xref: /freebsd/sys/vm/vm_mmap.c (revision 145992504973bd16cf3518af9ba5ce185fefa82a)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35  *
36  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
37  */
38 
39 /*
40  * Mapped file (mmap) interface to VM
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capability.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/racct.h>
60 #include <sys/resource.h>
61 #include <sys/resourcevar.h>
62 #include <sys/vnode.h>
63 #include <sys/fcntl.h>
64 #include <sys/file.h>
65 #include <sys/mman.h>
66 #include <sys/mount.h>
67 #include <sys/conf.h>
68 #include <sys/stat.h>
69 #include <sys/sysent.h>
70 #include <sys/vmmeter.h>
71 
72 #include <security/mac/mac_framework.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vm_pageout.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_page.h>
84 #include <vm/vnode_pager.h>
85 
86 #ifdef HWPMC_HOOKS
87 #include <sys/pmckern.h>
88 #endif
89 
90 #ifndef _SYS_SYSPROTO_H_
91 struct sbrk_args {
92 	int incr;
93 };
94 #endif
95 
96 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
97     int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
98 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
99     int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
100 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
101     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
102 
103 /*
104  * MPSAFE
105  */
106 /* ARGSUSED */
107 int
108 sys_sbrk(td, uap)
109 	struct thread *td;
110 	struct sbrk_args *uap;
111 {
112 	/* Not yet implemented */
113 	return (EOPNOTSUPP);
114 }
115 
116 #ifndef _SYS_SYSPROTO_H_
117 struct sstk_args {
118 	int incr;
119 };
120 #endif
121 
122 /*
123  * MPSAFE
124  */
125 /* ARGSUSED */
126 int
127 sys_sstk(td, uap)
128 	struct thread *td;
129 	struct sstk_args *uap;
130 {
131 	/* Not yet implemented */
132 	return (EOPNOTSUPP);
133 }
134 
135 #if defined(COMPAT_43)
136 #ifndef _SYS_SYSPROTO_H_
137 struct getpagesize_args {
138 	int dummy;
139 };
140 #endif
141 
142 int
143 ogetpagesize(td, uap)
144 	struct thread *td;
145 	struct getpagesize_args *uap;
146 {
147 	/* MP SAFE */
148 	td->td_retval[0] = PAGE_SIZE;
149 	return (0);
150 }
151 #endif				/* COMPAT_43 */
152 
153 
154 /*
155  * Memory Map (mmap) system call.  Note that the file offset
156  * and address are allowed to be NOT page aligned, though if
157  * the MAP_FIXED flag it set, both must have the same remainder
158  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
159  * page-aligned, the actual mapping starts at trunc_page(addr)
160  * and the return value is adjusted up by the page offset.
161  *
162  * Generally speaking, only character devices which are themselves
163  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
164  * there would be no cache coherency between a descriptor and a VM mapping
165  * both to the same character device.
166  */
167 #ifndef _SYS_SYSPROTO_H_
168 struct mmap_args {
169 	void *addr;
170 	size_t len;
171 	int prot;
172 	int flags;
173 	int fd;
174 	long pad;
175 	off_t pos;
176 };
177 #endif
178 
179 /*
180  * MPSAFE
181  */
182 int
183 sys_mmap(td, uap)
184 	struct thread *td;
185 	struct mmap_args *uap;
186 {
187 #ifdef HWPMC_HOOKS
188 	struct pmckern_map_in pkm;
189 #endif
190 	struct file *fp;
191 	struct vnode *vp;
192 	vm_offset_t addr;
193 	vm_size_t size, pageoff;
194 	vm_prot_t cap_maxprot, prot, maxprot;
195 	void *handle;
196 	objtype_t handle_type;
197 	int flags, error;
198 	off_t pos;
199 	struct vmspace *vms = td->td_proc->p_vmspace;
200 	cap_rights_t rights;
201 
202 	addr = (vm_offset_t) uap->addr;
203 	size = uap->len;
204 	prot = uap->prot & VM_PROT_ALL;
205 	flags = uap->flags;
206 	pos = uap->pos;
207 
208 	fp = NULL;
209 
210 	/*
211 	 * Enforce the constraints.
212 	 * Mapping of length 0 is only allowed for old binaries.
213 	 * Anonymous mapping shall specify -1 as filedescriptor and
214 	 * zero position for new code. Be nice to ancient a.out
215 	 * binaries and correct pos for anonymous mapping, since old
216 	 * ld.so sometimes issues anonymous map requests with non-zero
217 	 * pos.
218 	 */
219 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
220 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
221 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
222 			return (EINVAL);
223 	} else {
224 		if ((flags & MAP_ANON) != 0)
225 			pos = 0;
226 	}
227 
228 	if (flags & MAP_STACK) {
229 		if ((uap->fd != -1) ||
230 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
231 			return (EINVAL);
232 		flags |= MAP_ANON;
233 		pos = 0;
234 	}
235 
236 	/*
237 	 * Align the file position to a page boundary,
238 	 * and save its page offset component.
239 	 */
240 	pageoff = (pos & PAGE_MASK);
241 	pos -= pageoff;
242 
243 	/* Adjust size for rounding (on both ends). */
244 	size += pageoff;			/* low end... */
245 	size = (vm_size_t) round_page(size);	/* hi end */
246 
247 	/*
248 	 * Check for illegal addresses.  Watch out for address wrap... Note
249 	 * that VM_*_ADDRESS are not constants due to casts (argh).
250 	 */
251 	if (flags & MAP_FIXED) {
252 		/*
253 		 * The specified address must have the same remainder
254 		 * as the file offset taken modulo PAGE_SIZE, so it
255 		 * should be aligned after adjustment by pageoff.
256 		 */
257 		addr -= pageoff;
258 		if (addr & PAGE_MASK)
259 			return (EINVAL);
260 
261 		/* Address range must be all in user VM space. */
262 		if (addr < vm_map_min(&vms->vm_map) ||
263 		    addr + size > vm_map_max(&vms->vm_map))
264 			return (EINVAL);
265 		if (addr + size < addr)
266 			return (EINVAL);
267 	} else {
268 		/*
269 		 * XXX for non-fixed mappings where no hint is provided or
270 		 * the hint would fall in the potential heap space,
271 		 * place it after the end of the largest possible heap.
