xref: /dflybsd-src/sys/vm/vm_mmap.c (revision c9e3d8f96688a159959b1af2d4fef14b744173e3)
1 /*
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
39  *
40  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
41  * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $
42  * $DragonFly: src/sys/vm/vm_mmap.c,v 1.39 2007/04/30 07:18:57 dillon Exp $
43  */
44 
45 /*
46  * Mapped file (mmap) interface to VM
47  */
48 
49 #include <sys/param.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/sysproto.h>
53 #include <sys/filedesc.h>
54 #include <sys/kern_syscall.h>
55 #include <sys/proc.h>
56 #include <sys/priv.h>
57 #include <sys/resource.h>
58 #include <sys/resourcevar.h>
59 #include <sys/vnode.h>
60 #include <sys/fcntl.h>
61 #include <sys/file.h>
62 #include <sys/mman.h>
63 #include <sys/conf.h>
64 #include <sys/stat.h>
65 #include <sys/vmmeter.h>
66 #include <sys/sysctl.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_param.h>
70 #include <sys/lock.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_pager.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_kern.h>
80 
81 #include <sys/file2.h>
82 #include <sys/thread2.h>
83 #include <sys/mplock2.h>
84 
85 static int max_proc_mmap;
86 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
87 int vkernel_enable;
88 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, "");
89 
90 /*
91  * Set the maximum number of vm_map_entry structures per process.  Roughly
92  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
93  * of our KVM malloc space still results in generous limits.  We want a
94  * default that is good enough to prevent the kernel running out of resources
95  * if attacked from compromised user account but generous enough such that
96  * multi-threaded processes are not unduly inconvenienced.
97  */
98 
99 static void vmmapentry_rsrc_init (void *);
100 SYSINIT(vmmersrc, SI_BOOT1_POST, SI_ORDER_ANY, vmmapentry_rsrc_init, NULL)
101 
102 static void
103 vmmapentry_rsrc_init(void *dummy)
104 {
105     max_proc_mmap = KvaSize / sizeof(struct vm_map_entry);
106     max_proc_mmap /= 100;
107 }
108 
109 /*
110  * MPSAFE
111  */
112 int
113 sys_sbrk(struct sbrk_args *uap)
114 {
115 	/* Not yet implemented */
116 	return (EOPNOTSUPP);
117 }
118 
119 /*
120  * sstk_args(int incr)
121  *
122  * MPSAFE
123  */
124 int
125 sys_sstk(struct sstk_args *uap)
126 {
127 	/* Not yet implemented */
128 	return (EOPNOTSUPP);
129 }
130 
131 /*
132  * mmap_args(void *addr, size_t len, int prot, int flags, int fd,
133  *		long pad, off_t pos)
134  *
135  * Memory Map (mmap) system call.  Note that the file offset
136  * and address are allowed to be NOT page aligned, though if
137  * the MAP_FIXED flag it set, both must have the same remainder
138  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
139  * page-aligned, the actual mapping starts at trunc_page(addr)
140  * and the return value is adjusted up by the page offset.
141  *
142  * Generally speaking, only character devices which are themselves
143  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
144  * there would be no cache coherency between a descriptor and a VM mapping
145  * both to the same character device.
146  *
147  * Block devices can be mmap'd no matter what they represent.  Cache coherency
148  * is maintained as long as you do not write directly to the underlying
149  * character device.
150  */
151 
152 int
153 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen,
154 	  int uprot, int uflags, int fd, off_t upos, void **res)
155 {
156 	struct thread *td = curthread;
157  	struct proc *p = td->td_proc;
158 	struct file *fp = NULL;
159 	struct vnode *vp;
160 	vm_offset_t addr;
161 	vm_offset_t tmpaddr;
162 	vm_size_t size, pageoff;
163 	vm_prot_t prot, maxprot;
164 	void *handle;
165 	int flags, error;
166 	int disablexworkaround;
167 	off_t pos;
168 	vm_object_t obj;
169 
170 	KKASSERT(p);
171 
172 	addr = (vm_offset_t) uaddr;
173 	size = ulen;
174 	prot = uprot & VM_PROT_ALL;
175 	flags = uflags;
176 	pos = upos;
177 
178 	/*
179 	 * Make sure mapping fits into numeric range etc.
180 	 *
181 	 * NOTE: We support the full unsigned range for size now.
182 	 */
183 	if (((flags & MAP_ANON) && (fd != -1 || pos != 0)))
184 		return (EINVAL);
185 
186 	if (flags & MAP_STACK) {
187 		if ((fd != -1) ||
188 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
189 			return (EINVAL);
190 		flags |= MAP_ANON;
191 		pos = 0;
192 	}
193 
194 	/*
195 	 * Virtual page tables cannot be used with MAP_STACK.  Apart from
196 	 * it not making any sense, the aux union is used by both
197 	 * types.
198 	 *
199 	 * Because the virtual page table is stored in the backing object
200 	 * and might be updated by the kernel, the mapping must be R+W.
