xref: /freebsd/sys/vm/vm_mmap.c (revision eacee0ff7ec955b32e09515246bd97b6edcd2b0f)
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 
97 static void vmmapentry_rsrc_init __P((void *));
98 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
99 
100 static void
101 vmmapentry_rsrc_init(dummy)
102         void *dummy;
103 {
104     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
105     max_proc_mmap /= 100;
106 }
107 
108 /*
109  * MPSAFE
110  */
111 /* ARGSUSED */
112 int
113 sbrk(td, uap)
114 	struct thread *td;
115 	struct sbrk_args *uap;
116 {
117 	/* Not yet implemented */
118 	/* mtx_lock(&Giant); */
119 	/* mtx_unlock(&Giant); */
120 	return (EOPNOTSUPP);
121 }
122 
123 #ifndef _SYS_SYSPROTO_H_
124 struct sstk_args {
125 	int incr;
126 };
127 #endif
128 
129 /*
130  * MPSAFE
131  */
132 /* ARGSUSED */
133 int
134 sstk(td, uap)
135 	struct thread *td;
136 	struct sstk_args *uap;
137 {
138 	/* Not yet implemented */
139 	/* mtx_lock(&Giant); */
140 	/* mtx_unlock(&Giant); */
141 	return (EOPNOTSUPP);
142 }
143 
144 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
145 #ifndef _SYS_SYSPROTO_H_
146 struct getpagesize_args {
147 	int dummy;
148 };
149 #endif
150 
151 /* ARGSUSED */
152 int
153 ogetpagesize(td, uap)
154 	struct thread *td;
155 	struct getpagesize_args *uap;
156 {
157 	/* MP SAFE */
158 	td->td_retval[0] = PAGE_SIZE;
159 	return (0);
160 }
161 #endif				/* COMPAT_43 || COMPAT_SUNOS */
162 
163 
164 /*
165  * Memory Map (mmap) system call.  Note that the file offset
166  * and address are allowed to be NOT page aligned, though if
167  * the MAP_FIXED flag it set, both must have the same remainder
168  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
169  * page-aligned, the actual mapping starts at trunc_page(addr)
170  * and the return value is adjusted up by the page offset.
171  *
172  * Generally speaking, only character devices which are themselves
173  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
174  * there would be no cache coherency between a descriptor and a VM mapping
175  * both to the same character device.
176  *
177  * Block devices can be mmap'd no matter what they represent.  Cache coherency
178  * is maintained as long as you do not write directly to the underlying
179  * character device.
180  */
181 #ifndef _SYS_SYSPROTO_H_
182 struct mmap_args {
183 	void *addr;
184 	size_t len;
185 	int prot;
186 	int flags;
187 	int fd;
188 	long pad;
189 	off_t pos;
190 };
191 #endif
192 
193 /*
194  * MPSAFE
195  */
196 int
197 mmap(td, uap)
198 	struct thread *td;
199 	struct mmap_args *uap;
200 {
201 	struct file *fp = NULL;
202 	struct vnode *vp;
203 	vm_offset_t addr;
204 	vm_size_t size, pageoff;
205 	vm_prot_t prot, maxprot;
206 	void *handle;
207 	int flags, error;
208 	int disablexworkaround;
209 	off_t pos;
210 	struct vmspace *vms = td->td_proc->p_vmspace;
211 	vm_object_t obj;
212 
213 	addr = (vm_offset_t) uap->addr;
214 	size = uap->len;
215 	prot = uap->prot & VM_PROT_ALL;
216 	flags = uap->flags;
217 	pos = uap->pos;
218 
219 	fp = NULL;
220 	/* make sure mapping fits into numeric range etc */
221 	if ((ssize_t) uap->len < 0 ||
222 	    ((flags & MAP_ANON) && uap->fd != -1))
223 		return (EINVAL);
224 
225 	if (flags & MAP_STACK) {
226 		if ((uap->fd != -1) ||
227 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
228 			return (EINVAL);
229 		flags |= MAP_ANON;
230 		pos = 0;
231 	}
232 
233 	/*
234 	 * Align the file position to a page boundary,
235 	 * and save its page offset component.
236 	 */
237 	pageoff = (pos & PAGE_MASK);
238 	pos -= pageoff;
239 
240 	/* Adjust size for rounding (on both ends). */
241 	size += pageoff;			/* low end... */
242 	size = (vm_size_t) round_page(size);	/* hi end */
243 
244 	/*
245 	 * Check for illegal addresses.  Watch out for address wrap... Note
246 	 * that VM_*_ADDRESS are not constants due to casts (argh).
