xref: /freebsd/sys/vm/vm_mmap.c (revision eb69d1f144a6fcc765d1b9d44a5ae8082353e70b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1988 University of Utah.
5  * Copyright (c) 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
37  *
38  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
39  */
40 
41 /*
42  * Mapped file (mmap) interface to VM
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_compat.h"
49 #include "opt_hwpmc_hooks.h"
50 #include "opt_vm.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/capsicum.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/sysproto.h>
59 #include <sys/filedesc.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/procctl.h>
63 #include <sys/racct.h>
64 #include <sys/resource.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
69 #include <sys/fcntl.h>
70 #include <sys/file.h>
71 #include <sys/mman.h>
72 #include <sys/mount.h>
73 #include <sys/conf.h>
74 #include <sys/stat.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/vmmeter.h>
78 
79 #include <security/audit/audit.h>
80 #include <security/mac/mac_framework.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vm_pageout.h>
90 #include <vm/vm_extern.h>
91 #include <vm/vm_page.h>
92 #include <vm/vnode_pager.h>
93 
94 #ifdef HWPMC_HOOKS
95 #include <sys/pmckern.h>
96 #endif
97 
98 int old_mlock = 0;
99 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
100     "Do not apply RLIMIT_MEMLOCK on mlockall");
101 
102 #ifdef MAP_32BIT
103 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
104 #endif
105 
106 #ifndef _SYS_SYSPROTO_H_
107 struct sbrk_args {
108 	int incr;
109 };
110 #endif
111 
112 int
113 sys_sbrk(struct thread *td, struct sbrk_args *uap)
114 {
115 	/* Not yet implemented */
116 	return (EOPNOTSUPP);
117 }
118 
119 #ifndef _SYS_SYSPROTO_H_
120 struct sstk_args {
121 	int incr;
122 };
123 #endif
124 
125 int
126 sys_sstk(struct thread *td, struct sstk_args *uap)
127 {
128 	/* Not yet implemented */
129 	return (EOPNOTSUPP);
130 }
131 
132 #if defined(COMPAT_43)
133 #ifndef _SYS_SYSPROTO_H_
134 struct getpagesize_args {
135 	int dummy;
136 };
137 #endif
138 
139 int
140 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
141 {
142 
143 	td->td_retval[0] = PAGE_SIZE;
144 	return (0);
145 }
146 #endif				/* COMPAT_43 */
147 
148 
149 /*
150  * Memory Map (mmap) system call.  Note that the file offset
151  * and address are allowed to be NOT page aligned, though if
152  * the MAP_FIXED flag it set, both must have the same remainder
153  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
154  * page-aligned, the actual mapping starts at trunc_page(addr)
155  * and the return value is adjusted up by the page offset.
156  *
157  * Generally speaking, only character devices which are themselves
158  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
159  * there would be no cache coherency between a descriptor and a VM mapping
160  * both to the same character device.
161  */
162 #ifndef _SYS_SYSPROTO_H_
163 struct mmap_args {
164 	void *addr;
165 	size_t len;
166 	int prot;
167 	int flags;
168 	int fd;
169 	long pad;
170 	off_t pos;
171 };
172 #endif
173 
174 int
175 sys_mmap(struct thread *td, struct mmap_args *uap)
176 {
177 
178 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
179 	    uap->flags, uap->fd, uap->pos));
180 }
181 
182 int
183 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
184     int fd, off_t pos)
185 {
186 	struct vmspace *vms;
187 	struct file *fp;
188 	vm_offset_t addr;
189 	vm_size_t pageoff;
190 	vm_prot_t cap_maxprot;
191 	int align, error;
192 	cap_rights_t rights;
193 
194 	vms = td->td_proc->p_vmspace;
195 	fp = NULL;
196 	AUDIT_ARG_FD(fd);
197 	addr = addr0;
198 
199 	/*
200 	 * Ignore old flags that used to be defined but did not do anything.
201 	 */
202 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
203 
204 	/*
205 	 * Enforce the constraints.
206 	 * Mapping of length 0 is only allowed for old binaries.
207 	 * Anonymous mapping shall specify -1 as filedescriptor and
208 	 * zero position for new code. Be nice to ancient a.out
209 	 * binaries and correct pos for anonymous mapping, since old
210 	 * ld.so sometimes issues anonymous map requests with non-zero
211 	 * pos.
