xref: /freebsd/sys/vm/vm_mmap.c (revision 5e53a4f90f82c4345f277dd87cc9292f26e04a29)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
37  *
38  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
39  */
40 
41 /*
42  * Mapped file (mmap) interface to VM
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_compat.h"
49 #include "opt_hwpmc_hooks.h"
50 #include "opt_vm.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/capsicum.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/sysproto.h>
59 #include <sys/filedesc.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/procctl.h>
63 #include <sys/racct.h>
64 #include <sys/resource.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
69 #include <sys/fcntl.h>
70 #include <sys/file.h>
71 #include <sys/mman.h>
72 #include <sys/mount.h>
73 #include <sys/conf.h>
74 #include <sys/stat.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/vmmeter.h>
78 
79 #include <security/audit/audit.h>
80 #include <security/mac/mac_framework.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vm_pageout.h>
90 #include <vm/vm_extern.h>
91 #include <vm/vm_page.h>
92 #include <vm/vnode_pager.h>
93 
94 #ifdef HWPMC_HOOKS
95 #include <sys/pmckern.h>
96 #endif
97 
98 int old_mlock = 0;
99 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
100     "Do not apply RLIMIT_MEMLOCK on mlockall");
101 
102 #ifdef MAP_32BIT
103 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
104 #endif
105 
106 #ifndef _SYS_SYSPROTO_H_
107 struct sbrk_args {
108 	int incr;
109 };
110 #endif
111 
112 int
113 sys_sbrk(struct thread *td, struct sbrk_args *uap)
114 {
115 	/* Not yet implemented */
116 	return (EOPNOTSUPP);
117 }
118 
119 #ifndef _SYS_SYSPROTO_H_
120 struct sstk_args {
121 	int incr;
122 };
123 #endif
124 
125 int
126 sys_sstk(struct thread *td, struct sstk_args *uap)
127 {
128 	/* Not yet implemented */
129 	return (EOPNOTSUPP);
130 }
131 
132 #if defined(COMPAT_43)
133 #ifndef _SYS_SYSPROTO_H_
134 struct getpagesize_args {
135 	int dummy;
136 };
137 #endif
138 
139 int
140 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
141 {
142 
143 	td->td_retval[0] = PAGE_SIZE;
144 	return (0);
145 }
146 #endif				/* COMPAT_43 */
147 
148 
149 /*
150  * Memory Map (mmap) system call.  Note that the file offset
151  * and address are allowed to be NOT page aligned, though if
152  * the MAP_FIXED flag it set, both must have the same remainder
153  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
154  * page-aligned, the actual mapping starts at trunc_page(addr)
155  * and the return value is adjusted up by the page offset.
156  *
157  * Generally speaking, only character devices which are themselves
158  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
159  * there would be no cache coherency between a descriptor and a VM mapping
160  * both to the same character device.
161  */
162 #ifndef _SYS_SYSPROTO_H_
163 struct mmap_args {
164 	void *addr;
165 	size_t len;
166 	int prot;
167 	int flags;
168 	int fd;
169 	long pad;
170 	off_t pos;
171 };
172 #endif
173 
174 int
175 sys_mmap(struct thread *td, struct mmap_args *uap)
176 {
177 
178 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
179 	    uap->flags, uap->fd, uap->pos));
180 }
181 
182 int
183 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
184     int fd, off_t pos)
185 {
186 	struct vmspace *vms;
187 	struct file *fp;
188 	vm_offset_t addr;
189 	vm_size_t pageoff;
190 	vm_prot_t cap_maxprot;
191 	int align, error;
192 	cap_rights_t rights;
193 
194 	vms = td->td_proc->p_vmspace;
195 	fp = NULL;
196 	AUDIT_ARG_FD(fd);
197 	addr = addr0;
198 
199 	/*
200 	 * Ignore old flags that used to be defined but did not do anything.
201 	 */
202 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
203 
204 	/*
205 	 * Enforce the constraints.
206 	 * Mapping of length 0 is only allowed for old binaries.
207 	 * Anonymous mapping shall specify -1 as filedescriptor and
208 	 * zero position for new code. Be nice to ancient a.out
209 	 * binaries and correct pos for anonymous mapping, since old
210 	 * ld.so sometimes issues anonymous map requests with non-zero
211 	 * pos.
212 	 */
213 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
214 		if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
215 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
216 			return (EINVAL);
217 	} else {
218 		if ((flags & MAP_ANON) != 0)
219 			pos = 0;
220 	}
221 
222 	if (flags & MAP_STACK) {
223 		if ((fd != -1) ||
224 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
225 			return (EINVAL);
226 		flags |= MAP_ANON;
227 		pos = 0;
228 	}
229 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
230 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
231 	    MAP_PREFAULT_READ | MAP_GUARD |
232 #ifdef MAP_32BIT
233 	    MAP_32BIT |
234 #endif
235 	    MAP_ALIGNMENT_MASK)) != 0)
236 		return (EINVAL);
237 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
238 		return (EINVAL);
239 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
240 		return (EINVAL);
241 	if (prot != PROT_NONE &&
242 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
243 		return (EINVAL);
244 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
245 	    pos != 0 || (flags & (MAP_SHARED | MAP_PRIVATE | MAP_PREFAULT |
246 	    MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0))
247 		return (EINVAL);
248 
249 	/*
250 	 * Align the file position to a page boundary,
251 	 * and save its page offset component.
252 	 */
253 	pageoff = (pos & PAGE_MASK);
254 	pos -= pageoff;
255 
256 	/* Adjust size for rounding (on both ends). */
257 	size += pageoff;			/* low end... */
258 	size = (vm_size_t) round_page(size);	/* hi end */
259 
260 	/* Ensure alignment is at least a page and fits in a pointer. */
261 	align = flags & MAP_ALIGNMENT_MASK;
262 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
263 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
264 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
265 		return (EINVAL);
266 
267 	/*
268 	 * Check for illegal addresses.  Watch out for address wrap... Note
269 	 * that VM_*_ADDRESS are not constants due to casts (argh).
