xref: /freebsd/sys/vm/vm_mmap.c (revision a678f77962278b50da97a88aac8abe77c192a4f1)
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(struct thread *td, struct mmap_args *uap)
191 {
192 
193 	return (kern_vm_mmap(td, (vm_offset_t)uap->addr, uap->len,
194 	    uap->prot, uap->flags, uap->fd, uap->pos));
195 }
196 
197 int
198 kern_vm_mmap(struct thread *td, vm_offset_t addr, vm_size_t size,
199     vm_prot_t prot, int flags, int fd, off_t pos)
200 {
201 	struct file *fp;
202 	vm_size_t pageoff;
203 	vm_prot_t cap_maxprot;
204 	int align, error;
205 	struct vmspace *vms = td->td_proc->p_vmspace;
206 	cap_rights_t rights;
207 
208 	fp = NULL;
209 	AUDIT_ARG_FD(fd);
210 
211 	/*
212 	 * Ignore old flags that used to be defined but did not do anything.
213 	 */
214 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
215 
216 	/*
217 	 * Enforce the constraints.
218 	 * Mapping of length 0 is only allowed for old binaries.
219 	 * Anonymous mapping shall specify -1 as filedescriptor and
220 	 * zero position for new code. Be nice to ancient a.out
221 	 * binaries and correct pos for anonymous mapping, since old
222 	 * ld.so sometimes issues anonymous map requests with non-zero
223 	 * pos.
224 	 */
225 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
226 		if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
227 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
228 			return (EINVAL);
229 	} else {
230 		if ((flags & MAP_ANON) != 0)
231 			pos = 0;
232 	}
233 
234 	if (flags & MAP_STACK) {
235 		if ((fd != -1) ||
236 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
237 			return (EINVAL);
238 		flags |= MAP_ANON;
239 		pos = 0;
240 	}
241 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
242 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
243 	    MAP_PREFAULT_READ |
244 #ifdef MAP_32BIT
245 	    MAP_32BIT |
246 #endif
247 	    MAP_ALIGNMENT_MASK)) != 0)
248 		return (EINVAL);
249 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
250 		return (EINVAL);
251 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
252 		return (EINVAL);
253 	if (prot != PROT_NONE &&
254 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
255 		return (EINVAL);
256 
257 	/*
258 	 * Align the file position to a page boundary,
259 	 * and save its page offset component.
260 	 */
261 	pageoff = (pos & PAGE_MASK);
262 	pos -= pageoff;
263 
264 	/* Adjust size for rounding (on both ends). */
265 	size += pageoff;			/* low end... */
266 	size = (vm_size_t) round_page(size);	/* hi end */
267 
268 	/* Ensure alignment is at least a page and fits in a pointer. */
269 	align = flags & MAP_ALIGNMENT_MASK;
270 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
271 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
272 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
273 		return (EINVAL);
274 
275 	/*
276 	 * Check for illegal addresses.  Watch out for address wrap... Note
277 	 * that VM_*_ADDRESS are not constants due to casts (argh).
278 	 */
279 	if (flags & MAP_FIXED) {
280 		/*
281 		 * The specified address must have the same remainder
282 		 * as the file offset taken modulo PAGE_SIZE, so it
283 		 * should be aligned after adjustment by pageoff.
284 		 */
285 		addr -= pageoff;
286 		if (addr & PAGE_MASK)
287 			return (EINVAL);
288 
289 		/* Address range must be all in user VM space. */
290 		if (addr < vm_map_min(&vms->vm_map) ||
291 		    addr + size > vm_map_max(&vms->vm_map))
292 			return (EINVAL);
293 		if (addr + size < addr)
294 			return (EINVAL);
295 #ifdef MAP_32BIT
296 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
297 			return (EINVAL);
298 	} else if (flags & MAP_32BIT) {
299 		/*
300 		 * For MAP_32BIT, override the hint if it is too high and
301 		 * do not bother moving the mapping past the heap (since
302 		 * the heap is usually above 2GB).
303 		 */
304 		if (addr + size > MAP_32BIT_MAX_ADDR)
305 			addr = 0;
306 #endif
307 	} else {
308 		/*
309 		 * XXX for non-fixed mappings where no hint is provided or
310 		 * the hint would fall in the potential heap space,
311 		 * place it after the end of the largest possible heap.
312 		 *
313 		 * There should really be a pmap call to determine a reasonable
314 		 * location.
315 		 */
316 		if (addr == 0 ||
317 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
318 		    addr < round_page((vm_offset_t)vms->vm_daddr +
319 		    lim_max(td, RLIMIT_DATA))))
320 			addr = round_page((vm_offset_t)vms->vm_daddr +
321 			    lim_max(td, RLIMIT_DATA));
322 	}
323 	if (size == 0) {
324 		/*
325 		 * Return success without mapping anything for old
326 		 * binaries that request a page-aligned mapping of
327 		 * length 0.  For modern binaries, this function
328 		 * returns an error earlier.
329 		 */
330 		error = 0;
331 	} else if (flags & MAP_ANON) {
332 		/*
333 		 * Mapping blank space is trivial.
334 		 *
335 		 * This relies on VM_PROT_* matching PROT_*.
336 		 */
337 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
338 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
339 	} else {
340 		/*
341 		 * Mapping file, get fp for validation and don't let the
342 		 * descriptor disappear on us if we block. Check capability
343 		 * rights, but also return the maximum rights to be combined
344 		 * with maxprot later.