212 	 */
213 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
214 		if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
215 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
216 			return (EINVAL);
217 	} else {
218 		if ((flags & MAP_ANON) != 0)
219 			pos = 0;
220 	}
221 
222 	if (flags & MAP_STACK) {
223 		if ((fd != -1) ||
224 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
225 			return (EINVAL);
226 		flags |= MAP_ANON;
227 		pos = 0;
228 	}
229 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
230 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
231 	    MAP_PREFAULT_READ | MAP_GUARD |
232 #ifdef MAP_32BIT
233 	    MAP_32BIT |
234 #endif
235 	    MAP_ALIGNMENT_MASK)) != 0)
236 		return (EINVAL);
237 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
238 		return (EINVAL);
239 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
240 		return (EINVAL);
241 	if (prot != PROT_NONE &&
242 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
243 		return (EINVAL);
244 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
245 	    pos != 0 || (flags & (MAP_SHARED | MAP_PRIVATE | MAP_PREFAULT |
246 	    MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0))
247 		return (EINVAL);
248 
249 	/*
250 	 * Align the file position to a page boundary,
251 	 * and save its page offset component.
252 	 */
253 	pageoff = (pos & PAGE_MASK);
254 	pos -= pageoff;
255 
256 	/* Adjust size for rounding (on both ends). */
257 	size += pageoff;			/* low end... */
258 	size = (vm_size_t) round_page(size);	/* hi end */
259 
260 	/* Ensure alignment is at least a page and fits in a pointer. */
261 	align = flags & MAP_ALIGNMENT_MASK;
262 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
263 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
264 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
265 		return (EINVAL);
266 
267 	/*
268 	 * Check for illegal addresses.  Watch out for address wrap... Note
269 	 * that VM_*_ADDRESS are not constants due to casts (argh).
270 	 */
271 	if (flags & MAP_FIXED) {
272 		/*
273 		 * The specified address must have the same remainder
274 		 * as the file offset taken modulo PAGE_SIZE, so it
275 		 * should be aligned after adjustment by pageoff.
276 		 */
277 		addr -= pageoff;
278 		if (addr & PAGE_MASK)
279 			return (EINVAL);
280 
281 		/* Address range must be all in user VM space. */
282 		if (addr < vm_map_min(&vms->vm_map) ||
283 		    addr + size > vm_map_max(&vms->vm_map))
284 			return (EINVAL);
285 		if (addr + size < addr)
286 			return (EINVAL);
287 #ifdef MAP_32BIT
288 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
289 			return (EINVAL);
290 	} else if (flags & MAP_32BIT) {
291 		/*
292 		 * For MAP_32BIT, override the hint if it is too high and
293 		 * do not bother moving the mapping past the heap (since
294 		 * the heap is usually above 2GB).
295 		 */
296 		if (addr + size > MAP_32BIT_MAX_ADDR)
297 			addr = 0;
298 #endif
299 	} else {
300 		/*
301 		 * XXX for non-fixed mappings where no hint is provided or
302 		 * the hint would fall in the potential heap space,
303 		 * place it after the end of the largest possible heap.
304 		 *
305 		 * There should really be a pmap call to determine a reasonable
306 		 * location.
307 		 */
308 		if (addr == 0 ||
309 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
310 		    addr < round_page((vm_offset_t)vms->vm_daddr +
311 		    lim_max(td, RLIMIT_DATA))))
312 			addr = round_page((vm_offset_t)vms->vm_daddr +
313 			    lim_max(td, RLIMIT_DATA));
314 	}
315 	if (size == 0) {
316 		/*
317 		 * Return success without mapping anything for old
318 		 * binaries that request a page-aligned mapping of
319 		 * length 0.  For modern binaries, this function
320 		 * returns an error earlier.
321 		 */
322 		error = 0;
323 	} else if ((flags & MAP_GUARD) != 0) {
324 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
325 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
326 	} else if ((flags & MAP_ANON) != 0) {
327 		/*
328 		 * Mapping blank space is trivial.
329 		 *
330 		 * This relies on VM_PROT_* matching PROT_*.
331 		 */
332 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
333 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
334 	} else {
335 		/*
336 		 * Mapping file, get fp for validation and don't let the
337 		 * descriptor disappear on us if we block. Check capability
338 		 * rights, but also return the maximum rights to be combined
339 		 * with maxprot later.
