xref: /netbsd-src/sys/uvm/uvm_mmap.c (revision 8a8f936f250a330d54f8a24ed0e92aadf9743a7b)
1 /*	$NetBSD: uvm_mmap.c,v 1.56 2001/09/15 20:36:46 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 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/file.h>
55 #include <sys/filedesc.h>
56 #include <sys/resourcevar.h>
57 #include <sys/mman.h>
58 #include <sys/mount.h>
59 #include <sys/proc.h>
60 #include <sys/malloc.h>
61 #include <sys/vnode.h>
62 #include <sys/conf.h>
63 #include <sys/stat.h>
64 
65 #include <miscfs/specfs/specdev.h>
66 
67 #include <sys/syscallargs.h>
68 
69 #include <uvm/uvm.h>
70 #include <uvm/uvm_device.h>
71 
72 
73 /*
74  * unimplemented VM system calls:
75  */
76 
77 /*
78  * sys_sbrk: sbrk system call.
79  */
80 
81 /* ARGSUSED */
82 int
83 sys_sbrk(p, v, retval)
84 	struct proc *p;
85 	void *v;
86 	register_t *retval;
87 {
88 #if 0
89 	struct sys_sbrk_args /* {
90 		syscallarg(intptr_t) incr;
91 	} */ *uap = v;
92 #endif
93 
94 	return (ENOSYS);
95 }
96 
97 /*
98  * sys_sstk: sstk system call.
99  */
100 
101 /* ARGSUSED */
102 int
103 sys_sstk(p, v, retval)
104 	struct proc *p;
105 	void *v;
106 	register_t *retval;
107 {
108 #if 0
109 	struct sys_sstk_args /* {
110 		syscallarg(int) incr;
111 	} */ *uap = v;
112 #endif
113 
114 	return (ENOSYS);
115 }
116 
117 /*
118  * sys_mincore: determine if pages are in core or not.
119  */
120 
121 /* ARGSUSED */
122 int
123 sys_mincore(p, v, retval)
124 	struct proc *p;
125 	void *v;
126 	register_t *retval;
127 {
128 	struct sys_mincore_args /* {
129 		syscallarg(void *) addr;
130 		syscallarg(size_t) len;
131 		syscallarg(char *) vec;
132 	} */ *uap = v;
133 	struct vm_page *pg;
134 	char *vec, pgi;
135 	struct uvm_object *uobj;
136 	struct vm_amap *amap;
137 	struct vm_anon *anon;
138 	struct vm_map_entry *entry;
139 	vaddr_t start, end, lim;
140 	struct vm_map *map;
141 	vsize_t len;
142 	int error = 0, npgs;
143 
144 	map = &p->p_vmspace->vm_map;
145 
146 	start = (vaddr_t)SCARG(uap, addr);
147 	len = SCARG(uap, len);
148 	vec = SCARG(uap, vec);
149 
150 	if (start & PAGE_MASK)
151 		return (EINVAL);
152 	len = round_page(len);
153 	end = start + len;
154 	if (end <= start)
155 		return (EINVAL);
156 
157 	npgs = len >> PAGE_SHIFT;
158 
159 	if (uvm_useracc(vec, npgs, B_WRITE) == FALSE)
160 		return (EFAULT);
161 
162 	/*
163 	 * Lock down vec, so our returned status isn't outdated by
164 	 * storing the status byte for a page.
165 	 */
166 
167 	uvm_vslock(p, vec, npgs, VM_PROT_WRITE);
168 	vm_map_lock_read(map);
169 
170 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
171 		error = ENOMEM;
172 		goto out;
173 	}
174 
175 	for (/* nothing */;
176 	     entry != &map->header && entry->start < end;
177 	     entry = entry->next) {
178 		KASSERT(!UVM_ET_ISSUBMAP(entry));
179 		KASSERT(start >= entry->start);
180 
181 		/* Make sure there are no holes. */
182 		if (entry->end < end &&
183 		     (entry->next == &map->header ||
184 		      entry->next->start > entry->end)) {
185 			error = ENOMEM;
186 			goto out;
187 		}
188 
189 		lim = end < entry->end ? end : entry->end;
190 
191 		/*
192 		 * Special case for objects with no "real" pages.  Those
193 		 * are always considered resident (mapped devices).
