xref: /netbsd-src/sys/uvm/uvm_mmap.c (revision 220b5c059a84c51ea44107ea8951a57ffaecdc8c)
1 /*	$NetBSD: uvm_mmap.c,v 1.61 2001/11/25 06:42:47 chs Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1991, 1993 The Regents of the University of California.
6  * Copyright (c) 1988 University of Utah.
7  *
8  * All rights reserved.
9  *
10  * This code is derived from software contributed to Berkeley by
11  * the Systems Programming Group of the University of Utah Computer
12  * Science Department.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. All advertising materials mentioning features or use of this software
23  *    must display the following acknowledgement:
24  *      This product includes software developed by the Charles D. Cranor,
25  *	Washington University, University of California, Berkeley and
26  *	its contributors.
27  * 4. Neither the name of the University nor the names of its contributors
28  *    may be used to endorse or promote products derived from this software
29  *    without specific prior written permission.
30  *
31  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41  * SUCH DAMAGE.
42  *
43  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
44  *      @(#)vm_mmap.c   8.5 (Berkeley) 5/19/94
45  * from: Id: uvm_mmap.c,v 1.1.2.14 1998/01/05 21:04:26 chuck Exp
46  */
47 
48 /*
49  * uvm_mmap.c: system call interface into VM system, plus kernel vm_mmap
50  * function.
51  */
52 
53 #include <sys/cdefs.h>
54 __KERNEL_RCSID(0, "$NetBSD: uvm_mmap.c,v 1.61 2001/11/25 06:42:47 chs Exp $");
55 
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/file.h>
59 #include <sys/filedesc.h>
60 #include <sys/resourcevar.h>
61 #include <sys/mman.h>
62 #include <sys/mount.h>
63 #include <sys/proc.h>
64 #include <sys/malloc.h>
65 #include <sys/vnode.h>
66 #include <sys/conf.h>
67 #include <sys/stat.h>
68 
69 #include <miscfs/specfs/specdev.h>
70 
71 #include <sys/syscallargs.h>
72 
73 #include <uvm/uvm.h>
74 #include <uvm/uvm_device.h>
75 
76 
77 /*
78  * unimplemented VM system calls:
79  */
80 
81 /*
82  * sys_sbrk: sbrk system call.
83  */
84 
85 /* ARGSUSED */
86 int
87 sys_sbrk(p, v, retval)
88 	struct proc *p;
89 	void *v;
90 	register_t *retval;
91 {
92 #if 0
93 	struct sys_sbrk_args /* {
94 		syscallarg(intptr_t) incr;
95 	} */ *uap = v;
96 #endif
97 
98 	return (ENOSYS);
99 }
100 
101 /*
102  * sys_sstk: sstk system call.
103  */
104 
105 /* ARGSUSED */
106 int
107 sys_sstk(p, v, retval)
108 	struct proc *p;
109 	void *v;
110 	register_t *retval;
111 {
112 #if 0
113 	struct sys_sstk_args /* {
114 		syscallarg(int) incr;
115 	} */ *uap = v;
116 #endif
117 
118 	return (ENOSYS);
119 }
120 
121 /*
122  * sys_mincore: determine if pages are in core or not.
123  */
124 
125 /* ARGSUSED */
126 int
127 sys_mincore(p, v, retval)
128 	struct proc *p;
129 	void *v;
130 	register_t *retval;
131 {
132 	struct sys_mincore_args /* {
133 		syscallarg(void *) addr;
134 		syscallarg(size_t) len;
135 		syscallarg(char *) vec;
136 	} */ *uap = v;
137 	struct vm_page *pg;
138 	char *vec, pgi;
139 	struct uvm_object *uobj;
140 	struct vm_amap *amap;
141 	struct vm_anon *anon;
142 	struct vm_map_entry *entry;
143 	vaddr_t start, end, lim;
144 	struct vm_map *map;
145 	vsize_t len;
146 	int error = 0, npgs;
147 
148 	map = &p->p_vmspace->vm_map;
149 
150 	start = (vaddr_t)SCARG(uap, addr);
151 	len = SCARG(uap, len);
152 	vec = SCARG(uap, vec);
153 
154 	if (start & PAGE_MASK)
155 		return (EINVAL);
156 	len = round_page(len);
157 	end = start + len;
158 	if (end <= start)
159 		return (EINVAL);
160 
161 	npgs = len >> PAGE_SHIFT;
162 
163 	if (uvm_useracc(vec, npgs, B_WRITE) == FALSE)
164 		return (EFAULT);
165 
166 	/*
167 	 * Lock down vec, so our returned status isn't outdated by
168 	 * storing the status byte for a page.
