */
319 			if (VOP_GETVOBJECT(vp, &obj) != 0) {
320 				error = EINVAL;
321 				goto done;
322 			}
323 			vp = (struct vnode*)obj->handle;
324 		}
325 		/*
326 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
327 		 * SunOS).
328 		 */
329 		if ((vp->v_type == VCHR) &&
330 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
331 			handle = NULL;
332 			maxprot = VM_PROT_ALL;
333 			flags |= MAP_ANON;
334 			pos = 0;
335 		} else {
336 			/*
337 			 * cdevs does not provide private mappings of any kind.
338 			 */
339 			/*
340 			 * However, for XIG X server to continue to work,
341 			 * we should allow the superuser to do it anyway.
342 			 * We only allow it at securelevel < 1.
343 			 * (Because the XIG X server writes directly to video
344 			 * memory via /dev/mem, it should never work at any
345 			 * other securelevel.
346 			 * XXX this will have to go
347 			 */
348 			if (securelevel_ge(td->td_ucred, 1))
349 				disablexworkaround = 1;
350 			else
351 				disablexworkaround = suser(td);
352 			if (vp->v_type == VCHR && disablexworkaround &&
353 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
354 				error = EINVAL;
355 				goto done;
356 			}
357 			/*
358 			 * Ensure that file and memory protections are
359 			 * compatible.  Note that we only worry about
360 			 * writability if mapping is shared; in this case,
361 			 * current and max prot are dictated by the open file.
362 			 * XXX use the vnode instead?  Problem is: what
363 			 * credentials do we use for determination? What if
364 			 * proc does a setuid?
365 			 */
366 			maxprot = VM_PROT_EXECUTE;	/* ??? */
367 			if (fp->f_flag & FREAD) {
368 				maxprot |= VM_PROT_READ;
369 			} else if (prot & PROT_READ) {
370 				error = EACCES;
371 				goto done;
372 			}
373 			/*
374 			 * If we are sharing potential changes (either via
375 			 * MAP_SHARED or via the implicit sharing of character
376 			 * device mappings), and we are trying to get write
377 			 * permission although we opened it without asking
378 			 * for it, bail out.  Check for superuser, only if
379 			 * we're at securelevel < 1, to allow the XIG X server
380 			 * to continue to work.
381 			 */
382 			if ((flags & MAP_SHARED) != 0 ||
383 			    (vp->v_type == VCHR && disablexworkaround)) {
384 				if ((fp->f_flag & FWRITE) != 0) {
385 					struct vattr va;
386 					if ((error =
387 					    VOP_GETATTR(vp, &va,
388 						        td->td_ucred, td))) {
389 						goto done;
390 					}
391 					if ((va.va_flags &
392 					   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
393 						maxprot |= VM_PROT_WRITE;
394 					} else if (prot & PROT_WRITE) {
395 						error = EPERM;
396 						goto done;
397 					}
398 				} else if ((prot & PROT_WRITE) != 0) {
399 					error = EACCES;
400 					goto done;
401 				}
402 			} else {
403 				maxprot |= VM_PROT_WRITE;
404 			}
405 
406 			handle = (void *)vp;
407 		}
408 	}
409 
410 	/*
411 	 * Do not allow more then a certain number of vm_map_entry structures
412 	 * per process.  Scale with the number of rforks sharing the map
413 	 * to make the limit reasonable for threads.