201 	 */
202 	if (flags & MAP_VPAGETABLE) {
203 		if (vkernel_enable == 0)
204 			return (EOPNOTSUPP);
205 		if (flags & MAP_STACK)
206 			return (EINVAL);
207 		if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE))
208 			return (EINVAL);
209 	}
210 
211 	/*
212 	 * Align the file position to a page boundary,
213 	 * and save its page offset component.
214 	 */
215 	pageoff = (pos & PAGE_MASK);
216 	pos -= pageoff;
217 
218 	/* Adjust size for rounding (on both ends). */
219 	size += pageoff;			/* low end... */
220 	size = (vm_size_t) round_page(size);	/* hi end */
221 	if (size < ulen)			/* wrap */
222 		return(EINVAL);
223 
224 	/*
225 	 * Check for illegal addresses.  Watch out for address wrap... Note
226 	 * that VM_*_ADDRESS are not constants due to casts (argh).
227 	 */
228 	if (flags & (MAP_FIXED | MAP_TRYFIXED)) {
229 		/*
230 		 * The specified address must have the same remainder
231 		 * as the file offset taken modulo PAGE_SIZE, so it
232 		 * should be aligned after adjustment by pageoff.
233 		 */
234 		addr -= pageoff;
235 		if (addr & PAGE_MASK)
236 			return (EINVAL);
237 
238 		/*
239 		 * Address range must be all in user VM space and not wrap.
240 		 */
241 		tmpaddr = addr + size;
242 		if (tmpaddr < addr)
243 			return (EINVAL);
244 		if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
245 			return (EINVAL);
246 		if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
247 			return (EINVAL);
248 	} else {
249 		/*
250 		 * Set a reasonable start point for the hint if it was
251 		 * not specified or if it falls within the heap space.
252 		 * Hinted mmap()s do not allocate out of the heap space.
253 		 */
254 		if (addr == 0 ||
255 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
256 		     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
257 			addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
258 	}
259 
260 	if (flags & MAP_ANON) {
261 		/*
262 		 * Mapping blank space is trivial.
263 		 */
264 		handle = NULL;
265 		maxprot = VM_PROT_ALL;
266 	} else {
267 		/*
268 		 * Mapping file, get fp for validation. Obtain vnode and make
269 		 * sure it is of appropriate type.
270 		 */
271 		fp = holdfp(p->p_fd, fd, -1);
272 		if (fp == NULL)
273 			return (EBADF);
274 		if (fp->f_type != DTYPE_VNODE) {
275 			error = EINVAL;
276 			goto done;
277 		}
278 		/*
279 		 * POSIX shared-memory objects are defined to have
280 		 * kernel persistence, and are not defined to support
281 		 * read(2)/write(2) -- or even open(2).  Thus, we can
282 		 * use MAP_ASYNC to trade on-disk coherence for speed.
283 		 * The shm_open(3) library routine turns on the FPOSIXSHM
284 		 * flag to request this behavior.
285 		 */
286 		if (fp->f_flag & FPOSIXSHM)
287 			flags |= MAP_NOSYNC;
288 		vp = (struct vnode *) fp->f_data;
289 
290 		/*
291 		 * Validate the vnode for the operation.
292 		 */
293 		switch(vp->v_type) {
294 		case VREG:
295 			/*
296 			 * Get the proper underlying object
297 			 */
298 			if ((obj = vp->v_object) == NULL) {
299 				error = EINVAL;
300 				goto done;
301 			}
302 			KKASSERT((struct vnode *)obj->handle == vp);
303 			break;
304 		case VCHR:
305 			/*
306 			 * Make sure a device has not been revoked.
307 			 * Mappability is handled by the device layer.
308 			 */
309 			if (vp->v_rdev == NULL) {
310 				error = EBADF;
311 				goto done;
312 			}
313 			break;
314 		default:
315 			/*
316 			 * Nothing else is mappable.
317 			 */
318 			error = EINVAL;
319 			goto done;
320 		}
321 
322 		/*
323 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
324 		 * SunOS).
325 		 */
326 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
327 			handle = NULL;
328 			maxprot = VM_PROT_ALL;
329 			flags |= MAP_ANON;
330 			pos = 0;
331 		} else {
332 			/*
333 			 * cdevs does not provide private mappings of any kind.
334 			 */
335 			/*
336 			 * However, for XIG X server to continue to work,
337 			 * we should allow the superuser to do it anyway.
338 			 * We only allow it at securelevel < 1.
339 			 * (Because the XIG X server writes directly to video
340 			 * memory via /dev/mem, it should never work at any
341 			 * other securelevel.
342 			 * XXX this will have to go
343 			 */
344 			if (securelevel >= 1)
345 				disablexworkaround = 1;
346 			else
347 				disablexworkaround = priv_check(td, PRIV_ROOT);
348 			if (vp->v_type == VCHR && disablexworkaround &&
349 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
350 				error = EINVAL;
351 				goto done;
352 			}
353 			/*
354 			 * Ensure that file and memory protections are
355 			 * compatible.  Note that we only worry about
356 			 * writability if mapping is shared; in this case,
357 			 * current and max prot are dictated by the open file.
