xref: /freebsd/sys/vm/vm_mmap.c (revision 7660b554bc59a07be0431c17e0e33815818baa69)
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  */
42 
43 /*
44  * Mapped file (mmap) interface to VM
45  */
46 
47 #include <sys/cdefs.h>
48 __FBSDID("$FreeBSD$");
49 
50 #include "opt_compat.h"
51 #include "opt_mac.h"
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/sysproto.h>
59 #include <sys/filedesc.h>
60 #include <sys/proc.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/vnode.h>
64 #include <sys/fcntl.h>
65 #include <sys/file.h>
66 #include <sys/mac.h>
67 #include <sys/mman.h>
68 #include <sys/conf.h>
69 #include <sys/stat.h>
70 #include <sys/vmmeter.h>
71 #include <sys/sysctl.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vm_pageout.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_kern.h>
84 
85 #ifndef _SYS_SYSPROTO_H_
86 struct sbrk_args {
87 	int incr;
88 };
89 #endif
90 
91 static int max_proc_mmap;
92 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
93 
94 /*
95  * Set the maximum number of vm_map_entry structures per process.  Roughly
96  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
97  * of our KVM malloc space still results in generous limits.  We want a
98  * default that is good enough to prevent the kernel running out of resources
99  * if attacked from compromised user account but generous enough such that
100  * multi-threaded processes are not unduly inconvenienced.
101  */
102 static void vmmapentry_rsrc_init(void *);
103 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
104 
105 static void
106 vmmapentry_rsrc_init(dummy)
107         void *dummy;
108 {
109     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
110     max_proc_mmap /= 100;
111 }
112 
113 /*
114  * MPSAFE
115  */
116 /* ARGSUSED */
117 int
118 sbrk(td, uap)
119 	struct thread *td;
120 	struct sbrk_args *uap;
121 {
122 	/* Not yet implemented */
123 	/* mtx_lock(&Giant); */
124 	/* mtx_unlock(&Giant); */
125 	return (EOPNOTSUPP);
126 }
127 
128 #ifndef _SYS_SYSPROTO_H_
129 struct sstk_args {
130 	int incr;
131 };
132 #endif
133 
134 /*
135  * MPSAFE
136  */
137 /* ARGSUSED */
138 int
139 sstk(td, uap)
140 	struct thread *td;
141 	struct sstk_args *uap;
142 {
143 	/* Not yet implemented */
144 	/* mtx_lock(&Giant); */
145 	/* mtx_unlock(&Giant); */
146 	return (EOPNOTSUPP);
147 }
148 
149 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
150 #ifndef _SYS_SYSPROTO_H_
151 struct getpagesize_args {
152 	int dummy;
153 };
154 #endif
155 
156 /* ARGSUSED */
157 int
158 ogetpagesize(td, uap)
159 	struct thread *td;
160 	struct getpagesize_args *uap;
161 {
162 	/* MP SAFE */
163 	td->td_retval[0] = PAGE_SIZE;
164 	return (0);
165 }
166 #endif				/* COMPAT_43 || COMPAT_SUNOS */
167 
168 
169 /*
170  * Memory Map (mmap) system call.  Note that the file offset
171  * and address are allowed to be NOT page aligned, though if
172  * the MAP_FIXED flag it set, both must have the same remainder
173  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
174  * page-aligned, the actual mapping starts at trunc_page(addr)
175  * and the return value is adjusted up by the page offset.
176  *
177  * Generally speaking, only character devices which are themselves
178  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
179  * there would be no cache coherency between a descriptor and a VM mapping
180  * both to the same character device.
181  *
182  * Block devices can be mmap'd no matter what they represent.  Cache coherency
183  * is maintained as long as you do not write directly to the underlying
184  * character device.
185  */
186 #ifndef _SYS_SYSPROTO_H_
187 struct mmap_args {
188 	void *addr;
189 	size_t len;
190 	int prot;
191 	int flags;
192 	int fd;
193 	long pad;
194 	off_t pos;
195 };
196 #endif
197 
198 /*
199  * MPSAFE
200  */
201 int
202 mmap(td, uap)
203 	struct thread *td;
204 	struct mmap_args *uap;
205 {
206 	struct file *fp = NULL;
207 	struct vnode *vp;
208 	vm_offset_t addr;
209 	vm_size_t size, pageoff;
210 	vm_prot_t prot, maxprot;
211 	void *handle;
212 	int flags, error;
213 	int disablexworkaround;
214 	off_t pos;
215 	struct vmspace *vms = td->td_proc->p_vmspace;
216 	vm_object_t obj;
217 
218 	addr = (vm_offset_t) uap->addr;
219 	size = uap->len;
220 	prot = uap->prot & VM_PROT_ALL;
221 	flags = uap->flags;
222 	pos = uap->pos;
223 
224 	vp = NULL;
225 	fp = NULL;
226 	/* make sure mapping fits into numeric range etc */
227 	if ((ssize_t) uap->len < 0 ||
228 	    ((flags & MAP_ANON) && uap->fd != -1))
229 		return (EINVAL);
230 
231 	if (flags & MAP_STACK) {
232 		if ((uap->fd != -1) ||
233 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
234 			return (EINVAL);
235 		flags |= MAP_ANON;
236 		pos = 0;
237 	}
238 
239 	/*
240 	 * Align the file position to a page boundary,
241 	 * and save its page offset component.