169 	 */
170 
171 	uvm_vslock(p, vec, npgs, VM_PROT_WRITE);
172 	vm_map_lock_read(map);
173 
174 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
175 		error = ENOMEM;
176 		goto out;
177 	}
178 
179 	for (/* nothing */;
180 	     entry != &map->header && entry->start < end;
181 	     entry = entry->next) {
182 		KASSERT(!UVM_ET_ISSUBMAP(entry));
183 		KASSERT(start >= entry->start);
184 
185 		/* Make sure there are no holes. */
186 		if (entry->end < end &&
187 		     (entry->next == &map->header ||
188 		      entry->next->start > entry->end)) {
189 			error = ENOMEM;
190 			goto out;
191 		}
192 
193 		lim = end < entry->end ? end : entry->end;
194 
195 		/*
196 		 * Special case for objects with no "real" pages.  Those
197 		 * are always considered resident (mapped devices).
198 		 */
199 
200 		if (UVM_ET_ISOBJ(entry)) {
201 			KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj));
202 			if (!UVM_OBJ_IS_VNODE(entry->object.uvm_obj)) {
203 				for (/* nothing */; start < lim;
204 				     start += PAGE_SIZE, vec++)
205 					subyte(vec, 1);
206 				continue;
207 			}
208 		}
209 
210 		amap = entry->aref.ar_amap;	/* top layer */
211 		uobj = entry->object.uvm_obj;	/* bottom layer */
212 
213 		if (amap != NULL)
214 			amap_lock(amap);
215 		if (uobj != NULL)
216 			simple_lock(&uobj->vmobjlock);
217 
218 		for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) {
219 			pgi = 0;
220 			if (amap != NULL) {
221 				/* Check the top layer first. */
222 				anon = amap_lookup(&entry->aref,
223 				    start - entry->start);
224 				/* Don't need to lock anon here. */
225 				if (anon != NULL && anon->u.an_page != NULL) {
226 
227 					/*
228 					 * Anon has the page for this entry
229 					 * offset.
230 					 */
231 
232 					pgi = 1;
233 				}
234 			}
235 			if (uobj != NULL && pgi == 0) {
236 				/* Check the bottom layer. */
237 				pg = uvm_pagelookup(uobj,
238 				    entry->offset + (start - entry->start));
239 				if (pg != NULL) {
240 
241 					/*
242 					 * Object has the page for this entry
243 					 * offset.
244 					 */
245 
246 					pgi = 1;
247 				}
248 			}
249 			(void) subyte(vec, pgi);
250 		}
251 		if (uobj != NULL)
252 			simple_unlock(&uobj->vmobjlock);
253 		if (amap != NULL)
254 			amap_unlock(amap);
255 	}
256 
257  out:
258 	vm_map_unlock_read(map);
259 	uvm_vsunlock(p, SCARG(uap, vec), npgs);
260 	return (error);
261 }
262 
263 /*
264  * sys_mmap: mmap system call.
265  *
266  * => file offest and address may not be page aligned
267  *    - if MAP_FIXED, offset and address must have remainder mod PAGE_SIZE
268  *    - if address isn't page aligned the mapping starts at trunc_page(addr)
269  *      and the return value is adjusted up by the page offset.
270  */
271 
272 int
273 sys_mmap(p, v, retval)
274 	struct proc *p;
275 	void *v;
276 	register_t *retval;
277 {
278 	struct sys_mmap_args /* {
279 		syscallarg(caddr_t) addr;
280 		syscallarg(size_t) len;
281 		syscallarg(int) prot;
282 		syscallarg(int) flags;
283 		syscallarg(int) fd;
284 		syscallarg(long) pad;
285 		syscallarg(off_t) pos;
286 	} */ *uap = v;
287 	vaddr_t addr;
288 	struct vattr va;
289 	off_t pos;
290 	vsize_t size, pageoff;
291 	vm_prot_t prot, maxprot;
292 	int flags, fd;
293 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
294 	struct filedesc *fdp = p->p_fd;
295 	struct file *fp;
296 	struct vnode *vp;
297 	void *handle;
298 	int error;
299 
300 	/*
301 	 * first, extract syscall args from the uap.