414 	 */
415 	if (max_proc_mmap &&
416 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
417 		error = ENOMEM;
418 		goto done;
419 	}
420 
421 	mtx_unlock(&Giant);
422 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
423 	    flags, handle, pos);
424 	if (error == 0)
425 		td->td_retval[0] = (register_t) (addr + pageoff);
426 	mtx_lock(&Giant);
427 done:
428 	if (fp)
429 		fdrop(fp, td);
430 	mtx_unlock(&Giant);
431 	return (error);
432 }
433 
434 #ifdef COMPAT_43
435 #ifndef _SYS_SYSPROTO_H_
436 struct ommap_args {
437 	caddr_t addr;
438 	int len;
439 	int prot;
440 	int flags;
441 	int fd;
442 	long pos;
443 };
444 #endif
445 int
446 ommap(td, uap)
447 	struct thread *td;
448 	struct ommap_args *uap;
449 {
450 	struct mmap_args nargs;
451 	static const char cvtbsdprot[8] = {
452 		0,
453 		PROT_EXEC,
454 		PROT_WRITE,
455 		PROT_EXEC | PROT_WRITE,
456 		PROT_READ,
457 		PROT_EXEC | PROT_READ,
458 		PROT_WRITE | PROT_READ,
459 		PROT_EXEC | PROT_WRITE | PROT_READ,
460 	};
461 
462 #define	OMAP_ANON	0x0002
463 #define	OMAP_COPY	0x0020
464 #define	OMAP_SHARED	0x0010
465 #define	OMAP_FIXED	0x0100
466 
467 	nargs.addr = uap->addr;
468 	nargs.len = uap->len;
469 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
470 	nargs.flags = 0;
471 	if (uap->flags & OMAP_ANON)
472 		nargs.flags |= MAP_ANON;
473 	if (uap->flags & OMAP_COPY)
474 		nargs.flags |= MAP_COPY;
475 	if (uap->flags & OMAP_SHARED)
476 		nargs.flags |= MAP_SHARED;
477 	else
478 		nargs.flags |= MAP_PRIVATE;
479 	if (uap->flags & OMAP_FIXED)
480 		nargs.flags |= MAP_FIXED;
481 	nargs.fd = uap->fd;
482 	nargs.pos = uap->pos;
483 	return (mmap(td, &nargs));
484 }
485 #endif				/* COMPAT_43 */
486 
487 
488 #ifndef _SYS_SYSPROTO_H_
489 struct msync_args {
490 	void *addr;
491 	int len;
492 	int flags;
493 };
494 #endif
495 /*
496  * MPSAFE
497  */
498 int
499 msync(td, uap)
500 	struct thread *td;
501 	struct msync_args *uap;
502 {
503 	vm_offset_t addr;
504 	vm_size_t size, pageoff;
505 	int flags;
506 	vm_map_t map;
507 	int rv;
508 
509 	addr = (vm_offset_t) uap->addr;
510 	size = uap->len;
511 	flags = uap->flags;
512 
513 	pageoff = (addr & PAGE_MASK);
514 	addr -= pageoff;
515 	size += pageoff;
516 	size = (vm_size_t) round_page(size);
517 	if (addr + size < addr)
518 		return (EINVAL);
519 
520 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
521 		return (EINVAL);
522 
523 	mtx_lock(&Giant);
524 
525 	map = &td->td_proc->p_vmspace->vm_map;
526 
527 	/*
528 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
529 	 * pages with the region containing addr".  Unfortunately, we don't
530 	 * really keep track of individual mmaps so we approximate by flushing
531 	 * the range of the map entry containing addr. This can be incorrect
532 	 * if the region splits or is coalesced with a neighbor.
533 	 */
534 	if (size == 0) {
535 		vm_map_entry_t entry;
536 
537 		vm_map_lock_read(map);
538 		rv = vm_map_lookup_entry(map, addr, &entry);
539 		vm_map_unlock_read(map);
540 		if (rv == FALSE) {
541 			rv = -1;
542 			goto done2;
543 		}
544 		addr = entry->start;
545 		size = entry->end - entry->start;
546 	}
547 
548 	/*
549 	 * Clean the pages and interpret the return value.
550 	 */
551 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
552 	    (flags & MS_INVALIDATE) != 0);
553 
554 done2:
555 	mtx_unlock(&Giant);
556 
557 	switch (rv) {
558 	case KERN_SUCCESS:
559 		return (0);
560 	case KERN_INVALID_ADDRESS:
561 		return (EINVAL);	/* Sun returns ENOMEM? */
562 	case KERN_FAILURE:
563 		return (EIO);
564 	default:
565 		return (EINVAL);
566 	}
567 }
568 
569 #ifndef _SYS_SYSPROTO_H_
570 struct munmap_args {
571 	void *addr;
572 	size_t len;
573 };
574 #endif
575 /*
576  * MPSAFE
577  */
578 int
579 munmap(td, uap)
580 	struct thread *td;
581 	struct munmap_args *uap;
582 {
583 	vm_offset_t addr;
584 	vm_size_t size, pageoff;
585 	vm_map_t map;
586 
587 	addr = (vm_offset_t) uap->addr;
588 	size = uap->len;
589 
590 	pageoff = (addr & PAGE_MASK);
591 	addr -= pageoff;
592 	size += pageoff;
593 	size = (vm_size_t) round_page(size);
594 	if (addr + size < addr)
595 		return (EINVAL);
596 
597 	if (size == 0)
598 		return (0);
599 
600 	/*
601 	 * Check for illegal addresses.  Watch out for address wrap... Note
602 	 * that VM_*_ADDRESS are not constants due to casts (argh).