270 	 */
271 	if (flags & MAP_FIXED) {
272 		/*
273 		 * The specified address must have the same remainder
274 		 * as the file offset taken modulo PAGE_SIZE, so it
275 		 * should be aligned after adjustment by pageoff.
276 		 */
277 		addr -= pageoff;
278 		if (addr & PAGE_MASK)
279 			return (EINVAL);
280 
281 		/* Address range must be all in user VM space. */
282 		if (addr < vm_map_min(&vms->vm_map) ||
283 		    addr + size > vm_map_max(&vms->vm_map))
284 			return (EINVAL);
285 		if (addr + size < addr)
286 			return (EINVAL);
287 #ifdef MAP_32BIT
288 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
289 			return (EINVAL);
290 	} else if (flags & MAP_32BIT) {
291 		/*
292 		 * For MAP_32BIT, override the hint if it is too high and
293 		 * do not bother moving the mapping past the heap (since
294 		 * the heap is usually above 2GB).
295 		 */
296 		if (addr + size > MAP_32BIT_MAX_ADDR)
297 			addr = 0;
298 #endif
299 	} else {
300 		/*
301 		 * XXX for non-fixed mappings where no hint is provided or
302 		 * the hint would fall in the potential heap space,
303 		 * place it after the end of the largest possible heap.
304 		 *
305 		 * There should really be a pmap call to determine a reasonable
306 		 * location.
307 		 */
308 		if (addr == 0 ||
309 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
310 		    addr < round_page((vm_offset_t)vms->vm_daddr +
311 		    lim_max(td, RLIMIT_DATA))))
312 			addr = round_page((vm_offset_t)vms->vm_daddr +
313 			    lim_max(td, RLIMIT_DATA));
314 	}
315 	if (size == 0) {
316 		/*
317 		 * Return success without mapping anything for old
318 		 * binaries that request a page-aligned mapping of
319 		 * length 0.  For modern binaries, this function
320 		 * returns an error earlier.
321 		 */
322 		error = 0;
323 	} else if ((flags & MAP_GUARD) != 0) {
324 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
325 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
326 	} else if ((flags & MAP_ANON) != 0) {
327 		/*
328 		 * Mapping blank space is trivial.
329 		 *
330 		 * This relies on VM_PROT_* matching PROT_*.
331 		 */
332 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
333 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
334 	} else {
335 		/*
336 		 * Mapping file, get fp for validation and don't let the
337 		 * descriptor disappear on us if we block. Check capability
338 		 * rights, but also return the maximum rights to be combined
339 		 * with maxprot later.
340 		 */
341 		cap_rights_init(&rights, CAP_MMAP);
342 		if (prot & PROT_READ)
343 			cap_rights_set(&rights, CAP_MMAP_R);
344 		if ((flags & MAP_SHARED) != 0) {
345 			if (prot & PROT_WRITE)
346 				cap_rights_set(&rights, CAP_MMAP_W);
347 		}
348 		if (prot & PROT_EXEC)
349 			cap_rights_set(&rights, CAP_MMAP_X);
350 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
351 		if (error != 0)
352 			goto done;
353 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
354 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
355 			error = EINVAL;
356 			goto done;
357 		}
358 
359 		/* This relies on VM_PROT_* matching PROT_*. */
360 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
361 		    cap_maxprot, flags, pos, td);
362 	}
363 
364 	if (error == 0)
365 		td->td_retval[0] = (register_t) (addr + pageoff);
366 done:
367 	if (fp)
368 		fdrop(fp, td);
369 
370 	return (error);
371 }
372 
373 #if defined(COMPAT_FREEBSD6)
374 int
375 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
376 {
377 
378 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
379 	    uap->flags, uap->fd, uap->pos));
380 }
381 #endif
382 
383 #ifdef COMPAT_43
384 #ifndef _SYS_SYSPROTO_H_
385 struct ommap_args {
386 	caddr_t addr;
387 	int len;
388 	int prot;
389 	int flags;
390 	int fd;
391 	long pos;
392 };
393 #endif
394 int
395 ommap(struct thread *td, struct ommap_args *uap)
396 {
397 	static const char cvtbsdprot[8] = {
398 		0,
399 		PROT_EXEC,
400 		PROT_WRITE,
401 		PROT_EXEC | PROT_WRITE,
402 		PROT_READ,
403 		PROT_EXEC | PROT_READ,
404 		PROT_WRITE | PROT_READ,
405 		PROT_EXEC | PROT_WRITE | PROT_READ,
406 	};
407 	int flags, prot;
408 
409 #define	OMAP_ANON	0x0002
410 #define	OMAP_COPY	0x0020
411 #define	OMAP_SHARED	0x0010
412 #define	OMAP_FIXED	0x0100
413 
414 	prot = cvtbsdprot[uap->prot & 0x7];
415 #ifdef COMPAT_FREEBSD32
416 #if defined(__amd64__)
417 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
418 	    prot != 0)
419 		prot |= PROT_EXEC;
420 #endif
421 #endif
422 	flags = 0;
423 	if (uap->flags & OMAP_ANON)
424 		flags |= MAP_ANON;
425 	if (uap->flags & OMAP_COPY)
426 		flags |= MAP_COPY;
427 	if (uap->flags & OMAP_SHARED)
428 		flags |= MAP_SHARED;
429 	else
430 		flags |= MAP_PRIVATE;
431 	if (uap->flags & OMAP_FIXED)
432 		flags |= MAP_FIXED;
433 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
434 	    uap->fd, uap->pos));
435 }
436 #endif				/* COMPAT_43 */
437 
438 
439 #ifndef _SYS_SYSPROTO_H_
440 struct msync_args {
441 	void *addr;
442 	size_t len;
443 	int flags;
444 };
445 #endif
446 int
447 sys_msync(struct thread *td, struct msync_args *uap)
448 {
449 
450 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
451 }
452 
453 int
454 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
455 {
456 	vm_offset_t addr;
457 	vm_size_t pageoff;
458 	vm_map_t map;
459 	int rv;
460 
461 	addr = addr0;
462 	pageoff = (addr & PAGE_MASK);
463 	addr -= pageoff;
464 	size += pageoff;
465 	size = (vm_size_t) round_page(size);
466 	if (addr + size < addr)
467 		return (EINVAL);
468 
469 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
470 		return (EINVAL);
471 
472 	map = &td->td_proc->p_vmspace->vm_map;
473 
474 	/*
475 	 * Clean the pages and interpret the return value.