242 	 */
243 	pageoff = (pos & PAGE_MASK);
244 	pos -= pageoff;
245 
246 	/* Adjust size for rounding (on both ends). */
247 	size += pageoff;			/* low end... */
248 	size = (vm_size_t) round_page(size);	/* hi end */
249 
250 	/*
251 	 * Check for illegal addresses.  Watch out for address wrap... Note
252 	 * that VM_*_ADDRESS are not constants due to casts (argh).
253 	 */
254 	if (flags & MAP_FIXED) {
255 		/*
256 		 * The specified address must have the same remainder
257 		 * as the file offset taken modulo PAGE_SIZE, so it
258 		 * should be aligned after adjustment by pageoff.
259 		 */
260 		addr -= pageoff;
261 		if (addr & PAGE_MASK)
262 			return (EINVAL);
263 		/* Address range must be all in user VM space. */
264 		if (addr < vm_map_min(&vms->vm_map) ||
265 		    addr + size > vm_map_max(&vms->vm_map))
266 			return (EINVAL);
267 		if (addr + size < addr)
268 			return (EINVAL);
269 	}
270 	/*
271 	 * XXX for non-fixed mappings where no hint is provided or
272 	 * the hint would fall in the potential heap space,
273 	 * place it after the end of the largest possible heap.
274 	 *
275 	 * There should really be a pmap call to determine a reasonable
276 	 * location.
277 	 */
278 	else if (addr == 0 ||
279 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
280 	     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
281 		addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
282 
283 	mtx_lock(&Giant);	/* syscall marked mp-safe but isn't */
284 	do {
285 		if (flags & MAP_ANON) {
286 			/*
287 			 * Mapping blank space is trivial.
288 			 */
289 			handle = NULL;
290 			maxprot = VM_PROT_ALL;
291 			pos = 0;
292 			break;
293 		}
294 		/*
295 		 * Mapping file, get fp for validation. Obtain vnode and make
296 		 * sure it is of appropriate type.
297 		 * don't let the descriptor disappear on us if we block
298 		 */
299 		if ((error = fget(td, uap->fd, &fp)) != 0)
300 			goto done;
301 		if (fp->f_type != DTYPE_VNODE) {
302 			error = EINVAL;
303 			goto done;
304 		}
305 
306 		/*
307 		 * POSIX shared-memory objects are defined to have
308 		 * kernel persistence, and are not defined to support
309 		 * read(2)/write(2) -- or even open(2).  Thus, we can
310 		 * use MAP_ASYNC to trade on-disk coherence for speed.
311 		 * The shm_open(3) library routine turns on the FPOSIXSHM
312 		 * flag to request this behavior.
313 		 */
314 		if (fp->f_flag & FPOSIXSHM)
315 			flags |= MAP_NOSYNC;
316 		vp = fp->f_vnode;
317 		error = vget(vp, LK_EXCLUSIVE, td);
318 		if (error)
319 			goto done;
320 		if (vp->v_type != VREG && vp->v_type != VCHR) {
321 			error = EINVAL;
322 			goto done;
323 		}
324 		if (vp->v_type == VREG) {
325 			/*
326 			 * Get the proper underlying object
327 			 */
328 			if (VOP_GETVOBJECT(vp, &obj) != 0) {
329 				error = EINVAL;
330 				goto done;
331 			}
332 			if (obj->handle != vp) {
333 				vput(vp);
334 				vp = (struct vnode*)obj->handle;
335 				vget(vp, LK_EXCLUSIVE, td);
336 			}
337 		}
338 		/*
339 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
340 		 * SunOS).
341 		 */
342 		if ((vp->v_type == VCHR) &&
343 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
344 			handle = NULL;
345 			maxprot = VM_PROT_ALL;
346 			flags |= MAP_ANON;
347 			pos = 0;
348 			break;
349 		}
350 		/*
351 		 * cdevs does not provide private mappings of any kind.
352 		 */
353 		/*
354 		 * However, for XIG X server to continue to work,
355 		 * we should allow the superuser to do it anyway.
356 		 * We only allow it at securelevel < 1.
357 		 * (Because the XIG X server writes directly to video
358 		 * memory via /dev/mem, it should never work at any
359 		 * other securelevel.
360 		 * XXX this will have to go
361 		 */
362 		if (securelevel_ge(td->td_ucred, 1))
363 			disablexworkaround = 1;
364 		else
365 			disablexworkaround = suser(td);
366 		if (vp->v_type == VCHR && disablexworkaround &&
367 		    (flags & (MAP_PRIVATE|MAP_COPY))) {
368 			error = EINVAL;
369 			goto done;
370 		}
371 		/*
372 		 * Ensure that file and memory protections are
373 		 * compatible.  Note that we only worry about
374 		 * writability if mapping is shared; in this case,
375 		 * current and max prot are dictated by the open file.