603 	 */
604 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
605 		return (EINVAL);
606 #ifndef i386
607 	if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
608 		return (EINVAL);
609 #endif
610 	map = &td->td_proc->p_vmspace->vm_map;
611 	/*
612 	 * Make sure entire range is allocated.
613 	 */
614 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
615 		return (EINVAL);
616 
617 	/* returns nothing but KERN_SUCCESS anyway */
618 	mtx_lock(&Giant);
619 	(void) vm_map_remove(map, addr, addr + size);
620 	mtx_unlock(&Giant);
621 	return (0);
622 }
623 
624 #if 0
625 void
626 munmapfd(td, fd)
627 	struct thread *td;
628 	int fd;
629 {
630 	/*
631 	 * XXX should unmap any regions mapped to this file
632 	 */
633 	FILEDESC_LOCK(p->p_fd);
634 	td->td_proc->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
635 	FILEDESC_UNLOCK(p->p_fd);
636 }
637 #endif
638 
639 #ifndef _SYS_SYSPROTO_H_
640 struct mprotect_args {
641 	const void *addr;
642 	size_t len;
643 	int prot;
644 };
645 #endif
646 /*
647  * MPSAFE
648  */
649 int
650 mprotect(td, uap)
651 	struct thread *td;
652 	struct mprotect_args *uap;
653 {
654 	vm_offset_t addr;
655 	vm_size_t size, pageoff;
656 	vm_prot_t prot;
657 	int ret;
658 
659 	addr = (vm_offset_t) uap->addr;
660 	size = uap->len;
661 	prot = uap->prot & VM_PROT_ALL;
662 #if defined(VM_PROT_READ_IS_EXEC)
663 	if (prot & VM_PROT_READ)
664 		prot |= VM_PROT_EXECUTE;
665 #endif
666 
667 	pageoff = (addr & PAGE_MASK);
668 	addr -= pageoff;
669 	size += pageoff;
670 	size = (vm_size_t) round_page(size);
671 	if (addr + size < addr)
672 		return (EINVAL);
673 
674 	mtx_lock(&Giant);
675 	ret = vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
676 		     addr + size, prot, FALSE);
677 	mtx_unlock(&Giant);
678 	switch (ret) {
679 	case KERN_SUCCESS:
680 		return (0);
681 	case KERN_PROTECTION_FAILURE:
682 		return (EACCES);
683 	}
684 	return (EINVAL);
685 }
686 
687 #ifndef _SYS_SYSPROTO_H_
688 struct minherit_args {
689 	void *addr;
690 	size_t len;
691 	int inherit;
692 };
693 #endif
694 /*
695  * MPSAFE
696  */
697 int
698 minherit(td, uap)
699 	struct thread *td;
700 	struct minherit_args *uap;
701 {
702 	vm_offset_t addr;
703 	vm_size_t size, pageoff;
704 	vm_inherit_t inherit;
705 	int ret;
706 
707 	addr = (vm_offset_t)uap->addr;
708 	size = uap->len;
709 	inherit = uap->inherit;
710 
711 	pageoff = (addr & PAGE_MASK);
712 	addr -= pageoff;
713 	size += pageoff;
714 	size = (vm_size_t) round_page(size);
715 	if (addr + size < addr)
716 		return (EINVAL);
717 
718 	mtx_lock(&Giant);
719 	ret = vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, addr+size,
720 		    inherit);
721 	mtx_unlock(&Giant);
722 
723 	switch (ret) {
724 	case KERN_SUCCESS:
725 		return (0);
726 	case KERN_PROTECTION_FAILURE:
727 		return (EACCES);
728 	}
729 	return (EINVAL);
730 }
731 
732 #ifndef _SYS_SYSPROTO_H_
733 struct madvise_args {
734 	void *addr;
735 	size_t len;
736 	int behav;
737 };
738 #endif
739 
740 /*
741  * MPSAFE
742  */
743 /* ARGSUSED */
744 int
745 madvise(td, uap)
746 	struct thread *td;
747 	struct madvise_args *uap;
748 {
749 	vm_offset_t start, end;
750 	int ret;
751 
752 	/*
753 	 * Check for illegal behavior
754 	 */
755 	if (uap->behav < 0 || uap->behav > MADV_CORE)
756 		return (EINVAL);
757 	/*
758 	 * Check for illegal addresses.  Watch out for address wrap... Note
759 	 * that VM_*_ADDRESS are not constants due to casts (argh).
