xref: /freebsd/sys/vm/vm_mmap.c (revision 87b2cfcecfa519b11313951f10e6a4076d9f595a)
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 #include "opt_vm.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/capsicum.h>
53 #include <sys/kernel.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/sysproto.h>
57 #include <sys/filedesc.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procctl.h>
61 #include <sys/racct.h>
62 #include <sys/resource.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
68 #include <sys/file.h>
69 #include <sys/mman.h>
70 #include <sys/mount.h>
71 #include <sys/conf.h>
72 #include <sys/stat.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysent.h>
75 #include <sys/vmmeter.h>
76 
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_pageout.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_page.h>
90 #include <vm/vnode_pager.h>
91 
92 #ifdef HWPMC_HOOKS
93 #include <sys/pmckern.h>
94 #endif
95 
96 int old_mlock = 0;
97 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
98     "Do not apply RLIMIT_MEMLOCK on mlockall");
99 
100 #ifdef MAP_32BIT
101 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
102 #endif
103 
104 #ifndef _SYS_SYSPROTO_H_
105 struct sbrk_args {
106 	int incr;
107 };
108 #endif
109 
110 /*
111  * MPSAFE
112  */
113 /* ARGSUSED */
114 int
115 sys_sbrk(td, uap)
116 	struct thread *td;
117 	struct sbrk_args *uap;
118 {
119 	/* Not yet implemented */
120 	return (EOPNOTSUPP);
121 }
122 
123 #ifndef _SYS_SYSPROTO_H_
124 struct sstk_args {
125 	int incr;
126 };
127 #endif
128 
129 /*
130  * MPSAFE
131  */
132 /* ARGSUSED */
133 int
134 sys_sstk(td, uap)
135 	struct thread *td;
136 	struct sstk_args *uap;
137 {
138 	/* Not yet implemented */
139 	return (EOPNOTSUPP);
140 }
141 
142 #if defined(COMPAT_43)
143 #ifndef _SYS_SYSPROTO_H_
144 struct getpagesize_args {
145 	int dummy;
146 };
147 #endif
148 
149 int
150 ogetpagesize(td, uap)
151 	struct thread *td;
152 	struct getpagesize_args *uap;
153 {
154 	/* MP SAFE */
155 	td->td_retval[0] = PAGE_SIZE;
156 	return (0);
157 }
158 #endif				/* COMPAT_43 */
159 
160 
161 /*
162  * Memory Map (mmap) system call.  Note that the file offset
163  * and address are allowed to be NOT page aligned, though if
164  * the MAP_FIXED flag it set, both must have the same remainder
165  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
166  * page-aligned, the actual mapping starts at trunc_page(addr)
167  * and the return value is adjusted up by the page offset.
168  *
169  * Generally speaking, only character devices which are themselves
170  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
171  * there would be no cache coherency between a descriptor and a VM mapping
172  * both to the same character device.
173  */
174 #ifndef _SYS_SYSPROTO_H_
175 struct mmap_args {
176 	void *addr;
177 	size_t len;
178 	int prot;
179 	int flags;
180 	int fd;
181 	long pad;
182 	off_t pos;
183 };
184 #endif
185 
186 /*
187  * MPSAFE
188  */
189 int
190 sys_mmap(td, uap)
191 	struct thread *td;
192 	struct mmap_args *uap;
193 {
194 	struct file *fp;
195 	vm_offset_t addr;
196 	vm_size_t size, pageoff;
197 	vm_prot_t cap_maxprot;
198 	int align, error, flags, prot;
199 	off_t pos;
200 	struct vmspace *vms = td->td_proc->p_vmspace;
201 	cap_rights_t rights;
202 
203 	addr = (vm_offset_t) uap->addr;
204 	size = uap->len;
205 	prot = uap->prot;
206 	flags = uap->flags;
207 	pos = uap->pos;
208 
209 	fp = NULL;
210 	AUDIT_ARG_FD(uap->fd);
211 
212 	/*
213 	 * Ignore old flags that used to be defined but did not do anything.
214 	 */
215 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
216 
217 	/*
218 	 * Enforce the constraints.
219 	 * Mapping of length 0 is only allowed for old binaries.
220 	 * Anonymous mapping shall specify -1 as filedescriptor and
221 	 * zero position for new code. Be nice to ancient a.out
222 	 * binaries and correct pos for anonymous mapping, since old
223 	 * ld.so sometimes issues anonymous map requests with non-zero
224 	 * pos.
225 	 */
226 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
227 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
228 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
229 			return (EINVAL);
230 	} else {
231 		if ((flags & MAP_ANON) != 0)
232 			pos = 0;
233 	}
234 
235 	if (flags & MAP_STACK) {
236 		if ((uap->fd != -1) ||
237 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
238 			return (EINVAL);
239 		flags |= MAP_ANON;
240 		pos = 0;
241 	}
242 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
243 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
244 	    MAP_PREFAULT_READ |
245 #ifdef MAP_32BIT
246 	    MAP_32BIT |
247 #endif
248 	    MAP_ALIGNMENT_MASK)) != 0)
249 		return (EINVAL);
250 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
251 		return (EINVAL);
252 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
253 		return (EINVAL);
254 	if (prot != PROT_NONE &&
255 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
256 		return (EINVAL);
257 
258 	/*
259 	 * Align the file position to a page boundary,
260 	 * and save its page offset component.
261 	 */
262 	pageoff = (pos & PAGE_MASK);
263 	pos -= pageoff;
264 
265 	/* Adjust size for rounding (on both ends). */
266 	size += pageoff;			/* low end... */
267 	size = (vm_size_t) round_page(size);	/* hi end */
268 
269 	/* Ensure alignment is at least a page and fits in a pointer. */
270 	align = flags & MAP_ALIGNMENT_MASK;
271 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
272 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
273 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
274 		return (EINVAL);
275 
276 	/*
277 	 * Check for illegal addresses.  Watch out for address wrap... Note
278 	 * that VM_*_ADDRESS are not constants due to casts (argh).