302 	 */
303 
304 	addr = (vaddr_t)SCARG(uap, addr);
305 	size = (vsize_t)SCARG(uap, len);
306 	prot = SCARG(uap, prot) & VM_PROT_ALL;
307 	flags = SCARG(uap, flags);
308 	fd = SCARG(uap, fd);
309 	pos = SCARG(uap, pos);
310 
311 	/*
312 	 * Fixup the old deprecated MAP_COPY into MAP_PRIVATE, and
313 	 * validate the flags.
314 	 */
315 	if (flags & MAP_COPY)
316 		flags = (flags & ~MAP_COPY) | MAP_PRIVATE;
317 	if ((flags & (MAP_SHARED|MAP_PRIVATE)) == (MAP_SHARED|MAP_PRIVATE))
318 		return (EINVAL);
319 
320 	/*
321 	 * align file position and save offset.  adjust size.
322 	 */
323 
324 	pageoff = (pos & PAGE_MASK);
325 	pos  -= pageoff;
326 	size += pageoff;			/* add offset */
327 	size = (vsize_t)round_page(size);	/* round up */
328 	if ((ssize_t) size < 0)
329 		return (EINVAL);			/* don't allow wrap */
330 
331 	/*
332 	 * now check (MAP_FIXED) or get (!MAP_FIXED) the "addr"
333 	 */
334 
335 	if (flags & MAP_FIXED) {
336 
337 		/* ensure address and file offset are aligned properly */
338 		addr -= pageoff;
339 		if (addr & PAGE_MASK)
340 			return (EINVAL);
341 
342 		if (VM_MAXUSER_ADDRESS > 0 &&
343 		    (addr + size) > VM_MAXUSER_ADDRESS)
344 			return (EINVAL);
345 		if (vm_min_address > 0 && addr < vm_min_address)
346 			return (EINVAL);
347 		if (addr > addr + size)
348 			return (EINVAL);		/* no wrapping! */
349 
350 	} else {
351 
352 		/*
353 		 * not fixed: make sure we skip over the largest possible heap.
354 		 * we will refine our guess later (e.g. to account for VAC, etc)
355 		 */
356 
357 		addr = MAX(addr, round_page((vaddr_t)p->p_vmspace->vm_daddr +
358 					    MAXDSIZ));
359 	}
360 
361 	/*
362 	 * check for file mappings (i.e. not anonymous) and verify file.
363 	 */
364 
365 	if ((flags & MAP_ANON) == 0) {
366 
367 		if ((fp = fd_getfile(fdp, fd)) == NULL)
368 			return (EBADF);
369 
370 		if (fp->f_type != DTYPE_VNODE)
371 			return (ENODEV);		/* only mmap vnodes! */
372 		vp = (struct vnode *)fp->f_data;	/* convert to vnode */
373 
374 		if (vp->v_type != VREG && vp->v_type != VCHR &&
375 		    vp->v_type != VBLK)
376 			return (ENODEV);  /* only REG/CHR/BLK support mmap */
377 
378 		if (vp->v_type != VCHR && pos < 0)
379 			return (EINVAL);
380 
381 		if (vp->v_type != VCHR && (pos + size) < pos)
382 			return (EOVERFLOW);		/* no offset wrapping */
383 
384 		/* special case: catch SunOS style /dev/zero */
385 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
386 			flags |= MAP_ANON;
387 			goto is_anon;
388 		}
389 
390 		/*
391 		 * Old programs may not select a specific sharing type, so
392 		 * default to an appropriate one.
393 		 *
394 		 * XXX: how does MAP_ANON fit in the picture?
395 		 */
396 		if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
397 #if defined(DEBUG)
398 			printf("WARNING: defaulted mmap() share type to "
399 			   "%s (pid %d comm %s)\n", vp->v_type == VCHR ?
400 			   "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
401 			    p->p_comm);
402 #endif
403 			if (vp->v_type == VCHR)
404 				flags |= MAP_SHARED;	/* for a device */
405 			else
406 				flags |= MAP_PRIVATE;	/* for a file */
407 		}
408 
409 		/*
410 		 * MAP_PRIVATE device mappings don't make sense (and aren't
411 		 * supported anyway).  However, some programs rely on this,
412 		 * so just change it to MAP_SHARED.
413 		 */
414 		if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
415 			flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
416 		}
417 
418 		/*
419 		 * now check protection
420 		 */
421 
422 		maxprot = VM_PROT_EXECUTE;
423 
424 		/* check read access */
425 		if (fp->f_flag & FREAD)
426 			maxprot |= VM_PROT_READ;
427 		else if (prot & PROT_READ)
428 			return (EACCES);
429 
430 		/* check write access, shared case first */
431 		if (flags & MAP_SHARED) {
432 			/*
433 			 * if the file is writable, only add PROT_WRITE to
434 			 * maxprot if the file is not immutable, append-only.