247 	 */
248 	if (flags & MAP_FIXED) {
249 		/*
250 		 * The specified address must have the same remainder
251 		 * as the file offset taken modulo PAGE_SIZE, so it
252 		 * should be aligned after adjustment by pageoff.
253 		 */
254 		addr -= pageoff;
255 		if (addr & PAGE_MASK)
256 			return (EINVAL);
257 		/* Address range must be all in user VM space. */
258 		if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
259 			return (EINVAL);
260 #ifndef i386
261 		if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
262 			return (EINVAL);
263 #endif
264 		if (addr + size < addr)
265 			return (EINVAL);
266 	}
267 	/*
268 	 * XXX for non-fixed mappings where no hint is provided or
269 	 * the hint would fall in the potential heap space,
270 	 * place it after the end of the largest possible heap.
271 	 *
272 	 * There should really be a pmap call to determine a reasonable
273 	 * location.
274 	 */
275 	else if (addr == 0 ||
276 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
277 	     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
278 		addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
279 
280 	mtx_lock(&Giant);	/* syscall marked mp-safe but isn't */
281 	if (flags & MAP_ANON) {
282 		/*
283 		 * Mapping blank space is trivial.
284 		 */
285 		handle = NULL;
286 		maxprot = VM_PROT_ALL;
287 		pos = 0;
288 	} else {
289 		/*
290 		 * Mapping file, get fp for validation. Obtain vnode and make
291 		 * sure it is of appropriate type.
292 		 * don't let the descriptor disappear on us if we block
293 		 */
294 		if ((error = fget(td, uap->fd, &fp)) != 0)
295 			goto done;
296 		if (fp->f_type != DTYPE_VNODE) {
297 			error = EINVAL;
298 			goto done;
299 		}
300 
301 		/*
302 		 * POSIX shared-memory objects are defined to have
303 		 * kernel persistence, and are not defined to support
304 		 * read(2)/write(2) -- or even open(2).  Thus, we can
305 		 * use MAP_ASYNC to trade on-disk coherence for speed.
306 		 * The shm_open(3) library routine turns on the FPOSIXSHM
307 		 * flag to request this behavior.
308 		 */
309 		if (fp->f_flag & FPOSIXSHM)
310 			flags |= MAP_NOSYNC;
311 		vp = (struct vnode *) fp->f_data;
312 		if (vp->v_type != VREG && vp->v_type != VCHR) {
313 			error = EINVAL;
314 			goto done;
315 		}
316 		if (vp->v_type == VREG) {
317 			/*
318 			 * Get the proper underlying object
319 			 */
320 			if (VOP_GETVOBJECT(vp, &obj) != 0) {
321 				error = EINVAL;
322 				goto done;
323 			}
324 			vp = (struct vnode*)obj->handle;
325 		}
326 		/*
327 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
328 		 * SunOS).
329 		 */
330 		if ((vp->v_type == VCHR) &&
331 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
332 			handle = NULL;
333 			maxprot = VM_PROT_ALL;
334 			flags |= MAP_ANON;
335 			pos = 0;
336 		} else {
337 			/*
338 			 * cdevs does not provide private mappings of any kind.
339 			 */
340 			/*
341 			 * However, for XIG X server to continue to work,
342 			 * we should allow the superuser to do it anyway.
343 			 * We only allow it at securelevel < 1.
344 			 * (Because the XIG X server writes directly to video
345 			 * memory via /dev/mem, it should never work at any
346 			 * other securelevel.
347 			 * XXX this will have to go
348 			 */
349 			if (securelevel_ge(td->td_proc->p_ucred, 1))
350 				disablexworkaround = 1;
351 			else
352 				disablexworkaround = suser_td(td);
353 			if (vp->v_type == VCHR && disablexworkaround &&
354 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
355 				error = EINVAL;
356 				goto done;
357 			}
358 			/*
359 			 * Ensure that file and memory protections are
360 			 * compatible.  Note that we only worry about
361 			 * writability if mapping is shared; in this case,
362 			 * current and max prot are dictated by the open file.
363 			 * XXX use the vnode instead?  Problem is: what
364 			 * credentials do we use for determination? What if
365 			 * proc does a setuid?
