xref: /netbsd-src/sys/uvm/uvm_mmap.c (revision 06be8101a16cc95f40783b3cb7afd12112103a9a)
1 /*	$NetBSD: uvm_mmap.c,v 1.60 2001/11/10 07:37:00 lukem 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 1.60 2001/11/10 07:37:00 lukem 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 == VREG && (pos + size) < pos)
379 			return (EOVERFLOW);		/* no offset wrapping */
380 
381 		/* special case: catch SunOS style /dev/zero */
382 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
383 			flags |= MAP_ANON;
384 			goto is_anon;
385 		}
386 
387 		/*
388 		 * Old programs may not select a specific sharing type, so
389 		 * default to an appropriate one.
390 		 *
391 		 * XXX: how does MAP_ANON fit in the picture?
392 		 */
393 		if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
394 #if defined(DEBUG)
395 			printf("WARNING: defaulted mmap() share type to "
396 			   "%s (pid %d comm %s)\n", vp->v_type == VCHR ?
397 			   "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
398 			    p->p_comm);
399 #endif
400 			if (vp->v_type == VCHR)
401 				flags |= MAP_SHARED;	/* for a device */
402 			else
403 				flags |= MAP_PRIVATE;	/* for a file */
404 		}
405 
406 		/*
407 		 * MAP_PRIVATE device mappings don't make sense (and aren't
408 		 * supported anyway).  However, some programs rely on this,
409 		 * so just change it to MAP_SHARED.
410 		 */
411 		if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
412 			flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
413 		}
414 
415 		/*
416 		 * now check protection
417 		 */
418 
419 		maxprot = VM_PROT_EXECUTE;
420 
421 		/* check read access */
422 		if (fp->f_flag & FREAD)
423 			maxprot |= VM_PROT_READ;
424 		else if (prot & PROT_READ)
425 			return (EACCES);
426 
427 		/* check write access, shared case first */
428 		if (flags & MAP_SHARED) {
429 			/*
430 			 * if the file is writable, only add PROT_WRITE to
431 			 * maxprot if the file is not immutable, append-only.
432 			 * otherwise, if we have asked for PROT_WRITE, return
433 			 * EPERM.
434 			 */
435 			if (fp->f_flag & FWRITE) {
436 				if ((error =
437 				    VOP_GETATTR(vp, &va, p->p_ucred, p)))
438 					return (error);
439 				if ((va.va_flags & (IMMUTABLE|APPEND)) == 0)
440 					maxprot |= VM_PROT_WRITE;
441 				else if (prot & PROT_WRITE)
442 					return (EPERM);
443 			}
444 			else if (prot & PROT_WRITE)
445 				return (EACCES);
446 		} else {
447 			/* MAP_PRIVATE mappings can always write to */
448 			maxprot |= VM_PROT_WRITE;
449 		}
450 		handle = vp;
451 
452 	} else {		/* MAP_ANON case */
453 		/*
454 		 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0?
455 		 */
456 		if (fd != -1)
457 			return (EINVAL);
458 
459  is_anon:		/* label for SunOS style /dev/zero */
460 		handle = NULL;
461 		maxprot = VM_PROT_ALL;
462 		pos = 0;
463 	}
464 
465 	/*
466 	 * XXX (in)sanity check.  We don't do proper datasize checking
467 	 * XXX for anonymous (or private writable) mmap().  However,
468 	 * XXX know that if we're trying to allocate more than the amount
469 	 * XXX remaining under our current data size limit, _that_ should
470 	 * XXX be disallowed.
471 	 */
472 	if ((flags & MAP_ANON) != 0 ||
473 	    ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) {
474 		if (size >
475 		    (p->p_rlimit[RLIMIT_DATA].rlim_cur -
476 		     ctob(p->p_vmspace->vm_dsize))) {
477 			return (ENOMEM);
478 		}
479 	}
480 
481 	/*
482 	 * now let kernel internal function uvm_mmap do the work.