272 		 *
273 		 * There should really be a pmap call to determine a reasonable
274 		 * location.
275 		 */
276 		PROC_LOCK(td->td_proc);
277 		if (addr == 0 ||
278 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
279 		    addr < round_page((vm_offset_t)vms->vm_daddr +
280 		    lim_max(td->td_proc, RLIMIT_DATA))))
281 			addr = round_page((vm_offset_t)vms->vm_daddr +
282 			    lim_max(td->td_proc, RLIMIT_DATA));
283 		PROC_UNLOCK(td->td_proc);
284 	}
285 	if (flags & MAP_ANON) {
286 		/*
287 		 * Mapping blank space is trivial.
288 		 */
289 		handle = NULL;
290 		handle_type = OBJT_DEFAULT;
291 		maxprot = VM_PROT_ALL;
292 		cap_maxprot = VM_PROT_ALL;
293 	} else {
294 		/*
295 		 * Mapping file, get fp for validation and don't let the
296 		 * descriptor disappear on us if we block. Check capability
297 		 * rights, but also return the maximum rights to be combined
298 		 * with maxprot later.
299 		 */
300 		rights = CAP_MMAP;
301 		if (prot & PROT_READ)
302 			rights |= CAP_READ;
303 		if ((flags & MAP_SHARED) != 0) {
304 			if (prot & PROT_WRITE)
305 				rights |= CAP_WRITE;
306 		}
307 		if (prot & PROT_EXEC)
308 			rights |= CAP_MAPEXEC;
309 		if ((error = fget_mmap(td, uap->fd, rights, &cap_maxprot,
310 		    &fp)) != 0)
311 			goto done;
312 		if (fp->f_type == DTYPE_SHM) {
313 			handle = fp->f_data;
314 			handle_type = OBJT_SWAP;
315 			maxprot = VM_PROT_NONE;
316 
317 			/* FREAD should always be set. */
318 			if (fp->f_flag & FREAD)
319 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
320 			if (fp->f_flag & FWRITE)
321 				maxprot |= VM_PROT_WRITE;
322 			goto map;
323 		}
324 		if (fp->f_type != DTYPE_VNODE) {
325 			error = ENODEV;
326 			goto done;
327 		}
328 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
329     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
330 		/*
331 		 * POSIX shared-memory objects are defined to have
332 		 * kernel persistence, and are not defined to support
333 		 * read(2)/write(2) -- or even open(2).  Thus, we can
334 		 * use MAP_ASYNC to trade on-disk coherence for speed.
335 		 * The shm_open(3) library routine turns on the FPOSIXSHM
336 		 * flag to request this behavior.
337 		 */
338 		if (fp->f_flag & FPOSIXSHM)
339 			flags |= MAP_NOSYNC;
340 #endif
341 		vp = fp->f_vnode;
342 		/*
343 		 * Ensure that file and memory protections are
344 		 * compatible.  Note that we only worry about
345 		 * writability if mapping is shared; in this case,
346 		 * current and max prot are dictated by the open file.
347 		 * XXX use the vnode instead?  Problem is: what
348 		 * credentials do we use for determination? What if
349 		 * proc does a setuid?
350 		 */
351 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
352 			maxprot = VM_PROT_NONE;
353 		else
354 			maxprot = VM_PROT_EXECUTE;
355 		if (fp->f_flag & FREAD) {
356 			maxprot |= VM_PROT_READ;
357 		} else if (prot & PROT_READ) {
358 			error = EACCES;
359 			goto done;
360 		}
361 		/*
362 		 * If we are sharing potential changes (either via
363 		 * MAP_SHARED or via the implicit sharing of character
364 		 * device mappings), and we are trying to get write
365 		 * permission although we opened it without asking
366 		 * for it, bail out.
367 		 */
368 		if ((flags & MAP_SHARED) != 0) {
369 			if ((fp->f_flag & FWRITE) != 0) {
370 				maxprot |= VM_PROT_WRITE;
371 			} else if ((prot & PROT_WRITE) != 0) {
372 				error = EACCES;
373 				goto done;
374 			}
375 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
376 			maxprot |= VM_PROT_WRITE;
377 			cap_maxprot |= VM_PROT_WRITE;
378 		}
379 		handle = (void *)vp;
380 		handle_type = OBJT_VNODE;
381 	}
382 map:
383 	td->td_fpop = fp;
384 	maxprot &= cap_maxprot;
385 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
386 	    flags, handle_type, handle, pos);
387 	td->td_fpop = NULL;
388 #ifdef HWPMC_HOOKS
389 	/* inform hwpmc(4) if an executable is being mapped */
390 	if (error == 0 && handle_type == OBJT_VNODE &&
391 	    (prot & PROT_EXEC)) {
392 		pkm.pm_file = handle;
393 		pkm.pm_address = (uintptr_t) addr;
394 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
395 	}
396 #endif
397 	if (error == 0)