366 			 */
367 			maxprot = VM_PROT_EXECUTE;	/* ??? */
368 			if (fp->f_flag & FREAD) {
369 				maxprot |= VM_PROT_READ;
370 			} else if (prot & PROT_READ) {
371 				error = EACCES;
372 				goto done;
373 			}
374 			/*
375 			 * If we are sharing potential changes (either via
376 			 * MAP_SHARED or via the implicit sharing of character
377 			 * device mappings), and we are trying to get write
378 			 * permission although we opened it without asking
379 			 * for it, bail out.  Check for superuser, only if
380 			 * we're at securelevel < 1, to allow the XIG X server
381 			 * to continue to work.
382 			 */
383 
384 			if ((flags & MAP_SHARED) != 0 ||
385 			    (vp->v_type == VCHR && disablexworkaround)) {
386 				if ((fp->f_flag & FWRITE) != 0) {
387 					struct vattr va;
388 					if ((error =
389 					    VOP_GETATTR(vp, &va,
390 						        td->td_proc->p_ucred, td))) {
391 						goto done;
392 					}
393 					if ((va.va_flags &
394 					   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
395 						maxprot |= VM_PROT_WRITE;
396 					} else if (prot & PROT_WRITE) {
397 						error = EPERM;
398 						goto done;
399 					}
400 				} else if ((prot & PROT_WRITE) != 0) {
401 					error = EACCES;
402 					goto done;
403 				}
404 			} else {
405 				maxprot |= VM_PROT_WRITE;
406 			}
407 
408 			handle = (void *)vp;
409 		}
410 	}
411 
412 	/*
413 	 * Do not allow more then a certain number of vm_map_entry structures
414 	 * per process.  Scale with the number of rforks sharing the map
415 	 * to make the limit reasonable for threads.
416 	 */
417 	if (max_proc_mmap &&
418 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
419 		error = ENOMEM;
420 		goto done;
421 	}
422 
423 	mtx_unlock(&Giant);
424 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
425 	    flags, handle, pos);
426 	if (error == 0)
427 		td->td_retval[0] = (register_t) (addr + pageoff);
428 	mtx_lock(&Giant);
429 done:
430 	if (fp)
431 		fdrop(fp, td);
432 	mtx_unlock(&Giant);
433 	return (error);
434 }
435 
436 #ifdef COMPAT_43
437 #ifndef _SYS_SYSPROTO_H_
438 struct ommap_args {
439 	caddr_t addr;
440 	int len;
441 	int prot;
442 	int flags;
443 	int fd;
444 	long pos;
445 };
446 #endif
447 int
448 ommap(td, uap)
449 	struct thread *td;
450 	struct ommap_args *uap;
451 {
452 	struct mmap_args nargs;
453 	static const char cvtbsdprot[8] = {
454 		0,
455 		PROT_EXEC,
456 		PROT_WRITE,
457 		PROT_EXEC | PROT_WRITE,
458 		PROT_READ,
459 		PROT_EXEC | PROT_READ,
460 		PROT_WRITE | PROT_READ,
461 		PROT_EXEC | PROT_WRITE | PROT_READ,
462 	};
463 
464 #define	OMAP_ANON	0x0002
465 #define	OMAP_COPY	0x0020
466 #define	OMAP_SHARED	0x0010
467 #define	OMAP_FIXED	0x0100
468 
469 	nargs.addr = uap->addr;
470 	nargs.len = uap->len;
471 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
472 	nargs.flags = 0;
473 	if (uap->flags & OMAP_ANON)
474 		nargs.flags |= MAP_ANON;
475 	if (uap->flags & OMAP_COPY)
476 		nargs.flags |= MAP_COPY;
477 	if (uap->flags & OMAP_SHARED)
478 		nargs.flags |= MAP_SHARED;
479 	else
480 		nargs.flags |= MAP_PRIVATE;
481 	if (uap->flags & OMAP_FIXED)
482 		nargs.flags |= MAP_FIXED;
483 	nargs.fd = uap->fd;
484 	nargs.pos = uap->pos;
485 	return (mmap(td, &nargs));
486 }
487 #endif				/* COMPAT_43 */
488 
489 
490 #ifndef _SYS_SYSPROTO_H_
491 struct msync_args {
492 	void *addr;
493 	int len;
494 	int flags;
495 };
496 #endif
497 /*
498  * MPSAFE
499  */
500 int
501 msync(td, uap)
502 	struct thread *td;
503 	struct msync_args *uap;
504 {
505 	vm_offset_t addr;
506 	vm_size_t size, pageoff;
507 	int flags;
508 	vm_map_t map;
509 	int rv;
510 
511 	addr = (vm_offset_t) uap->addr;
512 	size = uap->len;
513 	flags = uap->flags;
514 
515 	pageoff = (addr & PAGE_MASK);
516 	addr -= pageoff;
517 	size += pageoff;
518 	size = (vm_size_t) round_page(size);
519 	if (addr + size < addr)
520 		return(EINVAL);
521 
522 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
523 		return (EINVAL);
524 
525 	mtx_lock(&Giant);
526 
527 	map = &td->td_proc->p_vmspace->vm_map;
528 
529 	/*
530 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
531 	 * pages with the region containing addr".  Unfortunately, we don't
532 	 * really keep track of individual mmaps so we approximate by flushing
533 	 * the range of the map entry containing addr. This can be incorrect
534 	 * if the region splits or is coalesced with a neighbor.
