xref: /openbsd-src/sys/uvm/uvm_mmap.c (revision a28daedfc357b214be5c701aa8ba8adb29a7f1c2)
1 /*	$OpenBSD: uvm_mmap.c,v 1.72 2009/03/20 15:19:04 oga 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(struct proc *p, void *v, register_t *retval)
105 {
106 #if 0
107 	struct sys_sbrk_args /* {
108 		syscallarg(int) incr;
109 	} */ *uap = v;
110 #endif
111 
112 	return (ENOSYS);
113 }
114 
115 /*
116  * sys_sstk: sstk system call.
117  */
118 
119 /* ARGSUSED */
120 int
121 sys_sstk(struct proc *p, void *v, register_t *retval)
122 {
123 #if 0
124 	struct sys_sstk_args /* {
125 		syscallarg(int) incr;
126 	} */ *uap = v;
127 #endif
128 
129 	return (ENOSYS);
130 }
131 
132 /*
133  * sys_mquery: provide mapping hints to applications that do fixed mappings
134  *
135  * flags: 0 or MAP_FIXED (MAP_FIXED - means that we insist on this addr and
136  *	don't care about PMAP_PREFER or such)
137  * addr: hint where we'd like to place the mapping.
138  * size: size of the mapping
139  * fd: fd of the file we want to map
140  * off: offset within the file
141  */
142 
143 int
144 sys_mquery(struct proc *p, void *v, register_t *retval)
145 {
146 	struct sys_mquery_args /* {
147 		syscallarg(void *) addr;
148 		syscallarg(size_t) len;
149 		syscallarg(int) prot;
150 		syscallarg(int) flags;
151 		syscallarg(int) fd;
152 		syscallarg(long) pad;
153 		syscallarg(off_t) pos;
154 	} */ *uap = v;
155 	struct file *fp;
156 	struct uvm_object *uobj;
157 	voff_t uoff;
158 	int error;
159 	vaddr_t vaddr;
160 	int flags = 0;
161 	vsize_t size;
162 	vm_prot_t prot;
163 	int fd;
164 
165 	vaddr = (vaddr_t) SCARG(uap, addr);
166 	prot = SCARG(uap, prot);
167 	size = (vsize_t) SCARG(uap, len);
168 	fd = SCARG(uap, fd);
169 
170 	if ((prot & VM_PROT_ALL) != prot)
171 		return (EINVAL);
172 
173 	if (SCARG(uap, flags) & MAP_FIXED)
174 		flags |= UVM_FLAG_FIXED;
175 
176 	if (fd >= 0) {
177 		if ((error = getvnode(p->p_fd, fd, &fp)) != 0)
178 			return (error);
179 		uobj = &((struct vnode *)fp->f_data)->v_uvm.u_obj;
180 		uoff = SCARG(uap, pos);
181 	} else {
182 		fp = NULL;
183 		uobj = NULL;
184 		uoff = 0;
185 	}
186 
187 	if (vaddr == 0)
188 		vaddr = uvm_map_hint(p, prot);
189 
190 	/* prevent a user requested address from falling in heap space */
191 	if ((vaddr + size > (vaddr_t)p->p_vmspace->vm_daddr) &&
192 	    (vaddr < (vaddr_t)p->p_vmspace->vm_daddr + MAXDSIZ)) {
193 		if (flags & UVM_FLAG_FIXED) {
194 			error = EINVAL;
195 			goto done;
196 		}
197 		vaddr = round_page((vaddr_t)p->p_vmspace->vm_daddr + MAXDSIZ);
198 	}
199 	vm_map_lock(&p->p_vmspace->vm_map);
200 
201 again:
202 	if (uvm_map_findspace(&p->p_vmspace->vm_map, vaddr, size,
203 	    &vaddr, uobj, uoff, 0, flags) == NULL) {
204 		if (flags & UVM_FLAG_FIXED)
205 			error = EINVAL;
206 		else
207 			error = ENOMEM;
208 	} else {
209 		/* prevent a returned address from falling in heap space */
210 		if ((vaddr + size > (vaddr_t)p->p_vmspace->vm_daddr)
211 		    && (vaddr < (vaddr_t)p->p_vmspace->vm_daddr + MAXDSIZ)) {
212 			vaddr = round_page((vaddr_t)p->p_vmspace->vm_daddr +
213 			    MAXDSIZ);
214 			goto again;
215 		}
216 		error = 0;
217 		*retval = (register_t)(vaddr);
218 	}
219 	vm_map_unlock(&p->p_vmspace->vm_map);
220 done:
221 	if (fp != NULL)
222 		FRELE(fp);
223 	return (error);
224 }
225 
226 /*
227  * sys_mincore: determine if pages are in core or not.
