xref: /openbsd/sys/uvm/uvm_mmap.c (revision 7b36286a)
1 /*	$OpenBSD: uvm_mmap.c,v 1.70 2007/09/01 15:14:44 martin Exp $	*/
2 /*	$NetBSD: uvm_mmap.c,v 1.49 2001/02/18 21:19:08 chs Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * Copyright (c) 1991, 1993 The Regents of the University of California.
7  * Copyright (c) 1988 University of Utah.
8  *
9  * All rights reserved.
10  *
11  * This code is derived from software contributed to Berkeley by
12  * the Systems Programming Group of the University of Utah Computer
13  * Science Department.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgement:
25  *      This product includes software developed by the Charles D. Cranor,
26  *	Washington University, University of California, Berkeley and
27  *	its contributors.
28  * 4. Neither the name of the University nor the names of its contributors
29  *    may be used to endorse or promote products derived from this software
30  *    without specific prior written permission.
31  *
32  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
33  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
34  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
35  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
36  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
37  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
38  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
39  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
40  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
41  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42  * SUCH DAMAGE.
43  *
44  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
45  *      @(#)vm_mmap.c   8.5 (Berkeley) 5/19/94
46  * from: Id: uvm_mmap.c,v 1.1.2.14 1998/01/05 21:04:26 chuck Exp
47  */
48 
49 /*
50  * uvm_mmap.c: system call interface into VM system, plus kernel vm_mmap
51  * function.
52  */
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/file.h>
56 #include <sys/filedesc.h>
57 #include <sys/resourcevar.h>
58 #include <sys/mman.h>
59 #include <sys/mount.h>
60 #include <sys/proc.h>
61 #include <sys/malloc.h>
62 #include <sys/vnode.h>
63 #include <sys/conf.h>
64 #include <sys/stat.h>
65 
66 #include <machine/exec.h>	/* for __LDPGSZ */
67 
68 #include <miscfs/specfs/specdev.h>
69 
70 #include <sys/syscallargs.h>
71 
72 #include <uvm/uvm.h>
73 #include <uvm/uvm_device.h>
74 #include <uvm/uvm_vnode.h>
75 
76 /*
77  * Page align addr and size, returning EINVAL on wraparound.
78  */
79 #define ALIGN_ADDR(addr, size, pageoff)	do {				\
80 	pageoff = (addr & PAGE_MASK);					\
81 	if (pageoff != 0) {						\
82 		if (size > SIZE_MAX - pageoff)				\
83 			return (EINVAL);	/* wraparound */	\
84 		addr -= pageoff;					\
85 		size += pageoff;					\
86 	}								\
87 	if (size != 0) {						\
88 		size = (vsize_t)round_page(size);			\
89 		if (size == 0)						\
90 			return (EINVAL);	/* wraparound */	\
91 	}								\
92 } while (0)
93 
94 /*
95  * unimplemented VM system calls:
96  */
97 
98 /*
99  * sys_sbrk: sbrk system call.
100  */
101 
102 /* ARGSUSED */
103 int
104 sys_sbrk(p, v, retval)
105 	struct proc *p;
106 	void *v;
107 	register_t *retval;
108 {
109 #if 0
110 	struct sys_sbrk_args /* {
111 		syscallarg(int) incr;
112 	} */ *uap = v;
113 #endif
114 
115 	return (ENOSYS);
116 }
117 
118 /*
119  * sys_sstk: sstk system call.
120  */
121 
122 /* ARGSUSED */
123 int
124 sys_sstk(p, v, retval)
125 	struct proc *p;
126 	void *v;
127 	register_t *retval;
128 {
129 #if 0
130 	struct sys_sstk_args /* {
131 		syscallarg(int) incr;
132 	} */ *uap = v;
133 #endif
134 
135 	return (ENOSYS);
136 }
137 
138 /*
139  * sys_mquery: provide mapping hints to applications that do fixed mappings
140  *
141  * flags: 0 or MAP_FIXED (MAP_FIXED - means that we insist on this addr and
142  *	don't care about PMAP_PREFER or such)
143  * addr: hint where we'd like to place the mapping.
