xref: /freebsd/sys/vm/vm_mmap.c (revision 7773002178c8dbc52b44e4d705f07706409af8e4)
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 +
281 	      td->td_proc->p_rlimit[RLIMIT_DATA].rlim_max)))
282 		addr = round_page((vm_offset_t)vms->vm_daddr +
283 		    td->td_proc->p_rlimit[RLIMIT_DATA].rlim_max);
284 
285 	mtx_lock(&Giant);	/* syscall marked mp-safe but isn't */
286 	do {
287 		if (flags & MAP_ANON) {
288 			/*
289 			 * Mapping blank space is trivial.
290 			 */
291 			handle = NULL;
292 			maxprot = VM_PROT_ALL;
293 			pos = 0;
294 			break;
295 		}
296 		/*
297 		 * Mapping file, get fp for validation. Obtain vnode and make
298 		 * sure it is of appropriate type.
299 		 * don't let the descriptor disappear on us if we block
300 		 */
301 		if ((error = fget(td, uap->fd, &fp)) != 0)
302 			goto done;
303 		if (fp->f_type != DTYPE_VNODE) {
304 			error = EINVAL;
305 			goto done;
306 		}
307 
308 		/*
309 		 * POSIX shared-memory objects are defined to have
310 		 * kernel persistence, and are not defined to support
311 		 * read(2)/write(2) -- or even open(2).  Thus, we can
312 		 * use MAP_ASYNC to trade on-disk coherence for speed.
313 		 * The shm_open(3) library routine turns on the FPOSIXSHM
314 		 * flag to request this behavior.
315 		 */
316 		if (fp->f_flag & FPOSIXSHM)
317 			flags |= MAP_NOSYNC;
318 		vp = fp->f_vnode;
319 		error = vget(vp, LK_EXCLUSIVE, td);
320 		if (error)
321 			goto done;
322 		if (vp->v_type != VREG && vp->v_type != VCHR) {
323 			error = EINVAL;
324 			goto done;
325 		}
326 		if (vp->v_type == VREG) {
327 			/*
328 			 * Get the proper underlying object
329 			 */
330 			if (VOP_GETVOBJECT(vp, &obj) != 0) {
331 				error = EINVAL;
332 				goto done;
333 			}
334 			if (obj->handle != vp) {
335 				vput(vp);
336 				vp = (struct vnode*)obj->handle;
337 				vget(vp, LK_EXCLUSIVE, td);
338 			}
339 		}
340 		/*
341 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
342 		 * SunOS).
343 		 */
344 		if ((vp->v_type == VCHR) &&
345 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
346 			handle = NULL;
347 			maxprot = VM_PROT_ALL;
348 			flags |= MAP_ANON;
349 			pos = 0;
350 			break;
351 		}
352 		/*
353 		 * cdevs does not provide private mappings of any kind.
354 		 */
355 		/*
356 		 * However, for XIG X server to continue to work,
357 		 * we should allow the superuser to do it anyway.
358 		 * We only allow it at securelevel < 1.
359 		 * (Because the XIG X server writes directly to video
360 		 * memory via /dev/mem, it should never work at any
361 		 * other securelevel.
362 		 * XXX this will have to go
363 		 */
364 		if (securelevel_ge(td->td_ucred, 1))
365 			disablexworkaround = 1;
366 		else
367 			disablexworkaround = suser(td);
368 		if (vp->v_type == VCHR && disablexworkaround &&
369 		    (flags & (MAP_PRIVATE|MAP_COPY))) {
370 			error = EINVAL;
371 			goto done;
372 		}
373 		/*
374 		 * Ensure that file and memory protections are
375 		 * compatible.  Note that we only worry about
376 		 * writability if mapping is shared; in this case,
377 		 * current and max prot are dictated by the open file.
378 		 * XXX use the vnode instead?  Problem is: what
379 		 * credentials do we use for determination? What if
380 		 * proc does a setuid?