476 	 */
477 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
478 	    (flags & MS_INVALIDATE) != 0);
479 	switch (rv) {
480 	case KERN_SUCCESS:
481 		return (0);
482 	case KERN_INVALID_ADDRESS:
483 		return (ENOMEM);
484 	case KERN_INVALID_ARGUMENT:
485 		return (EBUSY);
486 	case KERN_FAILURE:
487 		return (EIO);
488 	default:
489 		return (EINVAL);
490 	}
491 }
492 
493 #ifndef _SYS_SYSPROTO_H_
494 struct munmap_args {
495 	void *addr;
496 	size_t len;
497 };
498 #endif
499 int
500 sys_munmap(struct thread *td, struct munmap_args *uap)
501 {
502 
503 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
504 }
505 
506 int
507 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
508 {
509 #ifdef HWPMC_HOOKS
510 	struct pmckern_map_out pkm;
511 	vm_map_entry_t entry;
512 	bool pmc_handled;
513 #endif
514 	vm_offset_t addr;
515 	vm_size_t pageoff;
516 	vm_map_t map;
517 
518 	if (size == 0)
519 		return (EINVAL);
520 
521 	addr = addr0;
522 	pageoff = (addr & PAGE_MASK);
523 	addr -= pageoff;
524 	size += pageoff;
525 	size = (vm_size_t) round_page(size);
526 	if (addr + size < addr)
527 		return (EINVAL);
528 
529 	/*
530 	 * Check for illegal addresses.  Watch out for address wrap...
531 	 */
532 	map = &td->td_proc->p_vmspace->vm_map;
533 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
534 		return (EINVAL);
535 	vm_map_lock(map);
536 #ifdef HWPMC_HOOKS
537 	pmc_handled = false;
538 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
539 		pmc_handled = true;
540 		/*
541 		 * Inform hwpmc if the address range being unmapped contains
542 		 * an executable region.
543 		 */
544 		pkm.pm_address = (uintptr_t) NULL;
545 		if (vm_map_lookup_entry(map, addr, &entry)) {
546 			for (;
547 			    entry != &map->header && entry->start < addr + size;
548 			    entry = entry->next) {
549 				if (vm_map_check_protection(map, entry->start,
550 					entry->end, VM_PROT_EXECUTE) == TRUE) {
551 					pkm.pm_address = (uintptr_t) addr;
552 					pkm.pm_size = (size_t) size;
553 					break;
554 				}
555 			}
556 		}
557 	}
558 #endif
559 	vm_map_delete(map, addr, addr + size);
560 
561 #ifdef HWPMC_HOOKS
562 	if (__predict_false(pmc_handled)) {
563 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
564 		vm_map_lock_downgrade(map);
565 		if (pkm.pm_address != (uintptr_t) NULL)
566 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
567 		vm_map_unlock_read(map);
568 	} else
569 #endif
570 		vm_map_unlock(map);
571 
572 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
573 	return (0);
574 }
575 
576 #ifndef _SYS_SYSPROTO_H_
577 struct mprotect_args {
578 	const void *addr;
579 	size_t len;
580 	int prot;
581 };
582 #endif
583 int
584 sys_mprotect(struct thread *td, struct mprotect_args *uap)
585 {
586 
587 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
588 }
589 
590 int
591 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
592 {
593 	vm_offset_t addr;
594 	vm_size_t pageoff;
595 
596 	addr = addr0;
597 	prot = (prot & VM_PROT_ALL);
598 	pageoff = (addr & PAGE_MASK);
599 	addr -= pageoff;
600 	size += pageoff;
601 	size = (vm_size_t) round_page(size);
602 	if (addr + size < addr)
603 		return (EINVAL);
604 
605 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
606 	    addr + size, prot, FALSE)) {
607 	case KERN_SUCCESS:
608 		return (0);
609 	case KERN_PROTECTION_FAILURE:
610 		return (EACCES);
611 	case KERN_RESOURCE_SHORTAGE:
612 		return (ENOMEM);
613 	}
614 	return (EINVAL);
615 }
616 
617 #ifndef _SYS_SYSPROTO_H_
618 struct minherit_args {
619 	void *addr;
620 	size_t len;
621 	int inherit;
622 };
623 #endif
624 int
625 sys_minherit(struct thread *td, struct minherit_args *uap)
626 {
627 	vm_offset_t addr;
628 	vm_size_t size, pageoff;
629 	vm_inherit_t inherit;
630 
631 	addr = (vm_offset_t)uap->addr;
632 	size = uap->len;
633 	inherit = uap->inherit;
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_inherit(&td->td_proc->p_vmspace->vm_map, addr,
643 	    addr + size, inherit)) {
644 	case KERN_SUCCESS:
645 		return (0);
646 	case KERN_PROTECTION_FAILURE:
647 		return (EACCES);
648 	}
649 	return (EINVAL);
650 }
651 
652 #ifndef _SYS_SYSPROTO_H_
653 struct madvise_args {
654 	void *addr;
655 	size_t len;
656 	int behav;
657 };
658 #endif
659 
660 int
661 sys_madvise(struct thread *td, struct madvise_args *uap)
662 {
663 
664 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
665 }
666 
667 int
668 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
669 {
670 	vm_map_t map;
671 	vm_offset_t addr, end, start;
672 	int flags;
673 
674 	/*
675 	 * Check for our special case, advising the swap pager we are
676 	 * "immortal."