279 	 */
280 	if (flags & MAP_FIXED) {
281 		/*
282 		 * The specified address must have the same remainder
283 		 * as the file offset taken modulo PAGE_SIZE, so it
284 		 * should be aligned after adjustment by pageoff.
285 		 */
286 		addr -= pageoff;
287 		if (addr & PAGE_MASK)
288 			return (EINVAL);
289 
290 		/* Address range must be all in user VM space. */
291 		if (addr < vm_map_min(&vms->vm_map) ||
292 		    addr + size > vm_map_max(&vms->vm_map))
293 			return (EINVAL);
294 		if (addr + size < addr)
295 			return (EINVAL);
296 #ifdef MAP_32BIT
297 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
298 			return (EINVAL);
299 	} else if (flags & MAP_32BIT) {
300 		/*
301 		 * For MAP_32BIT, override the hint if it is too high and
302 		 * do not bother moving the mapping past the heap (since
303 		 * the heap is usually above 2GB).
304 		 */
305 		if (addr + size > MAP_32BIT_MAX_ADDR)
306 			addr = 0;
307 #endif
308 	} else {
309 		/*
310 		 * XXX for non-fixed mappings where no hint is provided or
311 		 * the hint would fall in the potential heap space,
312 		 * place it after the end of the largest possible heap.
313 		 *
314 		 * There should really be a pmap call to determine a reasonable
315 		 * location.
316 		 */
317 		if (addr == 0 ||
318 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
319 		    addr < round_page((vm_offset_t)vms->vm_daddr +
320 		    lim_max(td, RLIMIT_DATA))))
321 			addr = round_page((vm_offset_t)vms->vm_daddr +
322 			    lim_max(td, RLIMIT_DATA));
323 	}
324 	if (size == 0) {
325 		/*
326 		 * Return success without mapping anything for old
327 		 * binaries that request a page-aligned mapping of
328 		 * length 0.  For modern binaries, this function
329 		 * returns an error earlier.
330 		 */
331 		error = 0;
332 	} else if (flags & MAP_ANON) {
333 		/*
334 		 * Mapping blank space is trivial.
335 		 *
336 		 * This relies on VM_PROT_* matching PROT_*.
337 		 */
338 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
339 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
340 	} else {
341 		/*
342 		 * Mapping file, get fp for validation and don't let the
343 		 * descriptor disappear on us if we block. Check capability
344 		 * rights, but also return the maximum rights to be combined
345 		 * with maxprot later.
346 		 */
347 		cap_rights_init(&rights, CAP_MMAP);
348 		if (prot & PROT_READ)
349 			cap_rights_set(&rights, CAP_MMAP_R);
350 		if ((flags & MAP_SHARED) != 0) {
351 			if (prot & PROT_WRITE)
352 				cap_rights_set(&rights, CAP_MMAP_W);
353 		}
354 		if (prot & PROT_EXEC)
355 			cap_rights_set(&rights, CAP_MMAP_X);
356 		error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
357 		if (error != 0)
358 			goto done;
359 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
360 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
361 			error = EINVAL;
362 			goto done;
363 		}
364 
365 		/* This relies on VM_PROT_* matching PROT_*. */
366 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
367 		    cap_maxprot, flags, pos, td);
368 	}
369 
370 	if (error == 0)
371 		td->td_retval[0] = (register_t) (addr + pageoff);
372 done:
373 	if (fp)
374 		fdrop(fp, td);
375 
376 	return (error);
377 }
378 
379 #if defined(COMPAT_FREEBSD6)
380 int
381 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
382 {
383 	struct mmap_args oargs;
384 
385 	oargs.addr = uap->addr;
386 	oargs.len = uap->len;
387 	oargs.prot = uap->prot;
388 	oargs.flags = uap->flags;
389 	oargs.fd = uap->fd;
390 	oargs.pos = uap->pos;
391 	return (sys_mmap(td, &oargs));
392 }
393 #endif
394 
395 #ifdef COMPAT_43
396 #ifndef _SYS_SYSPROTO_H_
397 struct ommap_args {
398 	caddr_t addr;
399 	int len;
400 	int prot;
401 	int flags;
402 	int fd;
403 	long pos;
404 };
405 #endif
406 int
407 ommap(td, uap)
408 	struct thread *td;
409 	struct ommap_args *uap;
410 {
411 	struct mmap_args nargs;
412 	static const char cvtbsdprot[8] = {
413 		0,
414 		PROT_EXEC,
415 		PROT_WRITE,
416 		PROT_EXEC | PROT_WRITE,
417 		PROT_READ,
418 		PROT_EXEC | PROT_READ,
419 		PROT_WRITE | PROT_READ,
420 		PROT_EXEC | PROT_WRITE | PROT_READ,
421 	};
422 
423 #define	OMAP_ANON	0x0002
424 #define	OMAP_COPY	0x0020
425 #define	OMAP_SHARED	0x0010
426 #define	OMAP_FIXED	0x0100
427 
428 	nargs.addr = uap->addr;
429 	nargs.len = uap->len;
430 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
431 #ifdef COMPAT_FREEBSD32
432 #if defined(__amd64__)
433 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
434 	    nargs.prot != 0)
435 		nargs.prot |= PROT_EXEC;
436 #endif
437 #endif
438 	nargs.flags = 0;
439 	if (uap->flags & OMAP_ANON)
440 		nargs.flags |= MAP_ANON;
441 	if (uap->flags & OMAP_COPY)
442 		nargs.flags |= MAP_COPY;
443 	if (uap->flags & OMAP_SHARED)
444 		nargs.flags |= MAP_SHARED;
445 	else
446 		nargs.flags |= MAP_PRIVATE;
447 	if (uap->flags & OMAP_FIXED)
448 		nargs.flags |= MAP_FIXED;
449 	nargs.fd = uap->fd;
450 	nargs.pos = uap->pos;
451 	return (sys_mmap(td, &nargs));
452 }
453 #endif				/* COMPAT_43 */
454 
455 
456 #ifndef _SYS_SYSPROTO_H_
457 struct msync_args {
458 	void *addr;
459 	size_t len;
460 	int flags;
461 };
462 #endif
463 /*
464  * MPSAFE
465  */
466 int
467 sys_msync(td, uap)
468 	struct thread *td;
469 	struct msync_args *uap;
470 {
471 	vm_offset_t addr;
472 	vm_size_t size, pageoff;
473 	int flags;
474 	vm_map_t map;
475 	int rv;
476 
477 	addr = (vm_offset_t) uap->addr;
478 	size = uap->len;
479 	flags = uap->flags;
480 
481 	pageoff = (addr & PAGE_MASK);
482 	addr -= pageoff;
483 	size += pageoff;
484 	size = (vm_size_t) round_page(size);
485 	if (addr + size < addr)
486 		return (EINVAL);
487 
488 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
489 		return (EINVAL);
490 
491 	map = &td->td_proc->p_vmspace->vm_map;
492 
493 	/*
494 	 * Clean the pages and interpret the return value.