381 		 */
382 		maxprot = VM_PROT_EXECUTE;	/* ??? */
383 		if (fp->f_flag & FREAD) {
384 			maxprot |= VM_PROT_READ;
385 		} else if (prot & PROT_READ) {
386 			error = EACCES;
387 			goto done;
388 		}
389 		/*
390 		 * If we are sharing potential changes (either via
391 		 * MAP_SHARED or via the implicit sharing of character
392 		 * device mappings), and we are trying to get write
393 		 * permission although we opened it without asking
394 		 * for it, bail out.  Check for superuser, only if
395 		 * we're at securelevel < 1, to allow the XIG X server
396 		 * to continue to work.
397 		 */
398 		if ((flags & MAP_SHARED) != 0 ||
399 		    (vp->v_type == VCHR && disablexworkaround)) {
400 			if ((fp->f_flag & FWRITE) != 0) {
401 				struct vattr va;
402 				if ((error =
403 				    VOP_GETATTR(vp, &va,
404 						td->td_ucred, td))) {
405 					goto done;
406 				}
407 				if ((va.va_flags &
408 				   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
409 					maxprot |= VM_PROT_WRITE;
410 				} else if (prot & PROT_WRITE) {
411 					error = EPERM;
412 					goto done;
413 				}
414 			} else if ((prot & PROT_WRITE) != 0) {
415 				error = EACCES;
416 				goto done;
417 			}
418 		} else {
419 			maxprot |= VM_PROT_WRITE;
420 		}
421 
422 		handle = (void *)vp;
423 	} while (0);
424 
425 	/*
426 	 * Do not allow more then a certain number of vm_map_entry structures
427 	 * per process.  Scale with the number of rforks sharing the map
428 	 * to make the limit reasonable for threads.
429 	 */
430 	if (max_proc_mmap &&
431 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
432 		error = ENOMEM;
433 		goto done;
434 	}
435 
436 	mtx_unlock(&Giant);
437 	error = 0;
438 #ifdef MAC
439 	if (handle != NULL && (flags & MAP_SHARED) != 0) {
440 		error = mac_check_vnode_mmap(td->td_ucred,
441 		    (struct vnode *)handle, prot);
442 	}
443 #endif
444 	if (error == 0)
445 		error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
446 		    flags, handle, pos);
447 	mtx_lock(&Giant);
448 	if (error == 0)
449 		td->td_retval[0] = (register_t) (addr + pageoff);
450 done:
451 	if (vp)
452 		vput(vp);
453 	mtx_unlock(&Giant);
454 	if (fp)
455 		fdrop(fp, td);
456 
457 	return (error);
458 }
459 
460 #ifdef COMPAT_43
461 #ifndef _SYS_SYSPROTO_H_
462 struct ommap_args {
463 	caddr_t addr;
464 	int len;
465 	int prot;
466 	int flags;
467 	int fd;
468 	long pos;
469 };
470 #endif
471 int
472 ommap(td, uap)
473 	struct thread *td;
474 	struct ommap_args *uap;
475 {
476 	struct mmap_args nargs;
477 	static const char cvtbsdprot[8] = {
478 		0,
479 		PROT_EXEC,
480 		PROT_WRITE,
481 		PROT_EXEC | PROT_WRITE,
482 		PROT_READ,
483 		PROT_EXEC | PROT_READ,
484 		PROT_WRITE | PROT_READ,
485 		PROT_EXEC | PROT_WRITE | PROT_READ,
486 	};
487 
488 #define	OMAP_ANON	0x0002
489 #define	OMAP_COPY	0x0020
490 #define	OMAP_SHARED	0x0010
491 #define	OMAP_FIXED	0x0100
492 
493 	nargs.addr = uap->addr;
494 	nargs.len = uap->len;
495 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
496 	nargs.flags = 0;
497 	if (uap->flags & OMAP_ANON)
498 		nargs.flags |= MAP_ANON;