144  * size: size of the mapping
145  * fd: fd of the file we want to map
146  * off: offset within the file
147  */
148 
149 int
150 sys_mquery(p, v, retval)
151 	struct proc *p;
152 	void *v;
153 	register_t *retval;
154 {
155 	struct sys_mquery_args /* {
156 		syscallarg(void *) addr;
157 		syscallarg(size_t) len;
158 		syscallarg(int) prot;
159 		syscallarg(int) flags;
160 		syscallarg(int) fd;
161 		syscallarg(long) pad;
162 		syscallarg(off_t) pos;
163 	} */ *uap = v;
164 	struct file *fp;
165 	struct uvm_object *uobj;
166 	voff_t uoff;
167 	int error;
168 	vaddr_t vaddr;
169 	int flags = 0;
170 	vsize_t size;
171 	vm_prot_t prot;
172 	int fd;
173 
174 	vaddr = (vaddr_t) SCARG(uap, addr);
175 	prot = SCARG(uap, prot);
176 	size = (vsize_t) SCARG(uap, len);
177 	fd = SCARG(uap, fd);
178 
179 	if ((prot & VM_PROT_ALL) != prot)
180 		return (EINVAL);
181 
182 	if (SCARG(uap, flags) & MAP_FIXED)
183 		flags |= UVM_FLAG_FIXED;
184 
185 	if (fd >= 0) {
186 		if ((error = getvnode(p->p_fd, fd, &fp)) != 0)
187 			return (error);
188 		uobj = &((struct vnode *)fp->f_data)->v_uvm.u_obj;
189 		uoff = SCARG(uap, pos);
190 	} else {
191 		fp = NULL;
192 		uobj = NULL;
193 		uoff = 0;
194 	}
195 
196 	if (vaddr == 0)
197 		vaddr = uvm_map_hint(p, prot);
198 
199 	/* prevent a user requested address from falling in heap space */
200 	if ((vaddr + size > (vaddr_t)p->p_vmspace->vm_daddr) &&
201 	    (vaddr < (vaddr_t)p->p_vmspace->vm_daddr + MAXDSIZ)) {
202 		if (flags & UVM_FLAG_FIXED) {
203 			error = EINVAL;
204 			goto done;
205 		}
206 		vaddr = round_page((vaddr_t)p->p_vmspace->vm_daddr + MAXDSIZ);
207 	}
208 again:
209 
210 	if (uvm_map_findspace(&p->p_vmspace->vm_map, vaddr, size,
211 	    &vaddr, uobj, uoff, 0, flags) == NULL) {
212 		if (flags & UVM_FLAG_FIXED)
213 			error = EINVAL;
214 		else
215 			error = ENOMEM;
216 	} else {
217 		/* prevent a returned address from falling in heap space */
218 		if ((vaddr + size > (vaddr_t)p->p_vmspace->vm_daddr)
219 		    && (vaddr < (vaddr_t)p->p_vmspace->vm_daddr + MAXDSIZ)) {
220 			vaddr = round_page((vaddr_t)p->p_vmspace->vm_daddr +
221 			    MAXDSIZ);
222 			goto again;
223 		}
224 		error = 0;
225 		*retval = (register_t)(vaddr);
226 	}
227 done:
228 	if (fp != NULL)
229 		FRELE(fp);
230 	return (error);
231 }
232 
233 /*
234  * sys_mincore: determine if pages are in core or not.
235  */
236 
237 /* ARGSUSED */
238 int
239 sys_mincore(p, v, retval)
240 	struct proc *p;
241 	void *v;
242 	register_t *retval;
243 {
244 	struct sys_mincore_args /* {
245 		syscallarg(void *) addr;
246 		syscallarg(size_t) len;
247 		syscallarg(char *) vec;
248 	} */ *uap = v;
249 	vm_page_t m;
250 	char *vec, pgi;
251 	struct uvm_object *uobj;
252 	struct vm_amap *amap;
253 	struct vm_anon *anon;
254 	vm_map_entry_t entry;
255 	vaddr_t start, end, lim;
256 	vm_map_t map;
257 	vsize_t len, npgs;
258 	int error = 0;
259 
260 	map = &p->p_vmspace->vm_map;
261 
262 	start = (vaddr_t)SCARG(uap, addr);
263 	len = SCARG(uap, len);
264 	vec = SCARG(uap, vec);
265 
266 	if (start & PAGE_MASK)
267 		return (EINVAL);
268 	len = round_page(len);
269 	end = start + len;
270 	if (end <= start)
271 		return (EINVAL);
272 
273 	npgs = len >> PAGE_SHIFT;
274 
275 	/*
276 	 * Lock down vec, so our returned status isn't outdated by
277 	 * storing the status byte for a page.
278 	 */
279 	if ((error = uvm_vslock(p, vec, npgs, VM_PROT_WRITE)) != 0)
280 		return (error);
281 
282 	vm_map_lock_read(map);
283 
284 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
285 		error = ENOMEM;
286 		goto out;
287 	}
288 
289 	for (/* nothing */;
290 	     entry != &map->header && entry->start < end;
291 	     entry = entry->next) {
292 		KASSERT(!UVM_ET_ISSUBMAP(entry));
293 		KASSERT(start >= entry->start);
294 
295 		/* Make sure there are no holes. */
296 		if (entry->end < end &&
297 		     (entry->next == &map->header ||
298 		      entry->next->start > entry->end)) {
299 			error = ENOMEM;
300 			goto out;
301 		}
302 
303 		lim = end < entry->end ? end : entry->end;
304 
305 		/*
306 		 * Special case for objects with no "real" pages.  Those
307 		 * are always considered resident (mapped devices).
308 		 */
309 		if (UVM_ET_ISOBJ(entry)) {
310 			KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj));
311 			if (entry->object.uvm_obj->pgops->pgo_releasepg
312 			    == NULL) {
313 				pgi = 1;
314 				for (/* nothing */; start < lim;
315 				     start += PAGE_SIZE, vec++)
316 					copyout(&pgi, vec, sizeof(char));
317 				continue;
318 			}
319 		}
320 
321 		amap = entry->aref.ar_amap;	/* top layer */
322 		uobj = entry->object.uvm_obj;	/* bottom layer */
323 
324 		if (uobj != NULL)
325 			simple_lock(&uobj->vmobjlock);
326 
327 		for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) {
328 			pgi = 0;
329 			if (amap != NULL) {
330 				/* Check the top layer first. */
331 				anon = amap_lookup(&entry->aref,
332 				    start - entry->start);
333 				/* Don't need to lock anon here. */
334 				if (anon != NULL && anon->an_page != NULL) {
335 					/*
336 					 * Anon has the page for this entry
337 					 * offset.