483 	 */
484 
485 	error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot,
486 	    flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
487 
488 	if (error == 0)
489 		/* remember to add offset */
490 		*retval = (register_t)(addr + pageoff);
491 
492 	return (error);
493 }
494 
495 /*
496  * sys___msync13: the msync system call (a front-end for flush)
497  */
498 
499 int
500 sys___msync13(p, v, retval)
501 	struct proc *p;
502 	void *v;
503 	register_t *retval;
504 {
505 	struct sys___msync13_args /* {
506 		syscallarg(caddr_t) addr;
507 		syscallarg(size_t) len;
508 		syscallarg(int) flags;
509 	} */ *uap = v;
510 	vaddr_t addr;
511 	vsize_t size, pageoff;
512 	struct vm_map *map;
513 	int error, rv, flags, uvmflags;
514 
515 	/*
516 	 * extract syscall args from the uap
517 	 */
518 
519 	addr = (vaddr_t)SCARG(uap, addr);
520 	size = (vsize_t)SCARG(uap, len);
521 	flags = SCARG(uap, flags);
522 
523 	/* sanity check flags */
524 	if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 ||
525 			(flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 ||
526 			(flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC))
527 	  return (EINVAL);
528 	if ((flags & (MS_ASYNC | MS_SYNC)) == 0)
529 	  flags |= MS_SYNC;
530 
531 	/*
532 	 * align the address to a page boundary and adjust the size accordingly.
533 	 */
534 
535 	pageoff = (addr & PAGE_MASK);
536 	addr -= pageoff;
537 	size += pageoff;
538 	size = (vsize_t)round_page(size);
539 
540 	/* disallow wrap-around. */
541 	if (addr + size < addr)
542 		return (EINVAL);
543 
544 	/*
545 	 * get map
546 	 */
547 
548 	map = &p->p_vmspace->vm_map;
549 
550 	/*
551 	 * XXXCDC: do we really need this semantic?
552 	 *
553 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
554 	 * pages with the region containing addr".  Unfortunately, we
555 	 * don't really keep track of individual mmaps so we approximate
556 	 * by flushing the range of the map entry containing addr.
557 	 * This can be incorrect if the region splits or is coalesced
558 	 * with a neighbor.
559 	 */
560 
561 	if (size == 0) {
562 		struct vm_map_entry *entry;
563 
564 		vm_map_lock_read(map);
565 		rv = uvm_map_lookup_entry(map, addr, &entry);
566 		if (rv == TRUE) {
567 			addr = entry->start;
568 			size = entry->end - entry->start;
569 		}
570 		vm_map_unlock_read(map);
571 		if (rv == FALSE)
572 			return (EINVAL);
573 	}
574 
575 	/*
576 	 * translate MS_ flags into PGO_ flags
577 	 */
578 
579 	uvmflags = PGO_CLEANIT;
580 	if (flags & MS_INVALIDATE)
581 		uvmflags |= PGO_FREE;
582 	if (flags & MS_SYNC)
583 		uvmflags |= PGO_SYNCIO;
584 	else
585 		uvmflags |= PGO_SYNCIO;	 /* XXXCDC: force sync for now! */
586 
587 	error = uvm_map_clean(map, addr, addr+size, uvmflags);
588 	return error;
589 }
590 
591 /*
592  * sys_munmap: unmap a users memory
593  */
594 
595 int
596 sys_munmap(p, v, retval)
597 	struct proc *p;
598 	void *v;
599 	register_t *retval;
600 {
601 	struct sys_munmap_args /* {
602 		syscallarg(caddr_t) addr;
603 		syscallarg(size_t) len;
604 	} */ *uap = v;
605 	vaddr_t addr;
606 	vsize_t size, pageoff;
607 	struct vm_map *map;
608 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
609 	struct vm_map_entry *dead_entries;
610 
611 	/*
612 	 * get syscall args.
613 	 */
614 
615 	addr = (vaddr_t)SCARG(uap, addr);
616 	size = (vsize_t)SCARG(uap, len);
617 
618 	/*
619 	 * align the address to a page boundary and adjust the size accordingly.