228  */
229 
230 /* ARGSUSED */
231 int
232 sys_mincore(struct proc *p, void *v, register_t *retval)
233 {
234 	struct sys_mincore_args /* {
235 		syscallarg(void *) addr;
236 		syscallarg(size_t) len;
237 		syscallarg(char *) vec;
238 	} */ *uap = v;
239 	vm_page_t m;
240 	char *vec, pgi;
241 	struct uvm_object *uobj;
242 	struct vm_amap *amap;
243 	struct vm_anon *anon;
244 	vm_map_entry_t entry;
245 	vaddr_t start, end, lim;
246 	vm_map_t map;
247 	vsize_t len, npgs;
248 	int error = 0;
249 
250 	map = &p->p_vmspace->vm_map;
251 
252 	start = (vaddr_t)SCARG(uap, addr);
253 	len = SCARG(uap, len);
254 	vec = SCARG(uap, vec);
255 
256 	if (start & PAGE_MASK)
257 		return (EINVAL);
258 	len = round_page(len);
259 	end = start + len;
260 	if (end <= start)
261 		return (EINVAL);
262 
263 	npgs = len >> PAGE_SHIFT;
264 
265 	/*
266 	 * Lock down vec, so our returned status isn't outdated by
267 	 * storing the status byte for a page.
268 	 */
269 	if ((error = uvm_vslock(p, vec, npgs, VM_PROT_WRITE)) != 0)
270 		return (error);
271 
272 	vm_map_lock_read(map);
273 
274 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
275 		error = ENOMEM;
276 		goto out;
277 	}
278 
279 	for (/* nothing */;
280 	     entry != &map->header && entry->start < end;
281 	     entry = entry->next) {
282 		KASSERT(!UVM_ET_ISSUBMAP(entry));
283 		KASSERT(start >= entry->start);
284 
285 		/* Make sure there are no holes. */
286 		if (entry->end < end &&
287 		     (entry->next == &map->header ||
288 		      entry->next->start > entry->end)) {
289 			error = ENOMEM;
290 			goto out;
291 		}
292 
293 		lim = end < entry->end ? end : entry->end;
294 
295 		/*
296 		 * Special case for objects with no "real" pages.  Those
297 		 * are always considered resident (mapped devices).
298 		 */
299 		if (UVM_ET_ISOBJ(entry)) {
300 			KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj));
301 			if (entry->object.uvm_obj->pgops->pgo_releasepg
302 			    == NULL) {
303 				pgi = 1;
304 				for (/* nothing */; start < lim;
305 				     start += PAGE_SIZE, vec++)
306 					copyout(&pgi, vec, sizeof(char));
307 				continue;
308 			}
309 		}
310 
311 		amap = entry->aref.ar_amap;	/* top layer */
312 		uobj = entry->object.uvm_obj;	/* bottom layer */
313 
314 		if (uobj != NULL)
315 			simple_lock(&uobj->vmobjlock);
316 
317 		for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) {
318 			pgi = 0;
319 			if (amap != NULL) {
320 				/* Check the top layer first. */
321 				anon = amap_lookup(&entry->aref,
322 				    start - entry->start);
323 				/* Don't need to lock anon here. */
324 				if (anon != NULL && anon->an_page != NULL) {
325 					/*
326 					 * Anon has the page for this entry
327 					 * offset.
328 					 */
329 					pgi = 1;
330 				}
331 			}
332 
333 			if (uobj != NULL && pgi == 0) {
334 				/* Check the bottom layer. */
335 				m = uvm_pagelookup(uobj,
336 				    entry->offset + (start - entry->start));
337 				if (m != NULL) {
338 					/*
339 					 * Object has the page for this entry
340 					 * offset.
341 					 */
342 					pgi = 1;
343 				}
344 			}
345 
346 			copyout(&pgi, vec, sizeof(char));
347 		}
348 
349 		if (uobj != NULL)
350 			simple_unlock(&uobj->vmobjlock);
351 	}
352 
353  out:
354 	vm_map_unlock_read(map);
355 	uvm_vsunlock(p, SCARG(uap, vec), npgs);
356 	return (error);
357 }
358 
359 /*
360  * sys_mmap: mmap system call.
361  *
362  * => file offset and address may not be page aligned
363  *    - if MAP_FIXED, offset and address must have remainder mod PAGE_SIZE
364  *    - if address isn't page aligned the mapping starts at trunc_page(addr)
365  *      and the return value is adjusted up by the page offset.