495 	 */
496 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
497 	    (flags & MS_INVALIDATE) != 0);
498 	switch (rv) {
499 	case KERN_SUCCESS:
500 		return (0);
501 	case KERN_INVALID_ADDRESS:
502 		return (ENOMEM);
503 	case KERN_INVALID_ARGUMENT:
504 		return (EBUSY);
505 	case KERN_FAILURE:
506 		return (EIO);
507 	default:
508 		return (EINVAL);
509 	}
510 }
511 
512 #ifndef _SYS_SYSPROTO_H_
513 struct munmap_args {
514 	void *addr;
515 	size_t len;
516 };
517 #endif
518 /*
519  * MPSAFE
520  */
521 int
522 sys_munmap(td, uap)
523 	struct thread *td;
524 	struct munmap_args *uap;
525 {
526 #ifdef HWPMC_HOOKS
527 	struct pmckern_map_out pkm;
528 	vm_map_entry_t entry;
529 	bool pmc_handled;
530 #endif
531 	vm_offset_t addr;
532 	vm_size_t size, pageoff;
533 	vm_map_t map;
534 
535 	addr = (vm_offset_t) uap->addr;
536 	size = uap->len;
537 	if (size == 0)
538 		return (EINVAL);
539 
540 	pageoff = (addr & PAGE_MASK);
541 	addr -= pageoff;
542 	size += pageoff;
543 	size = (vm_size_t) round_page(size);
544 	if (addr + size < addr)
545 		return (EINVAL);
546 
547 	/*
548 	 * Check for illegal addresses.  Watch out for address wrap...
549 	 */
550 	map = &td->td_proc->p_vmspace->vm_map;
551 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
552 		return (EINVAL);
553 	vm_map_lock(map);
554 #ifdef HWPMC_HOOKS
555 	pmc_handled = false;
556 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
557 		pmc_handled = true;
558 		/*
559 		 * Inform hwpmc if the address range being unmapped contains
560 		 * an executable region.
561 		 */
562 		pkm.pm_address = (uintptr_t) NULL;
563 		if (vm_map_lookup_entry(map, addr, &entry)) {
564 			for (;
565 			    entry != &map->header && entry->start < addr + size;
566 			    entry = entry->next) {
567 				if (vm_map_check_protection(map, entry->start,
568 					entry->end, VM_PROT_EXECUTE) == TRUE) {
569 					pkm.pm_address = (uintptr_t) addr;
570 					pkm.pm_size = (size_t) size;
571 					break;
572 				}
573 			}
574 		}
575 	}
576 #endif
577 	vm_map_delete(map, addr, addr + size);
578 
579 #ifdef HWPMC_HOOKS
580 	if (__predict_false(pmc_handled)) {
581 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
582 		vm_map_lock_downgrade(map);
583 		if (pkm.pm_address != (uintptr_t) NULL)
584 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
585 		vm_map_unlock_read(map);
586 	} else
587 #endif
588 		vm_map_unlock(map);
589 
590 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
591 	return (0);
592 }
593 
594 #ifndef _SYS_SYSPROTO_H_
595 struct mprotect_args {
596 	const void *addr;
597 	size_t len;
598 	int prot;
599 };
600 #endif
601 /*
602  * MPSAFE
603  */
604 int
605 sys_mprotect(td, uap)
606 	struct thread *td;
607 	struct mprotect_args *uap;
608 {
609 	vm_offset_t addr;
610 	vm_size_t size, pageoff;
611 	vm_prot_t prot;
612 
613 	addr = (vm_offset_t) uap->addr;
614 	size = uap->len;
615 	prot = uap->prot & VM_PROT_ALL;
616 
617 	pageoff = (addr & PAGE_MASK);
618 	addr -= pageoff;
619 	size += pageoff;
620 	size = (vm_size_t) round_page(size);
621 	if (addr + size < addr)
622 		return (EINVAL);
623 
624 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
625 	    addr + size, prot, FALSE)) {
626 	case KERN_SUCCESS:
627 		return (0);
628 	case KERN_PROTECTION_FAILURE:
629 		return (EACCES);
630 	case KERN_RESOURCE_SHORTAGE:
631 		return (ENOMEM);
632 	}
633 	return (EINVAL);
634 }
635 
636 #ifndef _SYS_SYSPROTO_H_
637 struct minherit_args {
638 	void *addr;
639 	size_t len;
640 	int inherit;
641 };
642 #endif
643 /*
644  * MPSAFE
645  */
646 int
647 sys_minherit(td, uap)
648 	struct thread *td;
649 	struct minherit_args *uap;
650 {
651 	vm_offset_t addr;
652 	vm_size_t size, pageoff;
653 	vm_inherit_t inherit;
654 
655 	addr = (vm_offset_t)uap->addr;
656 	size = uap->len;
657 	inherit = uap->inherit;
658 
659 	pageoff = (addr & PAGE_MASK);
660 	addr -= pageoff;
661 	size += pageoff;
662 	size = (vm_size_t) round_page(size);
663 	if (addr + size < addr)
664 		return (EINVAL);
665 
666 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
667 	    addr + size, inherit)) {
668 	case KERN_SUCCESS:
669 		return (0);
670 	case KERN_PROTECTION_FAILURE:
671 		return (EACCES);
672 	}
673 	return (EINVAL);
674 }
675 
676 #ifndef _SYS_SYSPROTO_H_
677 struct madvise_args {
678 	void *addr;
679 	size_t len;
680 	int behav;
681 };
682 #endif
683 
684 /*
685  * MPSAFE
686  */
687 int
688 sys_madvise(td, uap)
689 	struct thread *td;
690 	struct madvise_args *uap;
691 {
692 	vm_offset_t start, end;
693 	vm_map_t map;
694 	int flags;
695 
696 	/*
697 	 * Check for our special case, advising the swap pager we are
698 	 * "immortal."