338 					 */
339 					pgi = 1;
340 				}
341 			}
342 
343 			if (uobj != NULL && pgi == 0) {
344 				/* Check the bottom layer. */
345 				m = uvm_pagelookup(uobj,
346 				    entry->offset + (start - entry->start));
347 				if (m != NULL) {
348 					/*
349 					 * Object has the page for this entry
350 					 * offset.
351 					 */
352 					pgi = 1;
353 				}
354 			}
355 
356 			copyout(&pgi, vec, sizeof(char));
357 		}
358 
359 		if (uobj != NULL)
360 			simple_unlock(&uobj->vmobjlock);
361 	}
362 
363  out:
364 	vm_map_unlock_read(map);
365 	uvm_vsunlock(p, SCARG(uap, vec), npgs);
366 	return (error);
367 }
368 
369 /*
370  * sys_mmap: mmap system call.
371  *
372  * => file offset and address may not be page aligned
373  *    - if MAP_FIXED, offset and address must have remainder mod PAGE_SIZE
374  *    - if address isn't page aligned the mapping starts at trunc_page(addr)
375  *      and the return value is adjusted up by the page offset.
376  */
377 
378 int
379 sys_mmap(p, v, retval)
380 	struct proc *p;
381 	void *v;
382 	register_t *retval;
383 {
384 	struct sys_mmap_args /* {
385 		syscallarg(void *) addr;
386 		syscallarg(size_t) len;
387 		syscallarg(int) prot;
388 		syscallarg(int) flags;
389 		syscallarg(int) fd;
390 		syscallarg(long) pad;
391 		syscallarg(off_t) pos;
392 	} */ *uap = v;
393 	vaddr_t addr;
394 	struct vattr va;
395 	off_t pos;
396 	vsize_t size, pageoff;
397 	vm_prot_t prot, maxprot;
398 	int flags, fd;
399 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
400 	struct filedesc *fdp = p->p_fd;
401 	struct file *fp = NULL;
402 	struct vnode *vp;
403 	caddr_t handle;
404 	int error;
405 
406 	/*
407 	 * first, extract syscall args from the uap.
408 	 */
409 
410 	addr = (vaddr_t) SCARG(uap, addr);
411 	size = (vsize_t) SCARG(uap, len);
412 	prot = SCARG(uap, prot);
413 	flags = SCARG(uap, flags);
414 	fd = SCARG(uap, fd);
415 	pos = SCARG(uap, pos);
416 
417 	/*
418 	 * Fixup the old deprecated MAP_COPY into MAP_PRIVATE, and
419 	 * validate the flags.
420 	 */
421 	if ((prot & VM_PROT_ALL) != prot)
422 		return (EINVAL);
423 	if ((flags & MAP_FLAGMASK) != flags)
424 		return (EINVAL);
425 	if (flags & MAP_COPY)
426 		flags = (flags & ~MAP_COPY) | MAP_PRIVATE;
427 	if ((flags & (MAP_SHARED|MAP_PRIVATE)) == (MAP_SHARED|MAP_PRIVATE))
428 		return (EINVAL);
429 
430 	/*
431 	 * align file position and save offset.  adjust size.
432 	 */
433 	ALIGN_ADDR(pos, size, pageoff);
434 
435 	/*
436 	 * now check (MAP_FIXED) or get (!MAP_FIXED) the "addr"
437 	 */
438 
439 	if (flags & MAP_FIXED) {
440 
441 		/* adjust address by the same amount as we did the offset */
442 		addr -= pageoff;
443 		if (addr & PAGE_MASK)
444 			return (EINVAL);		/* not page aligned */
445 
446 		if (addr > SIZE_MAX - size)
447 			return (EINVAL);		/* no wrapping! */
448 		if (VM_MAXUSER_ADDRESS > 0 &&
449 		    (addr + size) > VM_MAXUSER_ADDRESS)
450 			return (EINVAL);
451 		if (vm_min_address > 0 && addr < vm_min_address)
452 			return (EINVAL);
453 
454 	} else {
455 
456 		/*
457 		 * not fixed: make sure we skip over the largest possible heap.
458 		 * we will refine our guess later (e.g. to account for VAC, etc)
459 		 */
460 		if (addr == 0)
461 			addr = uvm_map_hint(p, prot);
462 		else if (!(flags & MAP_TRYFIXED) &&
463 		    addr < (vaddr_t)p->p_vmspace->vm_daddr)
464 			addr = uvm_map_hint(p, prot);
465 	}
466 
467 	/*
468 	 * check for file mappings (i.e. not anonymous) and verify file.