358 			 * XXX use the vnode instead?  Problem is: what
359 			 * credentials do we use for determination? What if
360 			 * proc does a setuid?
361 			 */
362 			maxprot = VM_PROT_EXECUTE;	/* ??? */
363 			if (fp->f_flag & FREAD) {
364 				maxprot |= VM_PROT_READ;
365 			} else if (prot & PROT_READ) {
366 				error = EACCES;
367 				goto done;
368 			}
369 			/*
370 			 * If we are sharing potential changes (either via
371 			 * MAP_SHARED or via the implicit sharing of character
372 			 * device mappings), and we are trying to get write
373 			 * permission although we opened it without asking
374 			 * for it, bail out.  Check for superuser, only if
375 			 * we're at securelevel < 1, to allow the XIG X server
376 			 * to continue to work.
377 			 */
378 
379 			if ((flags & MAP_SHARED) != 0 ||
380 			    (vp->v_type == VCHR && disablexworkaround)) {
381 				if ((fp->f_flag & FWRITE) != 0) {
382 					struct vattr va;
383 					if ((error = VOP_GETATTR(vp, &va))) {
384 						goto done;
385 					}
386 					if ((va.va_flags &
387 					    (IMMUTABLE|APPEND)) == 0) {
388 						maxprot |= VM_PROT_WRITE;
389 					} else if (prot & PROT_WRITE) {
390 						error = EPERM;
391 						goto done;
392 					}
393 				} else if ((prot & PROT_WRITE) != 0) {
394 					error = EACCES;
395 					goto done;
396 				}
397 			} else {
398 				maxprot |= VM_PROT_WRITE;
399 			}
400 			handle = (void *)vp;
401 		}
402 	}
403 
404 	/*
405 	 * Do not allow more then a certain number of vm_map_entry structures
406 	 * per process.  Scale with the number of rforks sharing the map
407 	 * to make the limit reasonable for threads.
408 	 */
409 	if (max_proc_mmap &&
410 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) {
411 		error = ENOMEM;
412 		goto done;
413 	}
414 
415 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
416 			flags, handle, pos);
417 	if (error == 0)
418 		*res = (void *)(addr + pageoff);
419 done:
420 	if (fp)
421 		fdrop(fp);
422 	return (error);
423 }
424 
425 /*
426  * MPALMOSTSAFE
427  */
428 int
429 sys_mmap(struct mmap_args *uap)
430 {
431 	int error;
432 
433 	get_mplock();
434 	error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
435 			  uap->prot, uap->flags,
436 			  uap->fd, uap->pos, &uap->sysmsg_resultp);
437 	rel_mplock();
438 
439 	return (error);
440 }
441 
442 /*
443  * msync_args(void *addr, size_t len, int flags)
444  *
445  * MPALMOSTSAFE
446  */
447 int
448 sys_msync(struct msync_args *uap)
449 {
450 	struct proc *p = curproc;
451 	vm_offset_t addr;
452 	vm_offset_t tmpaddr;
453 	vm_size_t size, pageoff;
454 	int flags;
455 	vm_map_t map;
456 	int rv;
457 
458 	addr = (vm_offset_t) uap->addr;
459 	size = uap->len;
460 	flags = uap->flags;
461 
462 	pageoff = (addr & PAGE_MASK);
463 	addr -= pageoff;
464 	size += pageoff;
465 	size = (vm_size_t) round_page(size);
466 	if (size < uap->len)		/* wrap */
467 		return(EINVAL);
468 	tmpaddr = addr + size;		/* workaround gcc4 opt */
469 	if (tmpaddr < addr)		/* wrap */
470 		return(EINVAL);
471 
472 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
473 		return (EINVAL);
474 
475 	get_mplock();
476 	map = &p->p_vmspace->vm_map;
477 
478 	/*
479 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
480 	 * pages with the region containing addr".  Unfortunately, we don't
481 	 * really keep track of individual mmaps so we approximate by flushing
482 	 * the range of the map entry containing addr. This can be incorrect
483 	 * if the region splits or is coalesced with a neighbor.
484 	 */
485 	if (size == 0) {
486 		vm_map_entry_t entry;
487 
488 		vm_map_lock_read(map);
489 		rv = vm_map_lookup_entry(map, addr, &entry);
490 		if (rv == FALSE) {
491 			vm_map_unlock_read(map);
492 			rv = KERN_INVALID_ADDRESS;
493 			goto done;
494 		}
495 		addr = entry->start;
496 		size = entry->end - entry->start;
497 		vm_map_unlock_read(map);
498 	}
499 
500 	/*
501 	 * Clean the pages and interpret the return value.