760 	 */
761 	if (VM_MAXUSER_ADDRESS > 0 &&
762 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
763 		return (EINVAL);
764 #ifndef i386
765 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
766 		return (EINVAL);
767 #endif
768 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
769 		return (EINVAL);
770 
771 	/*
772 	 * Since this routine is only advisory, we default to conservative
773 	 * behavior.
774 	 */
775 	start = trunc_page((vm_offset_t) uap->addr);
776 	end = round_page((vm_offset_t) uap->addr + uap->len);
777 
778 	mtx_lock(&Giant);
779 	ret = vm_map_madvise(&td->td_proc->p_vmspace->vm_map, start, end, uap->behav);
780 	mtx_unlock(&Giant);
781 	return (ret ? EINVAL : 0);
782 }
783 
784 #ifndef _SYS_SYSPROTO_H_
785 struct mincore_args {
786 	const void *addr;
787 	size_t len;
788 	char *vec;
789 };
790 #endif
791 
792 /*
793  * MPSAFE
794  */
795 /* ARGSUSED */
796 int
797 mincore(td, uap)
798 	struct thread *td;
799 	struct mincore_args *uap;
800 {
801 	vm_offset_t addr, first_addr;
802 	vm_offset_t end, cend;
803 	pmap_t pmap;
804 	vm_map_t map;
805 	char *vec;
806 	int error = 0;
807 	int vecindex, lastvecindex;
808 	vm_map_entry_t current;
809 	vm_map_entry_t entry;
810 	int mincoreinfo;
811 	unsigned int timestamp;
812 
813 	/*
814 	 * Make sure that the addresses presented are valid for user
815 	 * mode.
816 	 */
817 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
818 	end = addr + (vm_size_t)round_page(uap->len);
819 	if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
820 		return (EINVAL);
821 	if (end < addr)
822 		return (EINVAL);
823 
824 	/*
825 	 * Address of byte vector
826 	 */
827 	vec = uap->vec;
828 
829 	mtx_lock(&Giant);
830 	map = &td->td_proc->p_vmspace->vm_map;
831 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
832 
833 	vm_map_lock_read(map);
834 RestartScan:
835 	timestamp = map->timestamp;
836 
837 	if (!vm_map_lookup_entry(map, addr, &entry))
838 		entry = entry->next;
839 
840 	/*
841 	 * Do this on a map entry basis so that if the pages are not
842 	 * in the current processes address space, we can easily look
843 	 * up the pages elsewhere.
844 	 */
845 	lastvecindex = -1;
846 	for (current = entry;
847 	    (current != &map->header) && (current->start < end);
848 	    current = current->next) {
849 
850 		/*
851 		 * ignore submaps (for now) or null objects
852 		 */
853 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
854 			current->object.vm_object == NULL)
855 			continue;
856 
857 		/*
858 		 * limit this scan to the current map entry and the
859 		 * limits for the mincore call
860 		 */
861 		if (addr < current->start)
862 			addr = current->start;
863 		cend = current->end;
864 		if (cend > end)
865 			cend = end;
866 
867 		/*
868 		 * scan this entry one page at a time
869 		 */
870 		while (addr < cend) {
871 			/*
872 			 * Check pmap first, it is likely faster, also
873 			 * it can provide info as to whether we are the
874 			 * one referencing or modifying the page.