499 	if (uap->flags & OMAP_COPY)
500 		nargs.flags |= MAP_COPY;
501 	if (uap->flags & OMAP_SHARED)
502 		nargs.flags |= MAP_SHARED;
503 	else
504 		nargs.flags |= MAP_PRIVATE;
505 	if (uap->flags & OMAP_FIXED)
506 		nargs.flags |= MAP_FIXED;
507 	nargs.fd = uap->fd;
508 	nargs.pos = uap->pos;
509 	return (mmap(td, &nargs));
510 }
511 #endif				/* COMPAT_43 */
512 
513 
514 #ifndef _SYS_SYSPROTO_H_
515 struct msync_args {
516 	void *addr;
517 	int len;
518 	int flags;
519 };
520 #endif
521 /*
522  * MPSAFE
523  */
524 int
525 msync(td, uap)
526 	struct thread *td;
527 	struct msync_args *uap;
528 {
529 	vm_offset_t addr;
530 	vm_size_t size, pageoff;
531 	int flags;
532 	vm_map_t map;
533 	int rv;
534 
535 	addr = (vm_offset_t) uap->addr;
536 	size = uap->len;
537 	flags = uap->flags;
538 
539 	pageoff = (addr & PAGE_MASK);
540 	addr -= pageoff;
541 	size += pageoff;
542 	size = (vm_size_t) round_page(size);
543 	if (addr + size < addr)
544 		return (EINVAL);
545 
546 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
547 		return (EINVAL);
548 
549 	mtx_lock(&Giant);
550 
551 	map = &td->td_proc->p_vmspace->vm_map;
552 
553 	/*
554 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
555 	 * pages with the region containing addr".  Unfortunately, we don't
556 	 * really keep track of individual mmaps so we approximate by flushing
557 	 * the range of the map entry containing addr. This can be incorrect
558 	 * if the region splits or is coalesced with a neighbor.
559 	 */
560 	if (size == 0) {
561 		vm_map_entry_t entry;
562 
563 		vm_map_lock_read(map);
564 		rv = vm_map_lookup_entry(map, addr, &entry);
565 		vm_map_unlock_read(map);
566 		if (rv == FALSE) {
567 			rv = -1;
568 			goto done2;
569 		}
570 		addr = entry->start;
571 		size = entry->end - entry->start;
572 	}
573 
574 	/*
575 	 * Clean the pages and interpret the return value.
576 	 */
577 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
578 	    (flags & MS_INVALIDATE) != 0);
579 
580 done2:
581 	mtx_unlock(&Giant);
582 
583 	switch (rv) {
584 	case KERN_SUCCESS:
585 		return (0);
586 	case KERN_INVALID_ADDRESS:
587 		return (EINVAL);	/* Sun returns ENOMEM? */
588 	case KERN_FAILURE:
589 		return (EIO);
590 	default:
591 		return (EINVAL);
592 	}
593 }
594 
595 #ifndef _SYS_SYSPROTO_H_
596 struct munmap_args {
597 	void *addr;
598 	size_t len;
599 };
600 #endif
601 /*
602  * MPSAFE
603  */
604 int
605 munmap(td, uap)
606 	struct thread *td;
607 	struct munmap_args *uap;
608 {
609 	vm_offset_t addr;
610 	vm_size_t size, pageoff;
611 	vm_map_t map;
612 
613 	addr = (vm_offset_t) uap->addr;
614 	size = uap->len;
615 
616 	pageoff = (addr & PAGE_MASK);
617 	addr -= pageoff;
618 	size += pageoff;
619 	size = (vm_size_t) round_page(size);
620 	if (addr + size < addr)
621 		return (EINVAL);
622 
623 	if (size == 0)
624 		return (0);
625 
626 	/*
627 	 * Check for illegal addresses.  Watch out for address wrap...
628 	 */
629 	map = &td->td_proc->p_vmspace->vm_map;
630 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
631 		return (EINVAL);
632 	/*
633 	 * Make sure entire range is allocated.