677 	 */
678 	if (behav == MADV_PROTECT) {
679 		flags = PPROT_SET;
680 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
681 		    PROC_SPROTECT, &flags));
682 	}
683 
684 	/*
685 	 * Check for illegal behavior
686 	 */
687 	if (behav < 0 || behav > MADV_CORE)
688 		return (EINVAL);
689 	/*
690 	 * Check for illegal addresses.  Watch out for address wrap... Note
691 	 * that VM_*_ADDRESS are not constants due to casts (argh).
692 	 */
693 	map = &td->td_proc->p_vmspace->vm_map;
694 	addr = addr0;
695 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
696 		return (EINVAL);
697 	if ((addr + len) < addr)
698 		return (EINVAL);
699 
700 	/*
701 	 * Since this routine is only advisory, we default to conservative
702 	 * behavior.
703 	 */
704 	start = trunc_page(addr);
705 	end = round_page(addr + len);
706 
707 	if (vm_map_madvise(map, start, end, behav))
708 		return (EINVAL);
709 	return (0);
710 }
711 
712 #ifndef _SYS_SYSPROTO_H_
713 struct mincore_args {
714 	const void *addr;
715 	size_t len;
716 	char *vec;
717 };
718 #endif
719 
720 int
721 sys_mincore(struct thread *td, struct mincore_args *uap)
722 {
723 
724 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
725 }
726 
727 int
728 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
729 {
730 	vm_offset_t addr, first_addr;
731 	vm_offset_t end, cend;
732 	pmap_t pmap;
733 	vm_map_t map;
734 	int error = 0;
735 	int vecindex, lastvecindex;
736 	vm_map_entry_t current;
737 	vm_map_entry_t entry;
738 	vm_object_t object;
739 	vm_paddr_t locked_pa;
740 	vm_page_t m;
741 	vm_pindex_t pindex;
742 	int mincoreinfo;
743 	unsigned int timestamp;
744 	boolean_t locked;
745 
746 	/*
747 	 * Make sure that the addresses presented are valid for user
748 	 * mode.
749 	 */
750 	first_addr = addr = trunc_page(addr0);
751 	end = addr + (vm_size_t)round_page(len);
752 	map = &td->td_proc->p_vmspace->vm_map;
753 	if (end > vm_map_max(map) || end < addr)
754 		return (ENOMEM);
755 
756 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
757 
758 	vm_map_lock_read(map);
759 RestartScan:
760 	timestamp = map->timestamp;
761 
762 	if (!vm_map_lookup_entry(map, addr, &entry)) {
763 		vm_map_unlock_read(map);
764 		return (ENOMEM);
765 	}
766 
767 	/*
768 	 * Do this on a map entry basis so that if the pages are not
769 	 * in the current processes address space, we can easily look
770 	 * up the pages elsewhere.
771 	 */
772 	lastvecindex = -1;
773 	for (current = entry;
774 	    (current != &map->header) && (current->start < end);
775 	    current = current->next) {
776 
777 		/*
778 		 * check for contiguity
779 		 */
780 		if (current->end < end &&
781 		    (entry->next == &map->header ||
782 		     current->next->start > current->end)) {
783 			vm_map_unlock_read(map);
784 			return (ENOMEM);
785 		}
786 
787 		/*
788 		 * ignore submaps (for now) or null objects
789 		 */
790 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
791 			current->object.vm_object == NULL)
792 			continue;
793 
794 		/*
795 		 * limit this scan to the current map entry and the
796 		 * limits for the mincore call
797 		 */
798 		if (addr < current->start)
799 			addr = current->start;
800 		cend = current->end;
801 		if (cend > end)
802 			cend = end;
803 
804 		/*
805 		 * scan this entry one page at a time
806 		 */
807 		while (addr < cend) {
808 			/*
809 			 * Check pmap first, it is likely faster, also
810 			 * it can provide info as to whether we are the
811 			 * one referencing or modifying the page.
812 			 */
813 			object = NULL;
814 			locked_pa = 0;
815 		retry:
816 			m = NULL;
817 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
818 			if (locked_pa != 0) {
819 				/*
820 				 * The page is mapped by this process but not
821 				 * both accessed and modified.  It is also
822 				 * managed.  Acquire the object lock so that
823 				 * other mappings might be examined.
824 				 */
825 				m = PHYS_TO_VM_PAGE(locked_pa);
826 				if (m->object != object) {
827 					if (object != NULL)
828 						VM_OBJECT_WUNLOCK(object);
829 					object = m->object;
830 					locked = VM_OBJECT_TRYWLOCK(object);
831 					vm_page_unlock(m);
832 					if (!locked) {
833 						VM_OBJECT_WLOCK(object);
834 						vm_page_lock(m);
835 						goto retry;
836 					}
837 				} else
838 					vm_page_unlock(m);
839 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
840 				    ("mincore: page %p is mapped but invalid",
841 				    m));
842 			} else if (mincoreinfo == 0) {
843 				/*
844 				 * The page is not mapped by this process.  If
845 				 * the object implements managed pages, then
846 				 * determine if the page is resident so that
847 				 * the mappings might be examined.
