xref: /dflybsd-src/sys/vm/vm_mmap.c (revision 4d9022e3888d071c1d026d8aa01aecd099e7bd9b)
1 /*
2  * (MPSAFE)
3  *
4  * Copyright (c) 1988 University of Utah.
5  * Copyright (c) 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
41  *
42  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
43  * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $
44  * $DragonFly: src/sys/vm/vm_mmap.c,v 1.39 2007/04/30 07:18:57 dillon Exp $
45  */
46 
47 /*
48  * Mapped file (mmap) interface to VM
49  */
50 
51 #include <sys/param.h>
52 #include <sys/kernel.h>
53 #include <sys/systm.h>
54 #include <sys/sysproto.h>
55 #include <sys/filedesc.h>
56 #include <sys/kern_syscall.h>
57 #include <sys/proc.h>
58 #include <sys/priv.h>
59 #include <sys/resource.h>
60 #include <sys/resourcevar.h>
61 #include <sys/vnode.h>
62 #include <sys/fcntl.h>
63 #include <sys/file.h>
64 #include <sys/mman.h>
65 #include <sys/conf.h>
66 #include <sys/stat.h>
67 #include <sys/vmmeter.h>
68 #include <sys/sysctl.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <sys/lock.h>
73 #include <vm/pmap.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_pager.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_kern.h>
82 
83 #include <sys/file2.h>
84 #include <sys/thread.h>
85 #include <sys/thread2.h>
86 
87 static int max_proc_mmap;
88 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
89 int vkernel_enable;
90 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, "");
91 
92 /*
93  * Set the maximum number of vm_map_entry structures per process.  Roughly
94  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
95  * of our KVM malloc space still results in generous limits.  We want a
96  * default that is good enough to prevent the kernel running out of resources
97  * if attacked from compromised user account but generous enough such that
98  * multi-threaded processes are not unduly inconvenienced.
99  */
100 
101 static void vmmapentry_rsrc_init (void *);
102 SYSINIT(vmmersrc, SI_BOOT1_POST, SI_ORDER_ANY, vmmapentry_rsrc_init, NULL)
103 
104 static void
105 vmmapentry_rsrc_init(void *dummy)
106 {
107     max_proc_mmap = KvaSize / sizeof(struct vm_map_entry);
108     max_proc_mmap /= 100;
109 }
110 
111 /*
112  * MPSAFE
113  */
114 int
115 sys_sbrk(struct sbrk_args *uap)
116 {
117 	/* Not yet implemented */
118 	return (EOPNOTSUPP);
119 }
120 
121 /*
122  * sstk_args(int incr)
123  *
124  * MPSAFE
125  */
126 int
127 sys_sstk(struct sstk_args *uap)
128 {
129 	/* Not yet implemented */
130 	return (EOPNOTSUPP);
131 }
132 
133 /*
134  * mmap_args(void *addr, size_t len, int prot, int flags, int fd,
135  *		long pad, off_t pos)
136  *
137  * Memory Map (mmap) system call.  Note that the file offset
138  * and address are allowed to be NOT page aligned, though if
139  * the MAP_FIXED flag it set, both must have the same remainder
140  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
141  * page-aligned, the actual mapping starts at trunc_page(addr)
142  * and the return value is adjusted up by the page offset.
143  *
144  * Generally speaking, only character devices which are themselves
145  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
146  * there would be no cache coherency between a descriptor and a VM mapping
147  * both to the same character device.
148  *
149  * Block devices can be mmap'd no matter what they represent.  Cache coherency
150  * is maintained as long as you do not write directly to the underlying
151  * character device.
152  *
153  * Requires caller to hold vm_token.
154  */
155 int
156 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen,
157 	  int uprot, int uflags, int fd, off_t upos, void **res)
158 {
159 	struct thread *td = curthread;
160  	struct proc *p = td->td_proc;
161 	struct file *fp = NULL;
162 	struct vnode *vp;
163 	vm_offset_t addr;
164 	vm_offset_t tmpaddr;
165 	vm_size_t size, pageoff;
166 	vm_prot_t prot, maxprot;
167 	void *handle;
168 	int flags, error;
169 	off_t pos;
170 	vm_object_t obj;
171 
172 	KKASSERT(p);
173 	ASSERT_LWKT_TOKEN_HELD(&vm_token);
174 
175 	addr = (vm_offset_t) uaddr;
176 	size = ulen;
177 	prot = uprot & VM_PROT_ALL;
178 	flags = uflags;
179 	pos = upos;
180 
181 	/*
182 	 * Make sure mapping fits into numeric range etc.
183 	 *
184 	 * NOTE: We support the full unsigned range for size now.
185 	 */
186 	if (((flags & MAP_ANON) && (fd != -1 || pos != 0)))
187 		return (EINVAL);
188 
189 	if (flags & MAP_STACK) {
190 		if ((fd != -1) ||
191 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
192 			return (EINVAL);
193 		flags |= MAP_ANON;
194 		pos = 0;
195 	}
196 
197 	/*
198 	 * Virtual page tables cannot be used with MAP_STACK.  Apart from
199 	 * it not making any sense, the aux union is used by both
200 	 * types.