620 	 */
621 
622 	pageoff = (addr & PAGE_MASK);
623 	addr -= pageoff;
624 	size += pageoff;
625 	size = (vsize_t)round_page(size);
626 
627 	if ((int)size < 0)
628 		return (EINVAL);
629 	if (size == 0)
630 		return (0);
631 
632 	/*
633 	 * Check for illegal addresses.  Watch out for address wrap...
634 	 * Note that VM_*_ADDRESS are not constants due to casts (argh).
635 	 */
636 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
637 		return (EINVAL);
638 	if (vm_min_address > 0 && addr < vm_min_address)
639 		return (EINVAL);
640 	if (addr > addr + size)
641 		return (EINVAL);
642 	map = &p->p_vmspace->vm_map;
643 
644 	/*
645 	 * interesting system call semantic: make sure entire range is
646 	 * allocated before allowing an unmap.
647 	 */
648 
649 	vm_map_lock(map);
650 	if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
651 		vm_map_unlock(map);
652 		return (EINVAL);
653 	}
654 	uvm_unmap_remove(map, addr, addr + size, &dead_entries);
655 	vm_map_unlock(map);
656 	if (dead_entries != NULL)
657 		uvm_unmap_detach(dead_entries, 0);
658 	return (0);
659 }
660 
661 /*
662  * sys_mprotect: the mprotect system call
663  */
664 
665 int
666 sys_mprotect(p, v, retval)
667 	struct proc *p;
668 	void *v;
669 	register_t *retval;
670 {
671 	struct sys_mprotect_args /* {
672 		syscallarg(caddr_t) addr;
673 		syscallarg(int) len;
674 		syscallarg(int) prot;
675 	} */ *uap = v;
676 	vaddr_t addr;
677 	vsize_t size, pageoff;
678 	vm_prot_t prot;
679 	int error;
680 
681 	/*
682 	 * extract syscall args from uap
683 	 */
684 
685 	addr = (vaddr_t)SCARG(uap, addr);
686 	size = (vsize_t)SCARG(uap, len);
687 	prot = SCARG(uap, prot) & VM_PROT_ALL;
688 
689 	/*
690 	 * align the address to a page boundary and adjust the size accordingly.
691 	 */
692 
693 	pageoff = (addr & PAGE_MASK);
694 	addr -= pageoff;
695 	size += pageoff;
696 	size = (vsize_t)round_page(size);
697 
698 	if ((int)size < 0)
699 		return (EINVAL);
700 	error = uvm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
701 				FALSE);
702 	return error;
703 }
704 
705 /*
706  * sys_minherit: the minherit system call
707  */
708 
709 int
710 sys_minherit(p, v, retval)
711 	struct proc *p;
712 	void *v;
713 	register_t *retval;
714 {
715 	struct sys_minherit_args /* {
716 		syscallarg(caddr_t) addr;
717 		syscallarg(int) len;
718 		syscallarg(int) inherit;
719 	} */ *uap = v;
720 	vaddr_t addr;
721 	vsize_t size, pageoff;
722 	vm_inherit_t inherit;
723 	int error;
724 
725 	addr = (vaddr_t)SCARG(uap, addr);
726 	size = (vsize_t)SCARG(uap, len);
727 	inherit = SCARG(uap, inherit);
728 
729 	/*
730 	 * align the address to a page boundary and adjust the size accordingly.
731 	 */
732 
733 	pageoff = (addr & PAGE_MASK);
734 	addr -= pageoff;
735 	size += pageoff;
736 	size = (vsize_t)round_page(size);
737 
738 	if ((int)size < 0)
739 		return (EINVAL);
740 	error = uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr + size,
741 				inherit);
742 	return error;
743 }
744 
745 /*
746  * sys_madvise: give advice about memory usage.