535 	 */
536 	if (size == 0) {
537 		vm_map_entry_t entry;
538 
539 		vm_map_lock_read(map);
540 		rv = vm_map_lookup_entry(map, addr, &entry);
541 		vm_map_unlock_read(map);
542 		if (rv == FALSE) {
543 			rv = -1;
544 			goto done2;
545 		}
546 		addr = entry->start;
547 		size = entry->end - entry->start;
548 	}
549 
550 	/*
551 	 * Clean the pages and interpret the return value.
552 	 */
553 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
554 	    (flags & MS_INVALIDATE) != 0);
555 
556 done2:
557 	mtx_unlock(&Giant);
558 
559 	switch (rv) {
560 	case KERN_SUCCESS:
561 		return(0);
562 	case KERN_INVALID_ADDRESS:
563 		return (EINVAL);	/* Sun returns ENOMEM? */
564 	case KERN_FAILURE:
565 		return (EIO);
566 	default:
567 		return (EINVAL);
568 	}
569 }
570 
571 #ifndef _SYS_SYSPROTO_H_
572 struct munmap_args {
573 	void *addr;
574 	size_t len;
575 };
576 #endif
577 /*
578  * MPSAFE
579  */
580 int
581 munmap(td, uap)
582 	struct thread *td;
583 	struct munmap_args *uap;
584 {
585 	vm_offset_t addr;
586 	vm_size_t size, pageoff;
587 	vm_map_t map;
588 
589 	addr = (vm_offset_t) uap->addr;
590 	size = uap->len;
591 
592 	pageoff = (addr & PAGE_MASK);
593 	addr -= pageoff;
594 	size += pageoff;
595 	size = (vm_size_t) round_page(size);
596 	if (addr + size < addr)
597 		return(EINVAL);
598 
599 	if (size == 0)
600 		return (0);
601 
602 	/*
603 	 * Check for illegal addresses.  Watch out for address wrap... Note
604 	 * that VM_*_ADDRESS are not constants due to casts (argh).
605 	 */
606 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
607 		return (EINVAL);
608 #ifndef i386
609 	if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
610 		return (EINVAL);
611 #endif
612 	mtx_lock(&Giant);
613 	map = &td->td_proc->p_vmspace->vm_map;
614 	/*
615 	 * Make sure entire range is allocated.
616 	 */
617 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) {
618 		mtx_unlock(&Giant);
619 		return (EINVAL);
620 	}
621 	/* returns nothing but KERN_SUCCESS anyway */
622 	(void) vm_map_remove(map, addr, addr + size);
623 	mtx_unlock(&Giant);
624 	return (0);
625 }
626 
627 #if 0
628 void
629 munmapfd(td, fd)
630 	struct thread *td;
631 	int fd;
632 {
633 	/*
634 	 * XXX should unmap any regions mapped to this file
635 	 */
636 	FILEDESC_LOCK(p->p_fd);
637 	td->td_proc->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
638 	FILEDESC_UNLOCK(p->p_fd);
639 }
640 #endif
641 
642 #ifndef _SYS_SYSPROTO_H_
643 struct mprotect_args {
644 	const void *addr;
645 	size_t len;
646 	int prot;
647 };
648 #endif
649 /*
650  * MPSAFE
651  */
652 int
653 mprotect(td, uap)
654 	struct thread *td;
655 	struct mprotect_args *uap;
656 {
657 	vm_offset_t addr;
658 	vm_size_t size, pageoff;
659 	vm_prot_t prot;
660 	int ret;
661 
662 	addr = (vm_offset_t) uap->addr;
663 	size = uap->len;
664 	prot = uap->prot & VM_PROT_ALL;
665 #if defined(VM_PROT_READ_IS_EXEC)
666 	if (prot & VM_PROT_READ)
667 		prot |= VM_PROT_EXECUTE;
668 #endif
669 
670 	pageoff = (addr & PAGE_MASK);
671 	addr -= pageoff;
672 	size += pageoff;
673 	size = (vm_size_t) round_page(size);
674 	if (addr + size < addr)
675 		return(EINVAL);
676 
677 	mtx_lock(&Giant);