469 	 */
470 	if ((flags & MAP_ANON) == 0) {
471 
472 		if ((fp = fd_getfile(fdp, fd)) == NULL)
473 			return (EBADF);
474 
475 		FREF(fp);
476 
477 		if (fp->f_type != DTYPE_VNODE) {
478 			error = ENODEV;		/* only mmap vnodes! */
479 			goto out;
480 		}
481 		vp = (struct vnode *)fp->f_data;	/* convert to vnode */
482 
483 		if (vp->v_type != VREG && vp->v_type != VCHR &&
484 		    vp->v_type != VBLK) {
485 			error = ENODEV; /* only REG/CHR/BLK support mmap */
486 			goto out;
487 		}
488 
489 		if (vp->v_type == VREG && (pos + size) < pos) {
490 			error = EINVAL;		/* no offset wrapping */
491 			goto out;
492 		}
493 
494 		/* special case: catch SunOS style /dev/zero */
495 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
496 			flags |= MAP_ANON;
497 			FRELE(fp);
498 			fp = NULL;
499 			goto is_anon;
500 		}
501 
502 		/*
503 		 * Old programs may not select a specific sharing type, so
504 		 * default to an appropriate one.
505 		 *
506 		 * XXX: how does MAP_ANON fit in the picture?
507 		 */
508 		if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
509 #if defined(DEBUG)
510 			printf("WARNING: defaulted mmap() share type to "
511 			   "%s (pid %d comm %s)\n", vp->v_type == VCHR ?
512 			   "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
513 			    p->p_comm);
514 #endif
515 			if (vp->v_type == VCHR)
516 				flags |= MAP_SHARED;	/* for a device */
517 			else
518 				flags |= MAP_PRIVATE;	/* for a file */
519 		}
520 
521 		/*
522 		 * MAP_PRIVATE device mappings don't make sense (and aren't
523 		 * supported anyway).  However, some programs rely on this,
524 		 * so just change it to MAP_SHARED.
525 		 */
526 		if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
527 			flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
528 		}
529 
530 		/*
531 		 * now check protection
532 		 */
533 
534 		maxprot = VM_PROT_EXECUTE;
535 
536 		/* check read access */
537 		if (fp->f_flag & FREAD)
538 			maxprot |= VM_PROT_READ;
539 		else if (prot & PROT_READ) {
540 			error = EACCES;
541 			goto out;
542 		}
543 
544 		/* check write access, shared case first */
545 		if (flags & MAP_SHARED) {
546 			/*
547 			 * if the file is writable, only add PROT_WRITE to
548 			 * maxprot if the file is not immutable, append-only.
549 			 * otherwise, if we have asked for PROT_WRITE, return
550 			 * EPERM.
551 			 */
552 			if (fp->f_flag & FWRITE) {
553 				if ((error =
554 				    VOP_GETATTR(vp, &va, p->p_ucred, p)))
555 					goto out;
556 				if ((va.va_flags & (IMMUTABLE|APPEND)) == 0)
557 					maxprot |= VM_PROT_WRITE;
558 				else if (prot & PROT_WRITE) {
559 					error = EPERM;
560 					goto out;
561 				}
562 			} else if (prot & PROT_WRITE) {
563 				error = EACCES;
564 				goto out;
565 			}
566 		} else {
567 			/* MAP_PRIVATE mappings can always write to */
568 			maxprot |= VM_PROT_WRITE;
569 		}
570 
571 		/*
572 		 * set handle to vnode
573 		 */
574 
575 		handle = (caddr_t)vp;
576 
577 	} else {		/* MAP_ANON case */
578 		/*
579 		 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0?
580 		 */
581 		if (fd != -1) {
582 			error = EINVAL;
583 			goto out;
584 		}
585 
586  is_anon:		/* label for SunOS style /dev/zero */
587 		handle = NULL;
588 		maxprot = VM_PROT_ALL;
589 		pos = 0;
590 	}
591 
592 	if ((flags & MAP_ANON) != 0 ||
593 	    ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) {
594 		if (size >
595 		    (p->p_rlimit[RLIMIT_DATA].rlim_cur - ptoa(p->p_vmspace->vm_dused))) {
596 			error = ENOMEM;
597 			goto out;
598 		}
599 	}
600 
601 	/*
602 	 * now let kernel internal function uvm_mmap do the work.
603 	 */
604 
605 	error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot,
606 	    flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur, p);
607 
608 	if (error == 0)
609 		/* remember to add offset */
610 		*retval = (register_t)(addr + pageoff);
611 
612 out:
613 	if (fp)
614 		FRELE(fp);
615 	return (error);
616 }
617 
618 /*
619  * sys_msync: the msync system call (a front-end for flush)
620  */
621 
622 int
623 sys_msync(p, v, retval)
624 	struct proc *p;
625 	void *v;
626 	register_t *retval;
627 {
628 	struct sys_msync_args /* {
629 		syscallarg(void *) addr;
630 		syscallarg(size_t) len;
631 		syscallarg(int) flags;
632 	} */ *uap = v;
633 	vaddr_t addr;
634 	vsize_t size, pageoff;
635 	vm_map_t map;
636 	int rv, flags, uvmflags;
637 
638 	/*
639 	 * extract syscall args from the uap
640 	 */
641 
642 	addr = (vaddr_t)SCARG(uap, addr);
643 	size = (vsize_t)SCARG(uap, len);
644 	flags = SCARG(uap, flags);
645 
646 	/* sanity check flags */
647 	if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 ||
648 			(flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 ||
649 			(flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC))
650 		return (EINVAL);
651 	if ((flags & (MS_ASYNC | MS_SYNC)) == 0)
652 		flags |= MS_SYNC;
653 
654 	/*
655 	 * align the address to a page boundary, and adjust the size accordingly
656 	 */
657 	ALIGN_ADDR(addr, size, pageoff);
658 	if (addr > SIZE_MAX - size)
659 		return (EINVAL);		/* disallow wrap-around. */
660 
661 	/*
662 	 * get map
663 	 */
664 
665 	map = &p->p_vmspace->vm_map;
666 
667 	/*
668 	 * XXXCDC: do we really need this semantic?