366  */
367 
368 int
369 sys_mmap(struct proc *p, void *v, register_t *retval)
370 {
371 	struct sys_mmap_args /* {
372 		syscallarg(void *) addr;
373 		syscallarg(size_t) len;
374 		syscallarg(int) prot;
375 		syscallarg(int) flags;
376 		syscallarg(int) fd;
377 		syscallarg(long) pad;
378 		syscallarg(off_t) pos;
379 	} */ *uap = v;
380 	vaddr_t addr;
381 	struct vattr va;
382 	off_t pos;
383 	vsize_t size, pageoff;
384 	vm_prot_t prot, maxprot;
385 	int flags, fd;
386 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
387 	struct filedesc *fdp = p->p_fd;
388 	struct file *fp = NULL;
389 	struct vnode *vp;
390 	caddr_t handle;
391 	int error;
392 
393 	/*
394 	 * first, extract syscall args from the uap.
395 	 */
396 
397 	addr = (vaddr_t) SCARG(uap, addr);
398 	size = (vsize_t) SCARG(uap, len);
399 	prot = SCARG(uap, prot);
400 	flags = SCARG(uap, flags);
401 	fd = SCARG(uap, fd);
402 	pos = SCARG(uap, pos);
403 
404 	/*
405 	 * Fixup the old deprecated MAP_COPY into MAP_PRIVATE, and
406 	 * validate the flags.
407 	 */
408 	if ((prot & VM_PROT_ALL) != prot)
409 		return (EINVAL);
410 	if ((flags & MAP_FLAGMASK) != flags)
411 		return (EINVAL);
412 	if (flags & MAP_COPY)
413 		flags = (flags & ~MAP_COPY) | MAP_PRIVATE;
414 	if ((flags & (MAP_SHARED|MAP_PRIVATE)) == (MAP_SHARED|MAP_PRIVATE))
415 		return (EINVAL);
416 
417 	/*
418 	 * align file position and save offset.  adjust size.
419 	 */
420 	ALIGN_ADDR(pos, size, pageoff);
421 
422 	/*
423 	 * now check (MAP_FIXED) or get (!MAP_FIXED) the "addr"
424 	 */
425 
426 	if (flags & MAP_FIXED) {
427 
428 		/* adjust address by the same amount as we did the offset */
429 		addr -= pageoff;
430 		if (addr & PAGE_MASK)
431 			return (EINVAL);		/* not page aligned */
432 
433 		if (addr > SIZE_MAX - size)
434 			return (EINVAL);		/* no wrapping! */
435 		if (VM_MAXUSER_ADDRESS > 0 &&
436 		    (addr + size) > VM_MAXUSER_ADDRESS)
437 			return (EINVAL);
438 		if (vm_min_address > 0 && addr < vm_min_address)
439 			return (EINVAL);
440 
441 	} else {
442 
443 		/*
444 		 * not fixed: make sure we skip over the largest possible heap.
445 		 * we will refine our guess later (e.g. to account for VAC, etc)
446 		 */
447 		if (addr == 0)
448 			addr = uvm_map_hint(p, prot);
449 		else if (!(flags & MAP_TRYFIXED) &&
450 		    addr < (vaddr_t)p->p_vmspace->vm_daddr)
451 			addr = uvm_map_hint(p, prot);
452 	}
453 
454 	/*
455 	 * check for file mappings (i.e. not anonymous) and verify file.
456 	 */
457 	if ((flags & MAP_ANON) == 0) {
458 
459 		if ((fp = fd_getfile(fdp, fd)) == NULL)
460 			return (EBADF);
461 
462 		FREF(fp);
463 
464 		if (fp->f_type != DTYPE_VNODE) {
465 			error = ENODEV;		/* only mmap vnodes! */
466 			goto out;
467 		}
468 		vp = (struct vnode *)fp->f_data;	/* convert to vnode */
469 
470 		if (vp->v_type != VREG && vp->v_type != VCHR &&
471 		    vp->v_type != VBLK) {
472 			error = ENODEV; /* only REG/CHR/BLK support mmap */
473 			goto out;
474 		}
475 
476 		if (vp->v_type == VREG && (pos + size) < pos) {
477 			error = EINVAL;		/* no offset wrapping */
478 			goto out;
479 		}
480 
481 		/* special case: catch SunOS style /dev/zero */
482 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
483 			flags |= MAP_ANON;
484 			FRELE(fp);
485 			fp = NULL;
486 			goto is_anon;
487 		}
488 
489 		/*
490 		 * Old programs may not select a specific sharing type, so
491 		 * default to an appropriate one.
492 		 *
493 		 * XXX: how does MAP_ANON fit in the picture?
494 		 */
495 		if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
496 #if defined(DEBUG)
497 			printf("WARNING: defaulted mmap() share type to "
498 			   "%s (pid %d comm %s)\n", vp->v_type == VCHR ?
499 			   "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
500 			    p->p_comm);
501 #endif
502 			if (vp->v_type == VCHR)
503 				flags |= MAP_SHARED;	/* for a device */
504 			else
505 				flags |= MAP_PRIVATE;	/* for a file */
506 		}
507 
508 		/*
509 		 * MAP_PRIVATE device mappings don't make sense (and aren't
510 		 * supported anyway).  However, some programs rely on this,
511 		 * so just change it to MAP_SHARED.