376 		 * XXX use the vnode instead?  Problem is: what
377 		 * credentials do we use for determination? What if
378 		 * proc does a setuid?
379 		 */
380 		maxprot = VM_PROT_EXECUTE;	/* ??? */
381 		if (fp->f_flag & FREAD) {
382 			maxprot |= VM_PROT_READ;
383 		} else if (prot & PROT_READ) {
384 			error = EACCES;
385 			goto done;
386 		}
387 		/*
388 		 * If we are sharing potential changes (either via
389 		 * MAP_SHARED or via the implicit sharing of character
390 		 * device mappings), and we are trying to get write
391 		 * permission although we opened it without asking
392 		 * for it, bail out.  Check for superuser, only if
393 		 * we're at securelevel < 1, to allow the XIG X server
394 		 * to continue to work.
395 		 */
396 		if ((flags & MAP_SHARED) != 0 ||
397 		    (vp->v_type == VCHR && disablexworkaround)) {
398 			if ((fp->f_flag & FWRITE) != 0) {
399 				struct vattr va;
400 				if ((error =
401 				    VOP_GETATTR(vp, &va,
402 						td->td_ucred, td))) {
403 					goto done;
404 				}
405 				if ((va.va_flags &
406 				   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
407 					maxprot |= VM_PROT_WRITE;
408 				} else if (prot & PROT_WRITE) {
409 					error = EPERM;
410 					goto done;
411 				}
412 			} else if ((prot & PROT_WRITE) != 0) {
413 				error = EACCES;
414 				goto done;
415 			}
416 		} else {
417 			maxprot |= VM_PROT_WRITE;
418 		}
419 
420 		handle = (void *)vp;
421 	} while (0);
422 
423 	/*
424 	 * Do not allow more then a certain number of vm_map_entry structures
425 	 * per process.  Scale with the number of rforks sharing the map
426 	 * to make the limit reasonable for threads.
427 	 */
428 	if (max_proc_mmap &&
429 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
430 		error = ENOMEM;
431 		goto done;
432 	}
433 
434 	mtx_unlock(&Giant);
435 	error = 0;
436 #ifdef MAC
437 	if (handle != NULL && (flags & MAP_SHARED) != 0) {
438 		error = mac_check_vnode_mmap(td->td_ucred,
439 		    (struct vnode *)handle, prot);
440 	}
441 #endif
442 	if (error == 0)
443 		error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
444 		    flags, handle, pos);
445 	mtx_lock(&Giant);
446 	if (error == 0)
447 		td->td_retval[0] = (register_t) (addr + pageoff);
448 done:
449 	if (vp)
450 		vput(vp);
451 	mtx_unlock(&Giant);
452 	if (fp)
453 		fdrop(fp, td);
454 
455 	return (error);
456 }
457 
458 #ifdef COMPAT_43
459 #ifndef _SYS_SYSPROTO_H_
460 struct ommap_args {
461 	caddr_t addr;
462 	int len;
463 	int prot;
464 	int flags;
465 	int fd;
466 	long pos;
467 };
468 #endif
469 int
470 ommap(td, uap)
471 	struct thread *td;
472 	struct ommap_args *uap;
473 {
474 	struct mmap_args nargs;
475 	static const char cvtbsdprot[8] = {
476 		0,
477 		PROT_EXEC,
478 		PROT_WRITE,
479 		PROT_EXEC | PROT_WRITE,
480 		PROT_READ,
481 		PROT_EXEC | PROT_READ,
482 		PROT_WRITE | PROT_READ,
483 		PROT_EXEC | PROT_WRITE | PROT_READ,
484 	};
485 
486 #define	OMAP_ANON	0x0002
487 #define	OMAP_COPY	0x0020
488 #define	OMAP_SHARED	0x0010
489 #define	OMAP_FIXED	0x0100
490 
491 	nargs.addr = uap->addr;
492 	nargs.len = uap->len;
493 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
494 	nargs.flags = 0;
495 	if (uap->flags & OMAP_ANON)
496 		nargs.flags |= MAP_ANON;
497 	if (uap->flags & OMAP_COPY)
498 		nargs.flags |= MAP_COPY;
499 	if (uap->flags & OMAP_SHARED)
500 		nargs.flags |= MAP_SHARED;
501 	else
502 		nargs.flags |= MAP_PRIVATE;
503 	if (uap->flags & OMAP_FIXED)
504 		nargs.flags |= MAP_FIXED;
505 	nargs.fd = uap->fd;
506 	nargs.pos = uap->pos;
507 	return (mmap(td, &nargs));
508 }
509 #endif				/* COMPAT_43 */
510 
511 
512 #ifndef _SYS_SYSPROTO_H_
513 struct msync_args {
514 	void *addr;
515 	int len;
516 	int flags;
517 };
518 #endif
519 /*
520  * MPSAFE
521  */
522 int
523 msync(td, uap)
524 	struct thread *td;
525 	struct msync_args *uap;
526 {
527 	vm_offset_t addr;
528 	vm_size_t size, pageoff;
529 	int flags;
530 	vm_map_t map;
531 	int rv;
532 
533 	addr = (vm_offset_t) uap->addr;
534 	size = uap->len;
535 	flags = uap->flags;
536 
537 	pageoff = (addr & PAGE_MASK);
538 	addr -= pageoff;
539 	size += pageoff;
540 	size = (vm_size_t) round_page(size);
541 	if (addr + size < addr)
542 		return (EINVAL);
543 
544 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
545 		return (EINVAL);
546 
547 	mtx_lock(&Giant);
548 
549 	map = &td->td_proc->p_vmspace->vm_map;
550 
551 	/*
552 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
553 	 * pages with the region containing addr".  Unfortunately, we don't
554 	 * really keep track of individual mmaps so we approximate by flushing
555 	 * the range of the map entry containing addr. This can be incorrect
556 	 * if the region splits or is coalesced with a neighbor.