848 				 */
849 				if (current->object.vm_object != object) {
850 					if (object != NULL)
851 						VM_OBJECT_WUNLOCK(object);
852 					object = current->object.vm_object;
853 					VM_OBJECT_WLOCK(object);
854 				}
855 				if (object->type == OBJT_DEFAULT ||
856 				    object->type == OBJT_SWAP ||
857 				    object->type == OBJT_VNODE) {
858 					pindex = OFF_TO_IDX(current->offset +
859 					    (addr - current->start));
860 					m = vm_page_lookup(object, pindex);
861 					if (m != NULL && m->valid == 0)
862 						m = NULL;
863 					if (m != NULL)
864 						mincoreinfo = MINCORE_INCORE;
865 				}
866 			}
867 			if (m != NULL) {
868 				/* Examine other mappings to the page. */
869 				if (m->dirty == 0 && pmap_is_modified(m))
870 					vm_page_dirty(m);
871 				if (m->dirty != 0)
872 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
873 				/*
874 				 * The first test for PGA_REFERENCED is an
875 				 * optimization.  The second test is
876 				 * required because a concurrent pmap
877 				 * operation could clear the last reference
878 				 * and set PGA_REFERENCED before the call to
879 				 * pmap_is_referenced().
880 				 */
881 				if ((m->aflags & PGA_REFERENCED) != 0 ||
882 				    pmap_is_referenced(m) ||
883 				    (m->aflags & PGA_REFERENCED) != 0)
884 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
885 			}
886 			if (object != NULL)
887 				VM_OBJECT_WUNLOCK(object);
888 
889 			/*
890 			 * subyte may page fault.  In case it needs to modify
891 			 * the map, we release the lock.
892 			 */
893 			vm_map_unlock_read(map);
894 
895 			/*
896 			 * calculate index into user supplied byte vector
897 			 */
898 			vecindex = atop(addr - first_addr);
899 
900 			/*
901 			 * If we have skipped map entries, we need to make sure that
902 			 * the byte vector is zeroed for those skipped entries.
903 			 */
904 			while ((lastvecindex + 1) < vecindex) {
905 				++lastvecindex;
906 				error = subyte(vec + lastvecindex, 0);
907 				if (error) {
908 					error = EFAULT;
909 					goto done2;
910 				}
911 			}
912 
913 			/*
914 			 * Pass the page information to the user
915 			 */
916 			error = subyte(vec + vecindex, mincoreinfo);
917 			if (error) {
918 				error = EFAULT;
919 				goto done2;
920 			}
921 
922 			/*
923 			 * If the map has changed, due to the subyte, the previous
924 			 * output may be invalid.
925 			 */
926 			vm_map_lock_read(map);
927 			if (timestamp != map->timestamp)
928 				goto RestartScan;
929 
930 			lastvecindex = vecindex;
931 			addr += PAGE_SIZE;
932 		}
933 	}
934 
935 	/*
936 	 * subyte may page fault.  In case it needs to modify
937 	 * the map, we release the lock.
938 	 */
939 	vm_map_unlock_read(map);
940 
941 	/*
942 	 * Zero the last entries in the byte vector.
943 	 */
944 	vecindex = atop(end - first_addr);
945 	while ((lastvecindex + 1) < vecindex) {
946 		++lastvecindex;
947 		error = subyte(vec + lastvecindex, 0);
948 		if (error) {
949 			error = EFAULT;
950 			goto done2;
951 		}
952 	}
953 
954 	/*
955 	 * If the map has changed, due to the subyte, the previous
956 	 * output may be invalid.
957 	 */
958 	vm_map_lock_read(map);
959 	if (timestamp != map->timestamp)
960 		goto RestartScan;
961 	vm_map_unlock_read(map);
962 done2:
963 	return (error);
964 }
965 
966 #ifndef _SYS_SYSPROTO_H_
967 struct mlock_args {
968 	const void *addr;
969 	size_t len;
970 };
971 #endif
972 int
973 sys_mlock(struct thread *td, struct mlock_args *uap)
974 {
975 
976 	return (kern_mlock(td->td_proc, td->td_ucred,
977 	    __DECONST(uintptr_t, uap->addr), uap->len));
978 }
979 
980 int
981 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
982 {
983 	vm_offset_t addr, end, last, start;
984 	vm_size_t npages, size;
985 	vm_map_t map;
986 	unsigned long nsize;
987 	int error;
988 
989 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
990 	if (error)
991 		return (error);
992 	addr = addr0;
993 	size = len;
994 	last = addr + size;
995 	start = trunc_page(addr);
996 	end = round_page(last);
997 	if (last < addr || end < addr)
998 		return (EINVAL);
999 	npages = atop(end - start);
1000 	if (npages > vm_page_max_wired)
1001 		return (ENOMEM);
1002 	map = &proc->p_vmspace->vm_map;
1003 	PROC_LOCK(proc);
1004 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1005 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1006 		PROC_UNLOCK(proc);
1007 		return (ENOMEM);
1008 	}
1009 	PROC_UNLOCK(proc);
1010 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1011 		return (EAGAIN);
1012 #ifdef RACCT
1013 	if (racct_enable) {
1014 		PROC_LOCK(proc);
1015 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1016 		PROC_UNLOCK(proc);
1017 		if (error != 0)
1018 			return (ENOMEM);
1019 	}
1020 #endif
1021 	error = vm_map_wire(map, start, end,
1022 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1023 #ifdef RACCT
1024 	if (racct_enable && error != KERN_SUCCESS) {
1025 		PROC_LOCK(proc);
1026 		racct_set(proc, RACCT_MEMLOCK,
1027 		    ptoa(pmap_wired_count(map->pmap)));
1028 		PROC_UNLOCK(proc);
1029 	}
1030 #endif
1031 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1032 }
1033 
1034 #ifndef _SYS_SYSPROTO_H_
1035 struct mlockall_args {
1036 	int	how;
1037 };
1038 #endif
1039 
1040 int
1041 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1042 {
1043 	vm_map_t map;
1044 	int error;
1045 
1046 	map = &td->td_proc->p_vmspace->vm_map;
1047 	error = priv_check(td, PRIV_VM_MLOCK);
1048 	if (error)
1049 		return (error);
1050 
1051 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1052 		return (EINVAL);
1053 
1054 	/*
1055 	 * If wiring all pages in the process would cause it to exceed
1056 	 * a hard resource limit, return ENOMEM.