512 		 */
513 		if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
514 			flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
515 		}
516 
517 		/*
518 		 * now check protection
519 		 */
520 
521 		maxprot = VM_PROT_EXECUTE;
522 
523 		/* check read access */
524 		if (fp->f_flag & FREAD)
525 			maxprot |= VM_PROT_READ;
526 		else if (prot & PROT_READ) {
527 			error = EACCES;
528 			goto out;
529 		}
530 
531 		/* check write access, shared case first */
532 		if (flags & MAP_SHARED) {
533 			/*
534 			 * if the file is writable, only add PROT_WRITE to
535 			 * maxprot if the file is not immutable, append-only.
536 			 * otherwise, if we have asked for PROT_WRITE, return
537 			 * EPERM.
538 			 */
539 			if (fp->f_flag & FWRITE) {
540 				if ((error =
541 				    VOP_GETATTR(vp, &va, p->p_ucred, p)))
542 					goto out;
543 				if ((va.va_flags & (IMMUTABLE|APPEND)) == 0)
544 					maxprot |= VM_PROT_WRITE;
545 				else if (prot & PROT_WRITE) {
546 					error = EPERM;
547 					goto out;
548 				}
549 			} else if (prot & PROT_WRITE) {
550 				error = EACCES;
551 				goto out;
552 			}
553 		} else {
554 			/* MAP_PRIVATE mappings can always write to */
555 			maxprot |= VM_PROT_WRITE;
556 		}
557 
558 		/*
559 		 * set handle to vnode
560 		 */
561 
562 		handle = (caddr_t)vp;
563 
564 	} else {		/* MAP_ANON case */
565 		/*
566 		 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0?
567 		 */
568 		if (fd != -1) {
569 			error = EINVAL;
570 			goto out;
571 		}
572 
573  is_anon:		/* label for SunOS style /dev/zero */
574 		handle = NULL;
575 		maxprot = VM_PROT_ALL;
576 		pos = 0;
577 	}
578 
579 	if ((flags & MAP_ANON) != 0 ||
580 	    ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) {
581 		if (size >
582 		    (p->p_rlimit[RLIMIT_DATA].rlim_cur - ptoa(p->p_vmspace->vm_dused))) {
583 			error = ENOMEM;
584 			goto out;
585 		}
586 	}
587 
588 	/*
589 	 * now let kernel internal function uvm_mmap do the work.
590 	 */
591 
592 	error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot,
593 	    flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur, p);
594 
595 	if (error == 0)
596 		/* remember to add offset */
597 		*retval = (register_t)(addr + pageoff);
598 
599 out:
600 	if (fp)
601 		FRELE(fp);
602 	return (error);
603 }
604 
605 /*
606  * sys_msync: the msync system call (a front-end for flush)
607  */
608 
609 int
610 sys_msync(struct proc *p, void *v, register_t *retval)
611 {
612 	struct sys_msync_args /* {
613 		syscallarg(void *) addr;
614 		syscallarg(size_t) len;
615 		syscallarg(int) flags;
616 	} */ *uap = v;
617 	vaddr_t addr;
618 	vsize_t size, pageoff;
619 	vm_map_t map;
620 	int rv, flags, uvmflags;
621 
622 	/*
623 	 * extract syscall args from the uap
624 	 */
625 
626 	addr = (vaddr_t)SCARG(uap, addr);
627 	size = (vsize_t)SCARG(uap, len);
628 	flags = SCARG(uap, flags);
629 
630 	/* sanity check flags */
631 	if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 ||
632 			(flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 ||
633 			(flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC))
634 		return (EINVAL);
635 	if ((flags & (MS_ASYNC | MS_SYNC)) == 0)
636 		flags |= MS_SYNC;
637 
638 	/*
639 	 * align the address to a page boundary, and adjust the size accordingly
640 	 */
641 	ALIGN_ADDR(addr, size, pageoff);
642 	if (addr > SIZE_MAX - size)
643 		return (EINVAL);		/* disallow wrap-around. */
644 
645 	/*
646 	 * get map
647 	 */
648 
649 	map = &p->p_vmspace->vm_map;
650 
651 	/*
652 	 * XXXCDC: do we really need this semantic?
653 	 *
654 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
655 	 * pages with the region containing addr".  Unfortunately, we
656 	 * don't really keep track of individual mmaps so we approximate
657 	 * by flushing the range of the map entry containing addr.
658 	 * This can be incorrect if the region splits or is coalesced
659 	 * with a neighbor.