201 	 *
202 	 * Because the virtual page table is stored in the backing object
203 	 * and might be updated by the kernel, the mapping must be R+W.
204 	 */
205 	if (flags & MAP_VPAGETABLE) {
206 		if (vkernel_enable == 0)
207 			return (EOPNOTSUPP);
208 		if (flags & MAP_STACK)
209 			return (EINVAL);
210 		if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE))
211 			return (EINVAL);
212 	}
213 
214 	/*
215 	 * Align the file position to a page boundary,
216 	 * and save its page offset component.
217 	 */
218 	pageoff = (pos & PAGE_MASK);
219 	pos -= pageoff;
220 
221 	/* Adjust size for rounding (on both ends). */
222 	size += pageoff;			/* low end... */
223 	size = (vm_size_t) round_page(size);	/* hi end */
224 	if (size < ulen)			/* wrap */
225 		return(EINVAL);
226 
227 	/*
228 	 * Check for illegal addresses.  Watch out for address wrap... Note
229 	 * that VM_*_ADDRESS are not constants due to casts (argh).
230 	 */
231 	if (flags & (MAP_FIXED | MAP_TRYFIXED)) {
232 		/*
233 		 * The specified address must have the same remainder
234 		 * as the file offset taken modulo PAGE_SIZE, so it
235 		 * should be aligned after adjustment by pageoff.
236 		 */
237 		addr -= pageoff;
238 		if (addr & PAGE_MASK)
239 			return (EINVAL);
240 
241 		/*
242 		 * Address range must be all in user VM space and not wrap.
243 		 */
244 		tmpaddr = addr + size;
245 		if (tmpaddr < addr)
246 			return (EINVAL);
247 		if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
248 			return (EINVAL);
249 		if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
250 			return (EINVAL);
251 	} else {
252 		/*
253 		 * Set a reasonable start point for the hint if it was
254 		 * not specified or if it falls within the heap space.
255 		 * Hinted mmap()s do not allocate out of the heap space.
256 		 */
257 		if (addr == 0 ||
258 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
259 		     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
260 			addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
261 	}
262 
263 	if (flags & MAP_ANON) {
264 		/*
265 		 * Mapping blank space is trivial.
266 		 */
267 		handle = NULL;
268 		maxprot = VM_PROT_ALL;
269 	} else {
270 		/*
271 		 * Mapping file, get fp for validation. Obtain vnode and make
272 		 * sure it is of appropriate type.
273 		 */
274 		fp = holdfp(p->p_fd, fd, -1);
275 		if (fp == NULL)
276 			return (EBADF);
277 		if (fp->f_type != DTYPE_VNODE) {
278 			error = EINVAL;
279 			goto done;
280 		}
281 		/*
282 		 * POSIX shared-memory objects are defined to have
283 		 * kernel persistence, and are not defined to support
284 		 * read(2)/write(2) -- or even open(2).  Thus, we can
285 		 * use MAP_ASYNC to trade on-disk coherence for speed.
286 		 * The shm_open(3) library routine turns on the FPOSIXSHM
287 		 * flag to request this behavior.
288 		 */
289 		if (fp->f_flag & FPOSIXSHM)
290 			flags |= MAP_NOSYNC;
291 		vp = (struct vnode *) fp->f_data;
292 
293 		/*
294 		 * Validate the vnode for the operation.
295 		 */
296 		switch(vp->v_type) {
297 		case VREG:
298 			/*
299 			 * Get the proper underlying object
300 			 */
301 			if ((obj = vp->v_object) == NULL) {
302 				error = EINVAL;
303 				goto done;
304 			}
305 			KKASSERT((struct vnode *)obj->handle == vp);
306 			break;
307 		case VCHR:
308 			/*
309 			 * Make sure a device has not been revoked.
310 			 * Mappability is handled by the device layer.
311 			 */
312 			if (vp->v_rdev == NULL) {
313 				error = EBADF;
314 				goto done;
315 			}
316 			break;
317 		default:
318 			/*
319 			 * Nothing else is mappable.
320 			 */
321 			error = EINVAL;
322 			goto done;
323 		}
324 
325 		/*
326 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
327 		 * SunOS).
328 		 */
329 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
330 			handle = NULL;
331 			maxprot = VM_PROT_ALL;
332 			flags |= MAP_ANON;
333 			pos = 0;
334 		} else {
335 			/*
336 			 * cdevs does not provide private mappings of any kind.
337 			 */
338 			if (vp->v_type == VCHR &&
339 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
340 				error = EINVAL;
341 				goto done;
342 			}
343 			/*
344 			 * Ensure that file and memory protections are
345 			 * compatible.  Note that we only worry about
346 			 * writability if mapping is shared; in this case,
347 			 * current and max prot are dictated by the open file.
348 			 * XXX use the vnode instead?  Problem is: what
349 			 * credentials do we use for determination? What if
350 			 * proc does a setuid?