398 		td->td_retval[0] = (register_t) (addr + pageoff);
399 done:
400 	if (fp)
401 		fdrop(fp, td);
402 
403 	return (error);
404 }
405 
406 int
407 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
408 {
409 	struct mmap_args oargs;
410 
411 	oargs.addr = uap->addr;
412 	oargs.len = uap->len;
413 	oargs.prot = uap->prot;
414 	oargs.flags = uap->flags;
415 	oargs.fd = uap->fd;
416 	oargs.pos = uap->pos;
417 	return (sys_mmap(td, &oargs));
418 }
419 
420 #ifdef COMPAT_43
421 #ifndef _SYS_SYSPROTO_H_
422 struct ommap_args {
423 	caddr_t addr;
424 	int len;
425 	int prot;
426 	int flags;
427 	int fd;
428 	long pos;
429 };
430 #endif
431 int
432 ommap(td, uap)
433 	struct thread *td;
434 	struct ommap_args *uap;
435 {
436 	struct mmap_args nargs;
437 	static const char cvtbsdprot[8] = {
438 		0,
439 		PROT_EXEC,
440 		PROT_WRITE,
441 		PROT_EXEC | PROT_WRITE,
442 		PROT_READ,
443 		PROT_EXEC | PROT_READ,
444 		PROT_WRITE | PROT_READ,
445 		PROT_EXEC | PROT_WRITE | PROT_READ,
446 	};
447 
448 #define	OMAP_ANON	0x0002
449 #define	OMAP_COPY	0x0020
450 #define	OMAP_SHARED	0x0010
451 #define	OMAP_FIXED	0x0100
452 
453 	nargs.addr = uap->addr;
454 	nargs.len = uap->len;
455 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
456 #ifdef COMPAT_FREEBSD32
457 #if defined(__amd64__) || defined(__ia64__)
458 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
459 	    nargs.prot != 0)
460 		nargs.prot |= PROT_EXEC;
461 #endif
462 #endif
463 	nargs.flags = 0;
464 	if (uap->flags & OMAP_ANON)
465 		nargs.flags |= MAP_ANON;
466 	if (uap->flags & OMAP_COPY)
467 		nargs.flags |= MAP_COPY;
468 	if (uap->flags & OMAP_SHARED)
469 		nargs.flags |= MAP_SHARED;
470 	else
471 		nargs.flags |= MAP_PRIVATE;
472 	if (uap->flags & OMAP_FIXED)
473 		nargs.flags |= MAP_FIXED;
474 	nargs.fd = uap->fd;
475 	nargs.pos = uap->pos;
476 	return (sys_mmap(td, &nargs));
477 }
478 #endif				/* COMPAT_43 */
479 
480 
481 #ifndef _SYS_SYSPROTO_H_
482 struct msync_args {
483 	void *addr;
484 	size_t len;
485 	int flags;
486 };
487 #endif
488 /*
489  * MPSAFE
490  */
491 int
492 sys_msync(td, uap)
493 	struct thread *td;
494 	struct msync_args *uap;
495 {
496 	vm_offset_t addr;
497 	vm_size_t size, pageoff;
498 	int flags;
499 	vm_map_t map;
500 	int rv;
501 
502 	addr = (vm_offset_t) uap->addr;
503 	size = uap->len;
504 	flags = uap->flags;
505 
506 	pageoff = (addr & PAGE_MASK);
507 	addr -= pageoff;
508 	size += pageoff;
509 	size = (vm_size_t) round_page(size);
510 	if (addr + size < addr)
511 		return (EINVAL);
512 
513 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
514 		return (EINVAL);
515 
516 	map = &td->td_proc->p_vmspace->vm_map;
517 
518 	/*
519 	 * Clean the pages and interpret the return value.
520 	 */
521 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
522 	    (flags & MS_INVALIDATE) != 0);
523 	switch (rv) {
524 	case KERN_SUCCESS:
525 		return (0);
526 	case KERN_INVALID_ADDRESS:
527 		return (EINVAL);	/* Sun returns ENOMEM? */
528 	case KERN_INVALID_ARGUMENT:
529 		return (EBUSY);
530 	case KERN_FAILURE:
531 		return (EIO);
532 	default:
533 		return (EINVAL);
534 	}
535 }
536 
537 #ifndef _SYS_SYSPROTO_H_
538 struct munmap_args {
539 	void *addr;
540 	size_t len;
541 };
542 #endif
543 /*
544  * MPSAFE
545  */
546 int
547 sys_munmap(td, uap)
548 	struct thread *td;
549 	struct munmap_args *uap;
550 {
551 #ifdef HWPMC_HOOKS
552 	struct pmckern_map_out pkm;
553 	vm_map_entry_t entry;
554 #endif
555 	vm_offset_t addr;
556 	vm_size_t size, pageoff;
557 	vm_map_t map;
558 
559 	addr = (vm_offset_t) uap->addr;
560 	size = uap->len;
561 	if (size == 0)
562 		return (EINVAL);
563 
564 	pageoff = (addr & PAGE_MASK);
565 	addr -= pageoff;
566 	size += pageoff;
567 	size = (vm_size_t) round_page(size);
568 	if (addr + size < addr)
569 		return (EINVAL);
570 
571 	/*
572 	 * Check for illegal addresses.  Watch out for address wrap...
573 	 */
574 	map = &td->td_proc->p_vmspace->vm_map;
575 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
576 		return (EINVAL);
577 	vm_map_lock(map);
578 #ifdef HWPMC_HOOKS
579 	/*
580 	 * Inform hwpmc if the address range being unmapped contains
581 	 * an executable region.