634 	 */
635 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
636 		return (EINVAL);
637 
638 	/* returns nothing but KERN_SUCCESS anyway */
639 	(void) vm_map_remove(map, addr, addr + size);
640 	return (0);
641 }
642 
643 #if 0
644 void
645 munmapfd(td, fd)
646 	struct thread *td;
647 	int fd;
648 {
649 	/*
650 	 * XXX should unmap any regions mapped to this file
651 	 */
652 	FILEDESC_LOCK(p->p_fd);
653 	td->td_proc->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
654 	FILEDESC_UNLOCK(p->p_fd);
655 }
656 #endif
657 
658 #ifndef _SYS_SYSPROTO_H_
659 struct mprotect_args {
660 	const void *addr;
661 	size_t len;
662 	int prot;
663 };
664 #endif
665 /*
666  * MPSAFE
667  */
668 int
669 mprotect(td, uap)
670 	struct thread *td;
671 	struct mprotect_args *uap;
672 {
673 	vm_offset_t addr;
674 	vm_size_t size, pageoff;
675 	vm_prot_t prot;
676 
677 	addr = (vm_offset_t) uap->addr;
678 	size = uap->len;
679 	prot = uap->prot & VM_PROT_ALL;
680 #if defined(VM_PROT_READ_IS_EXEC)
681 	if (prot & VM_PROT_READ)
682 		prot |= VM_PROT_EXECUTE;
683 #endif
684 
685 	pageoff = (addr & PAGE_MASK);
686 	addr -= pageoff;
687 	size += pageoff;
688 	size = (vm_size_t) round_page(size);
689 	if (addr + size < addr)
690 		return (EINVAL);
691 
692 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
693 	    addr + size, prot, FALSE)) {
694 	case KERN_SUCCESS:
695 		return (0);
696 	case KERN_PROTECTION_FAILURE:
697 		return (EACCES);
698 	}
699 	return (EINVAL);
700 }
701 
702 #ifndef _SYS_SYSPROTO_H_
703 struct minherit_args {
704 	void *addr;
705 	size_t len;
706 	int inherit;
707 };
708 #endif
709 /*
710  * MPSAFE
711  */
712 int
713 minherit(td, uap)
714 	struct thread *td;
715 	struct minherit_args *uap;
716 {
717 	vm_offset_t addr;
718 	vm_size_t size, pageoff;
719 	vm_inherit_t inherit;
720 
721 	addr = (vm_offset_t)uap->addr;
722 	size = uap->len;
723 	inherit = uap->inherit;
724 
725 	pageoff = (addr & PAGE_MASK);
726 	addr -= pageoff;
727 	size += pageoff;
728 	size = (vm_size_t) round_page(size);
729 	if (addr + size < addr)
730 		return (EINVAL);
731 
732 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
733 	    addr + size, inherit)) {
734 	case KERN_SUCCESS:
735 		return (0);
736 	case KERN_PROTECTION_FAILURE:
737 		return (EACCES);
738 	}
739 	return (EINVAL);
740 }
741 
742 #ifndef _SYS_SYSPROTO_H_
743 struct madvise_args {
744 	void *addr;
745 	size_t len;
746 	int behav;
747 };
748 #endif
749 
750 /*
751  * MPSAFE
752  */
753 /* ARGSUSED */
754 int
755 madvise(td, uap)
756 	struct thread *td;
757 	struct madvise_args *uap;
758 {
759 	vm_offset_t start, end;
760 	vm_map_t map;
761 	struct proc *p;
762 	int error;
763 
764 	/*
765 	 * Check for our special case, advising the swap pager we are
766 	 * "immortal."
767 	 */
768 	if (uap->behav == MADV_PROTECT) {
769 		error = suser(td);
770 		if (error == 0) {
771 			p = td->td_proc;
772 			PROC_LOCK(p);
773 			p->p_flag |= P_PROTECTED;
774 			PROC_UNLOCK(p);
775 		}
776 		return (error);
777 	}
778 	/*
779 	 * Check for illegal behavior
780 	 */
781 	if (uap->behav < 0 || uap->behav > MADV_CORE)
782 		return (EINVAL);
783 	/*
784 	 * Check for illegal addresses.  Watch out for address wrap... Note
785 	 * that VM_*_ADDRESS are not constants due to casts (argh).
