xref: /freebsd/sys/vm/vm_mmap.c (revision 88640c0e8b6f503426cce9ea1337098c241d3801)
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_hwpmc_hooks.h"
49 #include "opt_vm.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/capsicum.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/sysproto.h>
58 #include <sys/filedesc.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procctl.h>
62 #include <sys/racct.h>
63 #include <sys/resource.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 #include <sys/fcntl.h>
69 #include <sys/file.h>
70 #include <sys/mman.h>
71 #include <sys/mount.h>
72 #include <sys/conf.h>
73 #include <sys/stat.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysent.h>
76 #include <sys/vmmeter.h>
77 
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_extern.h>
90 #include <vm/vm_page.h>
91 #include <vm/vnode_pager.h>
92 
93 #ifdef HWPMC_HOOKS
94 #include <sys/pmckern.h>
95 #endif
96 
97 int old_mlock = 0;
98 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
99     "Do not apply RLIMIT_MEMLOCK on mlockall");
100 static int mincore_mapped = 1;
101 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
102     "mincore reports mappings, not residency");
103 
104 #ifdef MAP_32BIT
105 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
106 #endif
107 
108 #ifndef _SYS_SYSPROTO_H_
109 struct sbrk_args {
110 	int incr;
111 };
112 #endif
113 
114 int
115 sys_sbrk(struct thread *td, struct sbrk_args *uap)
116 {
117 	/* Not yet implemented */
118 	return (EOPNOTSUPP);
119 }
120 
121 #ifndef _SYS_SYSPROTO_H_
122 struct sstk_args {
123 	int incr;
124 };
125 #endif
126 
127 int
128 sys_sstk(struct thread *td, struct sstk_args *uap)
129 {
130 	/* Not yet implemented */
131 	return (EOPNOTSUPP);
132 }
133 
134 #if defined(COMPAT_43)
135 int
136 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
137 {
138 
139 	td->td_retval[0] = PAGE_SIZE;
140 	return (0);
141 }
142 #endif				/* COMPAT_43 */
143 
144 
145 /*
146  * Memory Map (mmap) system call.  Note that the file offset
147  * and address are allowed to be NOT page aligned, though if
148  * the MAP_FIXED flag it set, both must have the same remainder
149  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
150  * page-aligned, the actual mapping starts at trunc_page(addr)
151  * and the return value is adjusted up by the page offset.
152  *
153  * Generally speaking, only character devices which are themselves
154  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
155  * there would be no cache coherency between a descriptor and a VM mapping
156  * both to the same character device.
157  */
158 #ifndef _SYS_SYSPROTO_H_
159 struct mmap_args {
160 	void *addr;
161 	size_t len;
162 	int prot;
163 	int flags;
164 	int fd;
165 	long pad;
166 	off_t pos;
167 };
168 #endif
169 
170 int
171 sys_mmap(struct thread *td, struct mmap_args *uap)
172 {
173 
174 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
175 	    uap->flags, uap->fd, uap->pos));
176 }
177 
178 int
179 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
180     int fd, off_t pos)
181 {
182 	struct vmspace *vms;
183 	struct file *fp;
184 	vm_offset_t addr;
185 	vm_size_t pageoff;
186 	vm_prot_t cap_maxprot;
187 	int align, error;
188 	cap_rights_t rights;
189 
190 	vms = td->td_proc->p_vmspace;
191 	fp = NULL;
192 	AUDIT_ARG_FD(fd);
193 	addr = addr0;
194 
195 	/*
196 	 * Ignore old flags that used to be defined but did not do anything.
197 	 */
198 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
199 
200 	/*
201 	 * Enforce the constraints.
202 	 * Mapping of length 0 is only allowed for old binaries.
203 	 * Anonymous mapping shall specify -1 as filedescriptor and
204 	 * zero position for new code. Be nice to ancient a.out
205 	 * binaries and correct pos for anonymous mapping, since old
206 	 * ld.so sometimes issues anonymous map requests with non-zero
207 	 * pos.
208 	 */
209 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
210 		if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
211 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
212 			return (EINVAL);
213 	} else {
214 		if ((flags & MAP_ANON) != 0)
215 			pos = 0;
216 	}
217 
218 	if (flags & MAP_STACK) {
219 		if ((fd != -1) ||
220 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
221 			return (EINVAL);
222 		flags |= MAP_ANON;
223 		pos = 0;
224 	}
225 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
226 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
227 	    MAP_PREFAULT_READ | MAP_GUARD |
228 #ifdef MAP_32BIT
229 	    MAP_32BIT |
230 #endif
231 	    MAP_ALIGNMENT_MASK)) != 0)
232 		return (EINVAL);
233 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
234 		return (EINVAL);
235 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
236 		return (EINVAL);
237 	if (prot != PROT_NONE &&
238 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
239 		return (EINVAL);
240 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
241 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
242 #ifdef MAP_32BIT
243 	    MAP_32BIT |
244 #endif
245 	    MAP_ALIGNMENT_MASK)) != 0))
246 		return (EINVAL);
247 
248 	/*
249 	 * Align the file position to a page boundary,
250 	 * and save its page offset component.
251 	 */
252 	pageoff = (pos & PAGE_MASK);
253 	pos -= pageoff;
254 
255 	/* Adjust size for rounding (on both ends). */
256 	size += pageoff;			/* low end... */
257 	size = (vm_size_t) round_page(size);	/* hi end */
258 
259 	/* Ensure alignment is at least a page and fits in a pointer. */
260 	align = flags & MAP_ALIGNMENT_MASK;
261 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
262 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
263 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
264 		return (EINVAL);
265 
266 	/*
267 	 * Check for illegal addresses.  Watch out for address wrap... Note
268 	 * that VM_*_ADDRESS are not constants due to casts (argh).