669 	 *
670 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
671 	 * pages with the region containing addr".  Unfortunately, we
672 	 * don't really keep track of individual mmaps so we approximate
673 	 * by flushing the range of the map entry containing addr.
674 	 * This can be incorrect if the region splits or is coalesced
675 	 * with a neighbor.
676 	 */
677 	if (size == 0) {
678 		vm_map_entry_t entry;
679 
680 		vm_map_lock_read(map);
681 		rv = uvm_map_lookup_entry(map, addr, &entry);
682 		if (rv == TRUE) {
683 			addr = entry->start;
684 			size = entry->end - entry->start;
685 		}
686 		vm_map_unlock_read(map);
687 		if (rv == FALSE)
688 			return (EINVAL);
689 	}
690 
691 	/*
692 	 * translate MS_ flags into PGO_ flags
693 	 */
694 	uvmflags = PGO_CLEANIT;
695 	if (flags & MS_INVALIDATE)
696 		uvmflags |= PGO_FREE;
697 	if (flags & MS_SYNC)
698 		uvmflags |= PGO_SYNCIO;
699 	else
700 		uvmflags |= PGO_SYNCIO;	 /* XXXCDC: force sync for now! */
701 
702 	return (uvm_map_clean(map, addr, addr+size, uvmflags));
703 }
704 
705 /*
706  * sys_munmap: unmap a users memory
707  */
708 
709 int
710 sys_munmap(p, v, retval)
711 	struct proc *p;
712 	void *v;
713 	register_t *retval;
714 {
715 	struct sys_munmap_args /* {
716 		syscallarg(void *) addr;
717 		syscallarg(size_t) len;
718 	} */ *uap = v;
719 	vaddr_t addr;
720 	vsize_t size, pageoff;
721 	vm_map_t map;
722 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
723 	struct vm_map_entry *dead_entries;
724 
725 	/*
726 	 * get syscall args...
727 	 */
728 
729 	addr = (vaddr_t) SCARG(uap, addr);
730 	size = (vsize_t) SCARG(uap, len);
731 
732 	/*
733 	 * align the address to a page boundary, and adjust the size accordingly
734 	 */
735 	ALIGN_ADDR(addr, size, pageoff);
736 
737 	/*
738 	 * Check for illegal addresses.  Watch out for address wrap...
739 	 * Note that VM_*_ADDRESS are not constants due to casts (argh).
740 	 */
741 	if (addr > SIZE_MAX - size)
742 		return (EINVAL);
743 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
744 		return (EINVAL);
745 	if (vm_min_address > 0 && addr < vm_min_address)
746 		return (EINVAL);
747 	map = &p->p_vmspace->vm_map;
748 
749 
750 	vm_map_lock(map);	/* lock map so we can checkprot */
751 
752 	/*
753 	 * interesting system call semantic: make sure entire range is
754 	 * allocated before allowing an unmap.
755 	 */
756 
757 	if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
758 		vm_map_unlock(map);
759 		return (EINVAL);
760 	}
761 
762 	/*
763 	 * doit!
764 	 */
765 	uvm_unmap_remove(map, addr, addr + size, &dead_entries, p);
766 
767 	vm_map_unlock(map);	/* and unlock */
768 
769 	if (dead_entries != NULL)
770 		uvm_unmap_detach(dead_entries, 0);
771 
772 	return (0);
773 }
774 
775 /*
776  * sys_mprotect: the mprotect system call
777  */
778 
779 int
780 sys_mprotect(p, v, retval)
781 	struct proc *p;
782 	void *v;
783 	register_t *retval;
784 {
785 	struct sys_mprotect_args /* {
786 		syscallarg(void *) addr;
787 		syscallarg(size_t) len;
788 		syscallarg(int) prot;
789 	} */ *uap = v;
790 	vaddr_t addr;
791 	vsize_t size, pageoff;
792 	vm_prot_t prot;
793 
794 	/*
795 	 * extract syscall args from uap
796 	 */
797 
798 	addr = (vaddr_t)SCARG(uap, addr);
799 	size = (vsize_t)SCARG(uap, len);
800 	prot = SCARG(uap, prot);
801 
802 	if ((prot & VM_PROT_ALL) != prot)
803 		return (EINVAL);
804 
805 	/*
806 	 * align the address to a page boundary, and adjust the size accordingly
807 	 */
808 	ALIGN_ADDR(addr, size, pageoff);
809 	if (addr > SIZE_MAX - size)
810 		return (EINVAL);		/* disallow wrap-around. */
811 
812 	return (uvm_map_protect(&p->p_vmspace->vm_map, addr, addr+size,
813 	    prot, FALSE));
814 }
815 
816 /*
817  * sys_minherit: the minherit system call
818  */
819 
820 int
821 sys_minherit(p, v, retval)
822 	struct proc *p;
823 	void *v;
824 	register_t *retval;
825 {
826 	struct sys_minherit_args /* {
827 		syscallarg(void *) addr;
828 		syscallarg(size_t) len;
829 		syscallarg(int) inherit;
830 	} */ *uap = v;
831 	vaddr_t addr;
832 	vsize_t size, pageoff;
833 	vm_inherit_t inherit;
834 
835 	addr = (vaddr_t)SCARG(uap, addr);
836 	size = (vsize_t)SCARG(uap, len);
837 	inherit = SCARG(uap, inherit);
838 
839 	/*
840 	 * align the address to a page boundary, and adjust the size accordingly
841 	 */
842 	ALIGN_ADDR(addr, size, pageoff);
843 	if (addr > SIZE_MAX - size)
844 		return (EINVAL);		/* disallow wrap-around. */
845 
846 	return (uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
847 	    inherit));
848 }
849 
850 /*
851  * sys_madvise: give advice about memory usage.