345 		 */
346 		cap_rights_init(&rights, CAP_MMAP);
347 		if (prot & PROT_READ)
348 			cap_rights_set(&rights, CAP_MMAP_R);
349 		if ((flags & MAP_SHARED) != 0) {
350 			if (prot & PROT_WRITE)
351 				cap_rights_set(&rights, CAP_MMAP_W);
352 		}
353 		if (prot & PROT_EXEC)
354 			cap_rights_set(&rights, CAP_MMAP_X);
355 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
356 		if (error != 0)
357 			goto done;
358 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
359 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
360 			error = EINVAL;
361 			goto done;
362 		}
363 
364 		/* This relies on VM_PROT_* matching PROT_*. */
365 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
366 		    cap_maxprot, flags, pos, td);
367 	}
368 
369 	if (error == 0)
370 		td->td_retval[0] = (register_t) (addr + pageoff);
371 done:
372 	if (fp)
373 		fdrop(fp, td);
374 
375 	return (error);
376 }
377 
378 #if defined(COMPAT_FREEBSD6)
379 int
380 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
381 {
382 
383 	return (kern_vm_mmap(td, (vm_offset_t)uap->addr, uap->len,
384 	    uap->prot, uap->flags, uap->fd, uap->pos));
385 }
386 #endif
387 
388 #ifdef COMPAT_43
389 #ifndef _SYS_SYSPROTO_H_
390 struct ommap_args {
391 	caddr_t addr;
392 	int len;
393 	int prot;
394 	int flags;
395 	int fd;
396 	long pos;
397 };
398 #endif
399 int
400 ommap(struct thread *td, struct ommap_args *uap)
401 {
402 	static const char cvtbsdprot[8] = {
403 		0,
404 		PROT_EXEC,
405 		PROT_WRITE,
406 		PROT_EXEC | PROT_WRITE,
407 		PROT_READ,
408 		PROT_EXEC | PROT_READ,
409 		PROT_WRITE | PROT_READ,
410 		PROT_EXEC | PROT_WRITE | PROT_READ,
411 	};
412 	int flags, prot;
413 
414 #define	OMAP_ANON	0x0002
415 #define	OMAP_COPY	0x0020
416 #define	OMAP_SHARED	0x0010
417 #define	OMAP_FIXED	0x0100
418 
419 	prot = cvtbsdprot[uap->prot & 0x7];
420 #ifdef COMPAT_FREEBSD32
421 #if defined(__amd64__)
422 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
423 	    prot != 0)
424 		prot |= PROT_EXEC;
425 #endif
426 #endif
427 	flags = 0;
428 	if (uap->flags & OMAP_ANON)
429 		flags |= MAP_ANON;
430 	if (uap->flags & OMAP_COPY)
431 		flags |= MAP_COPY;
432 	if (uap->flags & OMAP_SHARED)
433 		flags |= MAP_SHARED;
434 	else
435 		flags |= MAP_PRIVATE;
436 	if (uap->flags & OMAP_FIXED)
437 		flags |= MAP_FIXED;
438 	return (kern_vm_mmap(td, (vm_offset_t)uap->addr, uap->len,
439 	    prot, flags, uap->fd, uap->pos));
440 }
441 #endif				/* COMPAT_43 */
442 
443 
444 #ifndef _SYS_SYSPROTO_H_
445 struct msync_args {
446 	void *addr;
447 	size_t len;
448 	int flags;
449 };
450 #endif
451 /*
452  * MPSAFE
453  */
454 int
455 sys_msync(struct thread *td, struct msync_args *uap)
456 {
457 
458 	return (kern_vm_msync(td, (vm_offset_t)uap->addr, uap->len,
459 	    uap->flags));
460 }
461 
462 int
463 kern_vm_msync(struct thread *td, vm_offset_t addr, vm_size_t size, int flags)
464 {
465 	vm_size_t pageoff;
466 	vm_map_t map;
467 	int rv;
468 
469 	pageoff = (addr & PAGE_MASK);
470 	addr -= pageoff;
471 	size += pageoff;
472 	size = (vm_size_t) round_page(size);
473 	if (addr + size < addr)
474 		return (EINVAL);
475 
476 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
477 		return (EINVAL);
478 
479 	map = &td->td_proc->p_vmspace->vm_map;
480 
481 	/*
482 	 * Clean the pages and interpret the return value.
483 	 */
484 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
485 	    (flags & MS_INVALIDATE) != 0);
486 	switch (rv) {
487 	case KERN_SUCCESS:
488 		return (0);
489 	case KERN_INVALID_ADDRESS:
490 		return (ENOMEM);
491 	case KERN_INVALID_ARGUMENT:
492 		return (EBUSY);
493 	case KERN_FAILURE:
494 		return (EIO);
495 	default:
496 		return (EINVAL);
497 	}
498 }
499 
500 #ifndef _SYS_SYSPROTO_H_
501 struct munmap_args {
502 	void *addr;
503 	size_t len;
504 };
505 #endif
506 /*
507  * MPSAFE
508  */
509 int
510 sys_munmap(struct thread *td, struct munmap_args *uap)
511 {
512 
513 	return (kern_vm_munmap(td, (vm_offset_t)uap->addr, uap->len));
514 }
515 
516 int
517 kern_vm_munmap(struct thread *td, vm_offset_t addr, vm_size_t size)
518 {
519 #ifdef HWPMC_HOOKS
520 	struct pmckern_map_out pkm;
521 	vm_map_entry_t entry;
522 	bool pmc_handled;
523 #endif
524 	vm_size_t pageoff;
525 	vm_map_t map;
526 
527 	if (size == 0)
528 		return (EINVAL);
529 
530 	pageoff = (addr & PAGE_MASK);
531 	addr -= pageoff;
532 	size += pageoff;
533 	size = (vm_size_t) round_page(size);
534 	if (addr + size < addr)
535 		return (EINVAL);
536 
537 	/*
538 	 * Check for illegal addresses.  Watch out for address wrap...