582 	 */
583 	pkm.pm_address = (uintptr_t) NULL;
584 	if (vm_map_lookup_entry(map, addr, &entry)) {
585 		for (;
586 		     entry != &map->header && entry->start < addr + size;
587 		     entry = entry->next) {
588 			if (vm_map_check_protection(map, entry->start,
589 				entry->end, VM_PROT_EXECUTE) == TRUE) {
590 				pkm.pm_address = (uintptr_t) addr;
591 				pkm.pm_size = (size_t) size;
592 				break;
593 			}
594 		}
595 	}
596 #endif
597 	vm_map_delete(map, addr, addr + size);
598 
599 #ifdef HWPMC_HOOKS
600 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
601 	vm_map_lock_downgrade(map);
602 	if (pkm.pm_address != (uintptr_t) NULL)
603 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
604 	vm_map_unlock_read(map);
605 #else
606 	vm_map_unlock(map);
607 #endif
608 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
609 	return (0);
610 }
611 
612 #ifndef _SYS_SYSPROTO_H_
613 struct mprotect_args {
614 	const void *addr;
615 	size_t len;
616 	int prot;
617 };
618 #endif
619 /*
620  * MPSAFE
621  */
622 int
623 sys_mprotect(td, uap)
624 	struct thread *td;
625 	struct mprotect_args *uap;
626 {
627 	vm_offset_t addr;
628 	vm_size_t size, pageoff;
629 	vm_prot_t prot;
630 
631 	addr = (vm_offset_t) uap->addr;
632 	size = uap->len;
633 	prot = uap->prot & VM_PROT_ALL;
634 
635 	pageoff = (addr & PAGE_MASK);
636 	addr -= pageoff;
637 	size += pageoff;
638 	size = (vm_size_t) round_page(size);
639 	if (addr + size < addr)
640 		return (EINVAL);
641 
642 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
643 	    addr + size, prot, FALSE)) {
644 	case KERN_SUCCESS:
645 		return (0);
646 	case KERN_PROTECTION_FAILURE:
647 		return (EACCES);
648 	case KERN_RESOURCE_SHORTAGE:
649 		return (ENOMEM);
650 	}
651 	return (EINVAL);
652 }
653 
654 #ifndef _SYS_SYSPROTO_H_
655 struct minherit_args {
656 	void *addr;
657 	size_t len;
658 	int inherit;
659 };
660 #endif
661 /*
662  * MPSAFE
663  */
664 int
665 sys_minherit(td, uap)
666 	struct thread *td;
667 	struct minherit_args *uap;
668 {
669 	vm_offset_t addr;
670 	vm_size_t size, pageoff;
671 	vm_inherit_t inherit;
672 
673 	addr = (vm_offset_t)uap->addr;
674 	size = uap->len;
675 	inherit = uap->inherit;
676 
677 	pageoff = (addr & PAGE_MASK);
678 	addr -= pageoff;
679 	size += pageoff;
680 	size = (vm_size_t) round_page(size);
681 	if (addr + size < addr)
682 		return (EINVAL);
683 
684 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
685 	    addr + size, inherit)) {
686 	case KERN_SUCCESS:
687 		return (0);
688 	case KERN_PROTECTION_FAILURE:
689 		return (EACCES);
690 	}
691 	return (EINVAL);
692 }
693 
694 #ifndef _SYS_SYSPROTO_H_
695 struct madvise_args {
696 	void *addr;
697 	size_t len;
698 	int behav;
699 };
700 #endif
701 
702 /*
703  * MPSAFE
704  */
705 int
706 sys_madvise(td, uap)
707 	struct thread *td;
708 	struct madvise_args *uap;
709 {
710 	vm_offset_t start, end;
711 	vm_map_t map;
712 	struct proc *p;
713 	int error;
714 
715 	/*
716 	 * Check for our special case, advising the swap pager we are
717 	 * "immortal."
718 	 */
719 	if (uap->behav == MADV_PROTECT) {
720 		error = priv_check(td, PRIV_VM_MADV_PROTECT);
721 		if (error == 0) {
722 			p = td->td_proc;
723 			PROC_LOCK(p);
724 			p->p_flag |= P_PROTECTED;
725 			PROC_UNLOCK(p);
726 		}
727 		return (error);
728 	}
729 	/*
730 	 * Check for illegal behavior
731 	 */
732 	if (uap->behav < 0 || uap->behav > MADV_CORE)
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 	map = &td->td_proc->p_vmspace->vm_map;
739 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
740 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
741 		return (EINVAL);
742 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
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((vm_offset_t) uap->addr + uap->len);
751 
752 	if (vm_map_madvise(map, start, end, uap->behav))
753 		return (EINVAL);
754 	return (0);
755 }
756 
757 #ifndef _SYS_SYSPROTO_H_
758 struct mincore_args {
759 	const void *addr;
760 	size_t len;
761 	char *vec;
762 };
763 #endif
764 
765 /*
766  * MPSAFE
767  */
768 int
769 sys_mincore(td, uap)
770 	struct thread *td;
771 	struct mincore_args *uap;
772 {
773 	vm_offset_t addr, first_addr;
774 	vm_offset_t end, cend;
775 	pmap_t pmap;
776 	vm_map_t map;
777 	char *vec;
778 	int error = 0;
779 	int vecindex, lastvecindex;
780 	vm_map_entry_t current;
781 	vm_map_entry_t entry;
782 	vm_object_t object;
783 	vm_paddr_t locked_pa;
784 	vm_page_t m;
785 	vm_pindex_t pindex;
786 	int mincoreinfo;
787 	unsigned int timestamp;
788 	boolean_t locked;
789 
790 	/*
791 	 * Make sure that the addresses presented are valid for user
792 	 * mode.
793 	 */
794 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
795 	end = addr + (vm_size_t)round_page(uap->len);
796 	map = &td->td_proc->p_vmspace->vm_map;
797 	if (end > vm_map_max(map) || end < addr)
798 		return (ENOMEM);
799 
800 	/*
801 	 * Address of byte vector
802 	 */
803 	vec = uap->vec;
804 
805 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
806 
807 	vm_map_lock_read(map);
808 RestartScan:
809 	timestamp = map->timestamp;
810 
811 	if (!vm_map_lookup_entry(map, addr, &entry)) {
812 		vm_map_unlock_read(map);
813 		return (ENOMEM);
814 	}
815 
816 	/*
817 	 * Do this on a map entry basis so that if the pages are not
818 	 * in the current processes address space, we can easily look
819 	 * up the pages elsewhere.
820 	 */
821 	lastvecindex = -1;
822 	for (current = entry;
823 	    (current != &map->header) && (current->start < end);
824 	    current = current->next) {
825 
826 		/*
827 		 * check for contiguity
828 		 */
829 		if (current->end < end &&
830 		    (entry->next == &map->header ||
831 		     current->next->start > current->end)) {
832 			vm_map_unlock_read(map);
833 			return (ENOMEM);
834 		}
835 
836 		/*
837 		 * ignore submaps (for now) or null objects
838 		 */
839 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
840 			current->object.vm_object == NULL)
841 			continue;
842 
843 		/*
844 		 * limit this scan to the current map entry and the
845 		 * limits for the mincore call
846 		 */
847 		if (addr < current->start)
848 			addr = current->start;
849 		cend = current->end;
850 		if (cend > end)
851 			cend = end;
852 
853 		/*
854 		 * scan this entry one page at a time
855 		 */
856 		while (addr < cend) {
857 			/*
858 			 * Check pmap first, it is likely faster, also
859 			 * it can provide info as to whether we are the
860 			 * one referencing or modifying the page.