875 			 */
876 			mincoreinfo = pmap_mincore(pmap, addr);
877 			if (!mincoreinfo) {
878 				vm_pindex_t pindex;
879 				vm_ooffset_t offset;
880 				vm_page_t m;
881 				/*
882 				 * calculate the page index into the object
883 				 */
884 				offset = current->offset + (addr - current->start);
885 				pindex = OFF_TO_IDX(offset);
886 				m = vm_page_lookup(current->object.vm_object,
887 					pindex);
888 				/*
889 				 * if the page is resident, then gather information about
890 				 * it.
891 				 */
892 				if (m) {
893 					mincoreinfo = MINCORE_INCORE;
894 					if (m->dirty ||
895 						pmap_is_modified(m))
896 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
897 					if ((m->flags & PG_REFERENCED) ||
898 						pmap_ts_referenced(m)) {
899 						vm_page_flag_set(m, PG_REFERENCED);
900 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
901 					}
902 				}
903 			}
904 
905 			/*
906 			 * subyte may page fault.  In case it needs to modify
907 			 * the map, we release the lock.
908 			 */
909 			vm_map_unlock_read(map);
910 
911 			/*
912 			 * calculate index into user supplied byte vector
913 			 */
914 			vecindex = OFF_TO_IDX(addr - first_addr);
915 
916 			/*
917 			 * If we have skipped map entries, we need to make sure that
918 			 * the byte vector is zeroed for those skipped entries.
919 			 */
920 			while ((lastvecindex + 1) < vecindex) {
921 				error = subyte(vec + lastvecindex, 0);
922 				if (error) {
923 					error = EFAULT;
924 					goto done2;
925 				}
926 				++lastvecindex;
927 			}
928 
929 			/*
930 			 * Pass the page information to the user
931 			 */
932 			error = subyte(vec + vecindex, mincoreinfo);
933 			if (error) {
934 				error = EFAULT;
935 				goto done2;
936 			}
937 
938 			/*
939 			 * If the map has changed, due to the subyte, the previous
940 			 * output may be invalid.
941 			 */
942 			vm_map_lock_read(map);
943 			if (timestamp != map->timestamp)
944 				goto RestartScan;
945 
946 			lastvecindex = vecindex;
947 			addr += PAGE_SIZE;
948 		}
949 	}
950 
951 	/*
952 	 * subyte may page fault.  In case it needs to modify
953 	 * the map, we release the lock.
954 	 */
955 	vm_map_unlock_read(map);
956 
957 	/*
958 	 * Zero the last entries in the byte vector.
959 	 */
960 	vecindex = OFF_TO_IDX(end - first_addr);
961 	while ((lastvecindex + 1) < vecindex) {
962 		error = subyte(vec + lastvecindex, 0);
963 		if (error) {
964 			error = EFAULT;
965 			goto done2;
966 		}
967 		++lastvecindex;
968 	}
969 
970 	/*
971 	 * If the map has changed, due to the subyte, the previous
972 	 * output may be invalid.