557 	 */
558 	if (size == 0) {
559 		vm_map_entry_t entry;
560 
561 		vm_map_lock_read(map);
562 		rv = vm_map_lookup_entry(map, addr, &entry);
563 		vm_map_unlock_read(map);
564 		if (rv == FALSE) {
565 			rv = -1;
566 			goto done2;
567 		}
568 		addr = entry->start;
569 		size = entry->end - entry->start;
570 	}
571 
572 	/*
573 	 * Clean the pages and interpret the return value.
574 	 */
575 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
576 	    (flags & MS_INVALIDATE) != 0);
577 
578 done2:
579 	mtx_unlock(&Giant);
580 
581 	switch (rv) {
582 	case KERN_SUCCESS:
583 		return (0);
584 	case KERN_INVALID_ADDRESS:
585 		return (EINVAL);	/* Sun returns ENOMEM? */
586 	case KERN_FAILURE:
587 		return (EIO);
588 	default:
589 		return (EINVAL);
590 	}
591 }
592 
593 #ifndef _SYS_SYSPROTO_H_
594 struct munmap_args {
595 	void *addr;
596 	size_t len;
597 };
598 #endif
599 /*
600  * MPSAFE
601  */
602 int
603 munmap(td, uap)
604 	struct thread *td;
605 	struct munmap_args *uap;
606 {
607 	vm_offset_t addr;
608 	vm_size_t size, pageoff;
609 	vm_map_t map;
610 
611 	addr = (vm_offset_t) uap->addr;
612 	size = uap->len;
613 
614 	pageoff = (addr & PAGE_MASK);
615 	addr -= pageoff;
616 	size += pageoff;
617 	size = (vm_size_t) round_page(size);
618 	if (addr + size < addr)
619 		return (EINVAL);
620 
621 	if (size == 0)
622 		return (0);
623 
624 	/*
625 	 * Check for illegal addresses.  Watch out for address wrap...
626 	 */
627 	map = &td->td_proc->p_vmspace->vm_map;
628 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
629 		return (EINVAL);
630 	/*
631 	 * Make sure entire range is allocated.
632 	 */
633 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
634 		return (EINVAL);
635 
636 	/* returns nothing but KERN_SUCCESS anyway */
637 	(void) vm_map_remove(map, addr, addr + size);
638 	return (0);
639 }
640 
641 #if 0
642 void
643 munmapfd(td, fd)
644 	struct thread *td;
645 	int fd;
646 {
647 	/*
648 	 * XXX should unmap any regions mapped to this file
649 	 */
650 	FILEDESC_LOCK(p->p_fd);
651 	td->td_proc->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
652 	FILEDESC_UNLOCK(p->p_fd);
653 }
654 #endif
655 
656 #ifndef _SYS_SYSPROTO_H_
657 struct mprotect_args {
658 	const void *addr;
659 	size_t len;
660 	int prot;
661 };
662 #endif
663 /*
664  * MPSAFE
665  */
666 int
667 mprotect(td, uap)
668 	struct thread *td;
669 	struct mprotect_args *uap;
670 {
671 	vm_offset_t addr;
672 	vm_size_t size, pageoff;
673 	vm_prot_t prot;
674 
675 	addr = (vm_offset_t) uap->addr;
676 	size = uap->len;
677 	prot = uap->prot & VM_PROT_ALL;
678 #if defined(VM_PROT_READ_IS_EXEC)
679 	if (prot & VM_PROT_READ)
680 		prot |= VM_PROT_EXECUTE;
681 #endif
682 
683 	pageoff = (addr & PAGE_MASK);
684 	addr -= pageoff;
685 	size += pageoff;
686 	size = (vm_size_t) round_page(size);
687 	if (addr + size < addr)
688 		return (EINVAL);
689 
690 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
691 	    addr + size, prot, FALSE)) {
692 	case KERN_SUCCESS:
693 		return (0);
694 	case KERN_PROTECTION_FAILURE:
695 		return (EACCES);
696 	}
697 	return (EINVAL);
698 }
699 
700 #ifndef _SYS_SYSPROTO_H_
701 struct minherit_args {
702 	void *addr;
703 	size_t len;
704 	int inherit;
705 };
706 #endif
707 /*
708  * MPSAFE
709  */
710 int
711 minherit(td, uap)
712 	struct thread *td;
713 	struct minherit_args *uap;
714 {
715 	vm_offset_t addr;
716 	vm_size_t size, pageoff;
717 	vm_inherit_t inherit;