351 			 */
352 			maxprot = VM_PROT_EXECUTE;	/* ??? */
353 			if (fp->f_flag & FREAD) {
354 				maxprot |= VM_PROT_READ;
355 			} else if (prot & PROT_READ) {
356 				error = EACCES;
357 				goto done;
358 			}
359 			/*
360 			 * If we are sharing potential changes (either via
361 			 * MAP_SHARED or via the implicit sharing of character
362 			 * device mappings), and we are trying to get write
363 			 * permission although we opened it without asking
364 			 * for it, bail out.  Check for superuser, only if
365 			 * we're at securelevel < 1, to allow the XIG X server
366 			 * to continue to work.
367 			 */
368 			if ((flags & MAP_SHARED) != 0 || vp->v_type == VCHR) {
369 				if ((fp->f_flag & FWRITE) != 0) {
370 					struct vattr va;
371 					if ((error = VOP_GETATTR(vp, &va))) {
372 						goto done;
373 					}
374 					if ((va.va_flags &
375 					    (IMMUTABLE|APPEND)) == 0) {
376 						maxprot |= VM_PROT_WRITE;
377 					} else if (prot & PROT_WRITE) {
378 						error = EPERM;
379 						goto done;
380 					}
381 				} else if ((prot & PROT_WRITE) != 0) {
382 					error = EACCES;
383 					goto done;
384 				}
385 			} else {
386 				maxprot |= VM_PROT_WRITE;
387 			}
388 			handle = (void *)vp;
389 		}
390 	}
391 
392 	/*
393 	 * Do not allow more then a certain number of vm_map_entry structures
394 	 * per process.  Scale with the number of rforks sharing the map
395 	 * to make the limit reasonable for threads.
396 	 */
397 	if (max_proc_mmap &&
398 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) {
399 		error = ENOMEM;
400 		goto done;
401 	}
402 
403 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
404 			flags, handle, pos);
405 	if (error == 0)
406 		*res = (void *)(addr + pageoff);
407 done:
408 	if (fp)
409 		fdrop(fp);
410 	return (error);
411 }
412 
413 /*
414  * mmap system call handler
415  *
416  * No requirements.
417  */
418 int
419 sys_mmap(struct mmap_args *uap)
420 {
421 	int error;
422 
423 	lwkt_gettoken(&vm_token);
424 	error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
425 			  uap->prot, uap->flags,
426 			  uap->fd, uap->pos, &uap->sysmsg_resultp);
427 	lwkt_reltoken(&vm_token);
428 
429 	return (error);
430 }
431 
432 /*
433  * msync system call handler
434  *
435  * msync_args(void *addr, size_t len, int flags)
436  *
437  * No requirements
438  */
439 int
440 sys_msync(struct msync_args *uap)
441 {
442 	struct proc *p = curproc;
443 	vm_offset_t addr;
444 	vm_offset_t tmpaddr;
445 	vm_size_t size, pageoff;
446 	int flags;
447 	vm_map_t map;
448 	int rv;
449 
450 	addr = (vm_offset_t) uap->addr;
451 	size = uap->len;
452 	flags = uap->flags;
453 
454 	pageoff = (addr & PAGE_MASK);
455 	addr -= pageoff;
456 	size += pageoff;
457 	size = (vm_size_t) round_page(size);
458 	if (size < uap->len)		/* wrap */
459 		return(EINVAL);
460 	tmpaddr = addr + size;		/* workaround gcc4 opt */
461 	if (tmpaddr < addr)		/* wrap */
462 		return(EINVAL);
463 
464 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
465 		return (EINVAL);
466 
467 	lwkt_gettoken(&vm_token);
468 	map = &p->p_vmspace->vm_map;
469 
470 	/*
471 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
472 	 * pages with the region containing addr".  Unfortunately, we don't
473 	 * really keep track of individual mmaps so we approximate by flushing
474 	 * the range of the map entry containing addr. This can be incorrect
475 	 * if the region splits or is coalesced with a neighbor.
476 	 */
477 	if (size == 0) {
478 		vm_map_entry_t entry;
479 
480 		vm_map_lock_read(map);
481 		rv = vm_map_lookup_entry(map, addr, &entry);
482 		if (rv == FALSE) {
483 			vm_map_unlock_read(map);
484 			rv = KERN_INVALID_ADDRESS;
485 			goto done;
486 		}
487 		addr = entry->start;
488 		size = entry->end - entry->start;
489 		vm_map_unlock_read(map);
490 	}
491 
492 	/*
493 	 * Clean the pages and interpret the return value.