539 	 */
540 	map = &td->td_proc->p_vmspace->vm_map;
541 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
542 		return (EINVAL);
543 	vm_map_lock(map);
544 #ifdef HWPMC_HOOKS
545 	pmc_handled = false;
546 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
547 		pmc_handled = true;
548 		/*
549 		 * Inform hwpmc if the address range being unmapped contains
550 		 * an executable region.
551 		 */
552 		pkm.pm_address = (uintptr_t) NULL;
553 		if (vm_map_lookup_entry(map, addr, &entry)) {
554 			for (;
555 			    entry != &map->header && entry->start < addr + size;
556 			    entry = entry->next) {
557 				if (vm_map_check_protection(map, entry->start,
558 					entry->end, VM_PROT_EXECUTE) == TRUE) {
559 					pkm.pm_address = (uintptr_t) addr;
560 					pkm.pm_size = (size_t) size;
561 					break;
562 				}
563 			}
564 		}
565 	}
566 #endif
567 	vm_map_delete(map, addr, addr + size);
568 
569 #ifdef HWPMC_HOOKS
570 	if (__predict_false(pmc_handled)) {
571 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
572 		vm_map_lock_downgrade(map);
573 		if (pkm.pm_address != (uintptr_t) NULL)
574 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
575 		vm_map_unlock_read(map);
576 	} else
577 #endif
578 		vm_map_unlock(map);
579 
580 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
581 	return (0);
582 }
583 
584 #ifndef _SYS_SYSPROTO_H_
585 struct mprotect_args {
586 	const void *addr;
587 	size_t len;
588 	int prot;
589 };
590 #endif
591 /*
592  * MPSAFE
593  */
594 int
595 sys_mprotect(struct thread *td, struct mprotect_args *uap)
596 {
597 
598 	return (kern_vm_mprotect(td, (vm_offset_t)uap->addr, uap->len,
599 	    uap->prot));
600 }
601 
602 int
603 kern_vm_mprotect(struct thread *td, vm_offset_t addr, vm_size_t size,
604     vm_prot_t prot)
605 {
606 	vm_size_t pageoff;
607 
608 	prot = (prot & VM_PROT_ALL);
609 	pageoff = (addr & PAGE_MASK);
610 	addr -= pageoff;
611 	size += pageoff;
612 	size = (vm_size_t) round_page(size);
613 	if (addr + size < addr)
614 		return (EINVAL);
615 
616 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
617 	    addr + size, prot, FALSE)) {
618 	case KERN_SUCCESS:
619 		return (0);
620 	case KERN_PROTECTION_FAILURE:
621 		return (EACCES);
622 	case KERN_RESOURCE_SHORTAGE:
623 		return (ENOMEM);
624 	}
625 	return (EINVAL);
626 }
627 
628 #ifndef _SYS_SYSPROTO_H_
629 struct minherit_args {
630 	void *addr;
631 	size_t len;
632 	int inherit;
633 };
634 #endif
635 /*
636  * MPSAFE
637  */
638 int
639 sys_minherit(td, uap)
640 	struct thread *td;
641 	struct minherit_args *uap;
642 {
643 	vm_offset_t addr;
644 	vm_size_t size, pageoff;
645 	vm_inherit_t inherit;
646 
647 	addr = (vm_offset_t)uap->addr;
648 	size = uap->len;
649 	inherit = uap->inherit;
650 
651 	pageoff = (addr & PAGE_MASK);
652 	addr -= pageoff;
653 	size += pageoff;
654 	size = (vm_size_t) round_page(size);
655 	if (addr + size < addr)
656 		return (EINVAL);
657 
658 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
659 	    addr + size, inherit)) {
660 	case KERN_SUCCESS:
661 		return (0);
662 	case KERN_PROTECTION_FAILURE:
663 		return (EACCES);
664 	}
665 	return (EINVAL);
666 }
667 
668 #ifndef _SYS_SYSPROTO_H_
669 struct madvise_args {
670 	void *addr;
671 	size_t len;
672 	int behav;
673 };
674 #endif
675 
676 /*
677  * MPSAFE
678  */
679 int
680 sys_madvise(td, uap)
681 	struct thread *td;
682 	struct madvise_args *uap;
683 {
684 
685 	return (kern_vm_madvise(td, (vm_offset_t)uap->addr, uap->len,
686 	    uap->behav));
687 }
688 
689 int
690 kern_vm_madvise(struct thread *td, vm_offset_t addr, vm_size_t len, int behav)
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 (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 (behav < 0 || 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 (addr < vm_map_min(map) || addr + len > vm_map_max(map))
717 		return (EINVAL);
718 	if ((addr + len) < addr)
719 		return (EINVAL);
720 
721 	/*
722 	 * Since this routine is only advisory, we default to conservative
723 	 * behavior.