718 
719 	addr = (vm_offset_t)uap->addr;
720 	size = uap->len;
721 	inherit = uap->inherit;
722 
723 	pageoff = (addr & PAGE_MASK);
724 	addr -= pageoff;
725 	size += pageoff;
726 	size = (vm_size_t) round_page(size);
727 	if (addr + size < addr)
728 		return (EINVAL);
729 
730 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
731 	    addr + size, inherit)) {
732 	case KERN_SUCCESS:
733 		return (0);
734 	case KERN_PROTECTION_FAILURE:
735 		return (EACCES);
736 	}
737 	return (EINVAL);
738 }
739 
740 #ifndef _SYS_SYSPROTO_H_
741 struct madvise_args {
742 	void *addr;
743 	size_t len;
744 	int behav;
745 };
746 #endif
747 
748 /*
749  * MPSAFE
750  */
751 /* ARGSUSED */
752 int
753 madvise(td, uap)
754 	struct thread *td;
755 	struct madvise_args *uap;
756 {
757 	vm_offset_t start, end;
758 	vm_map_t map;
759 	struct proc *p;
760 	int error;
761 
762 	/*
763 	 * Check for our special case, advising the swap pager we are
764 	 * "immortal."
765 	 */
766 	if (uap->behav == MADV_PROTECT) {
767 		error = suser(td);
768 		if (error == 0) {
769 			p = td->td_proc;
770 			PROC_LOCK(p);
771 			p->p_flag |= P_PROTECTED;
772 			PROC_UNLOCK(p);
773 		}
774 		return (error);
775 	}
776 	/*
777 	 * Check for illegal behavior
778 	 */
779 	if (uap->behav < 0 || uap->behav > MADV_CORE)
780 		return (EINVAL);
781 	/*
782 	 * Check for illegal addresses.  Watch out for address wrap... Note
783 	 * that VM_*_ADDRESS are not constants due to casts (argh).
784 	 */
785 	map = &td->td_proc->p_vmspace->vm_map;
786 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
787 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
788 		return (EINVAL);
789 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
790 		return (EINVAL);
791 
792 	/*
793 	 * Since this routine is only advisory, we default to conservative
794 	 * behavior.
795 	 */
796 	start = trunc_page((vm_offset_t) uap->addr);
797 	end = round_page((vm_offset_t) uap->addr + uap->len);
798 
799 	if (vm_map_madvise(map, start, end, uap->behav))
800 		return (EINVAL);
801 	return (0);
802 }
803 
804 #ifndef _SYS_SYSPROTO_H_
805 struct mincore_args {
806 	const void *addr;
807 	size_t len;
808 	char *vec;
809 };
810 #endif
811 
812 /*
813  * MPSAFE
814  */
815 /* ARGSUSED */
816 int
817 mincore(td, uap)
818 	struct thread *td;
819 	struct mincore_args *uap;
820 {
821 	vm_offset_t addr, first_addr;
822 	vm_offset_t end, cend;
823 	pmap_t pmap;
824 	vm_map_t map;
825 	char *vec;
826 	int error = 0;
827 	int vecindex, lastvecindex;
828 	vm_map_entry_t current;
829 	vm_map_entry_t entry;
830 	int mincoreinfo;
831 	unsigned int timestamp;
832 
833 	/*
834 	 * Make sure that the addresses presented are valid for user
835 	 * mode.
836 	 */
837 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
838 	end = addr + (vm_size_t)round_page(uap->len);
839 	map = &td->td_proc->p_vmspace->vm_map;
840 	if (end > vm_map_max(map) || end < addr)
841 		return (EINVAL);
842 
843 	/*
844 	 * Address of byte vector
845 	 */
846 	vec = uap->vec;
847 
848 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
849 
850 	vm_map_lock_read(map);
851 RestartScan:
852 	timestamp = map->timestamp;
853 
854 	if (!vm_map_lookup_entry(map, addr, &entry))
855 		entry = entry->next;
856 
857 	/*
858 	 * Do this on a map entry basis so that if the pages are not
859 	 * in the current processes address space, we can easily look
860 	 * up the pages elsewhere.