494 	 */
495 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
496 			  (flags & MS_INVALIDATE) != 0);
497 done:
498 	lwkt_reltoken(&vm_token);
499 
500 	switch (rv) {
501 	case KERN_SUCCESS:
502 		break;
503 	case KERN_INVALID_ADDRESS:
504 		return (EINVAL);	/* Sun returns ENOMEM? */
505 	case KERN_FAILURE:
506 		return (EIO);
507 	default:
508 		return (EINVAL);
509 	}
510 
511 	return (0);
512 }
513 
514 /*
515  * munmap system call handler
516  *
517  * munmap_args(void *addr, size_t len)
518  *
519  * No requirements
520  */
521 int
522 sys_munmap(struct munmap_args *uap)
523 {
524 	struct proc *p = curproc;
525 	vm_offset_t addr;
526 	vm_offset_t tmpaddr;
527 	vm_size_t size, pageoff;
528 	vm_map_t map;
529 
530 	addr = (vm_offset_t) uap->addr;
531 	size = uap->len;
532 
533 	pageoff = (addr & PAGE_MASK);
534 	addr -= pageoff;
535 	size += pageoff;
536 	size = (vm_size_t) round_page(size);
537 	if (size < uap->len)		/* wrap */
538 		return(EINVAL);
539 	tmpaddr = addr + size;		/* workaround gcc4 opt */
540 	if (tmpaddr < addr)		/* wrap */
541 		return(EINVAL);
542 
543 	if (size == 0)
544 		return (0);
545 
546 	/*
547 	 * Check for illegal addresses.  Watch out for address wrap... Note
548 	 * that VM_*_ADDRESS are not constants due to casts (argh).
549 	 */
550 	if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
551 		return (EINVAL);
552 	if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
553 		return (EINVAL);
554 
555 	lwkt_gettoken(&vm_token);
556 	map = &p->p_vmspace->vm_map;
557 	/*
558 	 * Make sure entire range is allocated.
559 	 */
560 	if (!vm_map_check_protection(map, addr, addr + size,
561 				     VM_PROT_NONE, FALSE)) {
562 		lwkt_reltoken(&vm_token);
563 		return (EINVAL);
564 	}
565 	/* returns nothing but KERN_SUCCESS anyway */
566 	vm_map_remove(map, addr, addr + size);
567 	lwkt_reltoken(&vm_token);
568 	return (0);
569 }
570 
571 /*
572  * mprotect_args(const void *addr, size_t len, int prot)
573  *
574  * MPALMOSTSAFE
575  */
576 int
577 sys_mprotect(struct mprotect_args *uap)
578 {
579 	struct proc *p = curproc;
580 	vm_offset_t addr;
581 	vm_offset_t tmpaddr;
582 	vm_size_t size, pageoff;
583 	vm_prot_t prot;
584 	int error;
585 
586 	addr = (vm_offset_t) uap->addr;
587 	size = uap->len;
588 	prot = uap->prot & VM_PROT_ALL;
589 #if defined(VM_PROT_READ_IS_EXEC)
590 	if (prot & VM_PROT_READ)
591 		prot |= VM_PROT_EXECUTE;
592 #endif
593 
594 	pageoff = (addr & PAGE_MASK);
595 	addr -= pageoff;
596 	size += pageoff;
597 	size = (vm_size_t) round_page(size);
598 	if (size < uap->len)		/* wrap */
599 		return(EINVAL);
600 	tmpaddr = addr + size;		/* workaround gcc4 opt */
601 	if (tmpaddr < addr)		/* wrap */
602 		return(EINVAL);
603 
604 	lwkt_gettoken(&vm_token);
605 	switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size,
606 			       prot, FALSE)) {
607 	case KERN_SUCCESS:
608 		error = 0;
609 		break;
610 	case KERN_PROTECTION_FAILURE:
611 		error = EACCES;
612 		break;
613 	default:
614 		error = EINVAL;
615 		break;
616 	}
617 	lwkt_reltoken(&vm_token);
618 	return (error);
619 }
620 
621 /*
622  * minherit system call handler
623  *
624  * minherit_args(void *addr, size_t len, int inherit)
625  *
626  * No requirements.
627  */
628 int
629 sys_minherit(struct minherit_args *uap)
630 {
631 	struct proc *p = curproc;
632 	vm_offset_t addr;
633 	vm_offset_t tmpaddr;
634 	vm_size_t size, pageoff;
635 	vm_inherit_t inherit;
636 	int error;
637 
638 	addr = (vm_offset_t)uap->addr;
639 	size = uap->len;
640 	inherit = uap->inherit;
641 
642 	pageoff = (addr & PAGE_MASK);
643 	addr -= pageoff;
644 	size += pageoff;
645 	size = (vm_size_t) round_page(size);
646 	if (size < uap->len)		/* wrap */
647 		return(EINVAL);
648 	tmpaddr = addr + size;		/* workaround gcc4 opt */
649 	if (tmpaddr < addr)		/* wrap */
650 		return(EINVAL);
651 
652 	lwkt_gettoken(&vm_token);
653 
654 	switch (vm_map_inherit(&p->p_vmspace->vm_map, addr,
655 			       addr + size, inherit)) {
656 	case KERN_SUCCESS:
657 		error = 0;
658 		break;
659 	case KERN_PROTECTION_FAILURE:
660 		error = EACCES;
661 		break;
662 	default:
663 		error = EINVAL;
664 		break;
665 	}
666 	lwkt_reltoken(&vm_token);
667 	return (error);
668 }
669 
670 /*
671  * madvise system call handler
672  *
673  * madvise_args(void *addr, size_t len, int behav)
674  *
675  * No requirements.