699 	 */
700 	if (uap->behav == MADV_PROTECT) {
701 		flags = PPROT_SET;
702 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
703 		    PROC_SPROTECT, &flags));
704 	}
705 
706 	/*
707 	 * Check for illegal behavior
708 	 */
709 	if (uap->behav < 0 || uap->behav > MADV_CORE)
710 		return (EINVAL);
711 	/*
712 	 * Check for illegal addresses.  Watch out for address wrap... Note
713 	 * that VM_*_ADDRESS are not constants due to casts (argh).
714 	 */
715 	map = &td->td_proc->p_vmspace->vm_map;
716 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
717 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
718 		return (EINVAL);
719 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
720 		return (EINVAL);
721 
722 	/*
723 	 * Since this routine is only advisory, we default to conservative
724 	 * behavior.
725 	 */
726 	start = trunc_page((vm_offset_t) uap->addr);
727 	end = round_page((vm_offset_t) uap->addr + uap->len);
728 
729 	if (vm_map_madvise(map, start, end, uap->behav))
730 		return (EINVAL);
731 	return (0);
732 }
733 
734 #ifndef _SYS_SYSPROTO_H_
735 struct mincore_args {
736 	const void *addr;
737 	size_t len;
738 	char *vec;
739 };
740 #endif
741 
742 /*
743  * MPSAFE
744  */
745 int
746 sys_mincore(td, uap)
747 	struct thread *td;
748 	struct mincore_args *uap;
749 {
750 	vm_offset_t addr, first_addr;
751 	vm_offset_t end, cend;
752 	pmap_t pmap;
753 	vm_map_t map;
754 	char *vec;
755 	int error = 0;
756 	int vecindex, lastvecindex;
757 	vm_map_entry_t current;
758 	vm_map_entry_t entry;
759 	vm_object_t object;
760 	vm_paddr_t locked_pa;
761 	vm_page_t m;
762 	vm_pindex_t pindex;
763 	int mincoreinfo;
764 	unsigned int timestamp;
765 	boolean_t locked;
766 
767 	/*
768 	 * Make sure that the addresses presented are valid for user
769 	 * mode.
770 	 */
771 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
772 	end = addr + (vm_size_t)round_page(uap->len);
773 	map = &td->td_proc->p_vmspace->vm_map;
774 	if (end > vm_map_max(map) || end < addr)
775 		return (ENOMEM);
776 
777 	/*
778 	 * Address of byte vector
779 	 */
780 	vec = uap->vec;
781 
782 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
783 
784 	vm_map_lock_read(map);
785 RestartScan:
786 	timestamp = map->timestamp;
787 
788 	if (!vm_map_lookup_entry(map, addr, &entry)) {
789 		vm_map_unlock_read(map);
790 		return (ENOMEM);
791 	}
792 
793 	/*
794 	 * Do this on a map entry basis so that if the pages are not
795 	 * in the current processes address space, we can easily look
796 	 * up the pages elsewhere.
797 	 */
798 	lastvecindex = -1;
799 	for (current = entry;
800 	    (current != &map->header) && (current->start < end);
801 	    current = current->next) {
802 
803 		/*
804 		 * check for contiguity
805 		 */
806 		if (current->end < end &&
807 		    (entry->next == &map->header ||
808 		     current->next->start > current->end)) {
809 			vm_map_unlock_read(map);
810 			return (ENOMEM);
811 		}
812 
813 		/*
814 		 * ignore submaps (for now) or null objects
815 		 */
816 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
817 			current->object.vm_object == NULL)
818 			continue;
819 
820 		/*
821 		 * limit this scan to the current map entry and the
822 		 * limits for the mincore call
823 		 */
824 		if (addr < current->start)
825 			addr = current->start;
826 		cend = current->end;
827 		if (cend > end)
828 			cend = end;
829 
830 		/*
831 		 * scan this entry one page at a time
832 		 */
833 		while (addr < cend) {
834 			/*
835 			 * Check pmap first, it is likely faster, also
836 			 * it can provide info as to whether we are the
837 			 * one referencing or modifying the page.
838 			 */
839 			object = NULL;
840 			locked_pa = 0;
841 		retry:
842 			m = NULL;
843 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
844 			if (locked_pa != 0) {
845 				/*
846 				 * The page is mapped by this process but not
847 				 * both accessed and modified.  It is also
848 				 * managed.  Acquire the object lock so that
849 				 * other mappings might be examined.