861 			 */
862 			object = NULL;
863 			locked_pa = 0;
864 		retry:
865 			m = NULL;
866 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
867 			if (locked_pa != 0) {
868 				/*
869 				 * The page is mapped by this process but not
870 				 * both accessed and modified.  It is also
871 				 * managed.  Acquire the object lock so that
872 				 * other mappings might be examined.
873 				 */
874 				m = PHYS_TO_VM_PAGE(locked_pa);
875 				if (m->object != object) {
876 					if (object != NULL)
877 						VM_OBJECT_UNLOCK(object);
878 					object = m->object;
879 					locked = VM_OBJECT_TRYLOCK(object);
880 					vm_page_unlock(m);
881 					if (!locked) {
882 						VM_OBJECT_LOCK(object);
883 						vm_page_lock(m);
884 						goto retry;
885 					}
886 				} else
887 					vm_page_unlock(m);
888 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
889 				    ("mincore: page %p is mapped but invalid",
890 				    m));
891 			} else if (mincoreinfo == 0) {
892 				/*
893 				 * The page is not mapped by this process.  If
894 				 * the object implements managed pages, then
895 				 * determine if the page is resident so that
896 				 * the mappings might be examined.
897 				 */
898 				if (current->object.vm_object != object) {
899 					if (object != NULL)
900 						VM_OBJECT_UNLOCK(object);
901 					object = current->object.vm_object;
902 					VM_OBJECT_LOCK(object);
903 				}
904 				if (object->type == OBJT_DEFAULT ||
905 				    object->type == OBJT_SWAP ||
906 				    object->type == OBJT_VNODE) {
907 					pindex = OFF_TO_IDX(current->offset +
908 					    (addr - current->start));
909 					m = vm_page_lookup(object, pindex);
910 					if (m == NULL &&
911 					    vm_page_is_cached(object, pindex))
912 						mincoreinfo = MINCORE_INCORE;
913 					if (m != NULL && m->valid == 0)
914 						m = NULL;
915 					if (m != NULL)
916 						mincoreinfo = MINCORE_INCORE;
917 				}
918 			}
919 			if (m != NULL) {
920 				/* Examine other mappings to the page. */
921 				if (m->dirty == 0 && pmap_is_modified(m))
922 					vm_page_dirty(m);
923 				if (m->dirty != 0)
924 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
925 				/*
926 				 * The first test for PGA_REFERENCED is an
927 				 * optimization.  The second test is
928 				 * required because a concurrent pmap
929 				 * operation could clear the last reference
930 				 * and set PGA_REFERENCED before the call to
931 				 * pmap_is_referenced().
932 				 */
933 				if ((m->aflags & PGA_REFERENCED) != 0 ||
934 				    pmap_is_referenced(m) ||
935 				    (m->aflags & PGA_REFERENCED) != 0)
936 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
937 			}
938 			if (object != NULL)
939 				VM_OBJECT_UNLOCK(object);
940 
941 			/*
942 			 * subyte may page fault.  In case it needs to modify
943 			 * the map, we release the lock.
944 			 */
945 			vm_map_unlock_read(map);
946 
947 			/*
948 			 * calculate index into user supplied byte vector
949 			 */
950 			vecindex = OFF_TO_IDX(addr - first_addr);
951 
952 			/*
953 			 * If we have skipped map entries, we need to make sure that
954 			 * the byte vector is zeroed for those skipped entries.
955 			 */
956 			while ((lastvecindex + 1) < vecindex) {
957 				error = subyte(vec + lastvecindex, 0);
958 				if (error) {
959 					error = EFAULT;
960 					goto done2;
961 				}
962 				++lastvecindex;
963 			}
964 
965 			/*
966 			 * Pass the page information to the user
967 			 */
968 			error = subyte(vec + vecindex, mincoreinfo);
969 			if (error) {
970 				error = EFAULT;
971 				goto done2;
972 			}
973 
974 			/*
975 			 * If the map has changed, due to the subyte, the previous
976 			 * output may be invalid.
977 			 */
978 			vm_map_lock_read(map);
979 			if (timestamp != map->timestamp)
980 				goto RestartScan;
981 
982 			lastvecindex = vecindex;
983 			addr += PAGE_SIZE;
984 		}
985 	}
986 
987 	/*
988 	 * subyte may page fault.  In case it needs to modify
989 	 * the map, we release the lock.
990 	 */
991 	vm_map_unlock_read(map);
992 
993 	/*
994 	 * Zero the last entries in the byte vector.
995 	 */
996 	vecindex = OFF_TO_IDX(end - first_addr);
997 	while ((lastvecindex + 1) < vecindex) {
998 		error = subyte(vec + lastvecindex, 0);
999 		if (error) {
1000 			error = EFAULT;
1001 			goto done2;
1002 		}
1003 		++lastvecindex;
1004 	}
1005 
1006 	/*
1007 	 * If the map has changed, due to the subyte, the previous
1008 	 * output may be invalid.