435 			 * otherwise, if we have asked for PROT_WRITE, return
436 			 * EPERM.
437 			 */
438 			if (fp->f_flag & FWRITE) {
439 				if ((error =
440 				    VOP_GETATTR(vp, &va, p->p_ucred, p)))
441 					return (error);
442 				if ((va.va_flags & (IMMUTABLE|APPEND)) == 0)
443 					maxprot |= VM_PROT_WRITE;
444 				else if (prot & PROT_WRITE)
445 					return (EPERM);
446 			}
447 			else if (prot & PROT_WRITE)
448 				return (EACCES);
449 		} else {
450 			/* MAP_PRIVATE mappings can always write to */
451 			maxprot |= VM_PROT_WRITE;
452 		}
453 		handle = vp;
454 
455 	} else {		/* MAP_ANON case */
456 		/*
457 		 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0?
458 		 */
459 		if (fd != -1)
460 			return (EINVAL);
461 
462  is_anon:		/* label for SunOS style /dev/zero */
463 		handle = NULL;
464 		maxprot = VM_PROT_ALL;
465 		pos = 0;
466 	}
467 
468 	/*
469 	 * XXX (in)sanity check.  We don't do proper datasize checking
470 	 * XXX for anonymous (or private writable) mmap().  However,
471 	 * XXX know that if we're trying to allocate more than the amount
472 	 * XXX remaining under our current data size limit, _that_ should
473 	 * XXX be disallowed.
474 	 */
475 	if ((flags & MAP_ANON) != 0 ||
476 	    ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) {
477 		if (size >
478 		    (p->p_rlimit[RLIMIT_DATA].rlim_cur -
479 		     ctob(p->p_vmspace->vm_dsize))) {
480 			return (ENOMEM);
481 		}
482 	}
483 
484 	/*
485 	 * now let kernel internal function uvm_mmap do the work.
486 	 */
487 
488 	error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot,
489 	    flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
490 
491 	if (error == 0)
492 		/* remember to add offset */
493 		*retval = (register_t)(addr + pageoff);
494 
495 	return (error);
496 }
497 
498 /*
499  * sys___msync13: the msync system call (a front-end for flush)
500  */
501 
502 int
503 sys___msync13(p, v, retval)
504 	struct proc *p;
505 	void *v;
506 	register_t *retval;
507 {
508 	struct sys___msync13_args /* {
509 		syscallarg(caddr_t) addr;
510 		syscallarg(size_t) len;
511 		syscallarg(int) flags;
512 	} */ *uap = v;
513 	vaddr_t addr;
514 	vsize_t size, pageoff;
515 	struct vm_map *map;
516 	int error, rv, flags, uvmflags;
517 
518 	/*
519 	 * extract syscall args from the uap
520 	 */
521 
522 	addr = (vaddr_t)SCARG(uap, addr);
523 	size = (vsize_t)SCARG(uap, len);
524 	flags = SCARG(uap, flags);
525 
526 	/* sanity check flags */
527 	if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 ||
528 			(flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 ||
529 			(flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC))
530 	  return (EINVAL);
531 	if ((flags & (MS_ASYNC | MS_SYNC)) == 0)
532 	  flags |= MS_SYNC;
533 
534 	/*
535 	 * align the address to a page boundary and adjust the size accordingly.
536 	 */
537 
538 	pageoff = (addr & PAGE_MASK);
539 	addr -= pageoff;
540 	size += pageoff;
541 	size = (vsize_t)round_page(size);
542 
543 	/* disallow wrap-around. */
544 	if (addr + size < addr)
545 		return (EINVAL);
546 
547 	/*
548 	 * get map
549 	 */
550 
551 	map = &p->p_vmspace->vm_map;
552 
553 	/*
554 	 * XXXCDC: do we really need this semantic?
555 	 *
556 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
557 	 * pages with the region containing addr".  Unfortunately, we
558 	 * don't really keep track of individual mmaps so we approximate
559 	 * by flushing the range of the map entry containing addr.
560 	 * This can be incorrect if the region splits or is coalesced
561 	 * with a neighbor.