786 	 */
787 	map = &td->td_proc->p_vmspace->vm_map;
788 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
789 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
790 		return (EINVAL);
791 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
792 		return (EINVAL);
793 
794 	/*
795 	 * Since this routine is only advisory, we default to conservative
796 	 * behavior.
797 	 */
798 	start = trunc_page((vm_offset_t) uap->addr);
799 	end = round_page((vm_offset_t) uap->addr + uap->len);
800 
801 	if (vm_map_madvise(map, start, end, uap->behav))
802 		return (EINVAL);
803 	return (0);
804 }
805 
806 #ifndef _SYS_SYSPROTO_H_
807 struct mincore_args {
808 	const void *addr;
809 	size_t len;
810 	char *vec;
811 };
812 #endif
813 
814 /*
815  * MPSAFE
816  */
817 /* ARGSUSED */
818 int
819 mincore(td, uap)
820 	struct thread *td;
821 	struct mincore_args *uap;
822 {
823 	vm_offset_t addr, first_addr;
824 	vm_offset_t end, cend;
825 	pmap_t pmap;
826 	vm_map_t map;
827 	char *vec;
828 	int error = 0;
829 	int vecindex, lastvecindex;
830 	vm_map_entry_t current;
831 	vm_map_entry_t entry;
832 	int mincoreinfo;
833 	unsigned int timestamp;
834 
835 	/*
836 	 * Make sure that the addresses presented are valid for user
837 	 * mode.
838 	 */
839 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
840 	end = addr + (vm_size_t)round_page(uap->len);
841 	map = &td->td_proc->p_vmspace->vm_map;
842 	if (end > vm_map_max(map) || end < addr)
843 		return (EINVAL);
844 
845 	/*
846 	 * Address of byte vector
847 	 */
848 	vec = uap->vec;
849 
850 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
851 
852 	vm_map_lock_read(map);
853 RestartScan:
854 	timestamp = map->timestamp;
855 
856 	if (!vm_map_lookup_entry(map, addr, &entry))
857 		entry = entry->next;
858 
859 	/*
860 	 * Do this on a map entry basis so that if the pages are not
861 	 * in the current processes address space, we can easily look
862 	 * up the pages elsewhere.
863 	 */
864 	lastvecindex = -1;
865 	for (current = entry;
866 	    (current != &map->header) && (current->start < end);
867 	    current = current->next) {
868 
869 		/*
870 		 * ignore submaps (for now) or null objects
871 		 */
872 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
873 			current->object.vm_object == NULL)
874 			continue;
875 
876 		/*
877 		 * limit this scan to the current map entry and the
878 		 * limits for the mincore call
879 		 */
880 		if (addr < current->start)
881 			addr = current->start;
882 		cend = current->end;
883 		if (cend > end)
884 			cend = end;
885 
886 		/*
887 		 * scan this entry one page at a time
888 		 */
889 		while (addr < cend) {
890 			/*
891 			 * Check pmap first, it is likely faster, also
892 			 * it can provide info as to whether we are the
893 			 * one referencing or modifying the page.
894 			 */
895 			mtx_lock(&Giant);
896 			mincoreinfo = pmap_mincore(pmap, addr);
897 			mtx_unlock(&Giant);
898 			if (!mincoreinfo) {
899 				vm_pindex_t pindex;
900 				vm_ooffset_t offset;
901 				vm_page_t m;
902 				/*
903 				 * calculate the page index into the object
904 				 */
905 				offset = current->offset + (addr - current->start);
906 				pindex = OFF_TO_IDX(offset);
907 				VM_OBJECT_LOCK(current->object.vm_object);
908 				m = vm_page_lookup(current->object.vm_object,
909 					pindex);
910 				/*
911 				 * if the page is resident, then gather information about
912 				 * it.