850 				 */
851 				m = PHYS_TO_VM_PAGE(locked_pa);
852 				if (m->object != object) {
853 					if (object != NULL)
854 						VM_OBJECT_WUNLOCK(object);
855 					object = m->object;
856 					locked = VM_OBJECT_TRYWLOCK(object);
857 					vm_page_unlock(m);
858 					if (!locked) {
859 						VM_OBJECT_WLOCK(object);
860 						vm_page_lock(m);
861 						goto retry;
862 					}
863 				} else
864 					vm_page_unlock(m);
865 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
866 				    ("mincore: page %p is mapped but invalid",
867 				    m));
868 			} else if (mincoreinfo == 0) {
869 				/*
870 				 * The page is not mapped by this process.  If
871 				 * the object implements managed pages, then
872 				 * determine if the page is resident so that
873 				 * the mappings might be examined.
874 				 */
875 				if (current->object.vm_object != object) {
876 					if (object != NULL)
877 						VM_OBJECT_WUNLOCK(object);
878 					object = current->object.vm_object;
879 					VM_OBJECT_WLOCK(object);
880 				}
881 				if (object->type == OBJT_DEFAULT ||
882 				    object->type == OBJT_SWAP ||
883 				    object->type == OBJT_VNODE) {
884 					pindex = OFF_TO_IDX(current->offset +
885 					    (addr - current->start));
886 					m = vm_page_lookup(object, pindex);
887 					if (m != NULL && m->valid == 0)
888 						m = NULL;
889 					if (m != NULL)
890 						mincoreinfo = MINCORE_INCORE;
891 				}
892 			}
893 			if (m != NULL) {
894 				/* Examine other mappings to the page. */
895 				if (m->dirty == 0 && pmap_is_modified(m))
896 					vm_page_dirty(m);
897 				if (m->dirty != 0)
898 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
899 				/*
900 				 * The first test for PGA_REFERENCED is an
901 				 * optimization.  The second test is
902 				 * required because a concurrent pmap
903 				 * operation could clear the last reference
904 				 * and set PGA_REFERENCED before the call to
905 				 * pmap_is_referenced().
906 				 */
907 				if ((m->aflags & PGA_REFERENCED) != 0 ||
908 				    pmap_is_referenced(m) ||
909 				    (m->aflags & PGA_REFERENCED) != 0)
910 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
911 			}
912 			if (object != NULL)
913 				VM_OBJECT_WUNLOCK(object);
914 
915 			/*
916 			 * subyte may page fault.  In case it needs to modify
917 			 * the map, we release the lock.
918 			 */
919 			vm_map_unlock_read(map);
920 
921 			/*
922 			 * calculate index into user supplied byte vector
923 			 */
924 			vecindex = OFF_TO_IDX(addr - first_addr);
925 
926 			/*
927 			 * If we have skipped map entries, we need to make sure that
928 			 * the byte vector is zeroed for those skipped entries.
929 			 */
930 			while ((lastvecindex + 1) < vecindex) {
931 				++lastvecindex;
932 				error = subyte(vec + lastvecindex, 0);
933 				if (error) {
934 					error = EFAULT;
935 					goto done2;
936 				}
937 			}
938 
939 			/*
940 			 * Pass the page information to the user
941 			 */
942 			error = subyte(vec + vecindex, mincoreinfo);
943 			if (error) {
944 				error = EFAULT;
945 				goto done2;
946 			}
947 
948 			/*
949 			 * If the map has changed, due to the subyte, the previous
950 			 * output may be invalid.
951 			 */
952 			vm_map_lock_read(map);
953 			if (timestamp != map->timestamp)
954 				goto RestartScan;
955 
956 			lastvecindex = vecindex;
957 			addr += PAGE_SIZE;
958 		}
959 	}
960 
961 	/*
962 	 * subyte may page fault.  In case it needs to modify
963 	 * the map, we release the lock.
964 	 */
965 	vm_map_unlock_read(map);
966 
967 	/*
968 	 * Zero the last entries in the byte vector.
969 	 */
970 	vecindex = OFF_TO_IDX(end - first_addr);
971 	while ((lastvecindex + 1) < vecindex) {
972 		++lastvecindex;
973 		error = subyte(vec + lastvecindex, 0);
974 		if (error) {
975 			error = EFAULT;
976 			goto done2;
977 		}
978 	}
979 
980 	/*
981 	 * If the map has changed, due to the subyte, the previous
982 	 * output may be invalid.
983 	 */
984 	vm_map_lock_read(map);
985 	if (timestamp != map->timestamp)
986 		goto RestartScan;
987 	vm_map_unlock_read(map);
988 done2:
989 	return (error);
990 }
991 
992 #ifndef _SYS_SYSPROTO_H_
993 struct mlock_args {
994 	const void *addr;
995 	size_t len;
996 };
997 #endif
998 /*
999  * MPSAFE
1000  */
1001 int
1002 sys_mlock(td, uap)
1003 	struct thread *td;
1004 	struct mlock_args *uap;
1005 {
1006 
1007 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1008 }
1009 
1010 int
1011 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1012 {
1013 	vm_offset_t addr, end, last, start;
1014 	vm_size_t npages, size;
1015 	vm_map_t map;
1016 	unsigned long nsize;
1017 	int error;
1018 
1019 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1020 	if (error)
1021 		return (error);
1022 	addr = (vm_offset_t)addr0;
1023 	size = len;
1024 	last = addr + size;
1025 	start = trunc_page(addr);
1026 	end = round_page(last);
1027 	if (last < addr || end < addr)
1028 		return (EINVAL);
1029 	npages = atop(end - start);
1030 	if (npages > vm_page_max_wired)
1031 		return (ENOMEM);
1032 	map = &proc->p_vmspace->vm_map;
1033 	PROC_LOCK(proc);
1034 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1035 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1036 		PROC_UNLOCK(proc);
1037 		return (ENOMEM);
1038 	}
1039 	PROC_UNLOCK(proc);
1040 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1041 		return (EAGAIN);
1042 #ifdef RACCT
1043 	if (racct_enable) {
1044 		PROC_LOCK(proc);
1045 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1046 		PROC_UNLOCK(proc);
1047 		if (error != 0)
1048 			return (ENOMEM);
1049 	}
1050 #endif
1051 	error = vm_map_wire(map, start, end,
1052 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1053 #ifdef RACCT
1054 	if (racct_enable && error != KERN_SUCCESS) {
1055 		PROC_LOCK(proc);
1056 		racct_set(proc, RACCT_MEMLOCK,
1057 		    ptoa(pmap_wired_count(map->pmap)));
1058 		PROC_UNLOCK(proc);
1059 	}
1060 #endif
1061 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1062 }
1063 
1064 #ifndef _SYS_SYSPROTO_H_
1065 struct mlockall_args {
1066 	int	how;
1067 };
1068 #endif
1069 
1070 /*
1071  * MPSAFE
1072  */
1073 int
1074 sys_mlockall(td, uap)
1075 	struct thread *td;
1076 	struct mlockall_args *uap;
1077 {
1078 	vm_map_t map;
1079 	int error;
1080 
1081 	map = &td->td_proc->p_vmspace->vm_map;
1082 	error = priv_check(td, PRIV_VM_MLOCK);
1083 	if (error)
1084 		return (error);
1085 
1086 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1087 		return (EINVAL);
1088 
1089 	/*
1090 	 * If wiring all pages in the process would cause it to exceed
1091 	 * a hard resource limit, return ENOMEM.