194 		 */
195 
196 		if (UVM_ET_ISOBJ(entry)) {
197 			KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj));
198 			if (!UVM_OBJ_IS_VNODE(entry->object.uvm_obj)) {
199 				for (/* nothing */; start < lim;
200 				     start += PAGE_SIZE, vec++)
201 					subyte(vec, 1);
202 				continue;
203 			}
204 		}
205 
206 		amap = entry->aref.ar_amap;	/* top layer */
207 		uobj = entry->object.uvm_obj;	/* bottom layer */
208 
209 		if (amap != NULL)
210 			amap_lock(amap);
211 		if (uobj != NULL)
212 			simple_lock(&uobj->vmobjlock);
213 
214 		for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) {
215 			pgi = 0;
216 			if (amap != NULL) {
217 				/* Check the top layer first. */
218 				anon = amap_lookup(&entry->aref,
219 				    start - entry->start);
220 				/* Don't need to lock anon here. */
221 				if (anon != NULL && anon->u.an_page != NULL) {
222 
223 					/*
224 					 * Anon has the page for this entry
225 					 * offset.
226 					 */
227 
228 					pgi = 1;
229 				}
230 			}
231 			if (uobj != NULL && pgi == 0) {
232 				/* Check the bottom layer. */
233 				pg = uvm_pagelookup(uobj,
234 				    entry->offset + (start - entry->start));
235 				if (pg != NULL) {
236 
237 					/*
238 					 * Object has the page for this entry
239 					 * offset.
240 					 */
241 
242 					pgi = 1;
243 				}
244 			}
245 			(void) subyte(vec, pgi);
246 		}
247 		if (uobj != NULL)
248 			simple_unlock(&uobj->vmobjlock);
249 		if (amap != NULL)
250 			amap_unlock(amap);
251 	}
252 
253  out:
254 	vm_map_unlock_read(map);
255 	uvm_vsunlock(p, SCARG(uap, vec), npgs);
256 	return (error);
257 }
258 
259 /*
260  * sys_mmap: mmap system call.
261  *
262  * => file offest and address may not be page aligned
263  *    - if MAP_FIXED, offset and address must have remainder mod PAGE_SIZE
264  *    - if address isn't page aligned the mapping starts at trunc_page(addr)
265  *      and the return value is adjusted up by the page offset.
266  */
267 
268 int
269 sys_mmap(p, v, retval)
270 	struct proc *p;
271 	void *v;
272 	register_t *retval;
273 {
274 	struct sys_mmap_args /* {
275 		syscallarg(caddr_t) addr;
276 		syscallarg(size_t) len;
277 		syscallarg(int) prot;
278 		syscallarg(int) flags;
279 		syscallarg(int) fd;
280 		syscallarg(long) pad;
281 		syscallarg(off_t) pos;
282 	} */ *uap = v;
283 	vaddr_t addr;
284 	struct vattr va;
285 	off_t pos;
286 	vsize_t size, pageoff;
287 	vm_prot_t prot, maxprot;
288 	int flags, fd;
289 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
290 	struct filedesc *fdp = p->p_fd;
291 	struct file *fp;
292 	struct vnode *vp;
293 	void *handle;
294 	int error;
295 
296 	/*
297 	 * first, extract syscall args from the uap.
298 	 */
299 
300 	addr = (vaddr_t)SCARG(uap, addr);
301 	size = (vsize_t)SCARG(uap, len);
302 	prot = SCARG(uap, prot) & VM_PROT_ALL;
303 	flags = SCARG(uap, flags);
304 	fd = SCARG(uap, fd);
305 	pos = SCARG(uap, pos);
306 
307 	/*
308 	 * Fixup the old deprecated MAP_COPY into MAP_PRIVATE, and
309 	 * validate the flags.
310 	 */
311 	if (flags & MAP_COPY)
312 		flags = (flags & ~MAP_COPY) | MAP_PRIVATE;
313 	if ((flags & (MAP_SHARED|MAP_PRIVATE)) == (MAP_SHARED|MAP_PRIVATE))
314 		return (EINVAL);
315 
316 	/*
317 	 * align file position and save offset.  adjust size.