502 	 */
503 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
504 			  (flags & MS_INVALIDATE) != 0);
505 done:
506 	rel_mplock();
507 
508 	switch (rv) {
509 	case KERN_SUCCESS:
510 		break;
511 	case KERN_INVALID_ADDRESS:
512 		return (EINVAL);	/* Sun returns ENOMEM? */
513 	case KERN_FAILURE:
514 		return (EIO);
515 	default:
516 		return (EINVAL);
517 	}
518 
519 	return (0);
520 }
521 
522 /*
523  * munmap_args(void *addr, size_t len)
524  *
525  * MPALMOSTSAFE
526  */
527 int
528 sys_munmap(struct munmap_args *uap)
529 {
530 	struct proc *p = curproc;
531 	vm_offset_t addr;
532 	vm_offset_t tmpaddr;
533 	vm_size_t size, pageoff;
534 	vm_map_t map;
535 
536 	addr = (vm_offset_t) uap->addr;
537 	size = uap->len;
538 
539 	pageoff = (addr & PAGE_MASK);
540 	addr -= pageoff;
541 	size += pageoff;
542 	size = (vm_size_t) round_page(size);
543 	if (size < uap->len)		/* wrap */
544 		return(EINVAL);
545 	tmpaddr = addr + size;		/* workaround gcc4 opt */
546 	if (tmpaddr < addr)		/* wrap */
547 		return(EINVAL);
548 
549 	if (size == 0)
550 		return (0);
551 
552 	/*
553 	 * Check for illegal addresses.  Watch out for address wrap... Note
554 	 * that VM_*_ADDRESS are not constants due to casts (argh).
555 	 */
556 	if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
557 		return (EINVAL);
558 	if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
559 		return (EINVAL);
560 
561 	get_mplock();
562 	map = &p->p_vmspace->vm_map;
563 	/*
564 	 * Make sure entire range is allocated.
565 	 */
566 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) {
567 		rel_mplock();
568 		return (EINVAL);
569 	}
570 	/* returns nothing but KERN_SUCCESS anyway */
571 	vm_map_remove(map, addr, addr + size);
572 	rel_mplock();
573 	return (0);
574 }
575 
576 /*
577  * mprotect_args(const void *addr, size_t len, int prot)
578  *
579  * MPALMOSTSAFE
580  */
581 int
582 sys_mprotect(struct mprotect_args *uap)
583 {
584 	struct proc *p = curproc;
585 	vm_offset_t addr;
586 	vm_offset_t tmpaddr;
587 	vm_size_t size, pageoff;
588 	vm_prot_t prot;
589 	int error;
590 
591 	addr = (vm_offset_t) uap->addr;
592 	size = uap->len;
593 	prot = uap->prot & VM_PROT_ALL;
594 #if defined(VM_PROT_READ_IS_EXEC)
595 	if (prot & VM_PROT_READ)
596 		prot |= VM_PROT_EXECUTE;
597 #endif
598 
599 	pageoff = (addr & PAGE_MASK);
600 	addr -= pageoff;
601 	size += pageoff;
602 	size = (vm_size_t) round_page(size);
603 	if (size < uap->len)		/* wrap */
604 		return(EINVAL);
605 	tmpaddr = addr + size;		/* workaround gcc4 opt */
606 	if (tmpaddr < addr)		/* wrap */
607 		return(EINVAL);
608 
609 	get_mplock();
610 	switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size,
611 			       prot, FALSE)) {
612 	case KERN_SUCCESS:
613 		error = 0;
614 		break;
615 	case KERN_PROTECTION_FAILURE:
616 		error = EACCES;
617 		break;
618 	default:
619 		error = EINVAL;
620 		break;
621 	}
622 	rel_mplock();
623 	return (error);
624 }
625 
626 /*
627  * minherit_args(void *addr, size_t len, int inherit)
628  *
629  * MPALMOSTSAFE
630  */
631 int
632 sys_minherit(struct minherit_args *uap)
633 {
634 	struct proc *p = curproc;
635 	vm_offset_t addr;
636 	vm_offset_t tmpaddr;
637 	vm_size_t size, pageoff;
638 	vm_inherit_t inherit;
639 	int error;
640 
641 	addr = (vm_offset_t)uap->addr;
642 	size = uap->len;
643 	inherit = uap->inherit;
644 
645 	pageoff = (addr & PAGE_MASK);
646 	addr -= pageoff;
647 	size += pageoff;
648 	size = (vm_size_t) round_page(size);
649 	if (size < uap->len)		/* wrap */
650 		return(EINVAL);
651 	tmpaddr = addr + size;		/* workaround gcc4 opt */
652 	if (tmpaddr < addr)		/* wrap */
653 		return(EINVAL);
654 
655 	get_mplock();
656 
657 	switch (vm_map_inherit(&p->p_vmspace->vm_map, addr,
658 			       addr + size, inherit)) {
659 	case KERN_SUCCESS:
660 		error = 0;
661 		break;
662 	case KERN_PROTECTION_FAILURE:
663 		error = EACCES;
664 		break;
665 	default:
666 		error = EINVAL;
667 		break;
668 	}
669 	rel_mplock();
670 	return (error);
671 }
672 
673 /*
674  * madvise_args(void *addr, size_t len, int behav)
675  *
676  * MPALMOSTSAFE
677  */
678 int
679 sys_madvise(struct madvise_args *uap)
680 {
681 	struct proc *p = curproc;
682 	vm_offset_t start, end;
683 	vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
684 	int error;
685 
686 	/*
687 	 * Check for illegal behavior
688 	 */
689 	if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
690 		return (EINVAL);
691 	/*
692 	 * Check for illegal addresses.  Watch out for address wrap... Note
693 	 * that VM_*_ADDRESS are not constants due to casts (argh).