1092 	 */
1093 	if (!old_mlock && uap->how & MCL_CURRENT) {
1094 		PROC_LOCK(td->td_proc);
1095 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1096 			PROC_UNLOCK(td->td_proc);
1097 			return (ENOMEM);
1098 		}
1099 		PROC_UNLOCK(td->td_proc);
1100 	}
1101 #ifdef RACCT
1102 	if (racct_enable) {
1103 		PROC_LOCK(td->td_proc);
1104 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1105 		PROC_UNLOCK(td->td_proc);
1106 		if (error != 0)
1107 			return (ENOMEM);
1108 	}
1109 #endif
1110 
1111 	if (uap->how & MCL_FUTURE) {
1112 		vm_map_lock(map);
1113 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1114 		vm_map_unlock(map);
1115 		error = 0;
1116 	}
1117 
1118 	if (uap->how & MCL_CURRENT) {
1119 		/*
1120 		 * P1003.1-2001 mandates that all currently mapped pages
1121 		 * will be memory resident and locked (wired) upon return
1122 		 * from mlockall(). vm_map_wire() will wire pages, by
1123 		 * calling vm_fault_wire() for each page in the region.
1124 		 */
1125 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1126 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1127 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1128 	}
1129 #ifdef RACCT
1130 	if (racct_enable && error != KERN_SUCCESS) {
1131 		PROC_LOCK(td->td_proc);
1132 		racct_set(td->td_proc, RACCT_MEMLOCK,
1133 		    ptoa(pmap_wired_count(map->pmap)));
1134 		PROC_UNLOCK(td->td_proc);
1135 	}
1136 #endif
1137 
1138 	return (error);
1139 }
1140 
1141 #ifndef _SYS_SYSPROTO_H_
1142 struct munlockall_args {
1143 	register_t dummy;
1144 };
1145 #endif
1146 
1147 /*
1148  * MPSAFE
1149  */
1150 int
1151 sys_munlockall(td, uap)
1152 	struct thread *td;
1153 	struct munlockall_args *uap;
1154 {
1155 	vm_map_t map;
1156 	int error;
1157 
1158 	map = &td->td_proc->p_vmspace->vm_map;
1159 	error = priv_check(td, PRIV_VM_MUNLOCK);
1160 	if (error)
1161 		return (error);
1162 
1163 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1164 	vm_map_lock(map);
1165 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1166 	vm_map_unlock(map);
1167 
1168 	/* Forcibly unwire all pages. */
1169 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1170 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1171 #ifdef RACCT
1172 	if (racct_enable && error == KERN_SUCCESS) {
1173 		PROC_LOCK(td->td_proc);
1174 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1175 		PROC_UNLOCK(td->td_proc);
1176 	}
1177 #endif
1178 
1179 	return (error);
1180 }
1181 
1182 #ifndef _SYS_SYSPROTO_H_
1183 struct munlock_args {
1184 	const void *addr;
1185 	size_t len;
1186 };
1187 #endif
1188 /*
1189  * MPSAFE
1190  */
1191 int
1192 sys_munlock(td, uap)
1193 	struct thread *td;
1194 	struct munlock_args *uap;
1195 {
1196 	vm_offset_t addr, end, last, start;
1197 	vm_size_t size;
1198 #ifdef RACCT
1199 	vm_map_t map;
1200 #endif
1201 	int error;
1202 
1203 	error = priv_check(td, PRIV_VM_MUNLOCK);
1204 	if (error)
1205 		return (error);
1206 	addr = (vm_offset_t)uap->addr;
1207 	size = uap->len;
1208 	last = addr + size;
1209 	start = trunc_page(addr);
1210 	end = round_page(last);
1211 	if (last < addr || end < addr)
1212 		return (EINVAL);
1213 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1214 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1215 #ifdef RACCT
1216 	if (racct_enable && error == KERN_SUCCESS) {
1217 		PROC_LOCK(td->td_proc);
1218 		map = &td->td_proc->p_vmspace->vm_map;
1219 		racct_set(td->td_proc, RACCT_MEMLOCK,
1220 		    ptoa(pmap_wired_count(map->pmap)));
1221 		PROC_UNLOCK(td->td_proc);
1222 	}
1223 #endif
1224 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1225 }
1226 
1227 /*
1228  * vm_mmap_vnode()
1229  *
1230  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1231  * operations on vnodes.