852  */
853 
854 /* ARGSUSED */
855 int
856 sys_madvise(p, v, retval)
857 	struct proc *p;
858 	void *v;
859 	register_t *retval;
860 {
861 	struct sys_madvise_args /* {
862 		syscallarg(void *) addr;
863 		syscallarg(size_t) len;
864 		syscallarg(int) behav;
865 	} */ *uap = v;
866 	vaddr_t addr;
867 	vsize_t size, pageoff;
868 	int advice, error;
869 
870 	addr = (vaddr_t)SCARG(uap, addr);
871 	size = (vsize_t)SCARG(uap, len);
872 	advice = SCARG(uap, behav);
873 
874 	/*
875 	 * align the address to a page boundary, and adjust the size accordingly
876 	 */
877 	ALIGN_ADDR(addr, size, pageoff);
878 	if (addr > SIZE_MAX - size)
879 		return (EINVAL);		/* disallow wrap-around. */
880 
881 	switch (advice) {
882 	case MADV_NORMAL:
883 	case MADV_RANDOM:
884 	case MADV_SEQUENTIAL:
885 		error = uvm_map_advice(&p->p_vmspace->vm_map, addr,
886 		    addr + size, advice);
887 		break;
888 
889 	case MADV_WILLNEED:
890 		/*
891 		 * Activate all these pages, pre-faulting them in if
892 		 * necessary.
893 		 */
894 		/*
895 		 * XXX IMPLEMENT ME.
896 		 * Should invent a "weak" mode for uvm_fault()
897 		 * which would only do the PGO_LOCKED pgo_get().
898 		 */
899 		return (0);
900 
901 	case MADV_DONTNEED:
902 		/*
903 		 * Deactivate all these pages.  We don't need them
904 		 * any more.  We don't, however, toss the data in
905 		 * the pages.
906 		 */
907 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
908 		    PGO_DEACTIVATE);
909 		break;
910 
911 	case MADV_FREE:
912 		/*
913 		 * These pages contain no valid data, and may be
914 		 * garbage-collected.  Toss all resources, including
915 		 * any swap space in use.
916 		 */
917 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
918 		    PGO_FREE);
919 		break;
920 
921 	case MADV_SPACEAVAIL:
922 		/*
923 		 * XXXMRG What is this?  I think it's:
924 		 *
925 		 *	Ensure that we have allocated backing-store
926 		 *	for these pages.
927 		 *
928 		 * This is going to require changes to the page daemon,
929 		 * as it will free swap space allocated to pages in core.
930 		 * There's also what to do for device/file/anonymous memory.
931 		 */
932 		return (EINVAL);
933 
934 	default:
935 		return (EINVAL);
936 	}
937 
938 	return (error);
939 }
940 
941 /*
942  * sys_mlock: memory lock
943  */
944 
945 int
946 sys_mlock(p, v, retval)
947 	struct proc *p;
948 	void *v;
949 	register_t *retval;
950 {
951 	struct sys_mlock_args /* {
952 		syscallarg(const void *) addr;
953 		syscallarg(size_t) len;
954 	} */ *uap = v;
955 	vaddr_t addr;
956 	vsize_t size, pageoff;
957 	int error;
958 
959 	/*
960 	 * extract syscall args from uap
961 	 */
962 	addr = (vaddr_t)SCARG(uap, addr);
963 	size = (vsize_t)SCARG(uap, len);
964 
965 	/*
966 	 * align the address to a page boundary and adjust the size accordingly
967 	 */
968 	ALIGN_ADDR(addr, size, pageoff);
969 	if (addr > SIZE_MAX - size)
970 		return (EINVAL);		/* disallow wrap-around. */
971 
972 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
973 		return (EAGAIN);
974 
975 #ifdef pmap_wired_count
976 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
977 			p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
978 		return (EAGAIN);
979 #else
980 	if ((error = suser(p, 0)) != 0)
981 		return (error);
982 #endif
983 
984 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
985 	    0);
986 	return (error == 0 ? 0 : ENOMEM);
987 }
988 
989 /*
990  * sys_munlock: unlock wired pages
991  */
992 
993 int
994 sys_munlock(p, v, retval)
995 	struct proc *p;
996 	void *v;
997 	register_t *retval;
998 {
999 	struct sys_munlock_args /* {
1000 		syscallarg(const void *) addr;
1001 		syscallarg(size_t) len;
1002 	} */ *uap = v;
1003 	vaddr_t addr;
1004 	vsize_t size, pageoff;
1005 	int error;
1006 
1007 	/*
1008 	 * extract syscall args from uap
1009 	 */
1010 
1011 	addr = (vaddr_t)SCARG(uap, addr);
1012 	size = (vsize_t)SCARG(uap, len);
1013 
1014 	/*
1015 	 * align the address to a page boundary, and adjust the size accordingly
1016 	 */
1017 	ALIGN_ADDR(addr, size, pageoff);
1018 	if (addr > SIZE_MAX - size)
1019 		return (EINVAL);		/* disallow wrap-around. */
1020 
1021 #ifndef pmap_wired_count
1022 	if ((error = suser(p, 0)) != 0)
1023 		return (error);
1024 #endif
1025 
1026 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
1027 	    0);
1028 	return (error == 0 ? 0 : ENOMEM);
1029 }
1030 
1031 /*
1032  * sys_mlockall: lock all pages mapped into an address space.