861 	 */
862 	lastvecindex = -1;
863 	for (current = entry;
864 	    (current != &map->header) && (current->start < end);
865 	    current = current->next) {
866 
867 		/*
868 		 * ignore submaps (for now) or null objects
869 		 */
870 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
871 			current->object.vm_object == NULL)
872 			continue;
873 
874 		/*
875 		 * limit this scan to the current map entry and the
876 		 * limits for the mincore call
877 		 */
878 		if (addr < current->start)
879 			addr = current->start;
880 		cend = current->end;
881 		if (cend > end)
882 			cend = end;
883 
884 		/*
885 		 * scan this entry one page at a time
886 		 */
887 		while (addr < cend) {
888 			/*
889 			 * Check pmap first, it is likely faster, also
890 			 * it can provide info as to whether we are the
891 			 * one referencing or modifying the page.
892 			 */
893 			mtx_lock(&Giant);
894 			mincoreinfo = pmap_mincore(pmap, addr);
895 			mtx_unlock(&Giant);
896 			if (!mincoreinfo) {
897 				vm_pindex_t pindex;
898 				vm_ooffset_t offset;
899 				vm_page_t m;
900 				/*
901 				 * calculate the page index into the object
902 				 */
903 				offset = current->offset + (addr - current->start);
904 				pindex = OFF_TO_IDX(offset);
905 				VM_OBJECT_LOCK(current->object.vm_object);
906 				m = vm_page_lookup(current->object.vm_object,
907 					pindex);
908 				/*
909 				 * if the page is resident, then gather information about
910 				 * it.
911 				 */
912 				if (m) {
913 					mincoreinfo = MINCORE_INCORE;
914 					vm_page_lock_queues();
915 					if (m->dirty ||
916 						pmap_is_modified(m))
917 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
918 					if ((m->flags & PG_REFERENCED) ||
919 						pmap_ts_referenced(m)) {
920 						vm_page_flag_set(m, PG_REFERENCED);
921 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
922 					}
923 					vm_page_unlock_queues();
924 				}
925 				VM_OBJECT_UNLOCK(current->object.vm_object);
926 			}
927 
928 			/*
929 			 * subyte may page fault.  In case it needs to modify
930 			 * the map, we release the lock.
931 			 */
932 			vm_map_unlock_read(map);
933 
934 			/*
935 			 * calculate index into user supplied byte vector
936 			 */
937 			vecindex = OFF_TO_IDX(addr - first_addr);
938 
939 			/*
940 			 * If we have skipped map entries, we need to make sure that
941 			 * the byte vector is zeroed for those skipped entries.
942 			 */
943 			while ((lastvecindex + 1) < vecindex) {
944 				error = subyte(vec + lastvecindex, 0);
945 				if (error) {
946 					error = EFAULT;
947 					goto done2;
948 				}
949 				++lastvecindex;
950 			}
951 
952 			/*
953 			 * Pass the page information to the user
954 			 */
955 			error = subyte(vec + vecindex, mincoreinfo);
956 			if (error) {
957 				error = EFAULT;
958 				goto done2;
959 			}
960 
961 			/*
962 			 * If the map has changed, due to the subyte, the previous
963 			 * output may be invalid.
964 			 */
965 			vm_map_lock_read(map);
966 			if (timestamp != map->timestamp)
967 				goto RestartScan;
968 
969 			lastvecindex = vecindex;
970 			addr += PAGE_SIZE;
971 		}
972 	}
973 
974 	/*
975 	 * subyte may page fault.  In case it needs to modify
976 	 * the map, we release the lock.
977 	 */
978 	vm_map_unlock_read(map);
979 
980 	/*
981 	 * Zero the last entries in the byte vector.
982 	 */
983 	vecindex = OFF_TO_IDX(end - first_addr);
984 	while ((lastvecindex + 1) < vecindex) {
985 		error = subyte(vec + lastvecindex, 0);
986 		if (error) {
987 			error = EFAULT;
988 			goto done2;
989 		}
990 		++lastvecindex;
991 	}
992 
993 	/*
994 	 * If the map has changed, due to the subyte, the previous
995 	 * output may be invalid.