318 	 */
319 
320 	pageoff = (pos & PAGE_MASK);
321 	pos  -= pageoff;
322 	size += pageoff;			/* add offset */
323 	size = (vsize_t)round_page(size);	/* round up */
324 	if ((ssize_t) size < 0)
325 		return (EINVAL);			/* don't allow wrap */
326 
327 	/*
328 	 * now check (MAP_FIXED) or get (!MAP_FIXED) the "addr"
329 	 */
330 
331 	if (flags & MAP_FIXED) {
332 
333 		/* ensure address and file offset are aligned properly */
334 		addr -= pageoff;
335 		if (addr & PAGE_MASK)
336 			return (EINVAL);
337 
338 		if (VM_MAXUSER_ADDRESS > 0 &&
339 		    (addr + size) > VM_MAXUSER_ADDRESS)
340 			return (EINVAL);
341 		if (vm_min_address > 0 && addr < vm_min_address)
342 			return (EINVAL);
343 		if (addr > addr + size)
344 			return (EINVAL);		/* no wrapping! */
345 
346 	} else {
347 
348 		/*
349 		 * not fixed: make sure we skip over the largest possible heap.
350 		 * we will refine our guess later (e.g. to account for VAC, etc)
351 		 */
352 
353 		addr = MAX(addr, round_page((vaddr_t)p->p_vmspace->vm_daddr +
354 					    MAXDSIZ));
355 	}
356 
357 	/*
358 	 * check for file mappings (i.e. not anonymous) and verify file.
359 	 */
360 
361 	if ((flags & MAP_ANON) == 0) {
362 
363 		if ((fp = fd_getfile(fdp, fd)) == NULL)
364 			return (EBADF);
365 
366 		if (fp->f_type != DTYPE_VNODE)
367 			return (ENODEV);		/* only mmap vnodes! */
368 		vp = (struct vnode *)fp->f_data;	/* convert to vnode */
369 
370 		if (vp->v_type != VREG && vp->v_type != VCHR &&
371 		    vp->v_type != VBLK)
372 			return (ENODEV);  /* only REG/CHR/BLK support mmap */
373 
374 		if (vp->v_type == VREG && (pos + size) < pos)
375 			return (EOVERFLOW);		/* no offset wrapping */
376 
377 		/* special case: catch SunOS style /dev/zero */
378 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
379 			flags |= MAP_ANON;
380 			goto is_anon;
381 		}
382 
383 		/*
384 		 * Old programs may not select a specific sharing type, so
385 		 * default to an appropriate one.
386 		 *
387 		 * XXX: how does MAP_ANON fit in the picture?
388 		 */
389 		if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
390 #if defined(DEBUG)
391 			printf("WARNING: defaulted mmap() share type to "
392 			   "%s (pid %d comm %s)\n", vp->v_type == VCHR ?
393 			   "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
394 			    p->p_comm);
395 #endif
396 			if (vp->v_type == VCHR)
397 				flags |= MAP_SHARED;	/* for a device */
398 			else
399 				flags |= MAP_PRIVATE;	/* for a file */
400 		}
401 
402 		/*
403 		 * MAP_PRIVATE device mappings don't make sense (and aren't
404 		 * supported anyway).  However, some programs rely on this,
405 		 * so just change it to MAP_SHARED.
406 		 */
407 		if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
408 			flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
409 		}
410 
411 		/*
412 		 * now check protection
413 		 */
414 
415 		maxprot = VM_PROT_EXECUTE;
416 
417 		/* check read access */
418 		if (fp->f_flag & FREAD)
419 			maxprot |= VM_PROT_READ;
420 		else if (prot & PROT_READ)
421 			return (EACCES);
422 
423 		/* check write access, shared case first */
424 		if (flags & MAP_SHARED) {
425 			/*
426 			 * if the file is writable, only add PROT_WRITE to
427 			 * maxprot if the file is not immutable, append-only.
428 			 * otherwise, if we have asked for PROT_WRITE, return
429 			 * EPERM.
430 			 */
431 			if (fp->f_flag & FWRITE) {
432 				if ((error =
433 				    VOP_GETATTR(vp, &va, p->p_ucred, p)))
434 					return (error);
435 				if ((va.va_flags & (IMMUTABLE|APPEND)) == 0)
436 					maxprot |= VM_PROT_WRITE;
437 				else if (prot & PROT_WRITE)
438 					return (EPERM);
439 			}
440 			else if (prot & PROT_WRITE)
441 				return (EACCES);
442 		} else {
443 			/* MAP_PRIVATE mappings can always write to */
444 			maxprot |= VM_PROT_WRITE;
445 		}
446 		handle = vp;
447 
448 	} else {		/* MAP_ANON case */
449 		/*
450 		 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0?