724 	 */
725 	start = trunc_page(addr);
726 	end = round_page(addr + len);
727 
728 	if (vm_map_madvise(map, start, end, behav))
729 		return (EINVAL);
730 	return (0);
731 }
732 
733 #ifndef _SYS_SYSPROTO_H_
734 struct mincore_args {
735 	const void *addr;
736 	size_t len;
737 	char *vec;
738 };
739 #endif
740 
741 /*
742  * MPSAFE
743  */
744 int
745 sys_mincore(td, uap)
746 	struct thread *td;
747 	struct mincore_args *uap;
748 {
749 	vm_offset_t addr, first_addr;
750 	vm_offset_t end, cend;
751 	pmap_t pmap;
752 	vm_map_t map;
753 	char *vec;
754 	int error = 0;
755 	int vecindex, lastvecindex;
756 	vm_map_entry_t current;
757 	vm_map_entry_t entry;
758 	vm_object_t object;
759 	vm_paddr_t locked_pa;
760 	vm_page_t m;
761 	vm_pindex_t pindex;
762 	int mincoreinfo;
763 	unsigned int timestamp;
764 	boolean_t locked;
765 
766 	/*
767 	 * Make sure that the addresses presented are valid for user
768 	 * mode.
769 	 */
770 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
771 	end = addr + (vm_size_t)round_page(uap->len);
772 	map = &td->td_proc->p_vmspace->vm_map;
773 	if (end > vm_map_max(map) || end < addr)
774 		return (ENOMEM);
775 
776 	/*
777 	 * Address of byte vector
778 	 */
779 	vec = uap->vec;
780 
781 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
782 
783 	vm_map_lock_read(map);
784 RestartScan:
785 	timestamp = map->timestamp;
786 
787 	if (!vm_map_lookup_entry(map, addr, &entry)) {
788 		vm_map_unlock_read(map);
789 		return (ENOMEM);
790 	}
791 
792 	/*
793 	 * Do this on a map entry basis so that if the pages are not
794 	 * in the current processes address space, we can easily look
795 	 * up the pages elsewhere.
796 	 */
797 	lastvecindex = -1;
798 	for (current = entry;
799 	    (current != &map->header) && (current->start < end);
800 	    current = current->next) {
801 
802 		/*
803 		 * check for contiguity
804 		 */
805 		if (current->end < end &&
806 		    (entry->next == &map->header ||
807 		     current->next->start > current->end)) {
808 			vm_map_unlock_read(map);
809 			return (ENOMEM);
810 		}
811 
812 		/*
813 		 * ignore submaps (for now) or null objects
814 		 */
815 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
816 			current->object.vm_object == NULL)
817 			continue;
818 
819 		/*
820 		 * limit this scan to the current map entry and the
821 		 * limits for the mincore call
822 		 */
823 		if (addr < current->start)
824 			addr = current->start;
825 		cend = current->end;
826 		if (cend > end)
827 			cend = end;
828 
829 		/*
830 		 * scan this entry one page at a time
831 		 */
832 		while (addr < cend) {
833 			/*
834 			 * Check pmap first, it is likely faster, also
835 			 * it can provide info as to whether we are the
836 			 * one referencing or modifying the page.
837 			 */
838 			object = NULL;
839 			locked_pa = 0;
840 		retry:
841 			m = NULL;
842 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
843 			if (locked_pa != 0) {
844 				/*
845 				 * The page is mapped by this process but not
846 				 * both accessed and modified.  It is also
847 				 * managed.  Acquire the object lock so that
848 				 * other mappings might be examined.
849 				 */
850 				m = PHYS_TO_VM_PAGE(locked_pa);
851 				if (m->object != object) {
852 					if (object != NULL)
853 						VM_OBJECT_WUNLOCK(object);
854 					object = m->object;
855 					locked = VM_OBJECT_TRYWLOCK(object);
856 					vm_page_unlock(m);
857 					if (!locked) {
858 						VM_OBJECT_WLOCK(object);
859 						vm_page_lock(m);
860 						goto retry;
861 					}
862 				} else
863 					vm_page_unlock(m);
864 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
865 				    ("mincore: page %p is mapped but invalid",
866 				    m));
867 			} else if (mincoreinfo == 0) {
868 				/*
869 				 * The page is not mapped by this process.  If
870 				 * the object implements managed pages, then
871 				 * determine if the page is resident so that
872 				 * the mappings might be examined.
873 				 */
874 				if (current->object.vm_object != object) {
875 					if (object != NULL)
876 						VM_OBJECT_WUNLOCK(object);
877 					object = current->object.vm_object;
878 					VM_OBJECT_WLOCK(object);
879 				}
880 				if (object->type == OBJT_DEFAULT ||
881 				    object->type == OBJT_SWAP ||
882 				    object->type == OBJT_VNODE) {
883 					pindex = OFF_TO_IDX(current->offset +
884 					    (addr - current->start));
885 					m = vm_page_lookup(object, pindex);
886 					if (m != NULL && m->valid == 0)
887 						m = NULL;
888 					if (m != NULL)
889 						mincoreinfo = MINCORE_INCORE;
890 				}
891 			}
892 			if (m != NULL) {
893 				/* Examine other mappings to the page. */
894 				if (m->dirty == 0 && pmap_is_modified(m))
895 					vm_page_dirty(m);
896 				if (m->dirty != 0)
897 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
898 				/*
899 				 * The first test for PGA_REFERENCED is an
900 				 * optimization.  The second test is
901 				 * required because a concurrent pmap
902 				 * operation could clear the last reference
903 				 * and set PGA_REFERENCED before the call to
904 				 * pmap_is_referenced().
905 				 */
906 				if ((m->aflags & PGA_REFERENCED) != 0 ||
907 				    pmap_is_referenced(m) ||
908 				    (m->aflags & PGA_REFERENCED) != 0)
909 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
910 			}
911 			if (object != NULL)
912 				VM_OBJECT_WUNLOCK(object);
913 
914 			/*
915 			 * subyte may page fault.  In case it needs to modify
916 			 * the map, we release the lock.