660 	 */
661 	if (size == 0) {
662 		vm_map_entry_t entry;
663 
664 		vm_map_lock_read(map);
665 		rv = uvm_map_lookup_entry(map, addr, &entry);
666 		if (rv == TRUE) {
667 			addr = entry->start;
668 			size = entry->end - entry->start;
669 		}
670 		vm_map_unlock_read(map);
671 		if (rv == FALSE)
672 			return (EINVAL);
673 	}
674 
675 	/*
676 	 * translate MS_ flags into PGO_ flags
677 	 */
678 	uvmflags = PGO_CLEANIT;
679 	if (flags & MS_INVALIDATE)
680 		uvmflags |= PGO_FREE;
681 	if (flags & MS_SYNC)
682 		uvmflags |= PGO_SYNCIO;
683 	else
684 		uvmflags |= PGO_SYNCIO;	 /* XXXCDC: force sync for now! */
685 
686 	return (uvm_map_clean(map, addr, addr+size, uvmflags));
687 }
688 
689 /*
690  * sys_munmap: unmap a users memory
691  */
692 
693 int
694 sys_munmap(struct proc *p, void *v, register_t *retval)
695 {
696 	struct sys_munmap_args /* {
697 		syscallarg(void *) addr;
698 		syscallarg(size_t) len;
699 	} */ *uap = v;
700 	vaddr_t addr;
701 	vsize_t size, pageoff;
702 	vm_map_t map;
703 	vaddr_t vm_min_address = VM_MIN_ADDRESS;
704 	struct vm_map_entry *dead_entries;
705 
706 	/*
707 	 * get syscall args...
708 	 */
709 
710 	addr = (vaddr_t) SCARG(uap, addr);
711 	size = (vsize_t) SCARG(uap, len);
712 
713 	/*
714 	 * align the address to a page boundary, and adjust the size accordingly
715 	 */
716 	ALIGN_ADDR(addr, size, pageoff);
717 
718 	/*
719 	 * Check for illegal addresses.  Watch out for address wrap...
720 	 * Note that VM_*_ADDRESS are not constants due to casts (argh).
721 	 */
722 	if (addr > SIZE_MAX - size)
723 		return (EINVAL);
724 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
725 		return (EINVAL);
726 	if (vm_min_address > 0 && addr < vm_min_address)
727 		return (EINVAL);
728 	map = &p->p_vmspace->vm_map;
729 
730 
731 	vm_map_lock(map);	/* lock map so we can checkprot */
732 
733 	/*
734 	 * interesting system call semantic: make sure entire range is
735 	 * allocated before allowing an unmap.
736 	 */
737 
738 	if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
739 		vm_map_unlock(map);
740 		return (EINVAL);
741 	}
742 
743 	/*
744 	 * doit!
745 	 */
746 	uvm_unmap_remove(map, addr, addr + size, &dead_entries, p);
747 
748 	vm_map_unlock(map);	/* and unlock */
749 
750 	if (dead_entries != NULL)
751 		uvm_unmap_detach(dead_entries, 0);
752 
753 	return (0);
754 }
755 
756 /*
757  * sys_mprotect: the mprotect system call
758  */
759 
760 int
761 sys_mprotect(struct proc *p, void *v, register_t *retval)
762 {
763 	struct sys_mprotect_args /* {
764 		syscallarg(void *) addr;
765 		syscallarg(size_t) len;
766 		syscallarg(int) prot;
767 	} */ *uap = v;
768 	vaddr_t addr;
769 	vsize_t size, pageoff;
770 	vm_prot_t prot;
771 
772 	/*
773 	 * extract syscall args from uap
774 	 */
775 
776 	addr = (vaddr_t)SCARG(uap, addr);
777 	size = (vsize_t)SCARG(uap, len);
778 	prot = SCARG(uap, prot);
779 
780 	if ((prot & VM_PROT_ALL) != prot)
781 		return (EINVAL);
782 
783 	/*
784 	 * align the address to a page boundary, and adjust the size accordingly
785 	 */
786 	ALIGN_ADDR(addr, size, pageoff);
787 	if (addr > SIZE_MAX - size)
788 		return (EINVAL);		/* disallow wrap-around. */
789 
790 	return (uvm_map_protect(&p->p_vmspace->vm_map, addr, addr+size,
791 	    prot, FALSE));
792 }
793 
794 /*
795  * sys_minherit: the minherit system call
796  */
797 
798 int
799 sys_minherit(struct proc *p, void *v, register_t *retval)
800 {
801 	struct sys_minherit_args /* {
802 		syscallarg(void *) addr;
803 		syscallarg(size_t) len;
804 		syscallarg(int) inherit;
805 	} */ *uap = v;
806 	vaddr_t addr;
807 	vsize_t size, pageoff;
808 	vm_inherit_t inherit;
809 
810 	addr = (vaddr_t)SCARG(uap, addr);
811 	size = (vsize_t)SCARG(uap, len);
812 	inherit = SCARG(uap, inherit);
813 
814 	/*
815 	 * align the address to a page boundary, and adjust the size accordingly
816 	 */
817 	ALIGN_ADDR(addr, size, pageoff);
818 	if (addr > SIZE_MAX - size)
819 		return (EINVAL);		/* disallow wrap-around. */
820 
821 	return (uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
822 	    inherit));
823 }
824 
825 /*
826  * sys_madvise: give advice about memory usage.