451 		 */
452 		if (fd != -1)
453 			return (EINVAL);
454 
455  is_anon:		/* label for SunOS style /dev/zero */
456 		handle = NULL;
457 		maxprot = VM_PROT_ALL;
458 		pos = 0;
459 	}
460 
461 	/*
462 	 * XXX (in)sanity check.  We don't do proper datasize checking
463 	 * XXX for anonymous (or private writable) mmap().  However,
464 	 * XXX know that if we're trying to allocate more than the amount
465 	 * XXX remaining under our current data size limit, _that_ should
466 	 * XXX be disallowed.
467 	 */
468 	if ((flags & MAP_ANON) != 0 ||
469 	    ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) {
470 		if (size >
471 		    (p->p_rlimit[RLIMIT_DATA].rlim_cur -
472 		     ctob(p->p_vmspace->vm_dsize))) {
473 			return (ENOMEM);
474 		}
475 	}
476 
477 	/*
478 	 * now let kernel internal function uvm_mmap do the work.
479 	 */
480 
481 	error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot,
482 	    flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
483 
484 	if (error == 0)
485 		/* remember to add offset */
486 		*retval = (register_t)(addr + pageoff);
487 
488 	return (error);
489 }
490 
491 /*
492  * sys___msync13: the msync system call (a front-end for flush)
493  */
494 
495 int
496 sys___msync13(p, v, retval)
497 	struct proc *p;
498 	void *v;
499 	register_t *retval;
500 {
501 	struct sys___msync13_args /* {
502 		syscallarg(caddr_t) addr;
503 		syscallarg(size_t) len;
504 		syscallarg(int) flags;
505 	} */ *uap = v;
506 	vaddr_t addr;
507 	vsize_t size, pageoff;
508 	struct vm_map *map;
509 	int error, rv, flags, uvmflags;
510 
511 	/*
512 	 * extract syscall args from the uap
513 	 */
514 
515 	addr = (vaddr_t)SCARG(uap, addr);
516 	size = (vsize_t)SCARG(uap, len);
517 	flags = SCARG(uap, flags);
518 
519 	/* sanity check flags */
520 	if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 ||
521 			(flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 ||
522 			(flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC))
523 	  return (EINVAL);
524 	if ((flags & (MS_ASYNC | MS_SYNC)) == 0)
525 	  flags |= MS_SYNC;
526 
527 	/*
528 	 * align the address to a page boundary and adjust the size accordingly.
529 	 */
530 
531 	pageoff = (addr & PAGE_MASK);
532 	addr -= pageoff;
533 	size += pageoff;
534 	size = (vsize_t)round_page(size);
535 
536 	/* disallow wrap-around. */
537 	if (addr + size < addr)
538 		return (EINVAL);
539 
540 	/*
541 	 * get map
542 	 */
543 
544 	map = &p->p_vmspace->vm_map;
545 
546 	/*
547 	 * XXXCDC: do we really need this semantic?
548 	 *
549 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
550 	 * pages with the region containing addr".  Unfortunately, we
551 	 * don't really keep track of individual mmaps so we approximate
552 	 * by flushing the range of the map entry containing addr.
553 	 * This can be incorrect if the region splits or is coalesced
554 	 * with a neighbor.