1232  */
1233 int
1234 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1235     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1236     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1237     boolean_t *writecounted)
1238 {
1239 	struct vattr va;
1240 	vm_object_t obj;
1241 	vm_offset_t foff;
1242 	struct ucred *cred;
1243 	int error, flags, locktype;
1244 
1245 	cred = td->td_ucred;
1246 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1247 		locktype = LK_EXCLUSIVE;
1248 	else
1249 		locktype = LK_SHARED;
1250 	if ((error = vget(vp, locktype, td)) != 0)
1251 		return (error);
1252 	AUDIT_ARG_VNODE1(vp);
1253 	foff = *foffp;
1254 	flags = *flagsp;
1255 	obj = vp->v_object;
1256 	if (vp->v_type == VREG) {
1257 		/*
1258 		 * Get the proper underlying object
1259 		 */
1260 		if (obj == NULL) {
1261 			error = EINVAL;
1262 			goto done;
1263 		}
1264 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1265 			vput(vp);
1266 			vp = (struct vnode *)obj->handle;
1267 			/*
1268 			 * Bypass filesystems obey the mpsafety of the
1269 			 * underlying fs.  Tmpfs never bypasses.
1270 			 */
1271 			error = vget(vp, locktype, td);
1272 			if (error != 0)
1273 				return (error);
1274 		}
1275 		if (locktype == LK_EXCLUSIVE) {
1276 			*writecounted = TRUE;
1277 			vnode_pager_update_writecount(obj, 0, objsize);
1278 		}
1279 	} else {
1280 		error = EINVAL;
1281 		goto done;
1282 	}
1283 	if ((error = VOP_GETATTR(vp, &va, cred)))
1284 		goto done;
1285 #ifdef MAC
1286 	/* This relies on VM_PROT_* matching PROT_*. */
1287 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1288 	if (error != 0)
1289 		goto done;
1290 #endif
1291 	if ((flags & MAP_SHARED) != 0) {
1292 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1293 			if (prot & VM_PROT_WRITE) {
1294 				error = EPERM;
1295 				goto done;
1296 			}
1297 			*maxprotp &= ~VM_PROT_WRITE;
1298 		}
1299 	}
1300 	/*
1301 	 * If it is a regular file without any references
1302 	 * we do not need to sync it.
1303 	 * Adjust object size to be the size of actual file.
1304 	 */
1305 	objsize = round_page(va.va_size);
1306 	if (va.va_nlink == 0)
1307 		flags |= MAP_NOSYNC;
1308 	if (obj->type == OBJT_VNODE) {
1309 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1310 		    cred);
1311 		if (obj == NULL) {
1312 			error = ENOMEM;
1313 			goto done;
1314 		}
1315 	} else {
1316 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1317 		    ("wrong object type"));
1318 		VM_OBJECT_WLOCK(obj);
1319 		vm_object_reference_locked(obj);
1320 #if VM_NRESERVLEVEL > 0
1321 		vm_object_color(obj, 0);
1322 #endif
1323 		VM_OBJECT_WUNLOCK(obj);
1324 	}
1325 	*objp = obj;
1326 	*flagsp = flags;
1327 
1328 	vfs_mark_atime(vp, cred);
1329 
1330 done:
1331 	if (error != 0 && *writecounted) {
1332 		*writecounted = FALSE;
1333 		vnode_pager_update_writecount(obj, objsize, 0);
1334 	}
1335 	vput(vp);
1336 	return (error);
1337 }
1338 
1339 /*
1340  * vm_mmap_cdev()
1341  *
1342  * MPSAFE
1343  *
1344  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1345  * operations on cdevs.
1346  */
1347 int
1348 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1349     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1350     vm_ooffset_t *foff, vm_object_t *objp)
1351 {
1352 	vm_object_t obj;
1353 	int error, flags;
1354 
1355 	flags = *flagsp;
1356 
1357 	if (dsw->d_flags & D_MMAP_ANON) {
1358 		*objp = NULL;
1359 		*foff = 0;
1360 		*maxprotp = VM_PROT_ALL;
1361 		*flagsp |= MAP_ANON;
1362 		return (0);
1363 	}
1364 	/*
1365 	 * cdevs do not provide private mappings of any kind.
1366 	 */
1367 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1368 	    (prot & VM_PROT_WRITE) != 0)
1369 		return (EACCES);
1370 	if (flags & (MAP_PRIVATE|MAP_COPY))
1371 		return (EINVAL);
1372 	/*
1373 	 * Force device mappings to be shared.
1374 	 */
1375 	flags |= MAP_SHARED;
1376 #ifdef MAC_XXX
1377 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1378 	if (error != 0)
1379 		return (error);
1380 #endif
1381 	/*
1382 	 * First, try d_mmap_single().  If that is not implemented
1383 	 * (returns ENODEV), fall back to using the device pager.
1384 	 * Note that d_mmap_single() must return a reference to the
1385 	 * object (it needs to bump the reference count of the object
1386 	 * it returns somehow).
1387 	 *
1388 	 * XXX assumes VM_PROT_* == PROT_*
1389 	 */
1390 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1391 	if (error != ENODEV)
1392 		return (error);
1393 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1394 	    td->td_ucred);
1395 	if (obj == NULL)
1396 		return (EINVAL);
1397 	*objp = obj;
1398 	*flagsp = flags;
1399 	return (0);
1400 }
1401 
1402 /*
1403  * vm_mmap()
1404  *
1405  * Internal version of mmap used by exec, sys5 shared memory, and
1406  * various device drivers.  Handle is either a vnode pointer, a
1407  * character device, or NULL for MAP_ANON.
1408  */
1409 int
1410 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1411 	vm_prot_t maxprot, int flags,
1412 	objtype_t handle_type, void *handle,
1413 	vm_ooffset_t foff)
1414 {
1415 	vm_object_t object;
1416 	struct thread *td = curthread;
1417 	int error;
1418 	boolean_t writecounted;
1419 
1420 	if (size == 0)
1421 		return (EINVAL);
1422 
1423 	size = round_page(size);
1424 	object = NULL;
1425 	writecounted = FALSE;
1426 
1427 	/*
1428 	 * Lookup/allocate object.