747  */
748 
749 /* ARGSUSED */
750 int
751 sys_madvise(p, v, retval)
752 	struct proc *p;
753 	void *v;
754 	register_t *retval;
755 {
756 	struct sys_madvise_args /* {
757 		syscallarg(caddr_t) addr;
758 		syscallarg(size_t) len;
759 		syscallarg(int) behav;
760 	} */ *uap = v;
761 	vaddr_t addr;
762 	vsize_t size, pageoff;
763 	int advice, error;
764 
765 	addr = (vaddr_t)SCARG(uap, addr);
766 	size = (vsize_t)SCARG(uap, len);
767 	advice = SCARG(uap, behav);
768 
769 	/*
770 	 * align the address to a page boundary, and adjust the size accordingly
771 	 */
772 
773 	pageoff = (addr & PAGE_MASK);
774 	addr -= pageoff;
775 	size += pageoff;
776 	size = (vsize_t)round_page(size);
777 
778 	if ((ssize_t)size <= 0)
779 		return (EINVAL);
780 
781 	switch (advice) {
782 	case MADV_NORMAL:
783 	case MADV_RANDOM:
784 	case MADV_SEQUENTIAL:
785 		error = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size,
786 		    advice);
787 		break;
788 
789 	case MADV_WILLNEED:
790 
791 		/*
792 		 * Activate all these pages, pre-faulting them in if
793 		 * necessary.
794 		 */
795 		/*
796 		 * XXX IMPLEMENT ME.
797 		 * Should invent a "weak" mode for uvm_fault()
798 		 * which would only do the PGO_LOCKED pgo_get().
799 		 */
800 
801 		return (0);
802 
803 	case MADV_DONTNEED:
804 
805 		/*
806 		 * Deactivate all these pages.  We don't need them
807 		 * any more.  We don't, however, toss the data in
808 		 * the pages.
809 		 */
810 
811 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
812 		    PGO_DEACTIVATE);
813 		break;
814 
815 	case MADV_FREE:
816 
817 		/*
818 		 * These pages contain no valid data, and may be
819 		 * garbage-collected.  Toss all resources, including
820 		 * any swap space in use.
821 		 */
822 
823 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
824 		    PGO_FREE);
825 		break;
826 
827 	case MADV_SPACEAVAIL:
828 
829 		/*
830 		 * XXXMRG What is this?  I think it's:
831 		 *
832 		 *	Ensure that we have allocated backing-store
833 		 *	for these pages.
834 		 *
835 		 * This is going to require changes to the page daemon,
836 		 * as it will free swap space allocated to pages in core.
837 		 * There's also what to do for device/file/anonymous memory.
838 		 */
839 
840 		return (EINVAL);
841 
842 	default:
843 		return (EINVAL);
844 	}
845 
846 	return error;
847 }
848 
849 /*
850  * sys_mlock: memory lock
851  */
852 
853 int
854 sys_mlock(p, v, retval)
855 	struct proc *p;
856 	void *v;
857 	register_t *retval;
858 {
859 	struct sys_mlock_args /* {
860 		syscallarg(const void *) addr;
861 		syscallarg(size_t) len;
862 	} */ *uap = v;
863 	vaddr_t addr;
864 	vsize_t size, pageoff;
865 	int error;
866 
867 	/*
868 	 * extract syscall args from uap
869 	 */
870 
871 	addr = (vaddr_t)SCARG(uap, addr);
872 	size = (vsize_t)SCARG(uap, len);
873 
874 	/*
875 	 * align the address to a page boundary and adjust the size accordingly
876 	 */
877 
878 	pageoff = (addr & PAGE_MASK);
879 	addr -= pageoff;
880 	size += pageoff;
881 	size = (vsize_t)round_page(size);
882 
883 	/* disallow wrap-around. */
884 	if (addr + size < addr)
885 		return (EINVAL);
886 
887 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
888 		return (EAGAIN);
889 
890 #ifdef pmap_wired_count
891 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
892 			p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
893 		return (EAGAIN);
894 #else
895 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
896 		return (error);
897 #endif
898 
899 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
900 	    0);
901 	return error;
902 }
903 
904 /*
905  * sys_munlock: unlock wired pages
906  */
907 
908 int
909 sys_munlock(p, v, retval)
910 	struct proc *p;
911 	void *v;
912 	register_t *retval;
913 {
914 	struct sys_munlock_args /* {
915 		syscallarg(const void *) addr;
916 		syscallarg(size_t) len;
917 	} */ *uap = v;
918 	vaddr_t addr;
919 	vsize_t size, pageoff;
920 	int error;
921 
922 	/*
923 	 * extract syscall args from uap
924 	 */
925 
926 	addr = (vaddr_t)SCARG(uap, addr);
927 	size = (vsize_t)SCARG(uap, len);
928 
929 	/*
930 	 * align the address to a page boundary, and adjust the size accordingly
931 	 */
932 
933 	pageoff = (addr & PAGE_MASK);
934 	addr -= pageoff;
935 	size += pageoff;
936 	size = (vsize_t)round_page(size);
937 
938 	/* disallow wrap-around. */
939 	if (addr + size < addr)
940 		return (EINVAL);
941 
942 #ifndef pmap_wired_count
943 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
944 		return (error);
945 #endif
946 
947 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
948 	    0);
949 	return error;
950 }
951 
952 /*
953  * sys_mlockall: lock all pages mapped into an address space.