555 	 */
556 
557 	if (size == 0) {
558 		struct vm_map_entry *entry;
559 
560 		vm_map_lock_read(map);
561 		rv = uvm_map_lookup_entry(map, addr, &entry);
562 		if (rv == TRUE) {
563 			addr = entry->start;
564 			size = entry->end - entry->start;
565 		}
566 		vm_map_unlock_read(map);
567 		if (rv == FALSE)
568 			return (EINVAL);
569 	}
570 
571 	/*
572 	 * translate MS_ flags into PGO_ flags
573 	 */
574 
575 	uvmflags = PGO_CLEANIT;
576 	if (flags & MS_INVALIDATE)
577 		uvmflags |= PGO_FREE;
578 	if (flags & MS_SYNC)
579 		uvmflags |= PGO_SYNCIO;
580 	else
581 		uvmflags |= PGO_SYNCIO;	 /* XXXCDC: force sync for now! */
582 
583 	error = uvm_map_clean(map, addr, addr+size, uvmflags);
584 	return error;
585 }
586 
587 /*
588  * sys_munmap: unmap a users memory
589  */
590 
591 int
592 sys_munmap(p, v, retval)
593 	struct proc *p;
594 	void *v;
595 	register_t *retval;
596 {
597 	struct sys_munmap_args /* {
598 		syscallarg(caddr_t) addr;
599 		syscallarg(size_t) len;
600 	} */ *uap = v;
601 	vaddr_t addr;
602 	vsize_t size, pageoff;
603 	struct vm_map *map;
604 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
605 	struct vm_map_entry *dead_entries;
606 
607 	/*
608 	 * get syscall args.
609 	 */
610 
611 	addr = (vaddr_t)SCARG(uap, addr);
612 	size = (vsize_t)SCARG(uap, len);
613 
614 	/*
615 	 * align the address to a page boundary and adjust the size accordingly.
616 	 */
617 
618 	pageoff = (addr & PAGE_MASK);
619 	addr -= pageoff;
620 	size += pageoff;
621 	size = (vsize_t)round_page(size);
622 
623 	if ((int)size < 0)
624 		return (EINVAL);
625 	if (size == 0)
626 		return (0);
627 
628 	/*
629 	 * Check for illegal addresses.  Watch out for address wrap...
630 	 * Note that VM_*_ADDRESS are not constants due to casts (argh).
631 	 */
632 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
633 		return (EINVAL);
634 	if (vm_min_address > 0 && addr < vm_min_address)
635 		return (EINVAL);
636 	if (addr > addr + size)
637 		return (EINVAL);
638 	map = &p->p_vmspace->vm_map;
639 
640 	/*
641 	 * interesting system call semantic: make sure entire range is
642 	 * allocated before allowing an unmap.
643 	 */
644 
645 	vm_map_lock(map);
646 	if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
647 		vm_map_unlock(map);
648 		return (EINVAL);
649 	}
650 	uvm_unmap_remove(map, addr, addr + size, &dead_entries);
651 	vm_map_unlock(map);
652 	if (dead_entries != NULL)
653 		uvm_unmap_detach(dead_entries, 0);
654 	return (0);
655 }
656 
657 /*
658  * sys_mprotect: the mprotect system call
659  */
660 
661 int
662 sys_mprotect(p, v, retval)
663 	struct proc *p;
664 	void *v;
665 	register_t *retval;
666 {
667 	struct sys_mprotect_args /* {
668 		syscallarg(caddr_t) addr;
669 		syscallarg(int) len;
670 		syscallarg(int) prot;
671 	} */ *uap = v;
672 	vaddr_t addr;
673 	vsize_t size, pageoff;
674 	vm_prot_t prot;
675 	int error;
676 
677 	/*
678 	 * extract syscall args from uap
679 	 */
680 
681 	addr = (vaddr_t)SCARG(uap, addr);
682 	size = (vsize_t)SCARG(uap, len);
683 	prot = SCARG(uap, prot) & VM_PROT_ALL;
684 
685 	/*
686 	 * align the address to a page boundary and adjust the size accordingly.
687 	 */
688 
689 	pageoff = (addr & PAGE_MASK);
690 	addr -= pageoff;
691 	size += pageoff;
692 	size = (vsize_t)round_page(size);
693 
694 	if ((int)size < 0)
695 		return (EINVAL);
696 	error = uvm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
697 				FALSE);
698 	return error;
699 }
700 
701 /*
702  * sys_minherit: the minherit system call
703  */
704 
705 int
706 sys_minherit(p, v, retval)
707 	struct proc *p;
708 	void *v;
709 	register_t *retval;
710 {
711 	struct sys_minherit_args /* {
712 		syscallarg(caddr_t) addr;
713 		syscallarg(int) len;
714 		syscallarg(int) inherit;
715 	} */ *uap = v;
716 	vaddr_t addr;
717 	vsize_t size, pageoff;
718 	vm_inherit_t inherit;
719 	int error;
720 
721 	addr = (vaddr_t)SCARG(uap, addr);
722 	size = (vsize_t)SCARG(uap, len);
723 	inherit = SCARG(uap, inherit);
724 
725 	/*
726 	 * align the address to a page boundary and adjust the size accordingly.