269 	 */
270 	if (flags & MAP_FIXED) {
271 		/*
272 		 * The specified address must have the same remainder
273 		 * as the file offset taken modulo PAGE_SIZE, so it
274 		 * should be aligned after adjustment by pageoff.
275 		 */
276 		addr -= pageoff;
277 		if (addr & PAGE_MASK)
278 			return (EINVAL);
279 
280 		/* Address range must be all in user VM space. */
281 		if (addr < vm_map_min(&vms->vm_map) ||
282 		    addr + size > vm_map_max(&vms->vm_map))
283 			return (EINVAL);
284 		if (addr + size < addr)
285 			return (EINVAL);
286 #ifdef MAP_32BIT
287 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
288 			return (EINVAL);
289 	} else if (flags & MAP_32BIT) {
290 		/*
291 		 * For MAP_32BIT, override the hint if it is too high and
292 		 * do not bother moving the mapping past the heap (since
293 		 * the heap is usually above 2GB).
294 		 */
295 		if (addr + size > MAP_32BIT_MAX_ADDR)
296 			addr = 0;
297 #endif
298 	} else {
299 		/*
300 		 * XXX for non-fixed mappings where no hint is provided or
301 		 * the hint would fall in the potential heap space,
302 		 * place it after the end of the largest possible heap.
303 		 *
304 		 * There should really be a pmap call to determine a reasonable
305 		 * location.
306 		 */
307 		if (addr == 0 ||
308 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
309 		    addr < round_page((vm_offset_t)vms->vm_daddr +
310 		    lim_max(td, RLIMIT_DATA))))
311 			addr = round_page((vm_offset_t)vms->vm_daddr +
312 			    lim_max(td, RLIMIT_DATA));
313 	}
314 	if (size == 0) {
315 		/*
316 		 * Return success without mapping anything for old
317 		 * binaries that request a page-aligned mapping of
318 		 * length 0.  For modern binaries, this function
319 		 * returns an error earlier.
320 		 */
321 		error = 0;
322 	} else if ((flags & MAP_GUARD) != 0) {
323 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
324 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
325 	} else if ((flags & MAP_ANON) != 0) {
326 		/*
327 		 * Mapping blank space is trivial.
328 		 *
329 		 * This relies on VM_PROT_* matching PROT_*.
330 		 */
331 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
332 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
333 	} else {
334 		/*
335 		 * Mapping file, get fp for validation and don't let the
336 		 * descriptor disappear on us if we block. Check capability
337 		 * rights, but also return the maximum rights to be combined
338 		 * with maxprot later.
339 		 */
340 		cap_rights_init(&rights, CAP_MMAP);
341 		if (prot & PROT_READ)
342 			cap_rights_set(&rights, CAP_MMAP_R);
343 		if ((flags & MAP_SHARED) != 0) {
344 			if (prot & PROT_WRITE)
345 				cap_rights_set(&rights, CAP_MMAP_W);
346 		}
347 		if (prot & PROT_EXEC)
348 			cap_rights_set(&rights, CAP_MMAP_X);
349 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
350 		if (error != 0)
351 			goto done;
352 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
353 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
354 			error = EINVAL;
355 			goto done;
356 		}
357 
358 		/* This relies on VM_PROT_* matching PROT_*. */
359 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
360 		    cap_maxprot, flags, pos, td);
361 	}
362 
363 	if (error == 0)
364 		td->td_retval[0] = (register_t) (addr + pageoff);
365 done:
366 	if (fp)
367 		fdrop(fp, td);
368 
369 	return (error);
370 }
371 
372 #if defined(COMPAT_FREEBSD6)
373 int
374 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
375 {
376 
377 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
378 	    uap->flags, uap->fd, uap->pos));
379 }
380 #endif
381 
382 #ifdef COMPAT_43
383 #ifndef _SYS_SYSPROTO_H_
384 struct ommap_args {
385 	caddr_t addr;
386 	int len;
387 	int prot;
388 	int flags;
389 	int fd;
390 	long pos;
391 };
392 #endif
393 int
394 ommap(struct thread *td, struct ommap_args *uap)
395 {
396 	static const char cvtbsdprot[8] = {
397 		0,
398 		PROT_EXEC,
399 		PROT_WRITE,
400 		PROT_EXEC | PROT_WRITE,
401 		PROT_READ,
402 		PROT_EXEC | PROT_READ,
403 		PROT_WRITE | PROT_READ,
404 		PROT_EXEC | PROT_WRITE | PROT_READ,
405 	};
406 	int flags, prot;
407 
408 #define	OMAP_ANON	0x0002
409 #define	OMAP_COPY	0x0020
410 #define	OMAP_SHARED	0x0010
411 #define	OMAP_FIXED	0x0100
412 
413 	prot = cvtbsdprot[uap->prot & 0x7];
414 #ifdef COMPAT_FREEBSD32
415 #if defined(__amd64__)
416 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
417 	    prot != 0)
418 		prot |= PROT_EXEC;
419 #endif
420 #endif
421 	flags = 0;
422 	if (uap->flags & OMAP_ANON)
423 		flags |= MAP_ANON;
424 	if (uap->flags & OMAP_COPY)
425 		flags |= MAP_COPY;
426 	if (uap->flags & OMAP_SHARED)
427 		flags |= MAP_SHARED;
428 	else
429 		flags |= MAP_PRIVATE;
430 	if (uap->flags & OMAP_FIXED)
431 		flags |= MAP_FIXED;
432 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
433 	    uap->fd, uap->pos));
434 }
435 #endif				/* COMPAT_43 */
436 
437 
438 #ifndef _SYS_SYSPROTO_H_
439 struct msync_args {
440 	void *addr;
441 	size_t len;
442 	int flags;
443 };
444 #endif
445 int
446 sys_msync(struct thread *td, struct msync_args *uap)
447 {
448 
449 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
450 }
451 
452 int
453 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
454 {
455 	vm_offset_t addr;
456 	vm_size_t pageoff;
457 	vm_map_t map;
458 	int rv;
459 
460 	addr = addr0;
461 	pageoff = (addr & PAGE_MASK);
462 	addr -= pageoff;
463 	size += pageoff;
464 	size = (vm_size_t) round_page(size);
465 	if (addr + size < addr)
466 		return (EINVAL);
467 
468 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
469 		return (EINVAL);
470 
471 	map = &td->td_proc->p_vmspace->vm_map;
472 
473 	/*
474 	 * Clean the pages and interpret the return value.