973 	 */
974 	vm_map_lock_read(map);
975 	if (timestamp != map->timestamp)
976 		goto RestartScan;
977 	vm_map_unlock_read(map);
978 done2:
979 	mtx_unlock(&Giant);
980 	return (error);
981 }
982 
983 #ifndef _SYS_SYSPROTO_H_
984 struct mlock_args {
985 	const void *addr;
986 	size_t len;
987 };
988 #endif
989 /*
990  * MPSAFE
991  */
992 int
993 mlock(td, uap)
994 	struct thread *td;
995 	struct mlock_args *uap;
996 {
997 	vm_offset_t addr;
998 	vm_size_t size, pageoff;
999 	int error;
1000 
1001 	addr = (vm_offset_t) uap->addr;
1002 	size = uap->len;
1003 
1004 	pageoff = (addr & PAGE_MASK);
1005 	addr -= pageoff;
1006 	size += pageoff;
1007 	size = (vm_size_t) round_page(size);
1008 
1009 	/* disable wrap around */
1010 	if (addr + size < addr)
1011 		return (EINVAL);
1012 
1013 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
1014 		return (EAGAIN);
1015 
1016 #ifdef pmap_wired_count
1017 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) >
1018 	    td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
1019 		return (ENOMEM);
1020 #else
1021 	error = suser(td);
1022 	if (error)
1023 		return (error);
1024 #endif
1025 
1026 	mtx_lock(&Giant);
1027 	error = vm_map_user_pageable(&td->td_proc->p_vmspace->vm_map, addr,
1028 		     addr + size, FALSE);
1029 	mtx_unlock(&Giant);
1030 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1031 }
1032 
1033 #ifndef _SYS_SYSPROTO_H_
1034 struct mlockall_args {
1035 	int	how;
1036 };
1037 #endif
1038 
1039 /*
1040  * MPSAFE
1041  */
1042 int
1043 mlockall(td, uap)
1044 	struct thread *td;
1045 	struct mlockall_args *uap;
1046 {
1047 	/* mtx_lock(&Giant); */
1048 	/* mtx_unlock(&Giant); */
1049 	return 0;
1050 }
1051 
1052 #ifndef _SYS_SYSPROTO_H_
1053 struct mlockall_args {
1054 	int	how;
1055 };
1056 #endif
1057 
1058 /*
1059  * MPSAFE
1060  */
1061 int
1062 munlockall(td, uap)
1063 	struct thread *td;
1064 	struct munlockall_args *uap;
1065 {
1066 	/* mtx_lock(&Giant); */
1067 	/* mtx_unlock(&Giant); */
1068 	return 0;
1069 }
1070 
1071 #ifndef _SYS_SYSPROTO_H_
1072 struct munlock_args {
1073 	const void *addr;
1074 	size_t len;
1075 };
1076 #endif
1077 /*
1078  * MPSAFE
1079  */
1080 int
1081 munlock(td, uap)
1082 	struct thread *td;
1083 	struct munlock_args *uap;
1084 {
1085 	vm_offset_t addr;
1086 	vm_size_t size, pageoff;
1087 	int error;
1088 
1089 	addr = (vm_offset_t) uap->addr;
1090 	size = uap->len;
1091 
1092 	pageoff = (addr & PAGE_MASK);
1093 	addr -= pageoff;
1094 	size += pageoff;
1095 	size = (vm_size_t) round_page(size);
1096 
1097 	/* disable wrap around */
1098 	if (addr + size < addr)
1099 		return (EINVAL);
1100 
1101 #ifndef pmap_wired_count
1102 	error = suser(td);
1103 	if (error)
1104 		return (error);
1105 #endif
1106 
1107 	mtx_lock(&Giant);
1108 	error = vm_map_user_pageable(&td->td_proc->p_vmspace->vm_map, addr,
1109 		     addr + size, TRUE);
1110 	mtx_unlock(&Giant);
1111 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1112 }
1113 
1114 /*
1115  * vm_mmap()
1116  *
1117  * MPSAFE
1118  *
1119  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1120  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1121  */
1122 int
1123 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1124 	vm_prot_t maxprot, int flags,
1125 	void *handle,
1126 	vm_ooffset_t foff)
1127 {
1128 	boolean_t fitit;
1129 	vm_object_t object;
1130 	struct vnode *vp = NULL;
1131 	objtype_t type;
1132 	int rv = KERN_SUCCESS;
1133 	vm_ooffset_t objsize;
1134 	int docow;
1135 	struct thread *td = curthread;
1136 
1137 	if (size == 0)
1138 		return (0);
1139 
1140 	objsize = size = round_page(size);
1141 
1142 	/*
1143 	 * We currently can only deal with page aligned file offsets.
1144 	 * The check is here rather than in the syscall because the
1145 	 * kernel calls this function internally for other mmaping
1146 	 * operations (such as in exec) and non-aligned offsets will
1147 	 * cause pmap inconsistencies...so we want to be sure to
1148 	 * disallow this in all cases.