678 	ret = vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
679 		     addr + size, prot, FALSE);
680 	mtx_unlock(&Giant);
681 	switch (ret) {
682 	case KERN_SUCCESS:
683 		return (0);
684 	case KERN_PROTECTION_FAILURE:
685 		return (EACCES);
686 	}
687 	return (EINVAL);
688 }
689 
690 #ifndef _SYS_SYSPROTO_H_
691 struct minherit_args {
692 	void *addr;
693 	size_t len;
694 	int inherit;
695 };
696 #endif
697 /*
698  * MPSAFE
699  */
700 int
701 minherit(td, uap)
702 	struct thread *td;
703 	struct minherit_args *uap;
704 {
705 	vm_offset_t addr;
706 	vm_size_t size, pageoff;
707 	vm_inherit_t inherit;
708 	int ret;
709 
710 	addr = (vm_offset_t)uap->addr;
711 	size = uap->len;
712 	inherit = uap->inherit;
713 
714 	pageoff = (addr & PAGE_MASK);
715 	addr -= pageoff;
716 	size += pageoff;
717 	size = (vm_size_t) round_page(size);
718 	if (addr + size < addr)
719 		return(EINVAL);
720 
721 	mtx_lock(&Giant);
722 	ret = vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, addr+size,
723 		    inherit);
724 	mtx_unlock(&Giant);
725 
726 	switch (ret) {
727 	case KERN_SUCCESS:
728 		return (0);
729 	case KERN_PROTECTION_FAILURE:
730 		return (EACCES);
731 	}
732 	return (EINVAL);
733 }
734 
735 #ifndef _SYS_SYSPROTO_H_
736 struct madvise_args {
737 	void *addr;
738 	size_t len;
739 	int behav;
740 };
741 #endif
742 
743 /*
744  * MPSAFE
745  */
746 /* ARGSUSED */
747 int
748 madvise(td, uap)
749 	struct thread *td;
750 	struct madvise_args *uap;
751 {
752 	vm_offset_t start, end;
753 	int ret;
754 
755 	/*
756 	 * Check for illegal behavior
757 	 */
758 	if (uap->behav < 0 || uap->behav > MADV_CORE)
759 		return (EINVAL);
760 	/*
761 	 * Check for illegal addresses.  Watch out for address wrap... Note
762 	 * that VM_*_ADDRESS are not constants due to casts (argh).
763 	 */
764 	if (VM_MAXUSER_ADDRESS > 0 &&
765 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
766 		return (EINVAL);
767 #ifndef i386
768 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
769 		return (EINVAL);
770 #endif
771 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
772 		return (EINVAL);
773 
774 	/*
775 	 * Since this routine is only advisory, we default to conservative
776 	 * behavior.
777 	 */
778 	start = trunc_page((vm_offset_t) uap->addr);
779 	end = round_page((vm_offset_t) uap->addr + uap->len);
780 
781 	mtx_lock(&Giant);
782 	ret = vm_map_madvise(&td->td_proc->p_vmspace->vm_map, start, end, uap->behav);
783 	mtx_unlock(&Giant);
784 	return (ret ? EINVAL : 0);
785 }
786 
787 #ifndef _SYS_SYSPROTO_H_
788 struct mincore_args {
789 	const void *addr;
790 	size_t len;
791 	char *vec;
792 };
793 #endif
794 
795 /*
796  * MPSAFE
797  */
798 /* ARGSUSED */
799 int
800 mincore(td, uap)
801 	struct thread *td;
802 	struct mincore_args *uap;
803 {
804 	vm_offset_t addr, first_addr;
805 	vm_offset_t end, cend;
806 	pmap_t pmap;
807 	vm_map_t map;
808 	char *vec;
809 	int error = 0;
810 	int vecindex, lastvecindex;
811 	vm_map_entry_t current;
812 	vm_map_entry_t entry;
813 	int mincoreinfo;
814 	unsigned int timestamp;
815 
816 	/*
817 	 * Make sure that the addresses presented are valid for user
818 	 * mode.