917 			 */
918 			vm_map_unlock_read(map);
919 
920 			/*
921 			 * calculate index into user supplied byte vector
922 			 */
923 			vecindex = OFF_TO_IDX(addr - first_addr);
924 
925 			/*
926 			 * If we have skipped map entries, we need to make sure that
927 			 * the byte vector is zeroed for those skipped entries.
928 			 */
929 			while ((lastvecindex + 1) < vecindex) {
930 				++lastvecindex;
931 				error = subyte(vec + lastvecindex, 0);
932 				if (error) {
933 					error = EFAULT;
934 					goto done2;
935 				}
936 			}
937 
938 			/*
939 			 * Pass the page information to the user
940 			 */
941 			error = subyte(vec + vecindex, mincoreinfo);
942 			if (error) {
943 				error = EFAULT;
944 				goto done2;
945 			}
946 
947 			/*
948 			 * If the map has changed, due to the subyte, the previous
949 			 * output may be invalid.
950 			 */
951 			vm_map_lock_read(map);
952 			if (timestamp != map->timestamp)
953 				goto RestartScan;
954 
955 			lastvecindex = vecindex;
956 			addr += PAGE_SIZE;
957 		}
958 	}
959 
960 	/*
961 	 * subyte may page fault.  In case it needs to modify
962 	 * the map, we release the lock.
963 	 */
964 	vm_map_unlock_read(map);
965 
966 	/*
967 	 * Zero the last entries in the byte vector.
968 	 */
969 	vecindex = OFF_TO_IDX(end - first_addr);
970 	while ((lastvecindex + 1) < vecindex) {
971 		++lastvecindex;
972 		error = subyte(vec + lastvecindex, 0);
973 		if (error) {
974 			error = EFAULT;
975 			goto done2;
976 		}
977 	}
978 
979 	/*
980 	 * If the map has changed, due to the subyte, the previous
981 	 * output may be invalid.
982 	 */
983 	vm_map_lock_read(map);
984 	if (timestamp != map->timestamp)
985 		goto RestartScan;
986 	vm_map_unlock_read(map);
987 done2:
988 	return (error);
989 }
990 
991 #ifndef _SYS_SYSPROTO_H_
992 struct mlock_args {
993 	const void *addr;
994 	size_t len;
995 };
996 #endif
997 /*
998  * MPSAFE
999  */
1000 int
1001 sys_mlock(td, uap)
1002 	struct thread *td;
1003 	struct mlock_args *uap;
1004 {
1005 
1006 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1007 }
1008 
1009 int
1010 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1011 {
1012 	vm_offset_t addr, end, last, start;
1013 	vm_size_t npages, size;
1014 	vm_map_t map;
1015 	unsigned long nsize;
1016 	int error;
1017 
1018 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1019 	if (error)
1020 		return (error);
1021 	addr = (vm_offset_t)addr0;
1022 	size = len;
1023 	last = addr + size;
1024 	start = trunc_page(addr);
1025 	end = round_page(last);
1026 	if (last < addr || end < addr)
1027 		return (EINVAL);
1028 	npages = atop(end - start);
1029 	if (npages > vm_page_max_wired)
1030 		return (ENOMEM);
1031 	map = &proc->p_vmspace->vm_map;
1032 	PROC_LOCK(proc);
1033 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1034 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1035 		PROC_UNLOCK(proc);
1036 		return (ENOMEM);
1037 	}
1038 	PROC_UNLOCK(proc);
1039 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1040 		return (EAGAIN);
1041 #ifdef RACCT
1042 	if (racct_enable) {
1043 		PROC_LOCK(proc);
1044 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1045 		PROC_UNLOCK(proc);
1046 		if (error != 0)
1047 			return (ENOMEM);
1048 	}
1049 #endif
1050 	error = vm_map_wire(map, start, end,
1051 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1052 #ifdef RACCT
1053 	if (racct_enable && error != KERN_SUCCESS) {
1054 		PROC_LOCK(proc);
1055 		racct_set(proc, RACCT_MEMLOCK,
1056 		    ptoa(pmap_wired_count(map->pmap)));
1057 		PROC_UNLOCK(proc);
1058 	}
1059 #endif
1060 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1061 }
1062 
1063 #ifndef _SYS_SYSPROTO_H_
1064 struct mlockall_args {
1065 	int	how;
1066 };
1067 #endif
1068 
1069 /*
1070  * MPSAFE
1071  */
1072 int
1073 sys_mlockall(td, uap)
1074 	struct thread *td;
1075 	struct mlockall_args *uap;
1076 {
1077 	vm_map_t map;
1078 	int error;
1079 
1080 	map = &td->td_proc->p_vmspace->vm_map;
1081 	error = priv_check(td, PRIV_VM_MLOCK);
1082 	if (error)
1083 		return (error);
1084 
1085 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1086 		return (EINVAL);
1087 
1088 	/*
1089 	 * If wiring all pages in the process would cause it to exceed
1090 	 * a hard resource limit, return ENOMEM.