1009 	 */
1010 	vm_map_lock_read(map);
1011 	if (timestamp != map->timestamp)
1012 		goto RestartScan;
1013 	vm_map_unlock_read(map);
1014 done2:
1015 	return (error);
1016 }
1017 
1018 #ifndef _SYS_SYSPROTO_H_
1019 struct mlock_args {
1020 	const void *addr;
1021 	size_t len;
1022 };
1023 #endif
1024 /*
1025  * MPSAFE
1026  */
1027 int
1028 sys_mlock(td, uap)
1029 	struct thread *td;
1030 	struct mlock_args *uap;
1031 {
1032 	struct proc *proc;
1033 	vm_offset_t addr, end, last, start;
1034 	vm_size_t npages, size;
1035 	unsigned long nsize;
1036 	int error;
1037 
1038 	error = priv_check(td, PRIV_VM_MLOCK);
1039 	if (error)
1040 		return (error);
1041 	addr = (vm_offset_t)uap->addr;
1042 	size = uap->len;
1043 	last = addr + size;
1044 	start = trunc_page(addr);
1045 	end = round_page(last);
1046 	if (last < addr || end < addr)
1047 		return (EINVAL);
1048 	npages = atop(end - start);
1049 	if (npages > vm_page_max_wired)
1050 		return (ENOMEM);
1051 	proc = td->td_proc;
1052 	PROC_LOCK(proc);
1053 	nsize = ptoa(npages + vmspace_wired_count(proc->p_vmspace));
1054 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1055 		PROC_UNLOCK(proc);
1056 		return (ENOMEM);
1057 	}
1058 	PROC_UNLOCK(proc);
1059 	if (npages + cnt.v_wire_count > vm_page_max_wired)
1060 		return (EAGAIN);
1061 #ifdef RACCT
1062 	PROC_LOCK(proc);
1063 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1064 	PROC_UNLOCK(proc);
1065 	if (error != 0)
1066 		return (ENOMEM);
1067 #endif
1068 	error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1069 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1070 #ifdef RACCT
1071 	if (error != KERN_SUCCESS) {
1072 		PROC_LOCK(proc);
1073 		racct_set(proc, RACCT_MEMLOCK,
1074 		    ptoa(vmspace_wired_count(proc->p_vmspace)));
1075 		PROC_UNLOCK(proc);
1076 	}
1077 #endif
1078 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1079 }
1080 
1081 #ifndef _SYS_SYSPROTO_H_
1082 struct mlockall_args {
1083 	int	how;
1084 };
1085 #endif
1086 
1087 /*
1088  * MPSAFE
1089  */
1090 int
1091 sys_mlockall(td, uap)
1092 	struct thread *td;
1093 	struct mlockall_args *uap;
1094 {
1095 	vm_map_t map;
1096 	int error;
1097 
1098 	map = &td->td_proc->p_vmspace->vm_map;
1099 	error = 0;
1100 
1101 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1102 		return (EINVAL);
1103 
1104 #if 0
1105 	/*
1106 	 * If wiring all pages in the process would cause it to exceed
1107 	 * a hard resource limit, return ENOMEM.
1108 	 */
1109 	PROC_LOCK(td->td_proc);
1110 	if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1111 		PROC_UNLOCK(td->td_proc);
1112 		return (ENOMEM);
1113 	}
1114 	PROC_UNLOCK(td->td_proc);
1115 #else
1116 	error = priv_check(td, PRIV_VM_MLOCK);
1117 	if (error)
1118 		return (error);
1119 #endif
1120 #ifdef RACCT
1121 	PROC_LOCK(td->td_proc);
1122 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1123 	PROC_UNLOCK(td->td_proc);
1124 	if (error != 0)
1125 		return (ENOMEM);
1126 #endif
1127 
1128 	if (uap->how & MCL_FUTURE) {
1129 		vm_map_lock(map);
1130 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1131 		vm_map_unlock(map);
1132 		error = 0;
1133 	}
1134 
1135 	if (uap->how & MCL_CURRENT) {
1136 		/*
1137 		 * P1003.1-2001 mandates that all currently mapped pages
1138 		 * will be memory resident and locked (wired) upon return
1139 		 * from mlockall(). vm_map_wire() will wire pages, by
1140 		 * calling vm_fault_wire() for each page in the region.
1141 		 */
1142 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1143 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1144 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1145 	}
1146 #ifdef RACCT
1147 	if (error != KERN_SUCCESS) {
1148 		PROC_LOCK(td->td_proc);
1149 		racct_set(td->td_proc, RACCT_MEMLOCK,
1150 		    ptoa(vmspace_wired_count(td->td_proc->p_vmspace)));
1151 		PROC_UNLOCK(td->td_proc);
1152 	}
1153 #endif
1154 
1155 	return (error);
1156 }
1157 
1158 #ifndef _SYS_SYSPROTO_H_
1159 struct munlockall_args {
1160 	register_t dummy;
1161 };
1162 #endif
1163 
1164 /*
1165  * MPSAFE
1166  */
1167 int
1168 sys_munlockall(td, uap)
1169 	struct thread *td;
1170 	struct munlockall_args *uap;
1171 {
1172 	vm_map_t map;
1173 	int error;
1174 
1175 	map = &td->td_proc->p_vmspace->vm_map;
1176 	error = priv_check(td, PRIV_VM_MUNLOCK);
1177 	if (error)
1178 		return (error);
1179 
1180 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1181 	vm_map_lock(map);
1182 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1183 	vm_map_unlock(map);
1184 
1185 	/* Forcibly unwire all pages. */
1186 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1187 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1188 #ifdef RACCT
1189 	if (error == KERN_SUCCESS) {
1190 		PROC_LOCK(td->td_proc);
1191 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1192 		PROC_UNLOCK(td->td_proc);
1193 	}
1194 #endif
1195 
1196 	return (error);
1197 }
1198 
1199 #ifndef _SYS_SYSPROTO_H_
1200 struct munlock_args {
1201 	const void *addr;
1202 	size_t len;
1203 };
1204 #endif
1205 /*
1206  * MPSAFE
1207  */
1208 int
1209 sys_munlock(td, uap)
1210 	struct thread *td;
1211 	struct munlock_args *uap;
1212 {
1213 	vm_offset_t addr, end, last, start;
1214 	vm_size_t size;
1215 	int error;
1216 
1217 	error = priv_check(td, PRIV_VM_MUNLOCK);
1218 	if (error)
1219 		return (error);
1220 	addr = (vm_offset_t)uap->addr;
1221 	size = uap->len;
1222 	last = addr + size;
1223 	start = trunc_page(addr);
1224 	end = round_page(last);
1225 	if (last < addr || end < addr)
1226 		return (EINVAL);
1227 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1228 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1229 #ifdef RACCT
1230 	if (error == KERN_SUCCESS) {
1231 		PROC_LOCK(td->td_proc);
1232 		racct_sub(td->td_proc, RACCT_MEMLOCK, ptoa(end - start));
1233 		PROC_UNLOCK(td->td_proc);
1234 	}
1235 #endif
1236 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1237 }
1238 
1239 /*
1240  * vm_mmap_vnode()
1241  *
1242  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1243  * operations on vnodes.