819 	 */
820 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
821 	end = addr + (vm_size_t)round_page(uap->len);
822 	if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
823 		return (EINVAL);
824 	if (end < addr)
825 		return (EINVAL);
826 
827 	/*
828 	 * Address of byte vector
829 	 */
830 	vec = uap->vec;
831 
832 	mtx_lock(&Giant);
833 	map = &td->td_proc->p_vmspace->vm_map;
834 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
835 
836 	vm_map_lock_read(map);
837 RestartScan:
838 	timestamp = map->timestamp;
839 
840 	if (!vm_map_lookup_entry(map, addr, &entry))
841 		entry = entry->next;
842 
843 	/*
844 	 * Do this on a map entry basis so that if the pages are not
845 	 * in the current processes address space, we can easily look
846 	 * up the pages elsewhere.
847 	 */
848 	lastvecindex = -1;
849 	for (current = entry;
850 	    (current != &map->header) && (current->start < end);
851 	    current = current->next) {
852 
853 		/*
854 		 * ignore submaps (for now) or null objects
855 		 */
856 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
857 			current->object.vm_object == NULL)
858 			continue;
859 
860 		/*
861 		 * limit this scan to the current map entry and the
862 		 * limits for the mincore call
863 		 */
864 		if (addr < current->start)
865 			addr = current->start;
866 		cend = current->end;
867 		if (cend > end)
868 			cend = end;
869 
870 		/*
871 		 * scan this entry one page at a time
872 		 */
873 		while (addr < cend) {
874 			/*
875 			 * Check pmap first, it is likely faster, also
876 			 * it can provide info as to whether we are the
877 			 * one referencing or modifying the page.
878 			 */
879 			mincoreinfo = pmap_mincore(pmap, addr);
880 			if (!mincoreinfo) {
881 				vm_pindex_t pindex;
882 				vm_ooffset_t offset;
883 				vm_page_t m;
884 				/*
885 				 * calculate the page index into the object
886 				 */
887 				offset = current->offset + (addr - current->start);
888 				pindex = OFF_TO_IDX(offset);
889 				m = vm_page_lookup(current->object.vm_object,
890 					pindex);
891 				/*
892 				 * if the page is resident, then gather information about
893 				 * it.
894 				 */
895 				if (m) {
896 					mincoreinfo = MINCORE_INCORE;
897 					if (m->dirty ||
898 						pmap_is_modified(m))
899 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
900 					if ((m->flags & PG_REFERENCED) ||
901 						pmap_ts_referenced(m)) {
902 						vm_page_flag_set(m, PG_REFERENCED);
903 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
904 					}
905 				}
906 			}
907 
908 			/*
909 			 * subyte may page fault.  In case it needs to modify
910 			 * the map, we release the lock.
911 			 */
912 			vm_map_unlock_read(map);
913 
914 			/*
915 			 * calculate index into user supplied byte vector
916 			 */
917 			vecindex = OFF_TO_IDX(addr - first_addr);
918 
919 			/*
920 			 * If we have skipped map entries, we need to make sure that
921 			 * the byte vector is zeroed for those skipped entries.
922 			 */
923 			while ((lastvecindex + 1) < vecindex) {
924 				error = subyte( vec + lastvecindex, 0);
925 				if (error) {
926 					error = EFAULT;
927 					goto done2;
928 				}
929 				++lastvecindex;
930 			}
931 
932 			/*
933 			 * Pass the page information to the user
934 			 */
935 			error = subyte( vec + vecindex, mincoreinfo);
936 			if (error) {
937 				error = EFAULT;
938 				goto done2;
939 			}
940 
941 			/*
942 			 * If the map has changed, due to the subyte, the previous
943 			 * output may be invalid.
944 			 */
945 			vm_map_lock_read(map);
946 			if (timestamp != map->timestamp)
947 				goto RestartScan;
948 
949 			lastvecindex = vecindex;
950 			addr += PAGE_SIZE;
951 		}
952 	}
953 
954 	/*
955 	 * subyte may page fault.  In case it needs to modify
956 	 * the map, we release the lock.
957 	 */
958 	vm_map_unlock_read(map);
959 
960 	/*
961 	 * Zero the last entries in the byte vector.
962 	 */
963 	vecindex = OFF_TO_IDX(end - first_addr);
964 	while ((lastvecindex + 1) < vecindex) {
965 		error = subyte( vec + lastvecindex, 0);
966 		if (error) {
967 			error = EFAULT;
968 			goto done2;
969 		}
970 		++lastvecindex;
971 	}
972 
973 	/*
974 	 * If the map has changed, due to the subyte, the previous
975 	 * output may be invalid.