1057 	 */
1058 	if (!old_mlock && uap->how & MCL_CURRENT) {
1059 		PROC_LOCK(td->td_proc);
1060 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1061 			PROC_UNLOCK(td->td_proc);
1062 			return (ENOMEM);
1063 		}
1064 		PROC_UNLOCK(td->td_proc);
1065 	}
1066 #ifdef RACCT
1067 	if (racct_enable) {
1068 		PROC_LOCK(td->td_proc);
1069 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1070 		PROC_UNLOCK(td->td_proc);
1071 		if (error != 0)
1072 			return (ENOMEM);
1073 	}
1074 #endif
1075 
1076 	if (uap->how & MCL_FUTURE) {
1077 		vm_map_lock(map);
1078 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1079 		vm_map_unlock(map);
1080 		error = 0;
1081 	}
1082 
1083 	if (uap->how & MCL_CURRENT) {
1084 		/*
1085 		 * P1003.1-2001 mandates that all currently mapped pages
1086 		 * will be memory resident and locked (wired) upon return
1087 		 * from mlockall(). vm_map_wire() will wire pages, by
1088 		 * calling vm_fault_wire() for each page in the region.
1089 		 */
1090 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1091 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1092 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1093 	}
1094 #ifdef RACCT
1095 	if (racct_enable && error != KERN_SUCCESS) {
1096 		PROC_LOCK(td->td_proc);
1097 		racct_set(td->td_proc, RACCT_MEMLOCK,
1098 		    ptoa(pmap_wired_count(map->pmap)));
1099 		PROC_UNLOCK(td->td_proc);
1100 	}
1101 #endif
1102 
1103 	return (error);
1104 }
1105 
1106 #ifndef _SYS_SYSPROTO_H_
1107 struct munlockall_args {
1108 	register_t dummy;
1109 };
1110 #endif
1111 
1112 int
1113 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1114 {
1115 	vm_map_t map;
1116 	int error;
1117 
1118 	map = &td->td_proc->p_vmspace->vm_map;
1119 	error = priv_check(td, PRIV_VM_MUNLOCK);
1120 	if (error)
1121 		return (error);
1122 
1123 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1124 	vm_map_lock(map);
1125 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1126 	vm_map_unlock(map);
1127 
1128 	/* Forcibly unwire all pages. */
1129 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1130 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1131 #ifdef RACCT
1132 	if (racct_enable && error == KERN_SUCCESS) {
1133 		PROC_LOCK(td->td_proc);
1134 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1135 		PROC_UNLOCK(td->td_proc);
1136 	}
1137 #endif
1138 
1139 	return (error);
1140 }
1141 
1142 #ifndef _SYS_SYSPROTO_H_
1143 struct munlock_args {
1144 	const void *addr;
1145 	size_t len;
1146 };
1147 #endif
1148 int
1149 sys_munlock(struct thread *td, struct munlock_args *uap)
1150 {
1151 
1152 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1153 }
1154 
1155 int
1156 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1157 {
1158 	vm_offset_t addr, end, last, start;
1159 #ifdef RACCT
1160 	vm_map_t map;
1161 #endif
1162 	int error;
1163 
1164 	error = priv_check(td, PRIV_VM_MUNLOCK);
1165 	if (error)
1166 		return (error);
1167 	addr = addr0;
1168 	last = addr + size;
1169 	start = trunc_page(addr);
1170 	end = round_page(last);
1171 	if (last < addr || end < addr)
1172 		return (EINVAL);
1173 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1174 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1175 #ifdef RACCT
1176 	if (racct_enable && error == KERN_SUCCESS) {
1177 		PROC_LOCK(td->td_proc);
1178 		map = &td->td_proc->p_vmspace->vm_map;
1179 		racct_set(td->td_proc, RACCT_MEMLOCK,
1180 		    ptoa(pmap_wired_count(map->pmap)));
1181 		PROC_UNLOCK(td->td_proc);
1182 	}
1183 #endif
1184 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1185 }
1186 
1187 /*
1188  * vm_mmap_vnode()
1189  *
1190  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1191  * operations on vnodes.
1192  */
1193 int
1194 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1195     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1196     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1197     boolean_t *writecounted)
1198 {
1199 	struct vattr va;
1200 	vm_object_t obj;
1201 	vm_ooffset_t foff;
1202 	struct ucred *cred;
1203 	int error, flags, locktype;
1204 
1205 	cred = td->td_ucred;
1206 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1207 		locktype = LK_EXCLUSIVE;
1208 	else
1209 		locktype = LK_SHARED;
1210 	if ((error = vget(vp, locktype, td)) != 0)
1211 		return (error);
1212 	AUDIT_ARG_VNODE1(vp);
1213 	foff = *foffp;
1214 	flags = *flagsp;
1215 	obj = vp->v_object;
1216 	if (vp->v_type == VREG) {
1217 		/*
1218 		 * Get the proper underlying object
1219 		 */
1220 		if (obj == NULL) {
1221 			error = EINVAL;
1222 			goto done;
1223 		}
1224 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1225 			vput(vp);
1226 			vp = (struct vnode *)obj->handle;
1227 			/*
1228 			 * Bypass filesystems obey the mpsafety of the
1229 			 * underlying fs.  Tmpfs never bypasses.