475 	 */
476 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
477 	    (flags & MS_INVALIDATE) != 0);
478 	switch (rv) {
479 	case KERN_SUCCESS:
480 		return (0);
481 	case KERN_INVALID_ADDRESS:
482 		return (ENOMEM);
483 	case KERN_INVALID_ARGUMENT:
484 		return (EBUSY);
485 	case KERN_FAILURE:
486 		return (EIO);
487 	default:
488 		return (EINVAL);
489 	}
490 }
491 
492 #ifndef _SYS_SYSPROTO_H_
493 struct munmap_args {
494 	void *addr;
495 	size_t len;
496 };
497 #endif
498 int
499 sys_munmap(struct thread *td, struct munmap_args *uap)
500 {
501 
502 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
503 }
504 
505 int
506 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
507 {
508 #ifdef HWPMC_HOOKS
509 	struct pmckern_map_out pkm;
510 	vm_map_entry_t entry;
511 	bool pmc_handled;
512 #endif
513 	vm_offset_t addr;
514 	vm_size_t pageoff;
515 	vm_map_t map;
516 
517 	if (size == 0)
518 		return (EINVAL);
519 
520 	addr = addr0;
521 	pageoff = (addr & PAGE_MASK);
522 	addr -= pageoff;
523 	size += pageoff;
524 	size = (vm_size_t) round_page(size);
525 	if (addr + size < addr)
526 		return (EINVAL);
527 
528 	/*
529 	 * Check for illegal addresses.  Watch out for address wrap...
530 	 */
531 	map = &td->td_proc->p_vmspace->vm_map;
532 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
533 		return (EINVAL);
534 	vm_map_lock(map);
535 #ifdef HWPMC_HOOKS
536 	pmc_handled = false;
537 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
538 		pmc_handled = true;
539 		/*
540 		 * Inform hwpmc if the address range being unmapped contains
541 		 * an executable region.
542 		 */
543 		pkm.pm_address = (uintptr_t) NULL;
544 		if (vm_map_lookup_entry(map, addr, &entry)) {
545 			for (; entry->start < addr + size;
546 			    entry = entry->next) {
547 				if (vm_map_check_protection(map, entry->start,
548 					entry->end, VM_PROT_EXECUTE) == TRUE) {
549 					pkm.pm_address = (uintptr_t) addr;
550 					pkm.pm_size = (size_t) size;
551 					break;
552 				}
553 			}
554 		}
555 	}
556 #endif
557 	vm_map_delete(map, addr, addr + size);
558 
559 #ifdef HWPMC_HOOKS
560 	if (__predict_false(pmc_handled)) {
561 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
562 		vm_map_lock_downgrade(map);
563 		if (pkm.pm_address != (uintptr_t) NULL)
564 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
565 		vm_map_unlock_read(map);
566 	} else
567 #endif
568 		vm_map_unlock(map);
569 
570 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
571 	return (0);
572 }
573 
574 #ifndef _SYS_SYSPROTO_H_
575 struct mprotect_args {
576 	const void *addr;
577 	size_t len;
578 	int prot;
579 };
580 #endif
581 int
582 sys_mprotect(struct thread *td, struct mprotect_args *uap)
583 {
584 
585 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
586 }
587 
588 int
589 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
590 {
591 	vm_offset_t addr;
592 	vm_size_t pageoff;
593 
594 	addr = addr0;
595 	prot = (prot & VM_PROT_ALL);
596 	pageoff = (addr & PAGE_MASK);
597 	addr -= pageoff;
598 	size += pageoff;
599 	size = (vm_size_t) round_page(size);
600 #ifdef COMPAT_FREEBSD32
601 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
602 		if (((addr + size) & 0xffffffff) < addr)
603 			return (EINVAL);
604 	} else
605 #endif
606 	if (addr + size < addr)
607 		return (EINVAL);
608 
609 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
610 	    addr + size, prot, FALSE)) {
611 	case KERN_SUCCESS:
612 		return (0);
613 	case KERN_PROTECTION_FAILURE:
614 		return (EACCES);
615 	case KERN_RESOURCE_SHORTAGE:
616 		return (ENOMEM);
617 	}
618 	return (EINVAL);
619 }
620 
621 #ifndef _SYS_SYSPROTO_H_
622 struct minherit_args {
623 	void *addr;
624 	size_t len;
625 	int inherit;
626 };
627 #endif
628 int
629 sys_minherit(struct thread *td, struct minherit_args *uap)
630 {
631 	vm_offset_t addr;
632 	vm_size_t size, pageoff;
633 	vm_inherit_t inherit;
634 
635 	addr = (vm_offset_t)uap->addr;
636 	size = uap->len;
637 	inherit = uap->inherit;
638 
639 	pageoff = (addr & PAGE_MASK);
640 	addr -= pageoff;
641 	size += pageoff;
642 	size = (vm_size_t) round_page(size);
643 	if (addr + size < addr)
644 		return (EINVAL);
645 
646 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
647 	    addr + size, inherit)) {
648 	case KERN_SUCCESS:
649 		return (0);
650 	case KERN_PROTECTION_FAILURE:
651 		return (EACCES);
652 	}
653 	return (EINVAL);
654 }
655 
656 #ifndef _SYS_SYSPROTO_H_
657 struct madvise_args {
658 	void *addr;
659 	size_t len;
660 	int behav;
661 };
662 #endif
663 
664 int
665 sys_madvise(struct thread *td, struct madvise_args *uap)
666 {
667 
668 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
669 }
670 
671 int
672 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
673 {
674 	vm_map_t map;
675 	vm_offset_t addr, end, start;
676 	int flags;
677 
678 	/*
679 	 * Check for our special case, advising the swap pager we are
680 	 * "immortal."