996 	 */
997 	vm_map_lock_read(map);
998 	if (timestamp != map->timestamp)
999 		goto RestartScan;
1000 	vm_map_unlock_read(map);
1001 done2:
1002 	return (error);
1003 }
1004 
1005 #ifndef _SYS_SYSPROTO_H_
1006 struct mlock_args {
1007 	const void *addr;
1008 	size_t len;
1009 };
1010 #endif
1011 /*
1012  * MPSAFE
1013  */
1014 int
1015 mlock(td, uap)
1016 	struct thread *td;
1017 	struct mlock_args *uap;
1018 {
1019 	vm_offset_t addr;
1020 	vm_size_t size, pageoff;
1021 	int error;
1022 
1023 	addr = (vm_offset_t) uap->addr;
1024 	size = uap->len;
1025 
1026 	pageoff = (addr & PAGE_MASK);
1027 	addr -= pageoff;
1028 	size += pageoff;
1029 	size = (vm_size_t) round_page(size);
1030 
1031 	/* disable wrap around */
1032 	if (addr + size < addr)
1033 		return (EINVAL);
1034 
1035 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
1036 		return (EAGAIN);
1037 
1038 #ifdef pmap_wired_count
1039 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) >
1040 	    td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
1041 		return (ENOMEM);
1042 #else
1043 	error = suser(td);
1044 	if (error)
1045 		return (error);
1046 #endif
1047 
1048 	error = vm_map_wire(&td->td_proc->p_vmspace->vm_map, addr,
1049 		     addr + size, VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1050 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1051 }
1052 
1053 #ifndef _SYS_SYSPROTO_H_
1054 struct mlockall_args {
1055 	int	how;
1056 };
1057 #endif
1058 
1059 /*
1060  * MPSAFE
1061  */
1062 int
1063 mlockall(td, uap)
1064 	struct thread *td;
1065 	struct mlockall_args *uap;
1066 {
1067 	vm_map_t map;
1068 	int error;
1069 
1070 	map = &td->td_proc->p_vmspace->vm_map;
1071 	error = 0;
1072 
1073 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1074 		return (EINVAL);
1075 
1076 #ifdef pmap_wired_count
1077 	/*
1078 	 * If wiring all pages in the process would cause it to exceed
1079 	 * a hard resource limit, return ENOMEM.
1080 	 */
1081 	if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1082 		td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur))
1083 		return (ENOMEM);
1084 #else
1085 	error = suser(td);
1086 	if (error)
1087 		return (error);
1088 #endif
1089 
1090 	if (uap->how & MCL_FUTURE) {
1091 		vm_map_lock(map);
1092 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1093 		vm_map_unlock(map);
1094 		error = 0;
1095 	}
1096 
1097 	if (uap->how & MCL_CURRENT) {
1098 		/*
1099 		 * P1003.1-2001 mandates that all currently mapped pages
1100 		 * will be memory resident and locked (wired) upon return
1101 		 * from mlockall(). vm_map_wire() will wire pages, by
1102 		 * calling vm_fault_wire() for each page in the region.
1103 		 */
1104 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1105 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1106 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1107 	}
1108 
1109 	return (error);
1110 }
1111 
1112 #ifndef _SYS_SYSPROTO_H_
1113 struct munlockall_args {
1114 	register_t dummy;
1115 };
1116 #endif
1117 
1118 /*
1119  * MPSAFE
1120  */
1121 int
1122 munlockall(td, uap)
1123 	struct thread *td;
1124 	struct munlockall_args *uap;
1125 {
1126 	vm_map_t map;
1127 	int error;
1128 
1129 	map = &td->td_proc->p_vmspace->vm_map;
1130 #ifndef pmap_wired_count
1131 	error = suser(td);
1132 	if (error)
1133 		return (error);
1134 #endif
1135 
1136 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1137 	vm_map_lock(map);
1138 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1139 	vm_map_unlock(map);
1140 
1141 	/* Forcibly unwire all pages. */
1142 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1143 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1144 
1145 	return (error);
1146 }
1147 
1148 #ifndef _SYS_SYSPROTO_H_
1149 struct munlock_args {
1150 	const void *addr;
1151 	size_t len;
1152 };
1153 #endif
1154 /*
1155  * MPSAFE
1156  */
1157 int
1158 munlock(td, uap)
1159 	struct thread *td;
1160 	struct munlock_args *uap;
1161 {
1162 	vm_offset_t addr;
1163 	vm_size_t size, pageoff;
1164 	int error;
1165 
1166 	addr = (vm_offset_t) uap->addr;
1167 	size = uap->len;
1168 
1169 	pageoff = (addr & PAGE_MASK);
1170 	addr -= pageoff;
1171 	size += pageoff;
1172 	size = (vm_size_t) round_page(size);
1173 
1174 	/* disable wrap around */
1175 	if (addr + size < addr)
1176 		return (EINVAL);
1177 
1178 #ifndef pmap_wired_count
1179 	error = suser(td);
1180 	if (error)
1181 		return (error);
1182 #endif
1183 
1184 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, addr,
1185 		     addr + size, VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1186 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1187 }
1188 
1189 /*
1190  * vm_mmap()
1191  *
1192  * MPSAFE
1193  *
1194  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1195  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1196  */
1197 int
1198 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1199 	vm_prot_t maxprot, int flags,
1200 	void *handle,
1201 	vm_ooffset_t foff)
1202 {
1203 	boolean_t fitit;
1204 	vm_object_t object;
1205 	struct vnode *vp = NULL;
1206 	objtype_t type;
1207 	int rv = KERN_SUCCESS;
1208 	vm_ooffset_t objsize;
1209 	int docow;
1210 	struct thread *td = curthread;
1211 
1212 	if (size == 0)
1213 		return (0);
1214 
1215 	objsize = size = round_page(size);
1216 
1217 	if (td->td_proc->p_vmspace->vm_map.size + size >
1218 	    td->td_proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1219 		return(ENOMEM);
1220 	}
1221 
1222 	/*
1223 	 * We currently can only deal with page aligned file offsets.