727 	 */
728 
729 	pageoff = (addr & PAGE_MASK);
730 	addr -= pageoff;
731 	size += pageoff;
732 	size = (vsize_t)round_page(size);
733 
734 	if ((int)size < 0)
735 		return (EINVAL);
736 	error = uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr + size,
737 				inherit);
738 	return error;
739 }
740 
741 /*
742  * sys_madvise: give advice about memory usage.
743  */
744 
745 /* ARGSUSED */
746 int
747 sys_madvise(p, v, retval)
748 	struct proc *p;
749 	void *v;
750 	register_t *retval;
751 {
752 	struct sys_madvise_args /* {
753 		syscallarg(caddr_t) addr;
754 		syscallarg(size_t) len;
755 		syscallarg(int) behav;
756 	} */ *uap = v;
757 	vaddr_t addr;
758 	vsize_t size, pageoff;
759 	int advice, error;
760 
761 	addr = (vaddr_t)SCARG(uap, addr);
762 	size = (vsize_t)SCARG(uap, len);
763 	advice = SCARG(uap, behav);
764 
765 	/*
766 	 * align the address to a page boundary, and adjust the size accordingly
767 	 */
768 
769 	pageoff = (addr & PAGE_MASK);
770 	addr -= pageoff;
771 	size += pageoff;
772 	size = (vsize_t)round_page(size);
773 
774 	if ((ssize_t)size <= 0)
775 		return (EINVAL);
776 
777 	switch (advice) {
778 	case MADV_NORMAL:
779 	case MADV_RANDOM:
780 	case MADV_SEQUENTIAL:
781 		error = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size,
782 		    advice);
783 		break;
784 
785 	case MADV_WILLNEED:
786 
787 		/*
788 		 * Activate all these pages, pre-faulting them in if
789 		 * necessary.
790 		 */
791 		/*
792 		 * XXX IMPLEMENT ME.
793 		 * Should invent a "weak" mode for uvm_fault()
794 		 * which would only do the PGO_LOCKED pgo_get().
795 		 */
796 
797 		return (0);
798 
799 	case MADV_DONTNEED:
800 
801 		/*
802 		 * Deactivate all these pages.  We don't need them
803 		 * any more.  We don't, however, toss the data in
804 		 * the pages.
805 		 */
806 
807 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
808 		    PGO_DEACTIVATE);
809 		break;
810 
811 	case MADV_FREE:
812 
813 		/*
814 		 * These pages contain no valid data, and may be
815 		 * garbage-collected.  Toss all resources, including
816 		 * any swap space in use.
817 		 */
818 
819 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
820 		    PGO_FREE);
821 		break;
822 
823 	case MADV_SPACEAVAIL:
824 
825 		/*
826 		 * XXXMRG What is this?  I think it's:
827 		 *
828 		 *	Ensure that we have allocated backing-store
829 		 *	for these pages.
830 		 *
831 		 * This is going to require changes to the page daemon,
832 		 * as it will free swap space allocated to pages in core.
833 		 * There's also what to do for device/file/anonymous memory.