1149 	 */
1150 	if (foff & PAGE_MASK)
1151 		return (EINVAL);
1152 
1153 	if ((flags & MAP_FIXED) == 0) {
1154 		fitit = TRUE;
1155 		*addr = round_page(*addr);
1156 		mtx_lock(&Giant);
1157 	} else {
1158 		if (*addr != trunc_page(*addr))
1159 			return (EINVAL);
1160 		fitit = FALSE;
1161 		mtx_lock(&Giant);
1162 		(void) vm_map_remove(map, *addr, *addr + size);
1163 	}
1164 
1165 	/*
1166 	 * Lookup/allocate object.
1167 	 */
1168 	if (flags & MAP_ANON) {
1169 		type = OBJT_DEFAULT;
1170 		/*
1171 		 * Unnamed anonymous regions always start at 0.
1172 		 */
1173 		if (handle == 0)
1174 			foff = 0;
1175 	} else {
1176 		vp = (struct vnode *) handle;
1177 		if (vp->v_type == VCHR) {
1178 			type = OBJT_DEVICE;
1179 			handle = (void *)(intptr_t)vp->v_rdev;
1180 		} else {
1181 			struct vattr vat;
1182 			int error;
1183 
1184 			error = VOP_GETATTR(vp, &vat, td->td_ucred, td);
1185 			if (error) {
1186 				mtx_unlock(&Giant);
1187 				return (error);
1188 			}
1189 			objsize = round_page(vat.va_size);
1190 			type = OBJT_VNODE;
1191 			/*
1192 			 * if it is a regular file without any references
1193 			 * we do not need to sync it.
1194 			 */
1195 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1196 				flags |= MAP_NOSYNC;
1197 			}
1198 		}
1199 	}
1200 
1201 	if (handle == NULL) {
1202 		object = NULL;
1203 		docow = 0;
1204 	} else {
1205 		object = vm_pager_allocate(type,
1206 			handle, objsize, prot, foff);
1207 		if (object == NULL) {
1208 			mtx_unlock(&Giant);
1209 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1210 		}
1211 		docow = MAP_PREFAULT_PARTIAL;
1212 	}
1213 
1214 	/*
1215 	 * Force device mappings to be shared.
1216 	 */
1217 	if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1218 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1219 		flags |= MAP_SHARED;
1220 	}
1221 
1222 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1223 		docow |= MAP_COPY_ON_WRITE;
1224 	if (flags & MAP_NOSYNC)
1225 		docow |= MAP_DISABLE_SYNCER;
1226 	if (flags & MAP_NOCORE)
1227 		docow |= MAP_DISABLE_COREDUMP;
1228 
1229 #if defined(VM_PROT_READ_IS_EXEC)
1230 	if (prot & VM_PROT_READ)
1231 		prot |= VM_PROT_EXECUTE;
1232 
1233 	if (maxprot & VM_PROT_READ)
1234 		maxprot |= VM_PROT_EXECUTE;
1235 #endif
1236 
1237 	if (fitit)
1238 		*addr = pmap_addr_hint(object, *addr, size);
1239 
1240 	if (flags & MAP_STACK)
1241 		rv = vm_map_stack (map, *addr, size, prot,
1242 				   maxprot, docow);
1243 	else
1244 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1245 				 prot, maxprot, docow);
1246 
1247 	if (rv != KERN_SUCCESS) {
1248 		/*
1249 		 * Lose the object reference. Will destroy the
1250 		 * object if it's an unnamed anonymous mapping
1251 		 * or named anonymous without other references.
1252 		 */
1253 		vm_object_deallocate(object);
1254 	} else if (flags & MAP_SHARED) {
1255 		/*
1256 		 * Shared memory is also shared with children.
1257 		 */
1258 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1259 		if (rv != KERN_SUCCESS)
1260 			(void) vm_map_remove(map, *addr, *addr + size);
1261 	}
1262 	mtx_unlock(&Giant);
1263 	switch (rv) {
1264 	case KERN_SUCCESS:
1265 		return (0);
1266 	case KERN_INVALID_ADDRESS:
1267 	case KERN_NO_SPACE:
1268 		return (ENOMEM);
1269 	case KERN_PROTECTION_FAILURE:
1270 		return (EACCES);
1271 	default:
1272 		return (EINVAL);
1273 	}
1274 }
1275