694 	 */
695 	if (tmpaddr < (vm_offset_t)uap->addr)
696 		return (EINVAL);
697 	if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
698 		return (EINVAL);
699 	if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
700 		return (EINVAL);
701 
702 	/*
703 	 * Since this routine is only advisory, we default to conservative
704 	 * behavior.
705 	 */
706 	start = trunc_page((vm_offset_t)uap->addr);
707 	end = round_page(tmpaddr);
708 
709 	get_mplock();
710 	error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
711 			       uap->behav, 0);
712 	rel_mplock();
713 	return (error);
714 }
715 
716 /*
717  * mcontrol_args(void *addr, size_t len, int behav, off_t value)
718  *
719  * MPALMOSTSAFE
720  */
721 int
722 sys_mcontrol(struct mcontrol_args *uap)
723 {
724 	struct proc *p = curproc;
725 	vm_offset_t start, end;
726 	vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
727 	int error;
728 
729 	/*
730 	 * Check for illegal behavior
731 	 */
732 	if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
733 		return (EINVAL);
734 	/*
735 	 * Check for illegal addresses.  Watch out for address wrap... Note
736 	 * that VM_*_ADDRESS are not constants due to casts (argh).
737 	 */
738 	if (tmpaddr < (vm_offset_t) uap->addr)
739 		return (EINVAL);
740 	if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
741 		return (EINVAL);
742 	if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
743 		return (EINVAL);
744 
745 	/*
746 	 * Since this routine is only advisory, we default to conservative
747 	 * behavior.
748 	 */
749 	start = trunc_page((vm_offset_t)uap->addr);
750 	end = round_page(tmpaddr);
751 
752 	get_mplock();
753 	error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
754 			       uap->behav, uap->value);
755 	rel_mplock();
756 	return (error);
757 }
758 
759 
760 /*
761  * mincore_args(const void *addr, size_t len, char *vec)
762  *
763  * MPALMOSTSAFE
764  */
765 int
766 sys_mincore(struct mincore_args *uap)
767 {
768 	struct proc *p = curproc;
769 	vm_offset_t addr, first_addr;
770 	vm_offset_t end, cend;
771 	pmap_t pmap;
772 	vm_map_t map;
773 	char *vec;
774 	int error;
775 	int vecindex, lastvecindex;
776 	vm_map_entry_t current;
777 	vm_map_entry_t entry;
778 	int mincoreinfo;
779 	unsigned int timestamp;
780 
781 	/*
782 	 * Make sure that the addresses presented are valid for user
783 	 * mode.
784 	 */
785 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
786 	end = addr + (vm_size_t)round_page(uap->len);
787 	if (end < addr)
788 		return (EINVAL);
789 	if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS)
790 		return (EINVAL);
791 
792 	/*
793 	 * Address of byte vector
794 	 */
795 	vec = uap->vec;
796 
797 	map = &p->p_vmspace->vm_map;
798 	pmap = vmspace_pmap(p->p_vmspace);
799 
800 	get_mplock();
801 	vm_map_lock_read(map);
802 RestartScan:
803 	timestamp = map->timestamp;
804 
805 	if (!vm_map_lookup_entry(map, addr, &entry))
806 		entry = entry->next;
807 
808 	/*
809 	 * Do this on a map entry basis so that if the pages are not
810 	 * in the current processes address space, we can easily look
811 	 * up the pages elsewhere.
812 	 */
813 	lastvecindex = -1;
814 	for(current = entry;
815 		(current != &map->header) && (current->start < end);
816 		current = current->next) {
817 
818 		/*
819 		 * ignore submaps (for now) or null objects
820 		 */
821 		if (current->maptype != VM_MAPTYPE_NORMAL &&
822 		    current->maptype != VM_MAPTYPE_VPAGETABLE) {
823 			continue;
824 		}
825 		if (current->object.vm_object == NULL)
826 			continue;
827 
828 		/*
829 		 * limit this scan to the current map entry and the
830 		 * limits for the mincore call
831 		 */
832 		if (addr < current->start)
833 			addr = current->start;
834 		cend = current->end;
835 		if (cend > end)
836 			cend = end;
837 
838 		/*
839 		 * scan this entry one page at a time
840 		 */
841 		while (addr < cend) {
842 			/*
843 			 * Check pmap first, it is likely faster, also
844 			 * it can provide info as to whether we are the
845 			 * one referencing or modifying the page.