834 		 */
835 
836 		return (EINVAL);
837 
838 	default:
839 		return (EINVAL);
840 	}
841 
842 	return error;
843 }
844 
845 /*
846  * sys_mlock: memory lock
847  */
848 
849 int
850 sys_mlock(p, v, retval)
851 	struct proc *p;
852 	void *v;
853 	register_t *retval;
854 {
855 	struct sys_mlock_args /* {
856 		syscallarg(const void *) addr;
857 		syscallarg(size_t) len;
858 	} */ *uap = v;
859 	vaddr_t addr;
860 	vsize_t size, pageoff;
861 	int error;
862 
863 	/*
864 	 * extract syscall args from uap
865 	 */
866 
867 	addr = (vaddr_t)SCARG(uap, addr);
868 	size = (vsize_t)SCARG(uap, len);
869 
870 	/*
871 	 * align the address to a page boundary and adjust the size accordingly
872 	 */
873 
874 	pageoff = (addr & PAGE_MASK);
875 	addr -= pageoff;
876 	size += pageoff;
877 	size = (vsize_t)round_page(size);
878 
879 	/* disallow wrap-around. */
880 	if (addr + size < addr)
881 		return (EINVAL);
882 
883 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
884 		return (EAGAIN);
885 
886 #ifdef pmap_wired_count
887 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
888 			p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
889 		return (EAGAIN);
890 #else
891 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
892 		return (error);
893 #endif
894 
895 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
896 	    0);
897 	return error;
898 }
899 
900 /*
901  * sys_munlock: unlock wired pages
902  */
903 
904 int
905 sys_munlock(p, v, retval)
906 	struct proc *p;
907 	void *v;
908 	register_t *retval;
909 {
910 	struct sys_munlock_args /* {
911 		syscallarg(const void *) addr;
912 		syscallarg(size_t) len;
913 	} */ *uap = v;
914 	vaddr_t addr;
915 	vsize_t size, pageoff;
916 	int error;
917 
918 	/*
919 	 * extract syscall args from uap
920 	 */
921 
922 	addr = (vaddr_t)SCARG(uap, addr);
923 	size = (vsize_t)SCARG(uap, len);
924 
925 	/*
926 	 * align the address to a page boundary, and adjust the size accordingly
927 	 */
928 
929 	pageoff = (addr & PAGE_MASK);
930 	addr -= pageoff;
931 	size += pageoff;
932 	size = (vsize_t)round_page(size);
933 
934 	/* disallow wrap-around. */
935 	if (addr + size < addr)
936 		return (EINVAL);
937 
938 #ifndef pmap_wired_count
939 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
940 		return (error);
941 #endif
942 
943 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
944 	    0);
945 	return error;
946 }
947 
948 /*
949  * sys_mlockall: lock all pages mapped into an address space.
950  */
951 
952 int
953 sys_mlockall(p, v, retval)
954 	struct proc *p;
955 	void *v;
956 	register_t *retval;
957 {
958 	struct sys_mlockall_args /* {
959 		syscallarg(int) flags;
960 	} */ *uap = v;
961 	int error, flags;
962 
963 	flags = SCARG(uap, flags);
964 
965 	if (flags == 0 ||
966 	    (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
967 		return (EINVAL);
968 
969 #ifndef pmap_wired_count
970 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
971 		return (error);
972 #endif
973 
974 	error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
975 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
976 	return (error);
977 }
978 
979 /*
980  * sys_munlockall: unlock all pages mapped into an address space.
981  */
982 
983 int
984 sys_munlockall(p, v, retval)
985 	struct proc *p;
986 	void *v;
987 	register_t *retval;
988 {
989 
990 	(void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
991 	return (0);
992 }
993 
994 /*
995  * uvm_mmap: internal version of mmap
996  *
997  * - used by sys_mmap and various framebuffers
998  * - handle is a vnode pointer or NULL for MAP_ANON
999  * - caller must page-align the file offset
1000  */
1001 
1002 int
1003 uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit)
1004 	struct vm_map *map;
1005 	vaddr_t *addr;
1006 	vsize_t size;
1007 	vm_prot_t prot, maxprot;
1008 	int flags;
1009 	void *handle;
1010 	voff_t foff;
1011 	vsize_t locklimit;
1012 {
1013 	struct uvm_object *uobj;
1014 	struct vnode *vp;
1015 	int error;
1016 	int advice = UVM_ADV_NORMAL;
1017 	uvm_flag_t uvmflag = 0;
1018 
1019 	/*
1020 	 * check params
1021 	 */
1022 
1023 	if (size == 0)
1024 		return(0);
1025 	if (foff & PAGE_MASK)
1026 		return(EINVAL);
1027 	if ((prot & maxprot) != prot)
1028 		return(EINVAL);
1029 
1030 	/*
1031 	 * for non-fixed mappings, round off the suggested address.
1032 	 * for fixed mappings, check alignment and zap old mappings.
1033 	 */
1034 
1035 	if ((flags & MAP_FIXED) == 0) {
1036 		*addr = round_page(*addr);
1037 	} else {
1038 		if (*addr & PAGE_MASK)
1039 			return(EINVAL);
1040 		uvmflag |= UVM_FLAG_FIXED;
1041 		(void) uvm_unmap(map, *addr, *addr + size);
1042 	}
1043 
1044 	/*
1045 	 * handle anon vs. non-anon mappings.   for non-anon mappings attach
1046 	 * to underlying vm object.