676  */
677 int
678 sys_madvise(struct madvise_args *uap)
679 {
680 	struct proc *p = curproc;
681 	vm_offset_t start, end;
682 	vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
683 	int error;
684 
685 	/*
686 	 * Check for illegal behavior
687 	 */
688 	if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
689 		return (EINVAL);
690 	/*
691 	 * Check for illegal addresses.  Watch out for address wrap... Note
692 	 * that VM_*_ADDRESS are not constants due to casts (argh).
693 	 */
694 	if (tmpaddr < (vm_offset_t)uap->addr)
695 		return (EINVAL);
696 	if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
697 		return (EINVAL);
698 	if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
699 		return (EINVAL);
700 
701 	/*
702 	 * Since this routine is only advisory, we default to conservative
703 	 * behavior.
704 	 */
705 	start = trunc_page((vm_offset_t)uap->addr);
706 	end = round_page(tmpaddr);
707 
708 	lwkt_gettoken(&vm_token);
709 	error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
710 			       uap->behav, 0);
711 	lwkt_reltoken(&vm_token);
712 	return (error);
713 }
714 
715 /*
716  * mcontrol system call handler
717  *
718  * mcontrol_args(void *addr, size_t len, int behav, off_t value)
719  *
720  * No requirements
721  */
722 int
723 sys_mcontrol(struct mcontrol_args *uap)
724 {
725 	struct proc *p = curproc;
726 	vm_offset_t start, end;
727 	vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
728 	int error;
729 
730 	/*
731 	 * Check for illegal behavior
732 	 */
733 	if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
734 		return (EINVAL);
735 	/*
736 	 * Check for illegal addresses.  Watch out for address wrap... Note
737 	 * that VM_*_ADDRESS are not constants due to casts (argh).
738 	 */
739 	if (tmpaddr < (vm_offset_t) uap->addr)
740 		return (EINVAL);
741 	if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
742 		return (EINVAL);
743 	if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
744 		return (EINVAL);
745 
746 	/*
747 	 * Since this routine is only advisory, we default to conservative
748 	 * behavior.
749 	 */
750 	start = trunc_page((vm_offset_t)uap->addr);
751 	end = round_page(tmpaddr);
752 
753 	lwkt_gettoken(&vm_token);
754 	error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
755 			       uap->behav, uap->value);
756 	lwkt_reltoken(&vm_token);
757 	return (error);
758 }
759 
760 
761 /*
762  * mincore system call handler
763  *
764  * mincore_args(const void *addr, size_t len, char *vec)
765  *
766  * No requirements
767  */
768 int
769 sys_mincore(struct mincore_args *uap)
770 {
771 	struct proc *p = curproc;
772 	vm_offset_t addr, first_addr;
773 	vm_offset_t end, cend;
774 	pmap_t pmap;
775 	vm_map_t map;
776 	char *vec;
777 	int error;
778 	int vecindex, lastvecindex;
779 	vm_map_entry_t current;
780 	vm_map_entry_t entry;
781 	int mincoreinfo;
782 	unsigned int timestamp;
783 
784 	/*
785 	 * Make sure that the addresses presented are valid for user
786 	 * mode.
787 	 */
788 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
789 	end = addr + (vm_size_t)round_page(uap->len);
790 	if (end < addr)
791 		return (EINVAL);
792 	if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS)
793 		return (EINVAL);
794 
795 	/*
796 	 * Address of byte vector
797 	 */
798 	vec = uap->vec;
799 
800 	map = &p->p_vmspace->vm_map;
801 	pmap = vmspace_pmap(p->p_vmspace);
802 
803 	lwkt_gettoken(&vm_token);
804 	vm_map_lock_read(map);
805 RestartScan:
806 	timestamp = map->timestamp;
807 
808 	if (!vm_map_lookup_entry(map, addr, &entry))
809 		entry = entry->next;
810 
811 	/*
812 	 * Do this on a map entry basis so that if the pages are not
813 	 * in the current processes address space, we can easily look
814 	 * up the pages elsewhere.
815 	 */
816 	lastvecindex = -1;
817 	for(current = entry;
818 		(current != &map->header) && (current->start < end);
819 		current = current->next) {
820 
821 		/*
822 		 * ignore submaps (for now) or null objects
823 		 */
824 		if (current->maptype != VM_MAPTYPE_NORMAL &&
825 		    current->maptype != VM_MAPTYPE_VPAGETABLE) {
826 			continue;
827 		}
828 		if (current->object.vm_object == NULL)
829 			continue;
830 
831 		/*
832 		 * limit this scan to the current map entry and the
833 		 * limits for the mincore call
834 		 */
835 		if (addr < current->start)
836 			addr = current->start;
837 		cend = current->end;
838 		if (cend > end)
839 			cend = end;
840 
841 		/*
842 		 * scan this entry one page at a time
843 		 */
844 		while (addr < cend) {
845 			/*
846 			 * Check pmap first, it is likely faster, also
847 			 * it can provide info as to whether we are the
848 			 * one referencing or modifying the page.
849 			 *
850 			 * If we have to check the VM object, only mess
851 			 * around with normal maps.  Do not mess around
852 			 * with virtual page tables (XXX).