562 	 */
563 
564 	if (size == 0) {
565 		struct vm_map_entry *entry;
566 
567 		vm_map_lock_read(map);
568 		rv = uvm_map_lookup_entry(map, addr, &entry);
569 		if (rv == TRUE) {
570 			addr = entry->start;
571 			size = entry->end - entry->start;
572 		}
573 		vm_map_unlock_read(map);
574 		if (rv == FALSE)
575 			return (EINVAL);
576 	}
577 
578 	/*
579 	 * translate MS_ flags into PGO_ flags
580 	 */
581 
582 	uvmflags = PGO_CLEANIT;
583 	if (flags & MS_INVALIDATE)
584 		uvmflags |= PGO_FREE;
585 	if (flags & MS_SYNC)
586 		uvmflags |= PGO_SYNCIO;
587 	else
588 		uvmflags |= PGO_SYNCIO;	 /* XXXCDC: force sync for now! */
589 
590 	error = uvm_map_clean(map, addr, addr+size, uvmflags);
591 	return error;
592 }
593 
594 /*
595  * sys_munmap: unmap a users memory
596  */
597 
598 int
599 sys_munmap(p, v, retval)
600 	struct proc *p;
601 	void *v;
602 	register_t *retval;
603 {
604 	struct sys_munmap_args /* {
605 		syscallarg(caddr_t) addr;
606 		syscallarg(size_t) len;
607 	} */ *uap = v;
608 	vaddr_t addr;
609 	vsize_t size, pageoff;
610 	struct vm_map *map;
611 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
612 	struct vm_map_entry *dead_entries;
613 
614 	/*
615 	 * get syscall args.
616 	 */
617 
618 	addr = (vaddr_t)SCARG(uap, addr);
619 	size = (vsize_t)SCARG(uap, len);
620 
621 	/*
622 	 * align the address to a page boundary and adjust the size accordingly.
623 	 */
624 
625 	pageoff = (addr & PAGE_MASK);
626 	addr -= pageoff;
627 	size += pageoff;
628 	size = (vsize_t)round_page(size);
629 
630 	if ((int)size < 0)
631 		return (EINVAL);
632 	if (size == 0)
633 		return (0);
634 
635 	/*
636 	 * Check for illegal addresses.  Watch out for address wrap...
637 	 * Note that VM_*_ADDRESS are not constants due to casts (argh).
638 	 */
639 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
640 		return (EINVAL);
641 	if (vm_min_address > 0 && addr < vm_min_address)
642 		return (EINVAL);
643 	if (addr > addr + size)
644 		return (EINVAL);
645 	map = &p->p_vmspace->vm_map;
646 
647 	/*
648 	 * interesting system call semantic: make sure entire range is
649 	 * allocated before allowing an unmap.
650 	 */
651 
652 	vm_map_lock(map);
653 	if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
654 		vm_map_unlock(map);
655 		return (EINVAL);
656 	}
657 	uvm_unmap_remove(map, addr, addr + size, &dead_entries);
658 	vm_map_unlock(map);
659 	if (dead_entries != NULL)
660 		uvm_unmap_detach(dead_entries, 0);
661 	return (0);
662 }
663 
664 /*
665  * sys_mprotect: the mprotect system call
666  */
667 
668 int
669 sys_mprotect(p, v, retval)
670 	struct proc *p;
671 	void *v;
672 	register_t *retval;
673 {
674 	struct sys_mprotect_args /* {
675 		syscallarg(caddr_t) addr;
676 		syscallarg(int) len;
677 		syscallarg(int) prot;
678 	} */ *uap = v;
679 	vaddr_t addr;
680 	vsize_t size, pageoff;
681 	vm_prot_t prot;
682 	int error;
683 
684 	/*
685 	 * extract syscall args from uap
686 	 */
687 
688 	addr = (vaddr_t)SCARG(uap, addr);
689 	size = (vsize_t)SCARG(uap, len);
690 	prot = SCARG(uap, prot) & VM_PROT_ALL;
691 
692 	/*
693 	 * align the address to a page boundary and adjust the size accordingly.
694 	 */
695 
696 	pageoff = (addr & PAGE_MASK);
697 	addr -= pageoff;
698 	size += pageoff;
699 	size = (vsize_t)round_page(size);
700 
701 	if ((int)size < 0)
702 		return (EINVAL);
703 	error = uvm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
704 				FALSE);
705 	return error;
706 }
707 
708 /*
709  * sys_minherit: the minherit system call
710  */
711 
712 int
713 sys_minherit(p, v, retval)
714 	struct proc *p;
715 	void *v;
716 	register_t *retval;
717 {
718 	struct sys_minherit_args /* {
719 		syscallarg(caddr_t) addr;
720 		syscallarg(int) len;
721 		syscallarg(int) inherit;
722 	} */ *uap = v;
723 	vaddr_t addr;
724 	vsize_t size, pageoff;
725 	vm_inherit_t inherit;
726 	int error;
727 
728 	addr = (vaddr_t)SCARG(uap, addr);
729 	size = (vsize_t)SCARG(uap, len);
730 	inherit = SCARG(uap, inherit);
731 
732 	/*
733 	 * align the address to a page boundary and adjust the size accordingly.