1244  *
1245  * For VCHR vnodes, the vnode lock is held over the call to
1246  * vm_mmap_cdev() to keep vp->v_rdev valid.
1247  */
1248 int
1249 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1250     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1251     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1252     boolean_t *writecounted)
1253 {
1254 	struct vattr va;
1255 	vm_object_t obj;
1256 	vm_offset_t foff;
1257 	struct mount *mp;
1258 	struct ucred *cred;
1259 	int error, flags, locktype;
1260 
1261 	mp = vp->v_mount;
1262 	cred = td->td_ucred;
1263 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1264 		locktype = LK_EXCLUSIVE;
1265 	else
1266 		locktype = LK_SHARED;
1267 	if ((error = vget(vp, locktype, td)) != 0)
1268 		return (error);
1269 	foff = *foffp;
1270 	flags = *flagsp;
1271 	obj = vp->v_object;
1272 	if (vp->v_type == VREG) {
1273 		/*
1274 		 * Get the proper underlying object
1275 		 */
1276 		if (obj == NULL) {
1277 			error = EINVAL;
1278 			goto done;
1279 		}
1280 		if (obj->handle != vp) {
1281 			vput(vp);
1282 			vp = (struct vnode *)obj->handle;
1283 			/*
1284 			 * Bypass filesystems obey the mpsafety of the
1285 			 * underlying fs.
1286 			 */
1287 			error = vget(vp, locktype, td);
1288 			if (error != 0)
1289 				return (error);
1290 		}
1291 		if (locktype == LK_EXCLUSIVE) {
1292 			*writecounted = TRUE;
1293 			vnode_pager_update_writecount(obj, 0, objsize);
1294 		}
1295 	} else if (vp->v_type == VCHR) {
1296 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1297 		    vp->v_rdev, foffp, objp);
1298 		if (error == 0)
1299 			goto mark_atime;
1300 		goto done;
1301 	} else {
1302 		error = EINVAL;
1303 		goto done;
1304 	}
1305 	if ((error = VOP_GETATTR(vp, &va, cred)))
1306 		goto done;
1307 #ifdef MAC
1308 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1309 	if (error != 0)
1310 		goto done;
1311 #endif
1312 	if ((flags & MAP_SHARED) != 0) {
1313 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1314 			if (prot & PROT_WRITE) {
1315 				error = EPERM;
1316 				goto done;
1317 			}
1318 			*maxprotp &= ~VM_PROT_WRITE;
1319 		}
1320 	}
1321 	/*
1322 	 * If it is a regular file without any references
1323 	 * we do not need to sync it.
1324 	 * Adjust object size to be the size of actual file.
1325 	 */
1326 	objsize = round_page(va.va_size);
1327 	if (va.va_nlink == 0)
1328 		flags |= MAP_NOSYNC;
1329 	obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, cred);
1330 	if (obj == NULL) {
1331 		error = ENOMEM;
1332 		goto done;
1333 	}
1334 	*objp = obj;
1335 	*flagsp = flags;
1336 
1337 mark_atime:
1338 	vfs_mark_atime(vp, cred);
1339 
1340 done:
1341 	vput(vp);
1342 	return (error);
1343 }
1344 
1345 /*
1346  * vm_mmap_cdev()
1347  *
1348  * MPSAFE
1349  *
1350  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1351  * operations on cdevs.
1352  */
1353 int
1354 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1355     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1356     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1357 {
1358 	vm_object_t obj;
1359 	struct cdevsw *dsw;
1360 	int error, flags, ref;
1361 
1362 	flags = *flagsp;
1363 
1364 	dsw = dev_refthread(cdev, &ref);
1365 	if (dsw == NULL)
1366 		return (ENXIO);
1367 	if (dsw->d_flags & D_MMAP_ANON) {
1368 		dev_relthread(cdev, ref);
1369 		*maxprotp = VM_PROT_ALL;
1370 		*flagsp |= MAP_ANON;
1371 		return (0);
1372 	}
1373 	/*
1374 	 * cdevs do not provide private mappings of any kind.
1375 	 */
1376 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1377 	    (prot & PROT_WRITE) != 0) {
1378 		dev_relthread(cdev, ref);
1379 		return (EACCES);
1380 	}
1381 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1382 		dev_relthread(cdev, ref);
1383 		return (EINVAL);
1384 	}
1385 	/*
1386 	 * Force device mappings to be shared.
1387 	 */
1388 	flags |= MAP_SHARED;
1389 #ifdef MAC_XXX
1390 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1391 	if (error != 0) {
1392 		dev_relthread(cdev, ref);
1393 		return (error);
1394 	}
1395 #endif
1396 	/*
1397 	 * First, try d_mmap_single().  If that is not implemented
1398 	 * (returns ENODEV), fall back to using the device pager.
1399 	 * Note that d_mmap_single() must return a reference to the
1400 	 * object (it needs to bump the reference count of the object
1401 	 * it returns somehow).
1402 	 *
1403 	 * XXX assumes VM_PROT_* == PROT_*
1404 	 */
1405 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1406 	dev_relthread(cdev, ref);
1407 	if (error != ENODEV)
1408 		return (error);
1409 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1410 	    td->td_ucred);
1411 	if (obj == NULL)
1412 		return (EINVAL);
1413 	*objp = obj;
1414 	*flagsp = flags;
1415 	return (0);
1416 }
1417 
1418 /*
1419  * vm_mmap_shm()
1420  *
1421  * MPSAFE
1422  *
1423  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1424  * operations on shm file descriptors.