853 			 */
854 			mincoreinfo = pmap_mincore(pmap, addr);
855 			if (mincoreinfo == 0 &&
856 			    current->maptype == VM_MAPTYPE_NORMAL) {
857 				vm_pindex_t pindex;
858 				vm_ooffset_t offset;
859 				vm_page_t m;
860 
861 				/*
862 				 * calculate the page index into the object
863 				 */
864 				offset = current->offset + (addr - current->start);
865 				pindex = OFF_TO_IDX(offset);
866 
867 				/*
868 				 * if the page is resident, then gather
869 				 * information about it.  spl protection is
870 				 * required to maintain the object
871 				 * association.  And XXX what if the page is
872 				 * busy?  What's the deal with that?
873 				 */
874 				crit_enter();
875 				m = vm_page_lookup(current->object.vm_object,
876 						    pindex);
877 				if (m && m->valid) {
878 					mincoreinfo = MINCORE_INCORE;
879 					if (m->dirty ||
880 						pmap_is_modified(m))
881 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
882 					if ((m->flags & PG_REFERENCED) ||
883 						pmap_ts_referenced(m)) {
884 						vm_page_flag_set(m, PG_REFERENCED);
885 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
886 					}
887 				}
888 				crit_exit();
889 			}
890 
891 			/*
892 			 * subyte may page fault.  In case it needs to modify
893 			 * the map, we release the lock.
894 			 */
895 			vm_map_unlock_read(map);
896 
897 			/*
898 			 * calculate index into user supplied byte vector
899 			 */
900 			vecindex = OFF_TO_IDX(addr - first_addr);
901 
902 			/*
903 			 * If we have skipped map entries, we need to make sure that
904 			 * the byte vector is zeroed for those skipped entries.
905 			 */
906 			while((lastvecindex + 1) < vecindex) {
907 				error = subyte( vec + lastvecindex, 0);
908 				if (error) {
909 					error = EFAULT;
910 					goto done;
911 				}
912 				++lastvecindex;
913 			}
914 
915 			/*
916 			 * Pass the page information to the user
917 			 */
918 			error = subyte( vec + vecindex, mincoreinfo);
919 			if (error) {
920 				error = EFAULT;
921 				goto done;
922 			}
923 
924 			/*
925 			 * If the map has changed, due to the subyte, the previous
926 			 * output may be invalid.
927 			 */
928 			vm_map_lock_read(map);
929 			if (timestamp != map->timestamp)
930 				goto RestartScan;
931 
932 			lastvecindex = vecindex;
933 			addr += PAGE_SIZE;
934 		}
935 	}
936 
937 	/*
938 	 * subyte may page fault.  In case it needs to modify
939 	 * the map, we release the lock.
940 	 */
941 	vm_map_unlock_read(map);
942 
943 	/*
944 	 * Zero the last entries in the byte vector.
945 	 */
946 	vecindex = OFF_TO_IDX(end - first_addr);
947 	while((lastvecindex + 1) < vecindex) {
948 		error = subyte( vec + lastvecindex, 0);
949 		if (error) {
950 			error = EFAULT;
951 			goto done;
952 		}
953 		++lastvecindex;
954 	}
955 
956 	/*
957 	 * If the map has changed, due to the subyte, the previous
958 	 * output may be invalid.
959 	 */
960 	vm_map_lock_read(map);
961 	if (timestamp != map->timestamp)
962 		goto RestartScan;
963 	vm_map_unlock_read(map);
964 
965 	error = 0;
966 done:
967 	lwkt_reltoken(&vm_token);
968 	return (error);
969 }
970 
971 /*
972  * mlock system call handler
973  *
974  * mlock_args(const void *addr, size_t len)
975  *
976  * No requirements
977  */
978 int
979 sys_mlock(struct mlock_args *uap)
980 {
981 	vm_offset_t addr;
982 	vm_offset_t tmpaddr;
983 	vm_size_t size, pageoff;
984 	struct thread *td = curthread;
985 	struct proc *p = td->td_proc;
986 	int error;
987 
988 	addr = (vm_offset_t) uap->addr;
989 	size = uap->len;
990 
991 	pageoff = (addr & PAGE_MASK);
992 	addr -= pageoff;
993 	size += pageoff;
994 	size = (vm_size_t) round_page(size);
995 	if (size < uap->len)		/* wrap */
996 		return(EINVAL);
997 	tmpaddr = addr + size;		/* workaround gcc4 opt */
998 	if (tmpaddr < addr)		/* wrap */
999 		return (EINVAL);
1000 
1001 	if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
1002 		return (EAGAIN);
1003 
1004 	lwkt_gettoken(&vm_token);
1005 #ifdef pmap_wired_count
1006 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
1007 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) {
1008 		lwkt_reltoken(&vm_token);
1009 		return (ENOMEM);
1010 	}
1011 #else
1012 	error = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1013 	if (error) {
1014 		lwkt_reltoken(&vm_token);
1015 		return (error);
1016 	}
1017 #endif
1018 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
1019 	lwkt_reltoken(&vm_token);
1020 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1021 }
1022 
1023 /*
1024  * mlockall_args(int how)
1025  *
1026  * Dummy routine, doesn't actually do anything.