1230 			 */
1231 			error = vget(vp, locktype, td);
1232 			if (error != 0)
1233 				return (error);
1234 		}
1235 		if (locktype == LK_EXCLUSIVE) {
1236 			*writecounted = TRUE;
1237 			vnode_pager_update_writecount(obj, 0, objsize);
1238 		}
1239 	} else {
1240 		error = EINVAL;
1241 		goto done;
1242 	}
1243 	if ((error = VOP_GETATTR(vp, &va, cred)))
1244 		goto done;
1245 #ifdef MAC
1246 	/* This relies on VM_PROT_* matching PROT_*. */
1247 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1248 	if (error != 0)
1249 		goto done;
1250 #endif
1251 	if ((flags & MAP_SHARED) != 0) {
1252 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1253 			if (prot & VM_PROT_WRITE) {
1254 				error = EPERM;
1255 				goto done;
1256 			}
1257 			*maxprotp &= ~VM_PROT_WRITE;
1258 		}
1259 	}
1260 	/*
1261 	 * If it is a regular file without any references
1262 	 * we do not need to sync it.
1263 	 * Adjust object size to be the size of actual file.
1264 	 */
1265 	objsize = round_page(va.va_size);
1266 	if (va.va_nlink == 0)
1267 		flags |= MAP_NOSYNC;
1268 	if (obj->type == OBJT_VNODE) {
1269 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1270 		    cred);
1271 		if (obj == NULL) {
1272 			error = ENOMEM;
1273 			goto done;
1274 		}
1275 	} else {
1276 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1277 		    ("wrong object type"));
1278 		VM_OBJECT_WLOCK(obj);
1279 		vm_object_reference_locked(obj);
1280 #if VM_NRESERVLEVEL > 0
1281 		vm_object_color(obj, 0);
1282 #endif
1283 		VM_OBJECT_WUNLOCK(obj);
1284 	}
1285 	*objp = obj;
1286 	*flagsp = flags;
1287 
1288 	vfs_mark_atime(vp, cred);
1289 
1290 done:
1291 	if (error != 0 && *writecounted) {
1292 		*writecounted = FALSE;
1293 		vnode_pager_update_writecount(obj, objsize, 0);
1294 	}
1295 	vput(vp);
1296 	return (error);
1297 }
1298 
1299 /*
1300  * vm_mmap_cdev()
1301  *
1302  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1303  * operations on cdevs.
1304  */
1305 int
1306 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1307     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1308     vm_ooffset_t *foff, vm_object_t *objp)
1309 {
1310 	vm_object_t obj;
1311 	int error, flags;
1312 
1313 	flags = *flagsp;
1314 
1315 	if (dsw->d_flags & D_MMAP_ANON) {
1316 		*objp = NULL;
1317 		*foff = 0;
1318 		*maxprotp = VM_PROT_ALL;
1319 		*flagsp |= MAP_ANON;
1320 		return (0);
1321 	}
1322 	/*
1323 	 * cdevs do not provide private mappings of any kind.
1324 	 */
1325 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1326 	    (prot & VM_PROT_WRITE) != 0)
1327 		return (EACCES);
1328 	if (flags & (MAP_PRIVATE|MAP_COPY))
1329 		return (EINVAL);
1330 	/*
1331 	 * Force device mappings to be shared.
1332 	 */
1333 	flags |= MAP_SHARED;
1334 #ifdef MAC_XXX
1335 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1336 	if (error != 0)
1337 		return (error);
1338 #endif
1339 	/*
1340 	 * First, try d_mmap_single().  If that is not implemented
1341 	 * (returns ENODEV), fall back to using the device pager.
1342 	 * Note that d_mmap_single() must return a reference to the
1343 	 * object (it needs to bump the reference count of the object
1344 	 * it returns somehow).
1345 	 *
1346 	 * XXX assumes VM_PROT_* == PROT_*
1347 	 */
1348 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1349 	if (error != ENODEV)
1350 		return (error);
1351 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1352 	    td->td_ucred);
1353 	if (obj == NULL)
1354 		return (EINVAL);
1355 	*objp = obj;
1356 	*flagsp = flags;
1357 	return (0);
1358 }
1359 
1360 /*
1361  * vm_mmap()
1362  *
1363  * Internal version of mmap used by exec, sys5 shared memory, and
1364  * various device drivers.  Handle is either a vnode pointer, a
1365  * character device, or NULL for MAP_ANON.
1366  */
1367 int
1368 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1369 	vm_prot_t maxprot, int flags,
1370 	objtype_t handle_type, void *handle,
1371 	vm_ooffset_t foff)
1372 {
1373 	vm_object_t object;
1374 	struct thread *td = curthread;
1375 	int error;
1376 	boolean_t writecounted;
1377 
1378 	if (size == 0)
1379 		return (EINVAL);
1380 
1381 	size = round_page(size);
1382 	object = NULL;
1383 	writecounted = FALSE;
1384 
1385 	/*
1386 	 * Lookup/allocate object.
1387 	 */
1388 	switch (handle_type) {
1389 	case OBJT_DEVICE: {
1390 		struct cdevsw *dsw;
1391 		struct cdev *cdev;
1392 		int ref;
1393 
1394 		cdev = handle;
1395 		dsw = dev_refthread(cdev, &ref);
1396 		if (dsw == NULL)
1397 			return (ENXIO);
1398 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1399 		    dsw, &foff, &object);
1400 		dev_relthread(cdev, ref);
1401 		break;
1402 	}
1403 	case OBJT_VNODE:
1404 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1405 		    handle, &foff, &object, &writecounted);
1406 		break;
1407 	case OBJT_DEFAULT:
1408 		if (handle == NULL) {
1409 			error = 0;
1410 			break;
1411 		}
1412 		/* FALLTHROUGH */
1413 	default:
1414 		error = EINVAL;
1415 		break;
1416 	}
1417 	if (error)
1418 		return (error);
1419 
1420 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1421 	    foff, writecounted, td);
1422 	if (error != 0 && object != NULL) {
1423 		/*
1424 		 * If this mapping was accounted for in the vnode's
1425 		 * writecount, then undo that now.