1033  */
1034 
1035 int
1036 sys_mlockall(p, v, retval)
1037 	struct proc *p;
1038 	void *v;
1039 	register_t *retval;
1040 {
1041 	struct sys_mlockall_args /* {
1042 		syscallarg(int) flags;
1043 	} */ *uap = v;
1044 	int error, flags;
1045 
1046 	flags = SCARG(uap, flags);
1047 
1048 	if (flags == 0 ||
1049 	    (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
1050 		return (EINVAL);
1051 
1052 #ifndef pmap_wired_count
1053 	if ((error = suser(p, 0)) != 0)
1054 		return (error);
1055 #endif
1056 
1057 	error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
1058 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
1059 	if (error != 0 && error != ENOMEM)
1060 		return (EAGAIN);
1061 	return (error);
1062 }
1063 
1064 /*
1065  * sys_munlockall: unlock all pages mapped into an address space.
1066  */
1067 
1068 int
1069 sys_munlockall(p, v, retval)
1070 	struct proc *p;
1071 	void *v;
1072 	register_t *retval;
1073 {
1074 
1075 	(void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
1076 	return (0);
1077 }
1078 
1079 /*
1080  * uvm_mmap: internal version of mmap
1081  *
1082  * - used by sys_mmap, exec, and sysv shm
1083  * - handle is a vnode pointer or NULL for MAP_ANON (XXX: not true,
1084  *	sysv shm uses "named anonymous memory")
1085  * - caller must page-align the file offset
1086  */
1087 
1088 int
1089 uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit, p)
1090 	vm_map_t map;
1091 	vaddr_t *addr;
1092 	vsize_t size;
1093 	vm_prot_t prot, maxprot;
1094 	int flags;
1095 	caddr_t handle;		/* XXX: VNODE? */
1096 	voff_t foff;
1097 	vsize_t locklimit;
1098 	struct proc *p;
1099 {
1100 	struct uvm_object *uobj;
1101 	struct vnode *vp;
1102 	int error;
1103 	int advice = UVM_ADV_NORMAL;
1104 	uvm_flag_t uvmflag = 0;
1105 	vsize_t align = 0;	/* userland page size */
1106 
1107 	/*
1108 	 * check params
1109 	 */
1110 
1111 	if (size == 0)
1112 		return(0);
1113 	if (foff & PAGE_MASK)
1114 		return(EINVAL);
1115 	if ((prot & maxprot) != prot)
1116 		return(EINVAL);
1117 
1118 	/*
1119 	 * for non-fixed mappings, round off the suggested address.
1120 	 * for fixed mappings, check alignment and zap old mappings.
1121 	 */
1122 
1123 	if ((flags & MAP_FIXED) == 0) {
1124 		*addr = round_page(*addr);	/* round */
1125 	} else {
1126 		if (*addr & PAGE_MASK)
1127 			return(EINVAL);
1128 		uvmflag |= UVM_FLAG_FIXED;
1129 		uvm_unmap_p(map, *addr, *addr + size, p);	/* zap! */
1130 	}
1131 
1132 	/*
1133 	 * handle anon vs. non-anon mappings.   for non-anon mappings attach
1134 	 * to underlying vm object.
1135 	 */
1136 
1137 	if (flags & MAP_ANON) {
1138 		if ((flags & MAP_FIXED) == 0 && size >= __LDPGSZ)
1139 			align = __LDPGSZ;
1140 		foff = UVM_UNKNOWN_OFFSET;
1141 		uobj = NULL;
1142 		if ((flags & MAP_SHARED) == 0)
1143 			/* XXX: defer amap create */
1144 			uvmflag |= UVM_FLAG_COPYONW;
1145 		else
1146 			/* shared: create amap now */
1147 			uvmflag |= UVM_FLAG_OVERLAY;
1148 
1149 	} else {
1150 
1151 		vp = (struct vnode *) handle;	/* get vnode */
1152 		if (vp->v_type != VCHR) {
1153 			uobj = uvn_attach((void *) vp, (flags & MAP_SHARED) ?