954  */
955 
956 int
957 sys_mlockall(p, v, retval)
958 	struct proc *p;
959 	void *v;
960 	register_t *retval;
961 {
962 	struct sys_mlockall_args /* {
963 		syscallarg(int) flags;
964 	} */ *uap = v;
965 	int error, flags;
966 
967 	flags = SCARG(uap, flags);
968 
969 	if (flags == 0 ||
970 	    (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
971 		return (EINVAL);
972 
973 #ifndef pmap_wired_count
974 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
975 		return (error);
976 #endif
977 
978 	error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
979 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
980 	return (error);
981 }
982 
983 /*
984  * sys_munlockall: unlock all pages mapped into an address space.
985  */
986 
987 int
988 sys_munlockall(p, v, retval)
989 	struct proc *p;
990 	void *v;
991 	register_t *retval;
992 {
993 
994 	(void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
995 	return (0);
996 }
997 
998 /*
999  * uvm_mmap: internal version of mmap
1000  *
1001  * - used by sys_mmap and various framebuffers
1002  * - handle is a vnode pointer or NULL for MAP_ANON
1003  * - caller must page-align the file offset
1004  */
1005 
1006 int
1007 uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit)
1008 	struct vm_map *map;
1009 	vaddr_t *addr;
1010 	vsize_t size;
1011 	vm_prot_t prot, maxprot;
1012 	int flags;
1013 	void *handle;
1014 	voff_t foff;
1015 	vsize_t locklimit;
1016 {
1017 	struct uvm_object *uobj;
1018 	struct vnode *vp;
1019 	int error;
1020 	int advice = UVM_ADV_NORMAL;
1021 	uvm_flag_t uvmflag = 0;
1022 
1023 	/*
1024 	 * check params
1025 	 */
1026 
1027 	if (size == 0)
1028 		return(0);
1029 	if (foff & PAGE_MASK)
1030 		return(EINVAL);
1031 	if ((prot & maxprot) != prot)
1032 		return(EINVAL);
1033 
1034 	/*
1035 	 * for non-fixed mappings, round off the suggested address.
1036 	 * for fixed mappings, check alignment and zap old mappings.
1037 	 */
1038 
1039 	if ((flags & MAP_FIXED) == 0) {
1040 		*addr = round_page(*addr);
1041 	} else {
1042 		if (*addr & PAGE_MASK)
1043 			return(EINVAL);
1044 		uvmflag |= UVM_FLAG_FIXED;
1045 		(void) uvm_unmap(map, *addr, *addr + size);
1046 	}
1047 
1048 	/*
1049 	 * handle anon vs. non-anon mappings.   for non-anon mappings attach
1050 	 * to underlying vm object.
1051 	 */
1052 
1053 	if (flags & MAP_ANON) {
1054 		foff = UVM_UNKNOWN_OFFSET;
1055 		uobj = NULL;
1056 		if ((flags & MAP_SHARED) == 0)
1057 			/* XXX: defer amap create */
1058 			uvmflag |= UVM_FLAG_COPYONW;
1059 		else
1060 			/* shared: create amap now */
1061 			uvmflag |= UVM_FLAG_OVERLAY;
1062 
1063 	} else {
1064 		vp = (struct vnode *)handle;
1065 
1066 		/*
1067 		 * Don't allow mmap for EXEC if the file system
1068 		 * is mounted NOEXEC.