681 	 */
682 	if (behav == MADV_PROTECT) {
683 		flags = PPROT_SET;
684 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
685 		    PROC_SPROTECT, &flags));
686 	}
687 
688 	/*
689 	 * Check for illegal addresses.  Watch out for address wrap... Note
690 	 * that VM_*_ADDRESS are not constants due to casts (argh).
691 	 */
692 	map = &td->td_proc->p_vmspace->vm_map;
693 	addr = addr0;
694 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
695 		return (EINVAL);
696 	if ((addr + len) < addr)
697 		return (EINVAL);
698 
699 	/*
700 	 * Since this routine is only advisory, we default to conservative
701 	 * behavior.
702 	 */
703 	start = trunc_page(addr);
704 	end = round_page(addr + len);
705 
706 	/*
707 	 * vm_map_madvise() checks for illegal values of behav.
708 	 */
709 	return (vm_map_madvise(map, start, end, behav));
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; current->start < end; current = current->next) {
774 
775 		/*
776 		 * check for contiguity
777 		 */
778 		if (current->end < end && current->next->start > current->end) {
779 			vm_map_unlock_read(map);
780 			return (ENOMEM);
781 		}
782 
783 		/*
784 		 * ignore submaps (for now) or null objects
785 		 */
786 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
787 			current->object.vm_object == NULL)
788 			continue;
789 
790 		/*
791 		 * limit this scan to the current map entry and the
792 		 * limits for the mincore call
793 		 */
794 		if (addr < current->start)
795 			addr = current->start;
796 		cend = current->end;
797 		if (cend > end)
798 			cend = end;
799 
800 		/*
801 		 * scan this entry one page at a time
802 		 */
803 		while (addr < cend) {
804 			/*
805 			 * Check pmap first, it is likely faster, also
806 			 * it can provide info as to whether we are the
807 			 * one referencing or modifying the page.
808 			 */
809 			object = NULL;
810 			locked_pa = 0;
811 		retry:
812 			m = NULL;
813 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
814 			if (mincore_mapped) {
815 				/*
816 				 * We only care about this pmap's
817 				 * mapping of the page, if any.
818 				 */
819 				if (locked_pa != 0) {
820 					vm_page_unlock(PHYS_TO_VM_PAGE(
821 					    locked_pa));
822 				}
823 			} else if (locked_pa != 0) {
824 				/*
825 				 * The page is mapped by this process but not
826 				 * both accessed and modified.  It is also
827 				 * managed.  Acquire the object lock so that
828 				 * other mappings might be examined.
829 				 */
830 				m = PHYS_TO_VM_PAGE(locked_pa);
831 				if (m->object != object) {
832 					if (object != NULL)
833 						VM_OBJECT_WUNLOCK(object);
834 					object = m->object;
835 					locked = VM_OBJECT_TRYWLOCK(object);
836 					vm_page_unlock(m);
837 					if (!locked) {
838 						VM_OBJECT_WLOCK(object);
839 						vm_page_lock(m);
840 						goto retry;
841 					}
842 				} else
843 					vm_page_unlock(m);
844 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
845 				    ("mincore: page %p is mapped but invalid",
846 				    m));
847 			} else if (mincoreinfo == 0) {
848 				/*
849 				 * The page is not mapped by this process.  If
850 				 * the object implements managed pages, then
851 				 * determine if the page is resident so that
852 				 * the mappings might be examined.
853 				 */
854 				if (current->object.vm_object != object) {
855 					if (object != NULL)
856 						VM_OBJECT_WUNLOCK(object);
857 					object = current->object.vm_object;
858 					VM_OBJECT_WLOCK(object);
859 				}
860 				if (object->type == OBJT_DEFAULT ||
861 				    object->type == OBJT_SWAP ||
862 				    object->type == OBJT_VNODE) {
863 					pindex = OFF_TO_IDX(current->offset +
864 					    (addr - current->start));
865 					m = vm_page_lookup(object, pindex);
866 					if (m != NULL && m->valid == 0)
867 						m = NULL;
868 					if (m != NULL)
869 						mincoreinfo = MINCORE_INCORE;
870 				}
871 			}
872 			if (m != NULL) {
873 				/* Examine other mappings to the page. */
874 				if (m->dirty == 0 && pmap_is_modified(m))
875 					vm_page_dirty(m);
876 				if (m->dirty != 0)
877 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
878 				/*
879 				 * The first test for PGA_REFERENCED is an
880 				 * optimization.  The second test is
881 				 * required because a concurrent pmap
882 				 * operation could clear the last reference
883 				 * and set PGA_REFERENCED before the call to
884 				 * pmap_is_referenced().