913 				 */
914 				if (m) {
915 					mincoreinfo = MINCORE_INCORE;
916 					vm_page_lock_queues();
917 					if (m->dirty ||
918 						pmap_is_modified(m))
919 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
920 					if ((m->flags & PG_REFERENCED) ||
921 						pmap_ts_referenced(m)) {
922 						vm_page_flag_set(m, PG_REFERENCED);
923 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
924 					}
925 					vm_page_unlock_queues();
926 				}
927 				VM_OBJECT_UNLOCK(current->object.vm_object);
928 			}
929 
930 			/*
931 			 * subyte may page fault.  In case it needs to modify
932 			 * the map, we release the lock.
933 			 */
934 			vm_map_unlock_read(map);
935 
936 			/*
937 			 * calculate index into user supplied byte vector
938 			 */
939 			vecindex = OFF_TO_IDX(addr - first_addr);
940 
941 			/*
942 			 * If we have skipped map entries, we need to make sure that
943 			 * the byte vector is zeroed for those skipped entries.
944 			 */
945 			while ((lastvecindex + 1) < vecindex) {
946 				error = subyte(vec + lastvecindex, 0);
947 				if (error) {
948 					error = EFAULT;
949 					goto done2;
950 				}
951 				++lastvecindex;
952 			}
953 
954 			/*
955 			 * Pass the page information to the user
956 			 */
957 			error = subyte(vec + vecindex, mincoreinfo);
958 			if (error) {
959 				error = EFAULT;
960 				goto done2;
961 			}
962 
963 			/*
964 			 * If the map has changed, due to the subyte, the previous
965 			 * output may be invalid.
966 			 */
967 			vm_map_lock_read(map);
968 			if (timestamp != map->timestamp)
969 				goto RestartScan;
970 
971 			lastvecindex = vecindex;
972 			addr += PAGE_SIZE;
973 		}
974 	}
975 
976 	/*
977 	 * subyte may page fault.  In case it needs to modify
978 	 * the map, we release the lock.
979 	 */
980 	vm_map_unlock_read(map);
981 
982 	/*
983 	 * Zero the last entries in the byte vector.
984 	 */
985 	vecindex = OFF_TO_IDX(end - first_addr);
986 	while ((lastvecindex + 1) < vecindex) {
987 		error = subyte(vec + lastvecindex, 0);
988 		if (error) {
989 			error = EFAULT;
990 			goto done2;
991 		}
992 		++lastvecindex;
993 	}
994 
995 	/*
996 	 * If the map has changed, due to the subyte, the previous
997 	 * output may be invalid.
998 	 */
999 	vm_map_lock_read(map);
1000 	if (timestamp != map->timestamp)
1001 		goto RestartScan;
1002 	vm_map_unlock_read(map);
1003 done2:
1004 	return (error);
1005 }
1006 
1007 #ifndef _SYS_SYSPROTO_H_
1008 struct mlock_args {
1009 	const void *addr;
1010 	size_t len;
1011 };
1012 #endif
1013 /*
1014  * MPSAFE
1015  */
1016 int
1017 mlock(td, uap)
1018 	struct thread *td;
1019 	struct mlock_args *uap;
1020 {
1021 	vm_offset_t addr;
1022 	vm_size_t size, pageoff;
1023 	int error;
1024 
1025 	addr = (vm_offset_t) uap->addr;
1026 	size = uap->len;
1027 
1028 	pageoff = (addr & PAGE_MASK);
1029 	addr -= pageoff;
1030 	size += pageoff;
1031 	size = (vm_size_t) round_page(size);
1032 
1033 	/* disable wrap around */
1034 	if (addr + size < addr)
1035 		return (EINVAL);
1036 
1037 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
1038 		return (EAGAIN);
1039 
1040 #if 0
1041 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) >
1042 	    td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
1043 		return (ENOMEM);
1044 #else
1045 	error = suser(td);
1046 	if (error)
1047 		return (error);
1048 #endif
1049 
1050 	error = vm_map_wire(&td->td_proc->p_vmspace->vm_map, addr,
1051 		     addr + size, VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1052 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1053 }
1054 
1055 #ifndef _SYS_SYSPROTO_H_
1056 struct mlockall_args {
1057 	int	how;
1058 };
1059 #endif
1060 
1061 /*
1062  * MPSAFE
1063  */
1064 int
1065 mlockall(td, uap)
1066 	struct thread *td;
1067 	struct mlockall_args *uap;
1068 {
1069 	vm_map_t map;
1070 	int error;
1071 
1072 	map = &td->td_proc->p_vmspace->vm_map;
1073 	error = 0;
1074 
1075 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1076 		return (EINVAL);
1077 
1078 #if 0
1079 	/*
1080 	 * If wiring all pages in the process would cause it to exceed
1081 	 * a hard resource limit, return ENOMEM.