1425  */
1426 int
1427 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1428     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1429     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1430 {
1431 	int error;
1432 
1433 	if ((*flagsp & MAP_SHARED) != 0 &&
1434 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1435 	    (prot & PROT_WRITE) != 0)
1436 		return (EACCES);
1437 #ifdef MAC
1438 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1439 	if (error != 0)
1440 		return (error);
1441 #endif
1442 	error = shm_mmap(shmfd, objsize, foff, objp);
1443 	if (error)
1444 		return (error);
1445 	return (0);
1446 }
1447 
1448 /*
1449  * vm_mmap()
1450  *
1451  * MPSAFE
1452  *
1453  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1454  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1455  */
1456 int
1457 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1458 	vm_prot_t maxprot, int flags,
1459 	objtype_t handle_type, void *handle,
1460 	vm_ooffset_t foff)
1461 {
1462 	boolean_t fitit;
1463 	vm_object_t object = NULL;
1464 	struct thread *td = curthread;
1465 	int docow, error, rv;
1466 	boolean_t writecounted;
1467 
1468 	if (size == 0)
1469 		return (0);
1470 
1471 	size = round_page(size);
1472 
1473 	if (map == &td->td_proc->p_vmspace->vm_map) {
1474 		PROC_LOCK(td->td_proc);
1475 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1476 			PROC_UNLOCK(td->td_proc);
1477 			return (ENOMEM);
1478 		}
1479 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1480 			PROC_UNLOCK(td->td_proc);
1481 			return (ENOMEM);
1482 		}
1483 		PROC_UNLOCK(td->td_proc);
1484 	}
1485 
1486 	/*
1487 	 * We currently can only deal with page aligned file offsets.
1488 	 * The check is here rather than in the syscall because the
1489 	 * kernel calls this function internally for other mmaping
1490 	 * operations (such as in exec) and non-aligned offsets will
1491 	 * cause pmap inconsistencies...so we want to be sure to
1492 	 * disallow this in all cases.
1493 	 */
1494 	if (foff & PAGE_MASK)
1495 		return (EINVAL);
1496 
1497 	if ((flags & MAP_FIXED) == 0) {
1498 		fitit = TRUE;
1499 		*addr = round_page(*addr);
1500 	} else {
1501 		if (*addr != trunc_page(*addr))
1502 			return (EINVAL);
1503 		fitit = FALSE;
1504 	}
1505 	writecounted = FALSE;
1506 
1507 	/*
1508 	 * Lookup/allocate object.
1509 	 */
1510 	switch (handle_type) {
1511 	case OBJT_DEVICE:
1512 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1513 		    handle, &foff, &object);
1514 		break;
1515 	case OBJT_VNODE:
1516 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1517 		    handle, &foff, &object, &writecounted);
1518 		break;
1519 	case OBJT_SWAP:
1520 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1521 		    handle, foff, &object);
1522 		break;
1523 	case OBJT_DEFAULT:
1524 		if (handle == NULL) {
1525 			error = 0;
1526 			break;
1527 		}
1528 		/* FALLTHROUGH */
1529 	default:
1530 		error = EINVAL;
1531 		break;
1532 	}
1533 	if (error)
1534 		return (error);
1535 	if (flags & MAP_ANON) {
1536 		object = NULL;
1537 		docow = 0;
1538 		/*
1539 		 * Unnamed anonymous regions always start at 0.
1540 		 */
1541 		if (handle == 0)
1542 			foff = 0;
1543 	} else if (flags & MAP_PREFAULT_READ)
1544 		docow = MAP_PREFAULT;
1545 	else
1546 		docow = MAP_PREFAULT_PARTIAL;
1547 
1548 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1549 		docow |= MAP_COPY_ON_WRITE;
1550 	if (flags & MAP_NOSYNC)
1551 		docow |= MAP_DISABLE_SYNCER;
1552 	if (flags & MAP_NOCORE)
1553 		docow |= MAP_DISABLE_COREDUMP;
1554 	/* Shared memory is also shared with children. */
1555 	if (flags & MAP_SHARED)
1556 		docow |= MAP_INHERIT_SHARE;
1557 	if (writecounted)
1558 		docow |= MAP_VN_WRITECOUNT;
1559 
1560 	if (flags & MAP_STACK)
1561 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1562 		    docow | MAP_STACK_GROWS_DOWN);
1563 	else if (fitit)
1564 		rv = vm_map_find(map, object, foff, addr, size,
1565 		    object != NULL && object->type == OBJT_DEVICE ?
1566 		    VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1567 	else
1568 		rv = vm_map_fixed(map, object, foff, *addr, size,
1569 				 prot, maxprot, docow);
1570 
1571 	if (rv == KERN_SUCCESS) {
1572 		/*
1573 		 * If the process has requested that all future mappings
1574 		 * be wired, then heed this.
1575 		 */
1576 		if (map->flags & MAP_WIREFUTURE) {
1577 			vm_map_wire(map, *addr, *addr + size,
1578 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1579 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1580 		}
1581 	} else {
1582 		/*
1583 		 * If this mapping was accounted for in the vnode's
1584 		 * writecount, then undo that now.
1585 		 */
1586 		if (writecounted)
1587 			vnode_pager_release_writecount(object, 0, size);
1588 		/*
1589 		 * Lose the object reference.  Will destroy the
1590 		 * object if it's an unnamed anonymous mapping
1591 		 * or named anonymous without other references.
1592 		 */
1593 		vm_object_deallocate(object);
1594 	}
1595 	return (vm_mmap_to_errno(rv));
1596 }
1597 
1598 /*
1599  * Translate a Mach VM return code to zero on success or the appropriate errno
1600  * on failure.
1601  */
1602 int
1603 vm_mmap_to_errno(int rv)
1604 {
1605 
1606 	switch (rv) {
1607 	case KERN_SUCCESS:
1608 		return (0);
1609 	case KERN_INVALID_ADDRESS:
1610 	case KERN_NO_SPACE:
1611 		return (ENOMEM);
1612 	case KERN_PROTECTION_FAILURE:
1613 		return (EACCES);
1614 	default:
1615 		return (EINVAL);
1616 	}
1617 }
1618