1091 	 */
1092 	if (!old_mlock && uap->how & MCL_CURRENT) {
1093 		PROC_LOCK(td->td_proc);
1094 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1095 			PROC_UNLOCK(td->td_proc);
1096 			return (ENOMEM);
1097 		}
1098 		PROC_UNLOCK(td->td_proc);
1099 	}
1100 #ifdef RACCT
1101 	if (racct_enable) {
1102 		PROC_LOCK(td->td_proc);
1103 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1104 		PROC_UNLOCK(td->td_proc);
1105 		if (error != 0)
1106 			return (ENOMEM);
1107 	}
1108 #endif
1109 
1110 	if (uap->how & MCL_FUTURE) {
1111 		vm_map_lock(map);
1112 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1113 		vm_map_unlock(map);
1114 		error = 0;
1115 	}
1116 
1117 	if (uap->how & MCL_CURRENT) {
1118 		/*
1119 		 * P1003.1-2001 mandates that all currently mapped pages
1120 		 * will be memory resident and locked (wired) upon return
1121 		 * from mlockall(). vm_map_wire() will wire pages, by
1122 		 * calling vm_fault_wire() for each page in the region.
1123 		 */
1124 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1125 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1126 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1127 	}
1128 #ifdef RACCT
1129 	if (racct_enable && error != KERN_SUCCESS) {
1130 		PROC_LOCK(td->td_proc);
1131 		racct_set(td->td_proc, RACCT_MEMLOCK,
1132 		    ptoa(pmap_wired_count(map->pmap)));
1133 		PROC_UNLOCK(td->td_proc);
1134 	}
1135 #endif
1136 
1137 	return (error);
1138 }
1139 
1140 #ifndef _SYS_SYSPROTO_H_
1141 struct munlockall_args {
1142 	register_t dummy;
1143 };
1144 #endif
1145 
1146 /*
1147  * MPSAFE
1148  */
1149 int
1150 sys_munlockall(td, uap)
1151 	struct thread *td;
1152 	struct munlockall_args *uap;
1153 {
1154 	vm_map_t map;
1155 	int error;
1156 
1157 	map = &td->td_proc->p_vmspace->vm_map;
1158 	error = priv_check(td, PRIV_VM_MUNLOCK);
1159 	if (error)
1160 		return (error);
1161 
1162 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1163 	vm_map_lock(map);
1164 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1165 	vm_map_unlock(map);
1166 
1167 	/* Forcibly unwire all pages. */
1168 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1169 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1170 #ifdef RACCT
1171 	if (racct_enable && error == KERN_SUCCESS) {
1172 		PROC_LOCK(td->td_proc);
1173 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1174 		PROC_UNLOCK(td->td_proc);
1175 	}
1176 #endif
1177 
1178 	return (error);
1179 }
1180 
1181 #ifndef _SYS_SYSPROTO_H_
1182 struct munlock_args {
1183 	const void *addr;
1184 	size_t len;
1185 };
1186 #endif
1187 /*
1188  * MPSAFE
1189  */
1190 int
1191 sys_munlock(struct thread *td, struct munlock_args *uap)
1192 {
1193 
1194 	return (kern_vm_munlock(td, (vm_offset_t)uap->addr, uap->len));
1195 }
1196 
1197 int
1198 kern_vm_munlock(struct thread *td, vm_offset_t addr, vm_size_t size)
1199 {
1200 	vm_offset_t end, last, start;
1201 #ifdef RACCT
1202 	vm_map_t map;
1203 #endif
1204 	int error;
1205 
1206 	error = priv_check(td, PRIV_VM_MUNLOCK);
1207 	if (error)
1208 		return (error);
1209 	last = addr + size;
1210 	start = trunc_page(addr);
1211 	end = round_page(last);
1212 	if (last < addr || end < addr)
1213 		return (EINVAL);
1214 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1215 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1216 #ifdef RACCT
1217 	if (racct_enable && error == KERN_SUCCESS) {
1218 		PROC_LOCK(td->td_proc);
1219 		map = &td->td_proc->p_vmspace->vm_map;
1220 		racct_set(td->td_proc, RACCT_MEMLOCK,
1221 		    ptoa(pmap_wired_count(map->pmap)));
1222 		PROC_UNLOCK(td->td_proc);
1223 	}
1224 #endif
1225 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1226 }
1227 
1228 /*
1229  * vm_mmap_vnode()
1230  *
1231  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1232  * operations on vnodes.
1233  */
1234 int
1235 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1236     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1237     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1238     boolean_t *writecounted)
1239 {
1240 	struct vattr va;
1241 	vm_object_t obj;
1242 	vm_offset_t foff;
1243 	struct ucred *cred;
1244 	int error, flags, locktype;
1245 
1246 	cred = td->td_ucred;
1247 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1248 		locktype = LK_EXCLUSIVE;
1249 	else
1250 		locktype = LK_SHARED;
1251 	if ((error = vget(vp, locktype, td)) != 0)
1252 		return (error);
1253 	AUDIT_ARG_VNODE1(vp);
1254 	foff = *foffp;
1255 	flags = *flagsp;
1256 	obj = vp->v_object;
1257 	if (vp->v_type == VREG) {
1258 		/*
1259 		 * Get the proper underlying object
1260 		 */
1261 		if (obj == NULL) {
1262 			error = EINVAL;
1263 			goto done;
1264 		}
1265 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1266 			vput(vp);
1267 			vp = (struct vnode *)obj->handle;
1268 			/*
1269 			 * Bypass filesystems obey the mpsafety of the
1270 			 * underlying fs.  Tmpfs never bypasses.