1082 	 */
1083 	if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1084 		td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur))
1085 		return (ENOMEM);
1086 #else
1087 	error = suser(td);
1088 	if (error)
1089 		return (error);
1090 #endif
1091 
1092 	if (uap->how & MCL_FUTURE) {
1093 		vm_map_lock(map);
1094 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1095 		vm_map_unlock(map);
1096 		error = 0;
1097 	}
1098 
1099 	if (uap->how & MCL_CURRENT) {
1100 		/*
1101 		 * P1003.1-2001 mandates that all currently mapped pages
1102 		 * will be memory resident and locked (wired) upon return
1103 		 * from mlockall(). vm_map_wire() will wire pages, by
1104 		 * calling vm_fault_wire() for each page in the region.
1105 		 */
1106 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1107 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1108 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1109 	}
1110 
1111 	return (error);
1112 }
1113 
1114 #ifndef _SYS_SYSPROTO_H_
1115 struct munlockall_args {
1116 	register_t dummy;
1117 };
1118 #endif
1119 
1120 /*
1121  * MPSAFE
1122  */
1123 int
1124 munlockall(td, uap)
1125 	struct thread *td;
1126 	struct munlockall_args *uap;
1127 {
1128 	vm_map_t map;
1129 	int error;
1130 
1131 	map = &td->td_proc->p_vmspace->vm_map;
1132 	error = suser(td);
1133 	if (error)
1134 		return (error);
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 	error = suser(td);
1179 	if (error)
1180 		return (error);
1181 
1182 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, addr,
1183 		     addr + size, VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1184 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1185 }
1186 
1187 /*
1188  * vm_mmap()
1189  *
1190  * MPSAFE
1191  *
1192  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1193  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1194  */
1195 int
1196 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1197 	vm_prot_t maxprot, int flags,
1198 	void *handle,
1199 	vm_ooffset_t foff)
1200 {
1201 	boolean_t fitit;
1202 	vm_object_t object;
1203 	struct vnode *vp = NULL;
1204 	objtype_t type;
1205 	int rv = KERN_SUCCESS;
1206 	vm_ooffset_t objsize;
1207 	int docow;
1208 	struct thread *td = curthread;
1209 
1210 	if (size == 0)
1211 		return (0);
1212 
1213 	objsize = size = round_page(size);
1214 
1215 	if (td->td_proc->p_vmspace->vm_map.size + size >
1216 	    td->td_proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1217 		return(ENOMEM);
1218 	}
1219 
1220 	/*
1221 	 * We currently can only deal with page aligned file offsets.
1222 	 * The check is here rather than in the syscall because the
1223 	 * kernel calls this function internally for other mmaping
1224 	 * operations (such as in exec) and non-aligned offsets will
1225 	 * cause pmap inconsistencies...so we want to be sure to
1226 	 * disallow this in all cases.
1227 	 */
1228 	if (foff & PAGE_MASK)
1229 		return (EINVAL);
1230 
1231 	if ((flags & MAP_FIXED) == 0) {
1232 		fitit = TRUE;
1233 		*addr = round_page(*addr);
1234 	} else {
1235 		if (*addr != trunc_page(*addr))
1236 			return (EINVAL);
1237 		fitit = FALSE;
1238 		(void) vm_map_remove(map, *addr, *addr + size);
1239 	}
1240 
1241 	/*
1242 	 * Lookup/allocate object.