846 			 *
847 			 * If we have to check the VM object, only mess
848 			 * around with normal maps.  Do not mess around
849 			 * with virtual page tables (XXX).
850 			 */
851 			mincoreinfo = pmap_mincore(pmap, addr);
852 			if (mincoreinfo == 0 &&
853 			    current->maptype == VM_MAPTYPE_NORMAL) {
854 				vm_pindex_t pindex;
855 				vm_ooffset_t offset;
856 				vm_page_t m;
857 
858 				/*
859 				 * calculate the page index into the object
860 				 */
861 				offset = current->offset + (addr - current->start);
862 				pindex = OFF_TO_IDX(offset);
863 
864 				/*
865 				 * if the page is resident, then gather
866 				 * information about it.  spl protection is
867 				 * required to maintain the object
868 				 * association.  And XXX what if the page is
869 				 * busy?  What's the deal with that?
870 				 */
871 				crit_enter();
872 				m = vm_page_lookup(current->object.vm_object,
873 						    pindex);
874 				if (m && m->valid) {
875 					mincoreinfo = MINCORE_INCORE;
876 					if (m->dirty ||
877 						pmap_is_modified(m))
878 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
879 					if ((m->flags & PG_REFERENCED) ||
880 						pmap_ts_referenced(m)) {
881 						vm_page_flag_set(m, PG_REFERENCED);
882 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
883 					}
884 				}
885 				crit_exit();
886 			}
887 
888 			/*
889 			 * subyte may page fault.  In case it needs to modify
890 			 * the map, we release the lock.
891 			 */
892 			vm_map_unlock_read(map);
893 
894 			/*
895 			 * calculate index into user supplied byte vector
896 			 */
897 			vecindex = OFF_TO_IDX(addr - first_addr);
898 
899 			/*
900 			 * If we have skipped map entries, we need to make sure that
901 			 * the byte vector is zeroed for those skipped entries.
902 			 */
903 			while((lastvecindex + 1) < vecindex) {
904 				error = subyte( vec + lastvecindex, 0);
905 				if (error) {
906 					error = EFAULT;
907 					goto done;
908 				}
909 				++lastvecindex;
910 			}
911 
912 			/*
913 			 * Pass the page information to the user
914 			 */
915 			error = subyte( vec + vecindex, mincoreinfo);
916 			if (error) {
917 				error = EFAULT;
918 				goto done;
919 			}
920 
921 			/*
922 			 * If the map has changed, due to the subyte, the previous
923 			 * output may be invalid.
924 			 */
925 			vm_map_lock_read(map);
926 			if (timestamp != map->timestamp)
927 				goto RestartScan;
928 
929 			lastvecindex = vecindex;
930 			addr += PAGE_SIZE;
931 		}
932 	}
933 
934 	/*
935 	 * subyte may page fault.  In case it needs to modify
936 	 * the map, we release the lock.
937 	 */
938 	vm_map_unlock_read(map);
939 
940 	/*
941 	 * Zero the last entries in the byte vector.
942 	 */
943 	vecindex = OFF_TO_IDX(end - first_addr);
944 	while((lastvecindex + 1) < vecindex) {
945 		error = subyte( vec + lastvecindex, 0);
946 		if (error) {
947 			error = EFAULT;
948 			goto done;
949 		}
950 		++lastvecindex;
951 	}
952 
953 	/*
954 	 * If the map has changed, due to the subyte, the previous
955 	 * output may be invalid.
956 	 */
957 	vm_map_lock_read(map);
958 	if (timestamp != map->timestamp)
959 		goto RestartScan;
960 	vm_map_unlock_read(map);
961 
962 	error = 0;
963 done:
964 	rel_mplock();
965 	return (error);
966 }
967 
968 /*
969  * mlock_args(const void *addr, size_t len)
970  *
971  * MPALMOSTSAFE
972  */
973 int
974 sys_mlock(struct mlock_args *uap)
975 {
976 	vm_offset_t addr;
977 	vm_offset_t tmpaddr;
978 	vm_size_t size, pageoff;
979 	struct thread *td = curthread;
980 	struct proc *p = td->td_proc;
981 	int error;
982 
983 	addr = (vm_offset_t) uap->addr;
984 	size = uap->len;
985 
986 	pageoff = (addr & PAGE_MASK);
987 	addr -= pageoff;
988 	size += pageoff;
989 	size = (vm_size_t) round_page(size);
990 	if (size < uap->len)		/* wrap */
991 		return(EINVAL);
992 	tmpaddr = addr + size;		/* workaround gcc4 opt */
993 	if (tmpaddr < addr)		/* wrap */
994 		return (EINVAL);
995 
996 	if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
997 		return (EAGAIN);
998 
999 	get_mplock();
1000 #ifdef pmap_wired_count
1001 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
1002 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) {
1003 		rel_mplock();
1004 		return (ENOMEM);
1005 	}
1006 #else
1007 	error = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1008 	if (error) {
1009 		rel_mplock();
1010 		return (error);
1011 	}
1012 #endif
1013 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
1014 	rel_mplock();
1015 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1016 }
1017 
1018 /*
1019  * mlockall_args(int how)
1020  *
1021  * Dummy routine, doesn't actually do anything.