885 				 */
886 				if ((m->aflags & PGA_REFERENCED) != 0 ||
887 				    pmap_is_referenced(m) ||
888 				    (m->aflags & PGA_REFERENCED) != 0)
889 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
890 			}
891 			if (object != NULL)
892 				VM_OBJECT_WUNLOCK(object);
893 
894 			/*
895 			 * subyte may page fault.  In case it needs to modify
896 			 * the map, we release the lock.
897 			 */
898 			vm_map_unlock_read(map);
899 
900 			/*
901 			 * calculate index into user supplied byte vector
902 			 */
903 			vecindex = atop(addr - first_addr);
904 
905 			/*
906 			 * If we have skipped map entries, we need to make sure that
907 			 * the byte vector is zeroed for those skipped entries.
908 			 */
909 			while ((lastvecindex + 1) < vecindex) {
910 				++lastvecindex;
911 				error = subyte(vec + lastvecindex, 0);
912 				if (error) {
913 					error = EFAULT;
914 					goto done2;
915 				}
916 			}
917 
918 			/*
919 			 * Pass the page information to the user
920 			 */
921 			error = subyte(vec + vecindex, mincoreinfo);
922 			if (error) {
923 				error = EFAULT;
924 				goto done2;
925 			}
926 
927 			/*
928 			 * If the map has changed, due to the subyte, the previous
929 			 * output may be invalid.
930 			 */
931 			vm_map_lock_read(map);
932 			if (timestamp != map->timestamp)
933 				goto RestartScan;
934 
935 			lastvecindex = vecindex;
936 			addr += PAGE_SIZE;
937 		}
938 	}
939 
940 	/*
941 	 * subyte may page fault.  In case it needs to modify
942 	 * the map, we release the lock.
943 	 */
944 	vm_map_unlock_read(map);
945 
946 	/*
947 	 * Zero the last entries in the byte vector.
948 	 */
949 	vecindex = atop(end - first_addr);
950 	while ((lastvecindex + 1) < vecindex) {
951 		++lastvecindex;
952 		error = subyte(vec + lastvecindex, 0);
953 		if (error) {
954 			error = EFAULT;
955 			goto done2;
956 		}
957 	}
958 
959 	/*
960 	 * If the map has changed, due to the subyte, the previous
961 	 * output may be invalid.
962 	 */
963 	vm_map_lock_read(map);
964 	if (timestamp != map->timestamp)
965 		goto RestartScan;
966 	vm_map_unlock_read(map);
967 done2:
968 	return (error);
969 }
970 
971 #ifndef _SYS_SYSPROTO_H_
972 struct mlock_args {
973 	const void *addr;
974 	size_t len;
975 };
976 #endif
977 int
978 sys_mlock(struct thread *td, struct mlock_args *uap)
979 {
980 
981 	return (kern_mlock(td->td_proc, td->td_ucred,
982 	    __DECONST(uintptr_t, uap->addr), uap->len));
983 }
984 
985 int
986 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
987 {
988 	vm_offset_t addr, end, last, start;
989 	vm_size_t npages, size;
990 	vm_map_t map;
991 	unsigned long nsize;
992 	int error;
993 
994 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
995 	if (error)
996 		return (error);
997 	addr = addr0;
998 	size = len;
999 	last = addr + size;
1000 	start = trunc_page(addr);
1001 	end = round_page(last);
1002 	if (last < addr || end < addr)
1003 		return (EINVAL);
1004 	npages = atop(end - start);
1005 	if (npages > vm_page_max_wired)
1006 		return (ENOMEM);
1007 	map = &proc->p_vmspace->vm_map;
1008 	PROC_LOCK(proc);
1009 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1010 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1011 		PROC_UNLOCK(proc);
1012 		return (ENOMEM);
1013 	}
1014 	PROC_UNLOCK(proc);
1015 	if (npages + vm_wire_count() > vm_page_max_wired)
1016 		return (EAGAIN);
1017 #ifdef RACCT
1018 	if (racct_enable) {
1019 		PROC_LOCK(proc);
1020 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1021 		PROC_UNLOCK(proc);
1022 		if (error != 0)
1023 			return (ENOMEM);
1024 	}
1025 #endif
1026 	error = vm_map_wire(map, start, end,
1027 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1028 #ifdef RACCT
1029 	if (racct_enable && error != KERN_SUCCESS) {
1030 		PROC_LOCK(proc);
1031 		racct_set(proc, RACCT_MEMLOCK,
1032 		    ptoa(pmap_wired_count(map->pmap)));
1033 		PROC_UNLOCK(proc);
1034 	}
1035 #endif
1036 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1037 }
1038 
1039 #ifndef _SYS_SYSPROTO_H_
1040 struct mlockall_args {
1041 	int	how;
1042 };
1043 #endif
1044 
1045 int
1046 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1047 {
1048 	vm_map_t map;
1049 	int error;
1050 
1051 	map = &td->td_proc->p_vmspace->vm_map;
1052 	error = priv_check(td, PRIV_VM_MLOCK);
1053 	if (error)
1054 		return (error);
1055 
1056 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1057 		return (EINVAL);
1058 
1059 	/*
1060 	 * If wiring all pages in the process would cause it to exceed
1061 	 * a hard resource limit, return ENOMEM.