734 	 */
735 
736 	pageoff = (addr & PAGE_MASK);
737 	addr -= pageoff;
738 	size += pageoff;
739 	size = (vsize_t)round_page(size);
740 
741 	if ((int)size < 0)
742 		return (EINVAL);
743 	error = uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr + size,
744 				inherit);
745 	return error;
746 }
747 
748 /*
749  * sys_madvise: give advice about memory usage.
750  */
751 
752 /* ARGSUSED */
753 int
754 sys_madvise(p, v, retval)
755 	struct proc *p;
756 	void *v;
757 	register_t *retval;
758 {
759 	struct sys_madvise_args /* {
760 		syscallarg(caddr_t) addr;
761 		syscallarg(size_t) len;
762 		syscallarg(int) behav;
763 	} */ *uap = v;
764 	vaddr_t addr;
765 	vsize_t size, pageoff;
766 	int advice, error;
767 
768 	addr = (vaddr_t)SCARG(uap, addr);
769 	size = (vsize_t)SCARG(uap, len);
770 	advice = SCARG(uap, behav);
771 
772 	/*
773 	 * align the address to a page boundary, and adjust the size accordingly
774 	 */
775 
776 	pageoff = (addr & PAGE_MASK);
777 	addr -= pageoff;
778 	size += pageoff;
779 	size = (vsize_t)round_page(size);
780 
781 	if ((ssize_t)size <= 0)
782 		return (EINVAL);
783 
784 	switch (advice) {
785 	case MADV_NORMAL:
786 	case MADV_RANDOM:
787 	case MADV_SEQUENTIAL:
788 		error = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size,
789 		    advice);
790 		break;
791 
792 	case MADV_WILLNEED:
793 
794 		/*
795 		 * Activate all these pages, pre-faulting them in if
796 		 * necessary.
797 		 */
798 		/*
799 		 * XXX IMPLEMENT ME.
800 		 * Should invent a "weak" mode for uvm_fault()
801 		 * which would only do the PGO_LOCKED pgo_get().
802 		 */
803 
804 		return (0);
805 
806 	case MADV_DONTNEED:
807 
808 		/*
809 		 * Deactivate all these pages.  We don't need them
810 		 * any more.  We don't, however, toss the data in
811 		 * the pages.
812 		 */
813 
814 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
815 		    PGO_DEACTIVATE);
816 		break;
817 
818 	case MADV_FREE:
819 
820 		/*
821 		 * These pages contain no valid data, and may be
822 		 * garbage-collected.  Toss all resources, including
823 		 * any swap space in use.
824 		 */
825 
826 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
827 		    PGO_FREE);
828 		break;
829 
830 	case MADV_SPACEAVAIL:
831 
832 		/*
833 		 * XXXMRG What is this?  I think it's:
834 		 *
835 		 *	Ensure that we have allocated backing-store
836 		 *	for these pages.
837 		 *
838 		 * This is going to require changes to the page daemon,
839 		 * as it will free swap space allocated to pages in core.
840 		 * There's also what to do for device/file/anonymous memory.