1154 			   maxprot : (maxprot & ~VM_PROT_WRITE));
1155 
1156 #ifndef UBC
1157 			/*
1158 			 * XXXCDC: hack from old code
1159 			 * don't allow vnodes which have been mapped
1160 			 * shared-writeable to persist [forces them to be
1161 			 * flushed out when last reference goes].
1162 			 * XXXCDC: interesting side effect: avoids a bug.
1163 			 * note that in WRITE [ufs_readwrite.c] that we
1164 			 * allocate buffer, uncache, and then do the write.
1165 			 * the problem with this is that if the uncache causes
1166 			 * VM data to be flushed to the same area of the file
1167 			 * we are writing to... in that case we've got the
1168 			 * buffer locked and our process goes to sleep forever.
1169 			 *
1170 			 * XXXCDC: checking maxprot protects us from the
1171 			 * "persistbug" program but this is not a long term
1172 			 * solution.
1173 			 *
1174 			 * XXXCDC: we don't bother calling uncache with the vp
1175 			 * VOP_LOCKed since we know that we are already
1176 			 * holding a valid reference to the uvn (from the
1177 			 * uvn_attach above), and thus it is impossible for
1178 			 * the uncache to kill the uvn and trigger I/O.
1179 			 */
1180 			if (flags & MAP_SHARED) {
1181 				if ((prot & VM_PROT_WRITE) ||
1182 				    (maxprot & VM_PROT_WRITE)) {
1183 					uvm_vnp_uncache(vp);
1184 				}
1185 			}
1186 #else
1187 			/* XXX for now, attach doesn't gain a ref */
1188 			VREF(vp);
1189 #endif
1190 		} else {
1191 			uobj = udv_attach((void *) &vp->v_rdev,
1192 			    (flags & MAP_SHARED) ? maxprot :
1193 			    (maxprot & ~VM_PROT_WRITE), foff, size);
1194 			/*
1195 			 * XXX Some devices don't like to be mapped with
1196 			 * XXX PROT_EXEC, but we don't really have a
1197 			 * XXX better way of handling this, right now
1198 			 */
1199 			if (uobj == NULL && (prot & PROT_EXEC) == 0) {
1200 				maxprot &= ~VM_PROT_EXECUTE;
1201 				uobj = udv_attach((void *) &vp->v_rdev,
1202 				    (flags & MAP_SHARED) ? maxprot :
1203 				    (maxprot & ~VM_PROT_WRITE), foff, size);
1204 			}
1205 			advice = UVM_ADV_RANDOM;
1206 		}
1207 
1208 		if (uobj == NULL)
1209 			return((vp->v_type == VREG) ? ENOMEM : EINVAL);
1210 
1211 		if ((flags & MAP_SHARED) == 0)
1212 			uvmflag |= UVM_FLAG_COPYONW;
1213 	}
1214 
1215 	/*
1216 	 * set up mapping flags
1217 	 */
1218 
1219 	uvmflag = UVM_MAPFLAG(prot, maxprot,
1220 			(flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
1221 			advice, uvmflag);
1222 
1223 	error = uvm_map_p(map, addr, size, uobj, foff, align, uvmflag, p);
1224 
1225 	if (error == 0) {
1226 		/*
1227 		 * POSIX 1003.1b -- if our address space was configured
1228 		 * to lock all future mappings, wire the one we just made.
1229 		 */
1230 		if (prot == VM_PROT_NONE) {
1231 			/*
1232 			 * No more work to do in this case.
1233 			 */
1234 			return (0);
1235 		}
1236 
1237 		vm_map_lock(map);
1238 
1239 		if (map->flags & VM_MAP_WIREFUTURE) {
1240 			if ((atop(size) + uvmexp.wired) > uvmexp.wiredmax
1241 #ifdef pmap_wired_count
1242 			    || (locklimit != 0 && (size +
1243 			         ptoa(pmap_wired_count(vm_map_pmap(map)))) >
1244 			        locklimit)
1245 #endif
1246 			) {
1247 				error = ENOMEM;
1248 				vm_map_unlock(map);
1249 				/* unmap the region! */
1250 				uvm_unmap(map, *addr, *addr + size);
1251 				goto bad;
1252 			}
1253 			/*
1254 			 * uvm_map_pageable() always returns the map
1255 			 * unlocked.
1256 			 */
1257 			error = uvm_map_pageable(map, *addr, *addr + size,
1258 			    FALSE, UVM_LK_ENTER);
1259 			if (error != 0) {
1260 				/* unmap the region! */
1261 				uvm_unmap(map, *addr, *addr + size);
1262 				goto bad;
1263 			}
1264 			return (0);
1265 		}
1266 
1267 		vm_map_unlock(map);
1268 
1269 		return (0);
1270 	}
1271 
1272 	/*
1273 	 * errors: first detach from the uobj, if any.
1274 	 */
1275 
1276 	if (uobj)
1277 		uobj->pgops->pgo_detach(uobj);
1278 
1279 bad:
1280 	return (error);
1281 }
1282