1022  *
1023  * MPSAFE
1024  */
1025 int
1026 sys_mlockall(struct mlockall_args *uap)
1027 {
1028 	return (ENOSYS);
1029 }
1030 
1031 /*
1032  * munlockall_args(void)
1033  *
1034  * Dummy routine, doesn't actually do anything.
1035  *
1036  * MPSAFE
1037  */
1038 int
1039 sys_munlockall(struct munlockall_args *uap)
1040 {
1041 	return (ENOSYS);
1042 }
1043 
1044 /*
1045  * munlock_args(const void *addr, size_t len)
1046  *
1047  * MPALMOSTSAFE
1048  */
1049 int
1050 sys_munlock(struct munlock_args *uap)
1051 {
1052 	struct thread *td = curthread;
1053 	struct proc *p = td->td_proc;
1054 	vm_offset_t addr;
1055 	vm_offset_t tmpaddr;
1056 	vm_size_t size, pageoff;
1057 	int error;
1058 
1059 	addr = (vm_offset_t) uap->addr;
1060 	size = uap->len;
1061 
1062 	pageoff = (addr & PAGE_MASK);
1063 	addr -= pageoff;
1064 	size += pageoff;
1065 	size = (vm_size_t) round_page(size);
1066 
1067 	tmpaddr = addr + size;
1068 	if (tmpaddr < addr)		/* wrap */
1069 		return (EINVAL);
1070 
1071 #ifndef pmap_wired_count
1072 	error = priv_check(td, PRIV_ROOT);
1073 	if (error)
1074 		return (error);
1075 #endif
1076 
1077 	get_mplock();
1078 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
1079 	rel_mplock();
1080 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1081 }
1082 
1083 /*
1084  * Internal version of mmap.
1085  * Currently used by mmap, exec, and sys5 shared memory.
1086  * Handle is either a vnode pointer or NULL for MAP_ANON.
1087  */
1088 int
1089 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1090 	vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff)
1091 {
1092 	boolean_t fitit;
1093 	vm_object_t object;
1094 	vm_offset_t eaddr;
1095 	vm_size_t   esize;
1096 	struct vnode *vp;
1097 	struct thread *td = curthread;
1098 	struct proc *p;
1099 	int rv = KERN_SUCCESS;
1100 	off_t objsize;
1101 	int docow;
1102 
1103 	if (size == 0)
1104 		return (0);
1105 
1106 	objsize = round_page(size);
1107 	if (objsize < size)
1108 		return (EINVAL);
1109 	size = objsize;
1110 
1111 	/*
1112 	 * XXX messy code, fixme
1113 	 *
1114 	 * NOTE: Overflow checks require discrete statements or GCC4
1115 	 * will optimize it out.
1116 	 */
1117 	if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
1118 		esize = map->size + size;	/* workaround gcc4 opt */
1119 		if (esize < map->size ||
1120 		    esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1121 			return(ENOMEM);
1122 		}
1123 	}
1124 
1125 	/*
1126 	 * We currently can only deal with page aligned file offsets.
1127 	 * The check is here rather than in the syscall because the
1128 	 * kernel calls this function internally for other mmaping
1129 	 * operations (such as in exec) and non-aligned offsets will
1130 	 * cause pmap inconsistencies...so we want to be sure to
1131 	 * disallow this in all cases.
1132 	 *
1133 	 * NOTE: Overflow checks require discrete statements or GCC4
1134 	 * will optimize it out.
1135 	 */
1136 	if (foff & PAGE_MASK)
1137 		return (EINVAL);
1138 
1139 	if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
1140 		fitit = TRUE;
1141 		*addr = round_page(*addr);
1142 	} else {
1143 		if (*addr != trunc_page(*addr))
1144 			return (EINVAL);
1145 		eaddr = *addr + size;
1146 		if (eaddr < *addr)
1147 			return (EINVAL);
1148 		fitit = FALSE;
1149 		if ((flags & MAP_TRYFIXED) == 0)
1150 			vm_map_remove(map, *addr, *addr + size);
1151 	}
1152 
1153 	/*
1154 	 * Lookup/allocate object.
1155 	 */
1156 	if (flags & MAP_ANON) {
1157 		/*
1158 		 * Unnamed anonymous regions always start at 0.
1159 		 */
1160 		if (handle) {
1161 			/*
1162 			 * Default memory object
1163 			 */
1164 			object = default_pager_alloc(handle, objsize,
1165 						     prot, foff);
1166 			if (object == NULL)
1167 				return(ENOMEM);
1168 			docow = MAP_PREFAULT_PARTIAL;
1169 		} else {
1170 			/*
1171 			 * Implicit single instance of a default memory
1172 			 * object, so we don't need a VM object yet.