1271 			 */
1272 			error = vget(vp, locktype, td);
1273 			if (error != 0)
1274 				return (error);
1275 		}
1276 		if (locktype == LK_EXCLUSIVE) {
1277 			*writecounted = TRUE;
1278 			vnode_pager_update_writecount(obj, 0, objsize);
1279 		}
1280 	} else {
1281 		error = EINVAL;
1282 		goto done;
1283 	}
1284 	if ((error = VOP_GETATTR(vp, &va, cred)))
1285 		goto done;
1286 #ifdef MAC
1287 	/* This relies on VM_PROT_* matching PROT_*. */
1288 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1289 	if (error != 0)
1290 		goto done;
1291 #endif
1292 	if ((flags & MAP_SHARED) != 0) {
1293 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1294 			if (prot & VM_PROT_WRITE) {
1295 				error = EPERM;
1296 				goto done;
1297 			}
1298 			*maxprotp &= ~VM_PROT_WRITE;
1299 		}
1300 	}
1301 	/*
1302 	 * If it is a regular file without any references
1303 	 * we do not need to sync it.
1304 	 * Adjust object size to be the size of actual file.
1305 	 */
1306 	objsize = round_page(va.va_size);
1307 	if (va.va_nlink == 0)
1308 		flags |= MAP_NOSYNC;
1309 	if (obj->type == OBJT_VNODE) {
1310 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1311 		    cred);
1312 		if (obj == NULL) {
1313 			error = ENOMEM;
1314 			goto done;
1315 		}
1316 	} else {
1317 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1318 		    ("wrong object type"));
1319 		VM_OBJECT_WLOCK(obj);
1320 		vm_object_reference_locked(obj);
1321 #if VM_NRESERVLEVEL > 0
1322 		vm_object_color(obj, 0);
1323 #endif
1324 		VM_OBJECT_WUNLOCK(obj);
1325 	}
1326 	*objp = obj;
1327 	*flagsp = flags;
1328 
1329 	vfs_mark_atime(vp, cred);
1330 
1331 done:
1332 	if (error != 0 && *writecounted) {
1333 		*writecounted = FALSE;
1334 		vnode_pager_update_writecount(obj, objsize, 0);
1335 	}
1336 	vput(vp);
1337 	return (error);
1338 }
1339 
1340 /*
1341  * vm_mmap_cdev()
1342  *
1343  * MPSAFE
1344  *
1345  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1346  * operations on cdevs.
1347  */
1348 int
1349 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1350     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1351     vm_ooffset_t *foff, vm_object_t *objp)
1352 {
1353 	vm_object_t obj;
1354 	int error, flags;
1355 
1356 	flags = *flagsp;
1357 
1358 	if (dsw->d_flags & D_MMAP_ANON) {
1359 		*objp = NULL;
1360 		*foff = 0;
1361 		*maxprotp = VM_PROT_ALL;
1362 		*flagsp |= MAP_ANON;
1363 		return (0);
1364 	}
1365 	/*
1366 	 * cdevs do not provide private mappings of any kind.
1367 	 */
1368 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1369 	    (prot & VM_PROT_WRITE) != 0)
1370 		return (EACCES);
1371 	if (flags & (MAP_PRIVATE|MAP_COPY))
1372 		return (EINVAL);
1373 	/*
1374 	 * Force device mappings to be shared.
1375 	 */
1376 	flags |= MAP_SHARED;
1377 #ifdef MAC_XXX
1378 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1379 	if (error != 0)
1380 		return (error);
1381 #endif
1382 	/*
1383 	 * First, try d_mmap_single().  If that is not implemented
1384 	 * (returns ENODEV), fall back to using the device pager.
1385 	 * Note that d_mmap_single() must return a reference to the
1386 	 * object (it needs to bump the reference count of the object
1387 	 * it returns somehow).
1388 	 *
1389 	 * XXX assumes VM_PROT_* == PROT_*
1390 	 */
1391 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1392 	if (error != ENODEV)
1393 		return (error);
1394 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1395 	    td->td_ucred);
1396 	if (obj == NULL)
1397 		return (EINVAL);
1398 	*objp = obj;
1399 	*flagsp = flags;
1400 	return (0);
1401 }
1402 
1403 /*
1404  * vm_mmap()
1405  *
1406  * Internal version of mmap used by exec, sys5 shared memory, and
1407  * various device drivers.  Handle is either a vnode pointer, a
1408  * character device, or NULL for MAP_ANON.
1409  */
1410 int
1411 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1412 	vm_prot_t maxprot, int flags,
1413 	objtype_t handle_type, void *handle,
1414 	vm_ooffset_t foff)
1415 {
1416 	vm_object_t object;
1417 	struct thread *td = curthread;
1418 	int error;
1419 	boolean_t writecounted;
1420 
1421 	if (size == 0)
1422 		return (EINVAL);
1423 
1424 	size = round_page(size);
1425 	object = NULL;
1426 	writecounted = FALSE;
1427 
1428 	/*
1429 	 * Lookup/allocate object.
1430 	 */
1431 	switch (handle_type) {
1432 	case OBJT_DEVICE: {
1433 		struct cdevsw *dsw;
1434 		struct cdev *cdev;
1435 		int ref;
1436 
1437 		cdev = handle;
1438 		dsw = dev_refthread(cdev, &ref);
1439 		if (dsw == NULL)
1440 			return (ENXIO);
1441 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1442 		    dsw, &foff, &object);
1443 		dev_relthread(cdev, ref);
1444 		break;
1445 	}
1446 	case OBJT_VNODE:
1447 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1448 		    handle, &foff, &object, &writecounted);
1449 		break;
1450 	case OBJT_DEFAULT:
1451 		if (handle == NULL) {
1452 			error = 0;
1453 			break;
1454 		}
1455 		/* FALLTHROUGH */
1456 	default:
1457 		error = EINVAL;
1458 		break;
1459 	}
1460 	if (error)
1461 		return (error);
1462 
1463 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1464 	    foff, writecounted, td);
1465 	if (error != 0 && object != NULL) {
1466 		/*
1467 		 * If this mapping was accounted for in the vnode's
1468 		 * writecount, then undo that now.