827  */
828 
829 /* ARGSUSED */
830 int
831 sys_madvise(struct proc *p, void *v, register_t *retval)
832 {
833 	struct sys_madvise_args /* {
834 		syscallarg(void *) addr;
835 		syscallarg(size_t) len;
836 		syscallarg(int) behav;
837 	} */ *uap = v;
838 	vaddr_t addr;
839 	vsize_t size, pageoff;
840 	int advice, error;
841 
842 	addr = (vaddr_t)SCARG(uap, addr);
843 	size = (vsize_t)SCARG(uap, len);
844 	advice = SCARG(uap, behav);
845 
846 	/*
847 	 * align the address to a page boundary, and adjust the size accordingly
848 	 */
849 	ALIGN_ADDR(addr, size, pageoff);
850 	if (addr > SIZE_MAX - size)
851 		return (EINVAL);		/* disallow wrap-around. */
852 
853 	switch (advice) {
854 	case MADV_NORMAL:
855 	case MADV_RANDOM:
856 	case MADV_SEQUENTIAL:
857 		error = uvm_map_advice(&p->p_vmspace->vm_map, addr,
858 		    addr + size, advice);
859 		break;
860 
861 	case MADV_WILLNEED:
862 		/*
863 		 * Activate all these pages, pre-faulting them in if
864 		 * necessary.
865 		 */
866 		/*
867 		 * XXX IMPLEMENT ME.
868 		 * Should invent a "weak" mode for uvm_fault()
869 		 * which would only do the PGO_LOCKED pgo_get().
870 		 */
871 		return (0);
872 
873 	case MADV_DONTNEED:
874 		/*
875 		 * Deactivate all these pages.  We don't need them
876 		 * any more.  We don't, however, toss the data in
877 		 * the pages.
878 		 */
879 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
880 		    PGO_DEACTIVATE);
881 		break;
882 
883 	case MADV_FREE:
884 		/*
885 		 * These pages contain no valid data, and may be
886 		 * garbage-collected.  Toss all resources, including
887 		 * any swap space in use.
888 		 */
889 		error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
890 		    PGO_FREE);
891 		break;
892 
893 	case MADV_SPACEAVAIL:
894 		/*
895 		 * XXXMRG What is this?  I think it's:
896 		 *
897 		 *	Ensure that we have allocated backing-store
898 		 *	for these pages.
899 		 *
900 		 * This is going to require changes to the page daemon,
901 		 * as it will free swap space allocated to pages in core.
902 		 * There's also what to do for device/file/anonymous memory.
903 		 */
904 		return (EINVAL);
905 
906 	default:
907 		return (EINVAL);
908 	}
909 
910 	return (error);
911 }
912 
913 /*
914  * sys_mlock: memory lock
915  */
916 
917 int
918 sys_mlock(struct proc *p, void *v, register_t *retval)
919 {
920 	struct sys_mlock_args /* {
921 		syscallarg(const void *) addr;
922 		syscallarg(size_t) len;
923 	} */ *uap = v;
924 	vaddr_t addr;
925 	vsize_t size, pageoff;
926 	int error;
927 
928 	/*
929 	 * extract syscall args from uap
930 	 */
931 	addr = (vaddr_t)SCARG(uap, addr);
932 	size = (vsize_t)SCARG(uap, len);
933 
934 	/*
935 	 * align the address to a page boundary and adjust the size accordingly
936 	 */
937 	ALIGN_ADDR(addr, size, pageoff);
938 	if (addr > SIZE_MAX - size)
939 		return (EINVAL);		/* disallow wrap-around. */
940 
941 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
942 		return (EAGAIN);
943 
944 #ifdef pmap_wired_count
945 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
946 			p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
947 		return (EAGAIN);
948 #else
949 	if ((error = suser(p, 0)) != 0)
950 		return (error);
951 #endif
952 
953 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
954 	    0);
955 	return (error == 0 ? 0 : ENOMEM);
956 }
957 
958 /*
959  * sys_munlock: unlock wired pages
960  */
961 
962 int
963 sys_munlock(struct proc *p, void *v, register_t *retval)
964 {
965 	struct sys_munlock_args /* {
966 		syscallarg(const void *) addr;
967 		syscallarg(size_t) len;
968 	} */ *uap = v;
969 	vaddr_t addr;
970 	vsize_t size, pageoff;
971 	int error;
972 
973 	/*
974 	 * extract syscall args from uap
975 	 */
976 
977 	addr = (vaddr_t)SCARG(uap, addr);
978 	size = (vsize_t)SCARG(uap, len);
979 
980 	/*
981 	 * align the address to a page boundary, and adjust the size accordingly
982 	 */
983 	ALIGN_ADDR(addr, size, pageoff);
984 	if (addr > SIZE_MAX - size)
985 		return (EINVAL);		/* disallow wrap-around. */
986 
987 #ifndef pmap_wired_count
988 	if ((error = suser(p, 0)) != 0)
989 		return (error);
990 #endif
991 
992 	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
993 	    0);
994 	return (error == 0 ? 0 : ENOMEM);
995 }
996 
997 /*
998  * sys_mlockall: lock all pages mapped into an address space.