1047 	 */
1048 
1049 	if (flags & MAP_ANON) {
1050 		foff = UVM_UNKNOWN_OFFSET;
1051 		uobj = NULL;
1052 		if ((flags & MAP_SHARED) == 0)
1053 			/* XXX: defer amap create */
1054 			uvmflag |= UVM_FLAG_COPYONW;
1055 		else
1056 			/* shared: create amap now */
1057 			uvmflag |= UVM_FLAG_OVERLAY;
1058 
1059 	} else {
1060 		vp = (struct vnode *)handle;
1061 		if (vp->v_type != VCHR) {
1062 			error = VOP_MMAP(vp, 0, curproc->p_ucred, curproc);
1063 			if (error) {
1064 				return error;
1065 			}
1066 
1067 			uobj = uvn_attach((void *)vp, (flags & MAP_SHARED) ?
1068 			   maxprot : (maxprot & ~VM_PROT_WRITE));
1069 
1070 			/* XXX for now, attach doesn't gain a ref */
1071 			VREF(vp);
1072 		} else {
1073 			uobj = udv_attach((void *) &vp->v_rdev,
1074 			    (flags & MAP_SHARED) ? maxprot :
1075 			    (maxprot & ~VM_PROT_WRITE), foff, size);
1076 			/*
1077 			 * XXX Some devices don't like to be mapped with
1078 			 * XXX PROT_EXEC, but we don't really have a
1079 			 * XXX better way of handling this, right now
1080 			 */
1081 			if (uobj == NULL && (prot & PROT_EXEC) == 0) {
1082 				maxprot &= ~VM_PROT_EXECUTE;
1083 				uobj = udv_attach((void *)&vp->v_rdev,
1084 				    (flags & MAP_SHARED) ? maxprot :
1085 				    (maxprot & ~VM_PROT_WRITE), foff, size);
1086 			}
1087 			advice = UVM_ADV_RANDOM;
1088 		}
1089 		if (uobj == NULL)
1090 			return((vp->v_type == VREG) ? ENOMEM : EINVAL);
1091 		if ((flags & MAP_SHARED) == 0)
1092 			uvmflag |= UVM_FLAG_COPYONW;
1093 	}
1094 
1095 	uvmflag = UVM_MAPFLAG(prot, maxprot,
1096 			(flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
1097 			advice, uvmflag);
1098 	error = uvm_map(map, addr, size, uobj, foff, 0, uvmflag);
1099 	if (error) {
1100 		if (uobj)
1101 			uobj->pgops->pgo_detach(uobj);
1102 		return error;
1103 	}
1104 
1105 	/*
1106 	 * POSIX 1003.1b -- if our address space was configured
1107 	 * to lock all future mappings, wire the one we just made.
1108 	 */
1109 
1110 	if (prot == VM_PROT_NONE) {
1111 
1112 		/*
1113 		 * No more work to do in this case.
1114 		 */
1115 
1116 		return (0);
1117 	}
1118 	vm_map_lock(map);
1119 	if (map->flags & VM_MAP_WIREFUTURE) {
1120 		if ((atop(size) + uvmexp.wired) > uvmexp.wiredmax
1121 #ifdef pmap_wired_count
1122 		    || (locklimit != 0 && (size +
1123 		    ptoa(pmap_wired_count(vm_map_pmap(map)))) >
1124 			locklimit)
1125 #endif
1126 		) {
1127 			vm_map_unlock(map);
1128 			uvm_unmap(map, *addr, *addr + size);
1129 			return ENOMEM;
1130 		}
1131 
1132 		/*
1133 		 * uvm_map_pageable() always returns the map unlocked.
1134 		 */
1135 
1136 		error = uvm_map_pageable(map, *addr, *addr + size,
1137 					 FALSE, UVM_LK_ENTER);
1138 		if (error) {
1139 			uvm_unmap(map, *addr, *addr + size);
1140 			return error;
1141 		}
1142 		return (0);
1143 	}
1144 	vm_map_unlock(map);
1145 	return 0;
1146 }
1147