1027  *
1028  * No requirements
1029  */
1030 int
1031 sys_mlockall(struct mlockall_args *uap)
1032 {
1033 	return (ENOSYS);
1034 }
1035 
1036 /*
1037  * munlockall_args(void)
1038  *
1039  * Dummy routine, doesn't actually do anything.
1040  *
1041  * No requirements
1042  */
1043 int
1044 sys_munlockall(struct munlockall_args *uap)
1045 {
1046 	return (ENOSYS);
1047 }
1048 
1049 /*
1050  * munlock system call handler
1051  *
1052  * munlock_args(const void *addr, size_t len)
1053  *
1054  * No requirements
1055  */
1056 int
1057 sys_munlock(struct munlock_args *uap)
1058 {
1059 	struct thread *td = curthread;
1060 	struct proc *p = td->td_proc;
1061 	vm_offset_t addr;
1062 	vm_offset_t tmpaddr;
1063 	vm_size_t size, pageoff;
1064 	int error;
1065 
1066 	addr = (vm_offset_t) uap->addr;
1067 	size = uap->len;
1068 
1069 	pageoff = (addr & PAGE_MASK);
1070 	addr -= pageoff;
1071 	size += pageoff;
1072 	size = (vm_size_t) round_page(size);
1073 
1074 	tmpaddr = addr + size;
1075 	if (tmpaddr < addr)		/* wrap */
1076 		return (EINVAL);
1077 
1078 #ifndef pmap_wired_count
1079 	error = priv_check(td, PRIV_ROOT);
1080 	if (error)
1081 		return (error);
1082 #endif
1083 
1084 	lwkt_gettoken(&vm_token);
1085 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
1086 	lwkt_reltoken(&vm_token);
1087 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1088 }
1089 
1090 /*
1091  * Internal version of mmap.
1092  * Currently used by mmap, exec, and sys5 shared memory.
1093  * Handle is either a vnode pointer or NULL for MAP_ANON.
1094  *
1095  * No requirements; kern_mmap path holds the vm_token
1096  */
1097 int
1098 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1099 	vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff)
1100 {
1101 	boolean_t fitit;
1102 	vm_object_t object;
1103 	vm_offset_t eaddr;
1104 	vm_size_t   esize;
1105 	struct vnode *vp;
1106 	struct thread *td = curthread;
1107 	struct proc *p;
1108 	int rv = KERN_SUCCESS;
1109 	off_t objsize;
1110 	int docow;
1111 
1112 	if (size == 0)
1113 		return (0);
1114 
1115 	objsize = round_page(size);
1116 	if (objsize < size)
1117 		return (EINVAL);
1118 	size = objsize;
1119 
1120 	lwkt_gettoken(&vm_token);
1121 
1122 	/*
1123 	 * XXX messy code, fixme
1124 	 *
1125 	 * NOTE: Overflow checks require discrete statements or GCC4
1126 	 * will optimize it out.
1127 	 */
1128 	if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
1129 		esize = map->size + size;	/* workaround gcc4 opt */
1130 		if (esize < map->size ||
1131 		    esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1132 			lwkt_reltoken(&vm_token);
1133 			return(ENOMEM);
1134 		}
1135 	}
1136 
1137 	/*
1138 	 * We currently can only deal with page aligned file offsets.
1139 	 * The check is here rather than in the syscall because the
1140 	 * kernel calls this function internally for other mmaping
1141 	 * operations (such as in exec) and non-aligned offsets will
1142 	 * cause pmap inconsistencies...so we want to be sure to
1143 	 * disallow this in all cases.
1144 	 *
1145 	 * NOTE: Overflow checks require discrete statements or GCC4
1146 	 * will optimize it out.
1147 	 */
1148 	if (foff & PAGE_MASK) {
1149 		lwkt_reltoken(&vm_token);
1150 		return (EINVAL);
1151 	}
1152 
1153 	if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
1154 		fitit = TRUE;
1155 		*addr = round_page(*addr);
1156 	} else {
1157 		if (*addr != trunc_page(*addr)) {
1158 			lwkt_reltoken(&vm_token);
1159 			return (EINVAL);
1160 		}
1161 		eaddr = *addr + size;
1162 		if (eaddr < *addr) {
1163 			lwkt_reltoken(&vm_token);
1164 			return (EINVAL);
1165 		}
1166 		fitit = FALSE;
1167 		if ((flags & MAP_TRYFIXED) == 0)
1168 			vm_map_remove(map, *addr, *addr + size);
1169 	}
1170 
1171 	/*
1172 	 * Lookup/allocate object.
1173 	 */
1174 	if (flags & MAP_ANON) {
1175 		/*
1176 		 * Unnamed anonymous regions always start at 0.
1177 		 */
1178 		if (handle) {
1179 			/*
1180 			 * Default memory object
1181 			 */
1182 			object = default_pager_alloc(handle, objsize,
1183 						     prot, foff);
1184 			if (object == NULL) {
1185 				lwkt_reltoken(&vm_token);
1186 				return(ENOMEM);
1187 			}
1188 			docow = MAP_PREFAULT_PARTIAL;
1189 		} else {
1190 			/*
1191 			 * Implicit single instance of a default memory
1192 			 * object, so we don't need a VM object yet.