1469 		 */
1470 		if (writecounted)
1471 			vnode_pager_release_writecount(object, 0, size);
1472 		vm_object_deallocate(object);
1473 	}
1474 	return (error);
1475 }
1476 
1477 /*
1478  * Internal version of mmap that maps a specific VM object into an
1479  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1480  */
1481 int
1482 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1483     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1484     boolean_t writecounted, struct thread *td)
1485 {
1486 	boolean_t fitit;
1487 	int docow, error, findspace, rv;
1488 
1489 	if (map == &td->td_proc->p_vmspace->vm_map) {
1490 		PROC_LOCK(td->td_proc);
1491 		if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1492 			PROC_UNLOCK(td->td_proc);
1493 			return (ENOMEM);
1494 		}
1495 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1496 			PROC_UNLOCK(td->td_proc);
1497 			return (ENOMEM);
1498 		}
1499 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1500 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1501 			    lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1502 				racct_set_force(td->td_proc, RACCT_VMEM,
1503 				    map->size);
1504 				PROC_UNLOCK(td->td_proc);
1505 				return (ENOMEM);
1506 			}
1507 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1508 			    ptoa(pmap_wired_count(map->pmap)) + size);
1509 			if (error != 0) {
1510 				racct_set_force(td->td_proc, RACCT_VMEM,
1511 				    map->size);
1512 				PROC_UNLOCK(td->td_proc);
1513 				return (error);
1514 			}
1515 		}
1516 		PROC_UNLOCK(td->td_proc);
1517 	}
1518 
1519 	/*
1520 	 * We currently can only deal with page aligned file offsets.
1521 	 * The mmap() system call already enforces this by subtracting
1522 	 * the page offset from the file offset, but checking here
1523 	 * catches errors in device drivers (e.g. d_single_mmap()
1524 	 * callbacks) and other internal mapping requests (such as in
1525 	 * exec).
1526 	 */
1527 	if (foff & PAGE_MASK)
1528 		return (EINVAL);
1529 
1530 	if ((flags & MAP_FIXED) == 0) {
1531 		fitit = TRUE;
1532 		*addr = round_page(*addr);
1533 	} else {
1534 		if (*addr != trunc_page(*addr))
1535 			return (EINVAL);
1536 		fitit = FALSE;
1537 	}
1538 
1539 	if (flags & MAP_ANON) {
1540 		if (object != NULL || foff != 0)
1541 			return (EINVAL);
1542 		docow = 0;
1543 	} else if (flags & MAP_PREFAULT_READ)
1544 		docow = MAP_PREFAULT;
1545 	else
1546 		docow = MAP_PREFAULT_PARTIAL;
1547 
1548 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1549 		docow |= MAP_COPY_ON_WRITE;
1550 	if (flags & MAP_NOSYNC)
1551 		docow |= MAP_DISABLE_SYNCER;
1552 	if (flags & MAP_NOCORE)
1553 		docow |= MAP_DISABLE_COREDUMP;
1554 	/* Shared memory is also shared with children. */
1555 	if (flags & MAP_SHARED)
1556 		docow |= MAP_INHERIT_SHARE;
1557 	if (writecounted)
1558 		docow |= MAP_VN_WRITECOUNT;
1559 	if (flags & MAP_STACK) {
1560 		if (object != NULL)
1561 			return (EINVAL);
1562 		docow |= MAP_STACK_GROWS_DOWN;
1563 	}
1564 	if ((flags & MAP_EXCL) != 0)
1565 		docow |= MAP_CHECK_EXCL;
1566 
1567 	if (fitit) {
1568 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1569 			findspace = VMFS_SUPER_SPACE;
1570 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1571 			findspace = VMFS_ALIGNED_SPACE(flags >>
1572 			    MAP_ALIGNMENT_SHIFT);
1573 		else
1574 			findspace = VMFS_OPTIMAL_SPACE;
1575 		rv = vm_map_find(map, object, foff, addr, size,
1576 #ifdef MAP_32BIT
1577 		    flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1578 #endif
1579 		    0, findspace, prot, maxprot, docow);
1580 	} else {
1581 		rv = vm_map_fixed(map, object, foff, *addr, size,
1582 		    prot, maxprot, docow);
1583 	}
1584 
1585 	if (rv == KERN_SUCCESS) {
1586 		/*
1587 		 * If the process has requested that all future mappings
1588 		 * be wired, then heed this.
1589 		 */
1590 		if (map->flags & MAP_WIREFUTURE) {
1591 			vm_map_wire(map, *addr, *addr + size,
1592 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1593 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1594 		}
1595 	}
1596 	return (vm_mmap_to_errno(rv));
1597 }
1598 
1599 /*
1600  * Translate a Mach VM return code to zero on success or the appropriate errno
1601  * on failure.
1602  */
1603 int
1604 vm_mmap_to_errno(int rv)
1605 {
1606 
1607 	switch (rv) {
1608 	case KERN_SUCCESS:
1609 		return (0);
1610 	case KERN_INVALID_ADDRESS:
1611 	case KERN_NO_SPACE:
1612 		return (ENOMEM);
1613 	case KERN_PROTECTION_FAILURE:
1614 		return (EACCES);
1615 	default:
1616 		return (EINVAL);
1617 	}
1618 }
1619