1224 	 * The check is here rather than in the syscall because the
1225 	 * kernel calls this function internally for other mmaping
1226 	 * operations (such as in exec) and non-aligned offsets will
1227 	 * cause pmap inconsistencies...so we want to be sure to
1228 	 * disallow this in all cases.
1229 	 */
1230 	if (foff & PAGE_MASK)
1231 		return (EINVAL);
1232 
1233 	if ((flags & MAP_FIXED) == 0) {
1234 		fitit = TRUE;
1235 		*addr = round_page(*addr);
1236 	} else {
1237 		if (*addr != trunc_page(*addr))
1238 			return (EINVAL);
1239 		fitit = FALSE;
1240 		(void) vm_map_remove(map, *addr, *addr + size);
1241 	}
1242 
1243 	/*
1244 	 * Lookup/allocate object.
1245 	 */
1246 	if (flags & MAP_ANON) {
1247 		type = OBJT_DEFAULT;
1248 		/*
1249 		 * Unnamed anonymous regions always start at 0.
1250 		 */
1251 		if (handle == 0)
1252 			foff = 0;
1253 	} else {
1254 		vp = (struct vnode *) handle;
1255 		mtx_lock(&Giant);
1256 		ASSERT_VOP_LOCKED(vp, "vm_mmap");
1257 		if (vp->v_type == VCHR) {
1258 			type = OBJT_DEVICE;
1259 			handle = vp->v_rdev;
1260 		} else {
1261 			struct vattr vat;
1262 			int error;
1263 
1264 			error = VOP_GETATTR(vp, &vat, td->td_ucred, td);
1265 			if (error) {
1266 				mtx_unlock(&Giant);
1267 				return (error);
1268 			}
1269 			objsize = round_page(vat.va_size);
1270 			type = OBJT_VNODE;
1271 			/*
1272 			 * if it is a regular file without any references
1273 			 * we do not need to sync it.
1274 			 */
1275 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1276 				flags |= MAP_NOSYNC;
1277 			}
1278 		}
1279 		mtx_unlock(&Giant);
1280 	}
1281 
1282 	if (handle == NULL) {
1283 		object = NULL;
1284 		docow = 0;
1285 	} else {
1286 		object = vm_pager_allocate(type,
1287 			handle, objsize, prot, foff);
1288 		if (object == NULL) {
1289 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1290 		}
1291 		docow = MAP_PREFAULT_PARTIAL;
1292 	}
1293 
1294 	/*
1295 	 * Force device mappings to be shared.
1296 	 */
1297 	if (type == OBJT_DEVICE) {
1298 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1299 		flags |= MAP_SHARED;
1300 	}
1301 
1302 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1303 		docow |= MAP_COPY_ON_WRITE;
1304 	if (flags & MAP_NOSYNC)
1305 		docow |= MAP_DISABLE_SYNCER;
1306 	if (flags & MAP_NOCORE)
1307 		docow |= MAP_DISABLE_COREDUMP;
1308 
1309 #if defined(VM_PROT_READ_IS_EXEC)
1310 	if (prot & VM_PROT_READ)
1311 		prot |= VM_PROT_EXECUTE;
1312 
1313 	if (maxprot & VM_PROT_READ)
1314 		maxprot |= VM_PROT_EXECUTE;
1315 #endif
1316 
1317 	if (fitit)
1318 		*addr = pmap_addr_hint(object, *addr, size);
1319 
1320 	if (flags & MAP_STACK)
1321 		rv = vm_map_stack (map, *addr, size, prot,
1322 				   maxprot, docow);
1323 	else
1324 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1325 				 prot, maxprot, docow);
1326 
1327 	if (rv != KERN_SUCCESS) {
1328 		/*
1329 		 * Lose the object reference. Will destroy the
1330 		 * object if it's an unnamed anonymous mapping
1331 		 * or named anonymous without other references.
1332 		 */
1333 		vm_object_deallocate(object);
1334 	} else if (flags & MAP_SHARED) {
1335 		/*
1336 		 * Shared memory is also shared with children.
1337 		 */
1338 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1339 		if (rv != KERN_SUCCESS)
1340 			(void) vm_map_remove(map, *addr, *addr + size);
1341 	}
1342 
1343 	/*
1344 	 * If the process has requested that all future mappings
1345 	 * be wired, then heed this.
1346 	 */
1347 	if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1348 		vm_map_wire(map, *addr, *addr + size,
1349 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1350 
1351 	switch (rv) {
1352 	case KERN_SUCCESS:
1353 		return (0);
1354 	case KERN_INVALID_ADDRESS:
1355 	case KERN_NO_SPACE:
1356 		return (ENOMEM);
1357 	case KERN_PROTECTION_FAILURE:
1358 		return (EACCES);
1359 	default:
1360 		return (EINVAL);
1361 	}
1362 }
1363