1062 	 */
1063 	if (!old_mlock && uap->how & MCL_CURRENT) {
1064 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1065 			return (ENOMEM);
1066 	}
1067 #ifdef RACCT
1068 	if (racct_enable) {
1069 		PROC_LOCK(td->td_proc);
1070 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1071 		PROC_UNLOCK(td->td_proc);
1072 		if (error != 0)
1073 			return (ENOMEM);
1074 	}
1075 #endif
1076 
1077 	if (uap->how & MCL_FUTURE) {
1078 		vm_map_lock(map);
1079 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1080 		vm_map_unlock(map);
1081 		error = 0;
1082 	}
1083 
1084 	if (uap->how & MCL_CURRENT) {
1085 		/*
1086 		 * P1003.1-2001 mandates that all currently mapped pages
1087 		 * will be memory resident and locked (wired) upon return
1088 		 * from mlockall(). vm_map_wire() will wire pages, by
1089 		 * calling vm_fault_wire() for each page in the region.
1090 		 */
1091 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1092 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1093 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1094 	}
1095 #ifdef RACCT
1096 	if (racct_enable && error != KERN_SUCCESS) {
1097 		PROC_LOCK(td->td_proc);
1098 		racct_set(td->td_proc, RACCT_MEMLOCK,
1099 		    ptoa(pmap_wired_count(map->pmap)));
1100 		PROC_UNLOCK(td->td_proc);
1101 	}
1102 #endif
1103 
1104 	return (error);
1105 }
1106 
1107 #ifndef _SYS_SYSPROTO_H_
1108 struct munlockall_args {
1109 	register_t dummy;
1110 };
1111 #endif
1112 
1113 int
1114 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1115 {
1116 	vm_map_t map;
1117 	int error;
1118 
1119 	map = &td->td_proc->p_vmspace->vm_map;
1120 	error = priv_check(td, PRIV_VM_MUNLOCK);
1121 	if (error)
1122 		return (error);
1123 
1124 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1125 	vm_map_lock(map);
1126 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1127 	vm_map_unlock(map);
1128 
1129 	/* Forcibly unwire all pages. */
1130 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1131 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1132 #ifdef RACCT
1133 	if (racct_enable && error == KERN_SUCCESS) {
1134 		PROC_LOCK(td->td_proc);
1135 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1136 		PROC_UNLOCK(td->td_proc);
1137 	}
1138 #endif
1139 
1140 	return (error);
1141 }
1142 
1143 #ifndef _SYS_SYSPROTO_H_
1144 struct munlock_args {
1145 	const void *addr;
1146 	size_t len;
1147 };
1148 #endif
1149 int
1150 sys_munlock(struct thread *td, struct munlock_args *uap)
1151 {
1152 
1153 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1154 }
1155 
1156 int
1157 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1158 {
1159 	vm_offset_t addr, end, last, start;
1160 #ifdef RACCT
1161 	vm_map_t map;
1162 #endif
1163 	int error;
1164 
1165 	error = priv_check(td, PRIV_VM_MUNLOCK);
1166 	if (error)
1167 		return (error);
1168 	addr = addr0;
1169 	last = addr + size;
1170 	start = trunc_page(addr);
1171 	end = round_page(last);
1172 	if (last < addr || end < addr)
1173 		return (EINVAL);
1174 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1175 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1176 #ifdef RACCT
1177 	if (racct_enable && error == KERN_SUCCESS) {
1178 		PROC_LOCK(td->td_proc);
1179 		map = &td->td_proc->p_vmspace->vm_map;
1180 		racct_set(td->td_proc, RACCT_MEMLOCK,
1181 		    ptoa(pmap_wired_count(map->pmap)));
1182 		PROC_UNLOCK(td->td_proc);
1183 	}
1184 #endif
1185 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1186 }
1187 
1188 /*
1189  * vm_mmap_vnode()
1190  *
1191  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1192  * operations on vnodes.
1193  */
1194 int
1195 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1196     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1197     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1198     boolean_t *writecounted)
1199 {
1200 	struct vattr va;
1201 	vm_object_t obj;
1202 	vm_ooffset_t foff;
1203 	struct ucred *cred;
1204 	int error, flags, locktype;
1205 
1206 	cred = td->td_ucred;
1207 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1208 		locktype = LK_EXCLUSIVE;
1209 	else
1210 		locktype = LK_SHARED;
1211 	if ((error = vget(vp, locktype, td)) != 0)
1212 		return (error);
1213 	AUDIT_ARG_VNODE1(vp);
1214 	foff = *foffp;
1215 	flags = *flagsp;
1216 	obj = vp->v_object;
1217 	if (vp->v_type == VREG) {
1218 		/*
1219 		 * Get the proper underlying object
1220 		 */
1221 		if (obj == NULL) {
1222 			error = EINVAL;
1223 			goto done;
1224 		}
1225 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1226 			vput(vp);
1227 			vp = (struct vnode *)obj->handle;
1228 			/*
1229 			 * Bypass filesystems obey the mpsafety of the
1230 			 * underlying fs.  Tmpfs never bypasses.
1231 			 */
1232 			error = vget(vp, locktype, td);
1233 			if (error != 0)
1234 				return (error);
1235 		}
1236 		if (locktype == LK_EXCLUSIVE) {
1237 			*writecounted = TRUE;
1238 			vnode_pager_update_writecount(obj, 0, objsize);
1239 		}
1240 	} else {
1241 		error = EINVAL;
1242 		goto done;
1243 	}
1244 	if ((error = VOP_GETATTR(vp, &va, cred)))
1245 		goto done;
1246 #ifdef MAC
1247 	/* This relies on VM_PROT_* matching PROT_*. */
1248 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1249 	if (error != 0)
1250 		goto done;
1251 #endif
1252 	if ((flags & MAP_SHARED) != 0) {
1253 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1254 			if (prot & VM_PROT_WRITE) {
1255 				error = EPERM;
1256 				goto done;
1257 			}
1258 			*maxprotp &= ~VM_PROT_WRITE;
1259 		}
1260 	}
1261 	/*
1262 	 * If it is a regular file without any references
1263 	 * we do not need to sync it.