841 		 */
842 
843 		return (EINVAL);
844 
845 	default:
846 		return (EINVAL);
847 	}
848 
849 	return error;
850 }
851 
852 /*
853  * sys_mlock: memory lock
854  */
855 
856 int
857 sys_mlock(p, v, retval)
858 	struct proc *p;
859 	void *v;
860 	register_t *retval;
861 {
862 	struct sys_mlock_args /* {
863 		syscallarg(const void *) addr;
864 		syscallarg(size_t) len;
865 	} */ *uap = v;
866 	vaddr_t addr;
867 	vsize_t size, pageoff;
868 	int error;
869 
870 	/*
871 	 * extract syscall args from uap
872 	 */
873 
874 	addr = (vaddr_t)SCARG(uap, addr);
875 	size = (vsize_t)SCARG(uap, len);
876 
877 	/*
878 	 * align the address to a page boundary and adjust the size accordingly
879 	 */
880 
881 	pageoff = (addr & PAGE_MASK);
882 	addr -= pageoff;
883 	size += pageoff;
884 	size = (vsize_t)round_page(size);
885 
886 	/* disallow wrap-around. */
887 	if (addr + size < addr)
888 		return (EINVAL);
889 
890 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
891 		return (EAGAIN);
892 
893 #ifdef pmap_wired_count
894 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
895 			p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
896 		return (EAGAIN);
897 #else
898 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
899 		return (error);
900 #endif
901 
902 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
903 	    0);
904 	return error;
905 }
906 
907 /*
908  * sys_munlock: unlock wired pages
909  */
910 
911 int
912 sys_munlock(p, v, retval)
913 	struct proc *p;
914 	void *v;
915 	register_t *retval;
916 {
917 	struct sys_munlock_args /* {
918 		syscallarg(const void *) addr;
919 		syscallarg(size_t) len;
920 	} */ *uap = v;
921 	vaddr_t addr;
922 	vsize_t size, pageoff;
923 	int error;
924 
925 	/*
926 	 * extract syscall args from uap
927 	 */
928 
929 	addr = (vaddr_t)SCARG(uap, addr);
930 	size = (vsize_t)SCARG(uap, len);
931 
932 	/*
933 	 * align the address to a page boundary, and adjust the size accordingly
934 	 */
935 
936 	pageoff = (addr & PAGE_MASK);
937 	addr -= pageoff;
938 	size += pageoff;
939 	size = (vsize_t)round_page(size);
940 
941 	/* disallow wrap-around. */
942 	if (addr + size < addr)
943 		return (EINVAL);
944 
945 #ifndef pmap_wired_count
946 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
947 		return (error);
948 #endif
949 
950 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
951 	    0);
952 	return error;
953 }
954 
955 /*
956  * sys_mlockall: lock all pages mapped into an address space.
957  */
958 
959 int
960 sys_mlockall(p, v, retval)
961 	struct proc *p;
962 	void *v;
963 	register_t *retval;
964 {
965 	struct sys_mlockall_args /* {
966 		syscallarg(int) flags;
967 	} */ *uap = v;
968 	int error, flags;
969 
970 	flags = SCARG(uap, flags);
971 
972 	if (flags == 0 ||
973 	    (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
974 		return (EINVAL);
975 
976 #ifndef pmap_wired_count
977 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
978 		return (error);
979 #endif
980 
981 	error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
982 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
983 	return (error);
984 }
985 
986 /*
987  * sys_munlockall: unlock all pages mapped into an address space.
988  */
989 
990 int
991 sys_munlockall(p, v, retval)
992 	struct proc *p;
993 	void *v;
994 	register_t *retval;
995 {
996 
997 	(void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
998 	return (0);
999 }
1000 
1001 /*
1002  * uvm_mmap: internal version of mmap
1003  *
1004  * - used by sys_mmap and various framebuffers
1005  * - handle is a vnode pointer or NULL for MAP_ANON
1006  * - caller must page-align the file offset
1007  */
1008 
1009 int
1010 uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit)
1011 	struct vm_map *map;
1012 	vaddr_t *addr;
1013 	vsize_t size;
1014 	vm_prot_t prot, maxprot;
1015 	int flags;
1016 	void *handle;
1017 	voff_t foff;
1018 	vsize_t locklimit;
1019 {
1020 	struct uvm_object *uobj;
1021 	struct vnode *vp;
1022 	int error;
1023 	int advice = UVM_ADV_NORMAL;
1024 	uvm_flag_t uvmflag = 0;
1025 
1026 	/*
1027 	 * check params
1028 	 */
1029 
1030 	if (size == 0)
1031 		return(0);
1032 	if (foff & PAGE_MASK)
1033 		return(EINVAL);
1034 	if ((prot & maxprot) != prot)
1035 		return(EINVAL);
1036 
1037 	/*
1038 	 * for non-fixed mappings, round off the suggested address.
1039 	 * for fixed mappings, check alignment and zap old mappings.
1040 	 */
1041 
1042 	if ((flags & MAP_FIXED) == 0) {
1043 		*addr = round_page(*addr);
1044 	} else {
1045 		if (*addr & PAGE_MASK)
1046 			return(EINVAL);
1047 		uvmflag |= UVM_FLAG_FIXED;
1048 		(void) uvm_unmap(map, *addr, *addr + size);
1049 	}
1050 
1051 	/*
1052 	 * handle anon vs. non-anon mappings.   for non-anon mappings attach
1053 	 * to underlying vm object.