976 	 */
977 	vm_map_lock_read(map);
978 	if (timestamp != map->timestamp)
979 		goto RestartScan;
980 	vm_map_unlock_read(map);
981 done2:
982 	mtx_unlock(&Giant);
983 	return (error);
984 }
985 
986 #ifndef _SYS_SYSPROTO_H_
987 struct mlock_args {
988 	const void *addr;
989 	size_t len;
990 };
991 #endif
992 /*
993  * MPSAFE
994  */
995 int
996 mlock(td, uap)
997 	struct thread *td;
998 	struct mlock_args *uap;
999 {
1000 	vm_offset_t addr;
1001 	vm_size_t size, pageoff;
1002 	int error;
1003 
1004 	addr = (vm_offset_t) uap->addr;
1005 	size = uap->len;
1006 
1007 	pageoff = (addr & PAGE_MASK);
1008 	addr -= pageoff;
1009 	size += pageoff;
1010 	size = (vm_size_t) round_page(size);
1011 
1012 	/* disable wrap around */
1013 	if (addr + size < addr)
1014 		return (EINVAL);
1015 
1016 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
1017 		return (EAGAIN);
1018 
1019 #ifdef pmap_wired_count
1020 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) >
1021 	    td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
1022 		return (ENOMEM);
1023 #else
1024 	error = suser_td(td);
1025 	if (error)
1026 		return (error);
1027 #endif
1028 
1029 	mtx_lock(&Giant);
1030 	error = vm_map_user_pageable(&td->td_proc->p_vmspace->vm_map, addr,
1031 		     addr + size, FALSE);
1032 	mtx_unlock(&Giant);
1033 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1034 }
1035 
1036 #ifndef _SYS_SYSPROTO_H_
1037 struct mlockall_args {
1038 	int	how;
1039 };
1040 #endif
1041 
1042 /*
1043  * MPSAFE
1044  */
1045 int
1046 mlockall(td, uap)
1047 	struct thread *td;
1048 	struct mlockall_args *uap;
1049 {
1050 	/* mtx_lock(&Giant); */
1051 	/* mtx_unlock(&Giant); */
1052 	return 0;
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 munlockall(td, uap)
1066 	struct thread *td;
1067 	struct munlockall_args *uap;
1068 {
1069 	/* mtx_lock(&Giant); */
1070 	/* mtx_unlock(&Giant); */
1071 	return 0;
1072 }
1073 
1074 #ifndef _SYS_SYSPROTO_H_
1075 struct munlock_args {
1076 	const void *addr;
1077 	size_t len;
1078 };
1079 #endif
1080 /*
1081  * MPSAFE
1082  */
1083 int
1084 munlock(td, uap)
1085 	struct thread *td;
1086 	struct munlock_args *uap;
1087 {
1088 	vm_offset_t addr;
1089 	vm_size_t size, pageoff;
1090 	int error;
1091 
1092 	addr = (vm_offset_t) uap->addr;
1093 	size = uap->len;
1094 
1095 	pageoff = (addr & PAGE_MASK);
1096 	addr -= pageoff;
1097 	size += pageoff;
1098 	size = (vm_size_t) round_page(size);
1099 
1100 	/* disable wrap around */
1101 	if (addr + size < addr)
1102 		return (EINVAL);
1103 
1104 #ifndef pmap_wired_count
1105 	error = suser_td(td);
1106 	if (error)
1107 		return (error);
1108 #endif
1109 
1110 	mtx_lock(&Giant);
1111 	error = vm_map_user_pageable(&td->td_proc->p_vmspace->vm_map, addr,
1112 		     addr + size, TRUE);
1113 	mtx_unlock(&Giant);
1114 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1115 }
1116 
1117 /*
1118  * vm_mmap()
1119  *
1120  * MPSAFE
1121  *
1122  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1123  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1124  */
1125 int
1126 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1127 	vm_prot_t maxprot, int flags,
1128 	void *handle,
1129 	vm_ooffset_t foff)
1130 {
1131 	boolean_t fitit;
1132 	vm_object_t object;
1133 	struct vnode *vp = NULL;
1134 	objtype_t type;
1135 	int rv = KERN_SUCCESS;
1136 	vm_ooffset_t objsize;
1137 	int docow;
1138 	struct thread *td = curthread;
1139 
1140 	if (size == 0)
1141 		return (0);
1142 
1143 	objsize = size = round_page(size);
1144 
1145 	/*
1146 	 * We currently can only deal with page aligned file offsets.