1264 	 * Adjust object size to be the size of actual file.
1265 	 */
1266 	objsize = round_page(va.va_size);
1267 	if (va.va_nlink == 0)
1268 		flags |= MAP_NOSYNC;
1269 	if (obj->type == OBJT_VNODE) {
1270 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1271 		    cred);
1272 		if (obj == NULL) {
1273 			error = ENOMEM;
1274 			goto done;
1275 		}
1276 	} else {
1277 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1278 		    ("wrong object type"));
1279 		VM_OBJECT_WLOCK(obj);
1280 		vm_object_reference_locked(obj);
1281 #if VM_NRESERVLEVEL > 0
1282 		vm_object_color(obj, 0);
1283 #endif
1284 		VM_OBJECT_WUNLOCK(obj);
1285 	}
1286 	*objp = obj;
1287 	*flagsp = flags;
1288 
1289 	vfs_mark_atime(vp, cred);
1290 
1291 done:
1292 	if (error != 0 && *writecounted) {
1293 		*writecounted = FALSE;
1294 		vnode_pager_update_writecount(obj, objsize, 0);
1295 	}
1296 	vput(vp);
1297 	return (error);
1298 }
1299 
1300 /*
1301  * vm_mmap_cdev()
1302  *
1303  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1304  * operations on cdevs.
1305  */
1306 int
1307 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1308     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1309     vm_ooffset_t *foff, vm_object_t *objp)
1310 {
1311 	vm_object_t obj;
1312 	int error, flags;
1313 
1314 	flags = *flagsp;
1315 
1316 	if (dsw->d_flags & D_MMAP_ANON) {
1317 		*objp = NULL;
1318 		*foff = 0;
1319 		*maxprotp = VM_PROT_ALL;
1320 		*flagsp |= MAP_ANON;
1321 		return (0);
1322 	}
1323 	/*
1324 	 * cdevs do not provide private mappings of any kind.
1325 	 */
1326 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1327 	    (prot & VM_PROT_WRITE) != 0)
1328 		return (EACCES);
1329 	if (flags & (MAP_PRIVATE|MAP_COPY))
1330 		return (EINVAL);
1331 	/*
1332 	 * Force device mappings to be shared.
1333 	 */
1334 	flags |= MAP_SHARED;
1335 #ifdef MAC_XXX
1336 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1337 	if (error != 0)
1338 		return (error);
1339 #endif
1340 	/*
1341 	 * First, try d_mmap_single().  If that is not implemented
1342 	 * (returns ENODEV), fall back to using the device pager.
1343 	 * Note that d_mmap_single() must return a reference to the
1344 	 * object (it needs to bump the reference count of the object
1345 	 * it returns somehow).
1346 	 *
1347 	 * XXX assumes VM_PROT_* == PROT_*
1348 	 */
1349 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1350 	if (error != ENODEV)
1351 		return (error);
1352 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1353 	    td->td_ucred);
1354 	if (obj == NULL)
1355 		return (EINVAL);
1356 	*objp = obj;
1357 	*flagsp = flags;
1358 	return (0);
1359 }
1360 
1361 /*
1362  * vm_mmap()
1363  *
1364  * Internal version of mmap used by exec, sys5 shared memory, and
1365  * various device drivers.  Handle is either a vnode pointer, a
1366  * character device, or NULL for MAP_ANON.
1367  */
1368 int
1369 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1370 	vm_prot_t maxprot, int flags,
1371 	objtype_t handle_type, void *handle,
1372 	vm_ooffset_t foff)
1373 {
1374 	vm_object_t object;
1375 	struct thread *td = curthread;
1376 	int error;
1377 	boolean_t writecounted;
1378 
1379 	if (size == 0)
1380 		return (EINVAL);
1381 
1382 	size = round_page(size);
1383 	object = NULL;
1384 	writecounted = FALSE;
1385 
1386 	/*
1387 	 * Lookup/allocate object.
1388 	 */
1389 	switch (handle_type) {
1390 	case OBJT_DEVICE: {
1391 		struct cdevsw *dsw;
1392 		struct cdev *cdev;
1393 		int ref;
1394 
1395 		cdev = handle;
1396 		dsw = dev_refthread(cdev, &ref);
1397 		if (dsw == NULL)
1398 			return (ENXIO);
1399 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1400 		    dsw, &foff, &object);
1401 		dev_relthread(cdev, ref);
1402 		break;
1403 	}
1404 	case OBJT_VNODE:
1405 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1406 		    handle, &foff, &object, &writecounted);
1407 		break;
1408 	case OBJT_DEFAULT:
1409 		if (handle == NULL) {
1410 			error = 0;
1411 			break;
1412 		}
1413 		/* FALLTHROUGH */
1414 	default:
1415 		error = EINVAL;
1416 		break;
1417 	}
1418 	if (error)
1419 		return (error);
1420 
1421 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1422 	    foff, writecounted, td);
1423 	if (error != 0 && object != NULL) {
1424 		/*
1425 		 * If this mapping was accounted for in the vnode's
1426 		 * writecount, then undo that now.