1173 			 */
1174 			foff = 0;
1175 			object = NULL;
1176 			docow = 0;
1177 		}
1178 		vp = NULL;
1179 	} else {
1180 		vp = (struct vnode *)handle;
1181 		if (vp->v_type == VCHR) {
1182 			/*
1183 			 * Device mappings (device size unknown?).
1184 			 * Force them to be shared.
1185 			 */
1186 			handle = (void *)(intptr_t)vp->v_rdev;
1187 			object = dev_pager_alloc(handle, objsize, prot, foff);
1188 			if (object == NULL)
1189 				return(EINVAL);
1190 			docow = MAP_PREFAULT_PARTIAL;
1191 			flags &= ~(MAP_PRIVATE|MAP_COPY);
1192 			flags |= MAP_SHARED;
1193 		} else {
1194 			/*
1195 			 * Regular file mapping (typically).  The attribute
1196 			 * check is for the link count test only.  Mmapble
1197 			 * vnodes must already have a VM object assigned.
1198 			 */
1199 			struct vattr vat;
1200 			int error;
1201 
1202 			error = VOP_GETATTR(vp, &vat);
1203 			if (error)
1204 				return (error);
1205 			docow = MAP_PREFAULT_PARTIAL;
1206 			object = vnode_pager_reference(vp);
1207 			if (object == NULL && vp->v_type == VREG) {
1208 				kprintf("Warning: cannot mmap vnode %p, no "
1209 					"object\n", vp);
1210 				return(EINVAL);
1211 			}
1212 
1213 			/*
1214 			 * If it is a regular file without any references
1215 			 * we do not need to sync it.
1216 			 */
1217 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1218 				flags |= MAP_NOSYNC;
1219 			}
1220 		}
1221 	}
1222 
1223 	/*
1224 	 * Deal with the adjusted flags
1225 	 */
1226 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1227 		docow |= MAP_COPY_ON_WRITE;
1228 	if (flags & MAP_NOSYNC)
1229 		docow |= MAP_DISABLE_SYNCER;
1230 	if (flags & MAP_NOCORE)
1231 		docow |= MAP_DISABLE_COREDUMP;
1232 
1233 #if defined(VM_PROT_READ_IS_EXEC)
1234 	if (prot & VM_PROT_READ)
1235 		prot |= VM_PROT_EXECUTE;
1236 
1237 	if (maxprot & VM_PROT_READ)
1238 		maxprot |= VM_PROT_EXECUTE;
1239 #endif
1240 
1241 	/*
1242 	 * This may place the area in its own page directory if (size) is
1243 	 * large enough, otherwise it typically returns its argument.
1244 	 */
1245 	if (fitit) {
1246 		*addr = pmap_addr_hint(object, *addr, size);
1247 	}
1248 
1249 	/*
1250 	 * Stack mappings need special attention.
1251 	 *
1252 	 * Mappings that use virtual page tables will default to storing
1253 	 * the page table at offset 0.
1254 	 */
1255 	if (flags & MAP_STACK) {
1256 		rv = vm_map_stack(map, *addr, size, flags,
1257 				  prot, maxprot, docow);
1258 	} else if (flags & MAP_VPAGETABLE) {
1259 		rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1260 				 fitit, VM_MAPTYPE_VPAGETABLE,
1261 				 prot, maxprot, docow);
1262 	} else {
1263 		rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1264 				 fitit, VM_MAPTYPE_NORMAL,
1265 				 prot, maxprot, docow);
1266 	}
1267 
1268 	if (rv != KERN_SUCCESS) {
1269 		/*
1270 		 * Lose the object reference. Will destroy the
1271 		 * object if it's an unnamed anonymous mapping
1272 		 * or named anonymous without other references.
1273 		 */
1274 		vm_object_deallocate(object);
1275 		goto out;
1276 	}
1277 
1278 	/*
1279 	 * Shared memory is also shared with children.
1280 	 */
1281 	if (flags & (MAP_SHARED|MAP_INHERIT)) {
1282 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1283 		if (rv != KERN_SUCCESS) {
1284 			vm_map_remove(map, *addr, *addr + size);
1285 			goto out;
1286 		}
1287 	}
1288 
1289 	/*
1290 	 * Set the access time on the vnode
1291 	 */
1292 	if (vp != NULL)
1293 		vn_mark_atime(vp, td);
1294 out:
1295 	switch (rv) {
1296 	case KERN_SUCCESS:
1297 		return (0);
1298 	case KERN_INVALID_ADDRESS:
1299 	case KERN_NO_SPACE:
1300 		return (ENOMEM);
1301 	case KERN_PROTECTION_FAILURE:
1302 		return (EACCES);
1303 	default:
1304 		return (EINVAL);
1305 	}
1306 }
1307