1054 	 */
1055 
1056 	if (flags & MAP_ANON) {
1057 		foff = UVM_UNKNOWN_OFFSET;
1058 		uobj = NULL;
1059 		if ((flags & MAP_SHARED) == 0)
1060 			/* XXX: defer amap create */
1061 			uvmflag |= UVM_FLAG_COPYONW;
1062 		else
1063 			/* shared: create amap now */
1064 			uvmflag |= UVM_FLAG_OVERLAY;
1065 
1066 	} else {
1067 		vp = (struct vnode *)handle;
1068 
1069 		/*
1070 		 * Don't allow mmap for EXEC if the file system
1071 		 * is mounted NOEXEC.
1072 		 */
1073 		if ((prot & PROT_EXEC) != 0 &&
1074 		    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0)
1075 			return (EACCES);
1076 
1077 		if (vp->v_type != VCHR) {
1078 			error = VOP_MMAP(vp, 0, curproc->p_ucred, curproc);
1079 			if (error) {
1080 				return error;
1081 			}
1082 
1083 			uobj = uvn_attach((void *)vp, (flags & MAP_SHARED) ?
1084 			   maxprot : (maxprot & ~VM_PROT_WRITE));
1085 
1086 			/* XXX for now, attach doesn't gain a ref */
1087 			VREF(vp);
1088 
1089 			/*
1090 			 * If the vnode is being mapped with PROT_EXEC,
1091 			 * then mark it as text.
1092 			 */
1093 			if (prot & PROT_EXEC)
1094 				vn_markexec(vp);
1095 		} else {
1096 			uobj = udv_attach((void *) &vp->v_rdev,
1097 			    (flags & MAP_SHARED) ? maxprot :
1098 			    (maxprot & ~VM_PROT_WRITE), foff, size);
1099 			/*
1100 			 * XXX Some devices don't like to be mapped with
1101 			 * XXX PROT_EXEC, but we don't really have a
1102 			 * XXX better way of handling this, right now
1103 			 */
1104 			if (uobj == NULL && (prot & PROT_EXEC) == 0) {
1105 				maxprot &= ~VM_PROT_EXECUTE;
1106 				uobj = udv_attach((void *)&vp->v_rdev,
1107 				    (flags & MAP_SHARED) ? maxprot :
1108 				    (maxprot & ~VM_PROT_WRITE), foff, size);
1109 			}
1110 			advice = UVM_ADV_RANDOM;
1111 		}
1112 		if (uobj == NULL)
1113 			return((vp->v_type == VREG) ? ENOMEM : EINVAL);
1114 		if ((flags & MAP_SHARED) == 0)
1115 			uvmflag |= UVM_FLAG_COPYONW;
1116 	}
1117 
1118 	uvmflag = UVM_MAPFLAG(prot, maxprot,
1119 			(flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
1120 			advice, uvmflag);
1121 	error = uvm_map(map, addr, size, uobj, foff, 0, uvmflag);
1122 	if (error) {
1123 		if (uobj)
1124 			uobj->pgops->pgo_detach(uobj);
1125 		return error;
1126 	}
1127 
1128 	/*
1129 	 * POSIX 1003.1b -- if our address space was configured
1130 	 * to lock all future mappings, wire the one we just made.
1131 	 */
1132 
1133 	if (prot == VM_PROT_NONE) {
1134 
1135 		/*
1136 		 * No more work to do in this case.
1137 		 */
1138 
1139 		return (0);
1140 	}
1141 	vm_map_lock(map);
1142 	if (map->flags & VM_MAP_WIREFUTURE) {
1143 		if ((atop(size) + uvmexp.wired) > uvmexp.wiredmax
1144 #ifdef pmap_wired_count
1145 		    || (locklimit != 0 && (size +
1146 		    ptoa(pmap_wired_count(vm_map_pmap(map)))) >
1147 			locklimit)
1148 #endif
1149 		) {
1150 			vm_map_unlock(map);
1151 			uvm_unmap(map, *addr, *addr + size);
1152 			return ENOMEM;
1153 		}
1154 
1155 		/*
1156 		 * uvm_map_pageable() always returns the map unlocked.
1157 		 */
1158 
1159 		error = uvm_map_pageable(map, *addr, *addr + size,
1160 					 FALSE, UVM_LK_ENTER);
1161 		if (error) {
1162 			uvm_unmap(map, *addr, *addr + size);
1163 			return error;
1164 		}
1165 		return (0);
1166 	}
1167 	vm_map_unlock(map);
1168 	return 0;
1169 }
1170