1427 		 */
1428 		if (writecounted)
1429 			vnode_pager_release_writecount(object, 0, size);
1430 		vm_object_deallocate(object);
1431 	}
1432 	return (error);
1433 }
1434 
1435 /*
1436  * Internal version of mmap that maps a specific VM object into an
1437  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1438  */
1439 int
1440 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1441     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1442     boolean_t writecounted, struct thread *td)
1443 {
1444 	boolean_t curmap, fitit;
1445 	vm_offset_t max_addr;
1446 	int docow, error, findspace, rv;
1447 
1448 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1449 	if (curmap) {
1450 		RACCT_PROC_LOCK(td->td_proc);
1451 		if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1452 			RACCT_PROC_UNLOCK(td->td_proc);
1453 			return (ENOMEM);
1454 		}
1455 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1456 			RACCT_PROC_UNLOCK(td->td_proc);
1457 			return (ENOMEM);
1458 		}
1459 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1460 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1461 			    lim_cur(td, RLIMIT_MEMLOCK)) {
1462 				racct_set_force(td->td_proc, RACCT_VMEM,
1463 				    map->size);
1464 				RACCT_PROC_UNLOCK(td->td_proc);
1465 				return (ENOMEM);
1466 			}
1467 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1468 			    ptoa(pmap_wired_count(map->pmap)) + size);
1469 			if (error != 0) {
1470 				racct_set_force(td->td_proc, RACCT_VMEM,
1471 				    map->size);
1472 				RACCT_PROC_UNLOCK(td->td_proc);
1473 				return (error);
1474 			}
1475 		}
1476 		RACCT_PROC_UNLOCK(td->td_proc);
1477 	}
1478 
1479 	/*
1480 	 * We currently can only deal with page aligned file offsets.
1481 	 * The mmap() system call already enforces this by subtracting
1482 	 * the page offset from the file offset, but checking here
1483 	 * catches errors in device drivers (e.g. d_single_mmap()
1484 	 * callbacks) and other internal mapping requests (such as in
1485 	 * exec).
1486 	 */
1487 	if (foff & PAGE_MASK)
1488 		return (EINVAL);
1489 
1490 	if ((flags & MAP_FIXED) == 0) {
1491 		fitit = TRUE;
1492 		*addr = round_page(*addr);
1493 	} else {
1494 		if (*addr != trunc_page(*addr))
1495 			return (EINVAL);
1496 		fitit = FALSE;
1497 	}
1498 
1499 	if (flags & MAP_ANON) {
1500 		if (object != NULL || foff != 0)
1501 			return (EINVAL);
1502 		docow = 0;
1503 	} else if (flags & MAP_PREFAULT_READ)
1504 		docow = MAP_PREFAULT;
1505 	else
1506 		docow = MAP_PREFAULT_PARTIAL;
1507 
1508 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1509 		docow |= MAP_COPY_ON_WRITE;
1510 	if (flags & MAP_NOSYNC)
1511 		docow |= MAP_DISABLE_SYNCER;
1512 	if (flags & MAP_NOCORE)
1513 		docow |= MAP_DISABLE_COREDUMP;
1514 	/* Shared memory is also shared with children. */
1515 	if (flags & MAP_SHARED)
1516 		docow |= MAP_INHERIT_SHARE;
1517 	if (writecounted)
1518 		docow |= MAP_VN_WRITECOUNT;
1519 	if (flags & MAP_STACK) {
1520 		if (object != NULL)
1521 			return (EINVAL);
1522 		docow |= MAP_STACK_GROWS_DOWN;
1523 	}
1524 	if ((flags & MAP_EXCL) != 0)
1525 		docow |= MAP_CHECK_EXCL;
1526 	if ((flags & MAP_GUARD) != 0)
1527 		docow |= MAP_CREATE_GUARD;
1528 
1529 	if (fitit) {
1530 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1531 			findspace = VMFS_SUPER_SPACE;
1532 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1533 			findspace = VMFS_ALIGNED_SPACE(flags >>
1534 			    MAP_ALIGNMENT_SHIFT);
1535 		else
1536 			findspace = VMFS_OPTIMAL_SPACE;
1537 		max_addr = 0;
1538 #ifdef MAP_32BIT
1539 		if ((flags & MAP_32BIT) != 0)
1540 			max_addr = MAP_32BIT_MAX_ADDR;
1541 #endif
1542 		if (curmap) {
1543 			rv = vm_map_find_min(map, object, foff, addr, size,
1544 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1545 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1546 			    findspace, prot, maxprot, docow);
1547 		} else {
1548 			rv = vm_map_find(map, object, foff, addr, size,
1549 			    max_addr, findspace, prot, maxprot, docow);
1550 		}
1551 	} else {
1552 		rv = vm_map_fixed(map, object, foff, *addr, size,
1553 		    prot, maxprot, docow);
1554 	}
1555 
1556 	if (rv == KERN_SUCCESS) {
1557 		/*
1558 		 * If the process has requested that all future mappings
1559 		 * be wired, then heed this.
1560 		 */
1561 		if (map->flags & MAP_WIREFUTURE) {
1562 			vm_map_wire(map, *addr, *addr + size,
1563 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1564 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1565 		}
1566 	}
1567 	return (vm_mmap_to_errno(rv));
1568 }
1569 
1570 /*
1571  * Translate a Mach VM return code to zero on success or the appropriate errno
1572  * on failure.
1573  */
1574 int
1575 vm_mmap_to_errno(int rv)
1576 {
1577 
1578 	switch (rv) {
1579 	case KERN_SUCCESS:
1580 		return (0);
1581 	case KERN_INVALID_ADDRESS:
1582 	case KERN_NO_SPACE:
1583 		return (ENOMEM);
1584 	case KERN_PROTECTION_FAILURE:
1585 		return (EACCES);
1586 	default:
1587 		return (EINVAL);
1588 	}
1589 }
1590