1426 		 */
1427 		if (writecounted)
1428 			vnode_pager_release_writecount(object, 0, size);
1429 		vm_object_deallocate(object);
1430 	}
1431 	return (error);
1432 }
1433 
1434 /*
1435  * Internal version of mmap that maps a specific VM object into an
1436  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1437  */
1438 int
1439 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1440     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1441     boolean_t writecounted, struct thread *td)
1442 {
1443 	boolean_t curmap, fitit;
1444 	vm_offset_t max_addr;
1445 	int docow, error, findspace, rv;
1446 
1447 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1448 	if (curmap) {
1449 		PROC_LOCK(td->td_proc);
1450 		if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1451 			PROC_UNLOCK(td->td_proc);
1452 			return (ENOMEM);
1453 		}
1454 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1455 			PROC_UNLOCK(td->td_proc);
1456 			return (ENOMEM);
1457 		}
1458 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1459 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1460 			    lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1461 				racct_set_force(td->td_proc, RACCT_VMEM,
1462 				    map->size);
1463 				PROC_UNLOCK(td->td_proc);
1464 				return (ENOMEM);
1465 			}
1466 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1467 			    ptoa(pmap_wired_count(map->pmap)) + size);
1468 			if (error != 0) {
1469 				racct_set_force(td->td_proc, RACCT_VMEM,
1470 				    map->size);
1471 				PROC_UNLOCK(td->td_proc);
1472 				return (error);
1473 			}
1474 		}
1475 		PROC_UNLOCK(td->td_proc);
1476 	}
1477 
1478 	/*
1479 	 * We currently can only deal with page aligned file offsets.
1480 	 * The mmap() system call already enforces this by subtracting
1481 	 * the page offset from the file offset, but checking here
1482 	 * catches errors in device drivers (e.g. d_single_mmap()
1483 	 * callbacks) and other internal mapping requests (such as in
1484 	 * exec).
1485 	 */
1486 	if (foff & PAGE_MASK)
1487 		return (EINVAL);
1488 
1489 	if ((flags & MAP_FIXED) == 0) {
1490 		fitit = TRUE;
1491 		*addr = round_page(*addr);
1492 	} else {
1493 		if (*addr != trunc_page(*addr))
1494 			return (EINVAL);
1495 		fitit = FALSE;
1496 	}
1497 
1498 	if (flags & MAP_ANON) {
1499 		if (object != NULL || foff != 0)
1500 			return (EINVAL);
1501 		docow = 0;
1502 	} else if (flags & MAP_PREFAULT_READ)
1503 		docow = MAP_PREFAULT;
1504 	else
1505 		docow = MAP_PREFAULT_PARTIAL;
1506 
1507 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1508 		docow |= MAP_COPY_ON_WRITE;
1509 	if (flags & MAP_NOSYNC)
1510 		docow |= MAP_DISABLE_SYNCER;
1511 	if (flags & MAP_NOCORE)
1512 		docow |= MAP_DISABLE_COREDUMP;
1513 	/* Shared memory is also shared with children. */
1514 	if (flags & MAP_SHARED)
1515 		docow |= MAP_INHERIT_SHARE;
1516 	if (writecounted)
1517 		docow |= MAP_VN_WRITECOUNT;
1518 	if (flags & MAP_STACK) {
1519 		if (object != NULL)
1520 			return (EINVAL);
1521 		docow |= MAP_STACK_GROWS_DOWN;
1522 	}
1523 	if ((flags & MAP_EXCL) != 0)
1524 		docow |= MAP_CHECK_EXCL;
1525 	if ((flags & MAP_GUARD) != 0)
1526 		docow |= MAP_CREATE_GUARD;
1527 
1528 	if (fitit) {
1529 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1530 			findspace = VMFS_SUPER_SPACE;
1531 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1532 			findspace = VMFS_ALIGNED_SPACE(flags >>
1533 			    MAP_ALIGNMENT_SHIFT);
1534 		else
1535 			findspace = VMFS_OPTIMAL_SPACE;
1536 		max_addr = 0;
1537 #ifdef MAP_32BIT
1538 		if ((flags & MAP_32BIT) != 0)
1539 			max_addr = MAP_32BIT_MAX_ADDR;
1540 #endif
1541 		if (curmap) {
1542 			rv = vm_map_find_min(map, object, foff, addr, size,
1543 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1544 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1545 			    findspace, prot, maxprot, docow);
1546 		} else {
1547 			rv = vm_map_find(map, object, foff, addr, size,
1548 			    max_addr, findspace, prot, maxprot, docow);
1549 		}
1550 	} else {
1551 		rv = vm_map_fixed(map, object, foff, *addr, size,
1552 		    prot, maxprot, docow);
1553 	}
1554 
1555 	if (rv == KERN_SUCCESS) {
1556 		/*
1557 		 * If the process has requested that all future mappings
1558 		 * be wired, then heed this.
1559 		 */
1560 		if (map->flags & MAP_WIREFUTURE) {
1561 			vm_map_wire(map, *addr, *addr + size,
1562 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1563 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1564 		}
1565 	}
1566 	return (vm_mmap_to_errno(rv));
1567 }
1568 
1569 /*
1570  * Translate a Mach VM return code to zero on success or the appropriate errno
1571  * on failure.
1572  */
1573 int
1574 vm_mmap_to_errno(int rv)
1575 {
1576 
1577 	switch (rv) {
1578 	case KERN_SUCCESS:
1579 		return (0);
1580 	case KERN_INVALID_ADDRESS:
1581 	case KERN_NO_SPACE:
1582 		return (ENOMEM);
1583 	case KERN_PROTECTION_FAILURE:
1584 		return (EACCES);
1585 	default:
1586 		return (EINVAL);
1587 	}
1588 }
1589