999  */
1000 
1001 int
1002 sys_mlockall(struct proc *p, void *v, register_t *retval)
1003 {
1004 	struct sys_mlockall_args /* {
1005 		syscallarg(int) flags;
1006 	} */ *uap = v;
1007 	int error, flags;
1008 
1009 	flags = SCARG(uap, flags);
1010 
1011 	if (flags == 0 ||
1012 	    (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
1013 		return (EINVAL);
1014 
1015 #ifndef pmap_wired_count
1016 	if ((error = suser(p, 0)) != 0)
1017 		return (error);
1018 #endif
1019 
1020 	error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
1021 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
1022 	if (error != 0 && error != ENOMEM)
1023 		return (EAGAIN);
1024 	return (error);
1025 }
1026 
1027 /*
1028  * sys_munlockall: unlock all pages mapped into an address space.
1029  */
1030 
1031 int
1032 sys_munlockall(struct proc *p, void *v, register_t *retval)
1033 {
1034 
1035 	(void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
1036 	return (0);
1037 }
1038 
1039 /*
1040  * uvm_mmap: internal version of mmap
1041  *
1042  * - used by sys_mmap, exec, and sysv shm
1043  * - handle is a vnode pointer or NULL for MAP_ANON (XXX: not true,
1044  *	sysv shm uses "named anonymous memory")
1045  * - caller must page-align the file offset
1046  */
1047 
1048 int
1049 uvm_mmap(vm_map_t map, vaddr_t *addr, vsize_t size, vm_prot_t prot,
1050     vm_prot_t maxprot, int flags, caddr_t handle, voff_t foff,
1051     vsize_t locklimit, struct proc *p)
1052 {
1053 	struct uvm_object *uobj;
1054 	struct vnode *vp;
1055 	int error;
1056 	int advice = UVM_ADV_NORMAL;
1057 	uvm_flag_t uvmflag = 0;
1058 	vsize_t align = 0;	/* userland page size */
1059 
1060 	/*
1061 	 * check params
1062 	 */
1063 
1064 	if (size == 0)
1065 		return(0);
1066 	if (foff & PAGE_MASK)
1067 		return(EINVAL);
1068 	if ((prot & maxprot) != prot)
1069 		return(EINVAL);
1070 
1071 	/*
1072 	 * for non-fixed mappings, round off the suggested address.
1073 	 * for fixed mappings, check alignment and zap old mappings.
1074 	 */
1075 
1076 	if ((flags & MAP_FIXED) == 0) {
1077 		*addr = round_page(*addr);	/* round */
1078 	} else {
1079 		if (*addr & PAGE_MASK)
1080 			return(EINVAL);
1081 		uvmflag |= UVM_FLAG_FIXED;
1082 		uvm_unmap_p(map, *addr, *addr + size, p);	/* zap! */
1083 	}
1084 
1085 	/*
1086 	 * handle anon vs. non-anon mappings.   for non-anon mappings attach
1087 	 * to underlying vm object.
1088 	 */
1089 
1090 	if (flags & MAP_ANON) {
1091 		if ((flags & MAP_FIXED) == 0 && size >= __LDPGSZ)
1092 			align = __LDPGSZ;
1093 		foff = UVM_UNKNOWN_OFFSET;
1094 		uobj = NULL;
1095 		if ((flags & MAP_SHARED) == 0)
1096 			/* XXX: defer amap create */
1097 			uvmflag |= UVM_FLAG_COPYONW;
1098 		else
1099 			/* shared: create amap now */
1100 			uvmflag |= UVM_FLAG_OVERLAY;
1101 
1102 	} else {
1103 
1104 		vp = (struct vnode *) handle;	/* get vnode */
1105 		if (vp->v_type != VCHR) {
1106 			uobj = uvn_attach((void *) vp, (flags & MAP_SHARED) ?
1107 			   maxprot : (maxprot & ~VM_PROT_WRITE));
1108 
1109 #ifndef UBC
1110 			/*
1111 			 * XXXCDC: hack from old code
1112 			 * don't allow vnodes which have been mapped
1113 			 * shared-writeable to persist [forces them to be
1114 			 * flushed out when last reference goes].