1193 			 */
1194 			foff = 0;
1195 			object = NULL;
1196 			docow = 0;
1197 		}
1198 		vp = NULL;
1199 	} else {
1200 		vp = (struct vnode *)handle;
1201 		if (vp->v_type == VCHR) {
1202 			/*
1203 			 * Device mappings (device size unknown?).
1204 			 * Force them to be shared.
1205 			 */
1206 			handle = (void *)(intptr_t)vp->v_rdev;
1207 			object = dev_pager_alloc(handle, objsize, prot, foff);
1208 			if (object == NULL) {
1209 				lwkt_reltoken(&vm_token);
1210 				return(EINVAL);
1211 			}
1212 			docow = MAP_PREFAULT_PARTIAL;
1213 			flags &= ~(MAP_PRIVATE|MAP_COPY);
1214 			flags |= MAP_SHARED;
1215 		} else {
1216 			/*
1217 			 * Regular file mapping (typically).  The attribute
1218 			 * check is for the link count test only.  Mmapble
1219 			 * vnodes must already have a VM object assigned.
1220 			 */
1221 			struct vattr vat;
1222 			int error;
1223 
1224 			error = VOP_GETATTR(vp, &vat);
1225 			if (error) {
1226 				lwkt_reltoken(&vm_token);
1227 				return (error);
1228 			}
1229 			docow = MAP_PREFAULT_PARTIAL;
1230 			object = vnode_pager_reference(vp);
1231 			if (object == NULL && vp->v_type == VREG) {
1232 				lwkt_reltoken(&vm_token);
1233 				kprintf("Warning: cannot mmap vnode %p, no "
1234 					"object\n", vp);
1235 				return(EINVAL);
1236 			}
1237 
1238 			/*
1239 			 * If it is a regular file without any references
1240 			 * we do not need to sync it.
1241 			 */
1242 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1243 				flags |= MAP_NOSYNC;
1244 			}
1245 		}
1246 	}
1247 
1248 	/*
1249 	 * Deal with the adjusted flags
1250 	 */
1251 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1252 		docow |= MAP_COPY_ON_WRITE;
1253 	if (flags & MAP_NOSYNC)
1254 		docow |= MAP_DISABLE_SYNCER;
1255 	if (flags & MAP_NOCORE)
1256 		docow |= MAP_DISABLE_COREDUMP;
1257 
1258 #if defined(VM_PROT_READ_IS_EXEC)
1259 	if (prot & VM_PROT_READ)
1260 		prot |= VM_PROT_EXECUTE;
1261 
1262 	if (maxprot & VM_PROT_READ)
1263 		maxprot |= VM_PROT_EXECUTE;
1264 #endif
1265 
1266 	/*
1267 	 * This may place the area in its own page directory if (size) is
1268 	 * large enough, otherwise it typically returns its argument.
1269 	 */
1270 	if (fitit) {
1271 		*addr = pmap_addr_hint(object, *addr, size);
1272 	}
1273 
1274 	/*
1275 	 * Stack mappings need special attention.
1276 	 *
1277 	 * Mappings that use virtual page tables will default to storing
1278 	 * the page table at offset 0.
1279 	 */
1280 	if (flags & MAP_STACK) {
1281 		rv = vm_map_stack(map, *addr, size, flags,
1282 				  prot, maxprot, docow);
1283 	} else if (flags & MAP_VPAGETABLE) {
1284 		rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1285 				 fitit, VM_MAPTYPE_VPAGETABLE,
1286 				 prot, maxprot, docow);
1287 	} else {
1288 		rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1289 				 fitit, VM_MAPTYPE_NORMAL,
1290 				 prot, maxprot, docow);
1291 	}
1292 
1293 	if (rv != KERN_SUCCESS) {
1294 		/*
1295 		 * Lose the object reference. Will destroy the
1296 		 * object if it's an unnamed anonymous mapping
1297 		 * or named anonymous without other references.
1298 		 */
1299 		vm_object_deallocate(object);
1300 		goto out;
1301 	}
1302 
1303 	/*
1304 	 * Shared memory is also shared with children.
1305 	 */
1306 	if (flags & (MAP_SHARED|MAP_INHERIT)) {
1307 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1308 		if (rv != KERN_SUCCESS) {
1309 			vm_map_remove(map, *addr, *addr + size);
1310 			goto out;
1311 		}
1312 	}
1313 
1314 	/*
1315 	 * Set the access time on the vnode
1316 	 */
1317 	if (vp != NULL)
1318 		vn_mark_atime(vp, td);
1319 out:
1320 	lwkt_reltoken(&vm_token);
1321 
1322 	switch (rv) {
1323 	case KERN_SUCCESS:
1324 		return (0);
1325 	case KERN_INVALID_ADDRESS:
1326 	case KERN_NO_SPACE:
1327 		return (ENOMEM);
1328 	case KERN_PROTECTION_FAILURE:
1329 		return (EACCES);
1330 	default:
1331 		return (EINVAL);
1332 	}
1333 }
1334