1147 	 * The check is here rather than in the syscall because the
1148 	 * kernel calls this function internally for other mmaping
1149 	 * operations (such as in exec) and non-aligned offsets will
1150 	 * cause pmap inconsistencies...so we want to be sure to
1151 	 * disallow this in all cases.
1152 	 */
1153 	if (foff & PAGE_MASK)
1154 		return (EINVAL);
1155 
1156 	if ((flags & MAP_FIXED) == 0) {
1157 		fitit = TRUE;
1158 		*addr = round_page(*addr);
1159 		mtx_lock(&Giant);
1160 	} else {
1161 		if (*addr != trunc_page(*addr))
1162 			return (EINVAL);
1163 		fitit = FALSE;
1164 		mtx_lock(&Giant);
1165 		(void) vm_map_remove(map, *addr, *addr + size);
1166 	}
1167 
1168 	/*
1169 	 * Lookup/allocate object.
1170 	 */
1171 	if (flags & MAP_ANON) {
1172 		type = OBJT_DEFAULT;
1173 		/*
1174 		 * Unnamed anonymous regions always start at 0.
1175 		 */
1176 		if (handle == 0)
1177 			foff = 0;
1178 	} else {
1179 		vp = (struct vnode *) handle;
1180 		if (vp->v_type == VCHR) {
1181 			type = OBJT_DEVICE;
1182 			handle = (void *)(intptr_t)vp->v_rdev;
1183 		} else {
1184 			struct vattr vat;
1185 			int error;
1186 
1187 			error = VOP_GETATTR(vp, &vat, td->td_proc->p_ucred, td);
1188 			if (error) {
1189 				mtx_unlock(&Giant);
1190 				return (error);
1191 			}
1192 			objsize = round_page(vat.va_size);
1193 			type = OBJT_VNODE;
1194 			/*
1195 			 * if it is a regular file without any references
1196 			 * we do not need to sync it.
1197 			 */
1198 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1199 				flags |= MAP_NOSYNC;
1200 			}
1201 		}
1202 	}
1203 
1204 	if (handle == NULL) {
1205 		object = NULL;
1206 		docow = 0;
1207 	} else {
1208 		object = vm_pager_allocate(type,
1209 			handle, objsize, prot, foff);
1210 		if (object == NULL) {
1211 			mtx_unlock(&Giant);
1212 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1213 		}
1214 		docow = MAP_PREFAULT_PARTIAL;
1215 	}
1216 
1217 	/*
1218 	 * Force device mappings to be shared.
1219 	 */
1220 	if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1221 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1222 		flags |= MAP_SHARED;
1223 	}
1224 
1225 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1226 		docow |= MAP_COPY_ON_WRITE;
1227 	if (flags & MAP_NOSYNC)
1228 		docow |= MAP_DISABLE_SYNCER;
1229 	if (flags & MAP_NOCORE)
1230 		docow |= MAP_DISABLE_COREDUMP;
1231 
1232 #if defined(VM_PROT_READ_IS_EXEC)
1233 	if (prot & VM_PROT_READ)
1234 		prot |= VM_PROT_EXECUTE;
1235 
1236 	if (maxprot & VM_PROT_READ)
1237 		maxprot |= VM_PROT_EXECUTE;
1238 #endif
1239 
1240 	if (fitit)
1241 		*addr = pmap_addr_hint(object, *addr, size);
1242 
1243 	if (flags & MAP_STACK)
1244 		rv = vm_map_stack (map, *addr, size, prot,
1245 				   maxprot, docow);
1246 	else
1247 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1248 				 prot, maxprot, docow);
1249 
1250 	if (rv != KERN_SUCCESS) {
1251 		/*
1252 		 * Lose the object reference. Will destroy the
1253 		 * object if it's an unnamed anonymous mapping
1254 		 * or named anonymous without other references.
1255 		 */
1256 		vm_object_deallocate(object);
1257 	} else if (flags & MAP_SHARED) {
1258 		/*
1259 		 * Shared memory is also shared with children.
1260 		 */
1261 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1262 		if (rv != KERN_SUCCESS)
1263 			(void) vm_map_remove(map, *addr, *addr + size);
1264 	}
1265 	mtx_unlock(&Giant);
1266 	switch (rv) {
1267 	case KERN_SUCCESS:
1268 		return (0);
1269 	case KERN_INVALID_ADDRESS:
1270 	case KERN_NO_SPACE:
1271 		return (ENOMEM);
1272 	case KERN_PROTECTION_FAILURE:
1273 		return (EACCES);
1274 	default:
1275 		return (EINVAL);
1276 	}
1277 }
1278