1429 	 */
1430 	switch (handle_type) {
1431 	case OBJT_DEVICE: {
1432 		struct cdevsw *dsw;
1433 		struct cdev *cdev;
1434 		int ref;
1435 
1436 		cdev = handle;
1437 		dsw = dev_refthread(cdev, &ref);
1438 		if (dsw == NULL)
1439 			return (ENXIO);
1440 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1441 		    dsw, &foff, &object);
1442 		dev_relthread(cdev, ref);
1443 		break;
1444 	}
1445 	case OBJT_VNODE:
1446 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1447 		    handle, &foff, &object, &writecounted);
1448 		break;
1449 	case OBJT_DEFAULT:
1450 		if (handle == NULL) {
1451 			error = 0;
1452 			break;
1453 		}
1454 		/* FALLTHROUGH */
1455 	default:
1456 		error = EINVAL;
1457 		break;
1458 	}
1459 	if (error)
1460 		return (error);
1461 
1462 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1463 	    foff, writecounted, td);
1464 	if (error != 0 && object != NULL) {
1465 		/*
1466 		 * If this mapping was accounted for in the vnode's
1467 		 * writecount, then undo that now.
1468 		 */
1469 		if (writecounted)
1470 			vnode_pager_release_writecount(object, 0, size);
1471 		vm_object_deallocate(object);
1472 	}
1473 	return (error);
1474 }
1475 
1476 /*
1477  * Internal version of mmap that maps a specific VM object into an
1478  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1479  */
1480 int
1481 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1482     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1483     boolean_t writecounted, struct thread *td)
1484 {
1485 	boolean_t fitit;
1486 	int docow, error, findspace, rv;
1487 
1488 	if (map == &td->td_proc->p_vmspace->vm_map) {
1489 		PROC_LOCK(td->td_proc);
1490 		if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1491 			PROC_UNLOCK(td->td_proc);
1492 			return (ENOMEM);
1493 		}
1494 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1495 			PROC_UNLOCK(td->td_proc);
1496 			return (ENOMEM);
1497 		}
1498 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1499 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1500 			    lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1501 				racct_set_force(td->td_proc, RACCT_VMEM,
1502 				    map->size);
1503 				PROC_UNLOCK(td->td_proc);
1504 				return (ENOMEM);
1505 			}
1506 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1507 			    ptoa(pmap_wired_count(map->pmap)) + size);
1508 			if (error != 0) {
1509 				racct_set_force(td->td_proc, RACCT_VMEM,
1510 				    map->size);
1511 				PROC_UNLOCK(td->td_proc);
1512 				return (error);
1513 			}
1514 		}
1515 		PROC_UNLOCK(td->td_proc);
1516 	}
1517 
1518 	/*
1519 	 * We currently can only deal with page aligned file offsets.
1520 	 * The mmap() system call already enforces this by subtracting
1521 	 * the page offset from the file offset, but checking here
1522 	 * catches errors in device drivers (e.g. d_single_mmap()
1523 	 * callbacks) and other internal mapping requests (such as in
1524 	 * exec).
1525 	 */
1526 	if (foff & PAGE_MASK)
1527 		return (EINVAL);
1528 
1529 	if ((flags & MAP_FIXED) == 0) {
1530 		fitit = TRUE;
1531 		*addr = round_page(*addr);
1532 	} else {
1533 		if (*addr != trunc_page(*addr))
1534 			return (EINVAL);
1535 		fitit = FALSE;
1536 	}
1537 
1538 	if (flags & MAP_ANON) {
1539 		if (object != NULL || foff != 0)
1540 			return (EINVAL);
1541 		docow = 0;
1542 	} else if (flags & MAP_PREFAULT_READ)
1543 		docow = MAP_PREFAULT;
1544 	else
1545 		docow = MAP_PREFAULT_PARTIAL;
1546 
1547 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1548 		docow |= MAP_COPY_ON_WRITE;
1549 	if (flags & MAP_NOSYNC)
1550 		docow |= MAP_DISABLE_SYNCER;
1551 	if (flags & MAP_NOCORE)
1552 		docow |= MAP_DISABLE_COREDUMP;
1553 	/* Shared memory is also shared with children. */
1554 	if (flags & MAP_SHARED)
1555 		docow |= MAP_INHERIT_SHARE;
1556 	if (writecounted)
1557 		docow |= MAP_VN_WRITECOUNT;
1558 	if (flags & MAP_STACK) {
1559 		if (object != NULL)
1560 			return (EINVAL);
1561 		docow |= MAP_STACK_GROWS_DOWN;
1562 	}
1563 	if ((flags & MAP_EXCL) != 0)
1564 		docow |= MAP_CHECK_EXCL;
1565 
1566 	if (fitit) {
1567 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1568 			findspace = VMFS_SUPER_SPACE;
1569 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1570 			findspace = VMFS_ALIGNED_SPACE(flags >>
1571 			    MAP_ALIGNMENT_SHIFT);
1572 		else
1573 			findspace = VMFS_OPTIMAL_SPACE;
1574 		rv = vm_map_find(map, object, foff, addr, size,
1575 #ifdef MAP_32BIT
1576 		    flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1577 #endif
1578 		    0, findspace, prot, maxprot, docow);
1579 	} else {
1580 		rv = vm_map_fixed(map, object, foff, *addr, size,
1581 		    prot, maxprot, docow);
1582 	}
1583 
1584 	if (rv == KERN_SUCCESS) {
1585 		/*
1586 		 * If the process has requested that all future mappings
1587 		 * be wired, then heed this.
1588 		 */
1589 		if (map->flags & MAP_WIREFUTURE) {
1590 			vm_map_wire(map, *addr, *addr + size,
1591 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1592 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1593 		}
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