1243 	 */
1244 	if (flags & MAP_ANON) {
1245 		type = OBJT_DEFAULT;
1246 		/*
1247 		 * Unnamed anonymous regions always start at 0.
1248 		 */
1249 		if (handle == 0)
1250 			foff = 0;
1251 	} else {
1252 		vp = (struct vnode *) handle;
1253 		mtx_lock(&Giant);
1254 		ASSERT_VOP_LOCKED(vp, "vm_mmap");
1255 		if (vp->v_type == VCHR) {
1256 			type = OBJT_DEVICE;
1257 			handle = vp->v_rdev;
1258 		} else {
1259 			struct vattr vat;
1260 			int error;
1261 
1262 			error = VOP_GETATTR(vp, &vat, td->td_ucred, td);
1263 			if (error) {
1264 				mtx_unlock(&Giant);
1265 				return (error);
1266 			}
1267 			objsize = round_page(vat.va_size);
1268 			type = OBJT_VNODE;
1269 			/*
1270 			 * if it is a regular file without any references
1271 			 * we do not need to sync it.
1272 			 */
1273 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1274 				flags |= MAP_NOSYNC;
1275 			}
1276 		}
1277 		mtx_unlock(&Giant);
1278 	}
1279 
1280 	if (handle == NULL) {
1281 		object = NULL;
1282 		docow = 0;
1283 	} else {
1284 		object = vm_pager_allocate(type,
1285 			handle, objsize, prot, foff);
1286 		if (object == NULL) {
1287 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1288 		}
1289 		docow = MAP_PREFAULT_PARTIAL;
1290 	}
1291 
1292 	/*
1293 	 * Force device mappings to be shared.
1294 	 */
1295 	if (type == OBJT_DEVICE) {
1296 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1297 		flags |= MAP_SHARED;
1298 	}
1299 
1300 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1301 		docow |= MAP_COPY_ON_WRITE;
1302 	if (flags & MAP_NOSYNC)
1303 		docow |= MAP_DISABLE_SYNCER;
1304 	if (flags & MAP_NOCORE)
1305 		docow |= MAP_DISABLE_COREDUMP;
1306 
1307 #if defined(VM_PROT_READ_IS_EXEC)
1308 	if (prot & VM_PROT_READ)
1309 		prot |= VM_PROT_EXECUTE;
1310 
1311 	if (maxprot & VM_PROT_READ)
1312 		maxprot |= VM_PROT_EXECUTE;
1313 #endif
1314 
1315 	if (fitit)
1316 		*addr = pmap_addr_hint(object, *addr, size);
1317 
1318 	if (flags & MAP_STACK)
1319 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1320 		    docow | MAP_STACK_GROWS_DOWN);
1321 	else
1322 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1323 				 prot, maxprot, docow);
1324 
1325 	if (rv != KERN_SUCCESS) {
1326 		/*
1327 		 * Lose the object reference. Will destroy the
1328 		 * object if it's an unnamed anonymous mapping
1329 		 * or named anonymous without other references.
1330 		 */
1331 		vm_object_deallocate(object);
1332 	} else if (flags & MAP_SHARED) {
1333 		/*
1334 		 * Shared memory is also shared with children.
1335 		 */
1336 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1337 		if (rv != KERN_SUCCESS)
1338 			(void) vm_map_remove(map, *addr, *addr + size);
1339 	}
1340 
1341 	/*
1342 	 * If the process has requested that all future mappings
1343 	 * be wired, then heed this.
1344 	 */
1345 	if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1346 		vm_map_wire(map, *addr, *addr + size,
1347 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1348 
1349 	switch (rv) {
1350 	case KERN_SUCCESS:
1351 		return (0);
1352 	case KERN_INVALID_ADDRESS:
1353 	case KERN_NO_SPACE:
1354 		return (ENOMEM);
1355 	case KERN_PROTECTION_FAILURE:
1356 		return (EACCES);
1357 	default:
1358 		return (EINVAL);
1359 	}
1360 }
1361