1115 			 * XXXCDC: interesting side effect: avoids a bug.
1116 			 * note that in WRITE [ufs_readwrite.c] that we
1117 			 * allocate buffer, uncache, and then do the write.
1118 			 * the problem with this is that if the uncache causes
1119 			 * VM data to be flushed to the same area of the file
1120 			 * we are writing to... in that case we've got the
1121 			 * buffer locked and our process goes to sleep forever.
1122 			 *
1123 			 * XXXCDC: checking maxprot protects us from the
1124 			 * "persistbug" program but this is not a long term
1125 			 * solution.
1126 			 *
1127 			 * XXXCDC: we don't bother calling uncache with the vp
1128 			 * VOP_LOCKed since we know that we are already
1129 			 * holding a valid reference to the uvn (from the
1130 			 * uvn_attach above), and thus it is impossible for
1131 			 * the uncache to kill the uvn and trigger I/O.
1132 			 */
1133 			if (flags & MAP_SHARED) {
1134 				if ((prot & VM_PROT_WRITE) ||
1135 				    (maxprot & VM_PROT_WRITE)) {
1136 					uvm_vnp_uncache(vp);
1137 				}
1138 			}
1139 #else
1140 			/* XXX for now, attach doesn't gain a ref */
1141 			VREF(vp);
1142 #endif
1143 		} else {
1144 			uobj = udv_attach((void *) &vp->v_rdev,
1145 			    (flags & MAP_SHARED) ? maxprot :
1146 			    (maxprot & ~VM_PROT_WRITE), foff, size);
1147 			/*
1148 			 * XXX Some devices don't like to be mapped with
1149 			 * XXX PROT_EXEC, but we don't really have a
1150 			 * XXX better way of handling this, right now
1151 			 */
1152 			if (uobj == NULL && (prot & PROT_EXEC) == 0) {
1153 				maxprot &= ~VM_PROT_EXECUTE;
1154 				uobj = udv_attach((void *) &vp->v_rdev,
1155 				    (flags & MAP_SHARED) ? maxprot :
1156 				    (maxprot & ~VM_PROT_WRITE), foff, size);
1157 			}
1158 			advice = UVM_ADV_RANDOM;
1159 		}
1160 
1161 		if (uobj == NULL)
1162 			return((vp->v_type == VREG) ? ENOMEM : EINVAL);
1163 
1164 		if ((flags & MAP_SHARED) == 0)
1165 			uvmflag |= UVM_FLAG_COPYONW;
1166 	}
1167 
1168 	/*
1169 	 * set up mapping flags
1170 	 */
1171 
1172 	uvmflag = UVM_MAPFLAG(prot, maxprot,
1173 			(flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
1174 			advice, uvmflag);
1175 
1176 	error = uvm_map_p(map, addr, size, uobj, foff, align, uvmflag, p);
1177 
1178 	if (error == 0) {
1179 		/*
1180 		 * POSIX 1003.1b -- if our address space was configured
1181 		 * to lock all future mappings, wire the one we just made.
1182 		 */
1183 		if (prot == VM_PROT_NONE) {
1184 			/*
1185 			 * No more work to do in this case.
1186 			 */
1187 			return (0);
1188 		}
1189 
1190 		vm_map_lock(map);
1191 
1192 		if (map->flags & VM_MAP_WIREFUTURE) {
1193 			if ((atop(size) + uvmexp.wired) > uvmexp.wiredmax
1194 #ifdef pmap_wired_count
1195 			    || (locklimit != 0 && (size +
1196 			         ptoa(pmap_wired_count(vm_map_pmap(map)))) >
1197 			        locklimit)
1198 #endif
1199 			) {
1200 				error = ENOMEM;
1201 				vm_map_unlock(map);
1202 				/* unmap the region! */
1203 				uvm_unmap(map, *addr, *addr + size);
1204 				goto bad;
1205 			}
1206 			/*
1207 			 * uvm_map_pageable() always returns the map
1208 			 * unlocked.
1209 			 */
1210 			error = uvm_map_pageable(map, *addr, *addr + size,
1211 			    FALSE, UVM_LK_ENTER);
1212 			if (error != 0) {
1213 				/* unmap the region! */
1214 				uvm_unmap(map, *addr, *addr + size);
1215 				goto bad;
1216 			}
1217 			return (0);
1218 		}
1219 
1220 		vm_map_unlock(map);
1221 
1222 		return (0);
1223 	}
1224 
1225 	/*
1226 	 * errors: first detach from the uobj, if any.
1227 	 */
1228 
1229 	if (uobj)
1230 		uobj->pgops->pgo_detach(uobj);
1231 
1232 bad:
1233 	return (error);
1234 }
1235