1069 		 */
1070 		if ((prot & PROT_EXEC) != 0 &&
1071 		    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0)
1072 			return (EACCES);
1073 
1074 		if (vp->v_type != VCHR) {
1075 			error = VOP_MMAP(vp, 0, curproc->p_ucred, curproc);
1076 			if (error) {
1077 				return error;
1078 			}
1079 
1080 			uobj = uvn_attach((void *)vp, (flags & MAP_SHARED) ?
1081 			   maxprot : (maxprot & ~VM_PROT_WRITE));
1082 
1083 			/* XXX for now, attach doesn't gain a ref */
1084 			VREF(vp);
1085 
1086 			/*
1087 			 * If the vnode is being mapped with PROT_EXEC,
1088 			 * then mark it as text.
1089 			 */
1090 			if (prot & PROT_EXEC)
1091 				vn_markexec(vp);
1092 		} else {
1093 			uobj = udv_attach((void *) &vp->v_rdev,
1094 			    (flags & MAP_SHARED) ? maxprot :
1095 			    (maxprot & ~VM_PROT_WRITE), foff, size);
1096 			/*
1097 			 * XXX Some devices don't like to be mapped with
1098 			 * XXX PROT_EXEC, but we don't really have a
1099 			 * XXX better way of handling this, right now
1100 			 */
1101 			if (uobj == NULL && (prot & PROT_EXEC) == 0) {
1102 				maxprot &= ~VM_PROT_EXECUTE;
1103 				uobj = udv_attach((void *)&vp->v_rdev,
1104 				    (flags & MAP_SHARED) ? maxprot :
1105 				    (maxprot & ~VM_PROT_WRITE), foff, size);
1106 			}
1107 			advice = UVM_ADV_RANDOM;
1108 		}
1109 		if (uobj == NULL)
1110 			return((vp->v_type == VREG) ? ENOMEM : EINVAL);
1111 		if ((flags & MAP_SHARED) == 0)
1112 			uvmflag |= UVM_FLAG_COPYONW;
1113 	}
1114 
1115 	uvmflag = UVM_MAPFLAG(prot, maxprot,
1116 			(flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
1117 			advice, uvmflag);
1118 	error = uvm_map(map, addr, size, uobj, foff, 0, uvmflag);
1119 	if (error) {
1120 		if (uobj)
1121 			uobj->pgops->pgo_detach(uobj);
1122 		return error;
1123 	}
1124 
1125 	/*
1126 	 * POSIX 1003.1b -- if our address space was configured
1127 	 * to lock all future mappings, wire the one we just made.
1128 	 */
1129 
1130 	if (prot == VM_PROT_NONE) {
1131 
1132 		/*
1133 		 * No more work to do in this case.
1134 		 */
1135 
1136 		return (0);
1137 	}
1138 	vm_map_lock(map);
1139 	if (map->flags & VM_MAP_WIREFUTURE) {
1140 		if ((atop(size) + uvmexp.wired) > uvmexp.wiredmax
1141 #ifdef pmap_wired_count
1142 		    || (locklimit != 0 && (size +
1143 		    ptoa(pmap_wired_count(vm_map_pmap(map)))) >
1144 			locklimit)
1145 #endif
1146 		) {
1147 			vm_map_unlock(map);
1148 			uvm_unmap(map, *addr, *addr + size);
1149 			return ENOMEM;
1150 		}
1151 
1152 		/*
1153 		 * uvm_map_pageable() always returns the map unlocked.
1154 		 */
1155 
1156 		error = uvm_map_pageable(map, *addr, *addr + size,
1157 					 FALSE, UVM_LK_ENTER);
1158 		if (error) {
1159 			uvm_unmap(map, *addr, *addr + size);
1160 			return error;
1161 		}
1162 		return (0);
1163 	}
1164 	vm_map_unlock(map);
1165 	return 0;
1166 }
1167