340 		 */
341 		cap_rights_init(&rights, CAP_MMAP);
342 		if (prot & PROT_READ)
343 			cap_rights_set(&rights, CAP_MMAP_R);
344 		if ((flags & MAP_SHARED) != 0) {
345 			if (prot & PROT_WRITE)
346 				cap_rights_set(&rights, CAP_MMAP_W);
347 		}
348 		if (prot & PROT_EXEC)
349 			cap_rights_set(&rights, CAP_MMAP_X);
350 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
351 		if (error != 0)
352 			goto done;
353 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
354 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
355 			error = EINVAL;
356 			goto done;
357 		}
358 
359 		/* This relies on VM_PROT_* matching PROT_*. */
360 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
361 		    cap_maxprot, flags, pos, td);
362 	}
363 
364 	if (error == 0)
365 		td->td_retval[0] = (register_t) (addr + pageoff);
366 done:
367 	if (fp)
368 		fdrop(fp, td);
369 
370 	return (error);
371 }
372 
373 #if defined(COMPAT_FREEBSD6)
374 int
375 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
376 {
377 
378 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
379 	    uap->flags, uap->fd, uap->pos));
380 }
381 #endif
382 
383 #ifdef COMPAT_43
384 #ifndef _SYS_SYSPROTO_H_
385 struct ommap_args {
386 	caddr_t addr;
387 	int len;
388 	int prot;
389 	int flags;
390 	int fd;
391 	long pos;
392 };
393 #endif
394 int
395 ommap(struct thread *td, struct ommap_args *uap)
396 {
397 	static const char cvtbsdprot[8] = {
398 		0,
399 		PROT_EXEC,
400 		PROT_WRITE,
401 		PROT_EXEC | PROT_WRITE,
402 		PROT_READ,
403 		PROT_EXEC | PROT_READ,
404 		PROT_WRITE | PROT_READ,
405 		PROT_EXEC | PROT_WRITE | PROT_READ,
406 	};
407 	int flags, prot;
408 
409 #define	OMAP_ANON	0x0002
410 #define	OMAP_COPY	0x0020
411 #define	OMAP_SHARED	0x0010
412 #define	OMAP_FIXED	0x0100
413 
414 	prot = cvtbsdprot[uap->prot & 0x7];
415 #ifdef COMPAT_FREEBSD32
416 #if defined(__amd64__)
417 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
418 	    prot != 0)
419 		prot |= PROT_EXEC;
420 #endif
421 #endif
422 	flags = 0;
423 	if (uap->flags & OMAP_ANON)
424 		flags |= MAP_ANON;
425 	if (uap->flags & OMAP_COPY)
426 		flags |= MAP_COPY;
427 	if (uap->flags & OMAP_SHARED)
428 		flags |= MAP_SHARED;
429 	else
430 		flags |= MAP_PRIVATE;
431 	if (uap->flags & OMAP_FIXED)
432 		flags |= MAP_FIXED;
433 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
434 	    uap->fd, uap->pos));
435 }
436 #endif				/* COMPAT_43 */
437 
438 
439 #ifndef _SYS_SYSPROTO_H_
440 struct msync_args {
441 	void *addr;
442 	size_t len;
443 	int flags;
444 };
445 #endif
446 int
447 sys_msync(struct thread *td, struct msync_args *uap)
448 {
449 
450 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
451 }
452 
453 int
454 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
455 {
456 	vm_offset_t addr;
457 	vm_size_t pageoff;
458 	vm_map_t map;
459 	int rv;
460 
461 	addr = addr0;
462 	pageoff = (addr & PAGE_MASK);
463 	addr -= pageoff;
464 	size += pageoff;
465 	size = (vm_size_t) round_page(size);
466 	if (addr + size < addr)
467 		return (EINVAL);
468 
469 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
470 		return (EINVAL);
471 
472 	map = &td->td_proc->p_vmspace->vm_map;
473 
474 	/*
475 	 * Clean the pages and interpret the return value.
476 	 */
477 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
478 	    (flags & MS_INVALIDATE) != 0);
479 	switch (rv) {
480 	case KERN_SUCCESS:
481 		return (0);
482 	case KERN_INVALID_ADDRESS:
483 		return (ENOMEM);
484 	case KERN_INVALID_ARGUMENT:
485 		return (EBUSY);
486 	case KERN_FAILURE:
487 		return (EIO);
488 	default:
489 		return (EINVAL);
490 	}
491 }
492 
493 #ifndef _SYS_SYSPROTO_H_
494 struct munmap_args {
495 	void *addr;
496 	size_t len;
497 };
498 #endif
499 int
500 sys_munmap(struct thread *td, struct munmap_args *uap)
501 {
502 
503 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
504 }
505 
506 int
507 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
508 {
509 #ifdef HWPMC_HOOKS
510 	struct pmckern_map_out pkm;
511 	vm_map_entry_t entry;
512 	bool pmc_handled;
513 #endif
514 	vm_offset_t addr;
515 	vm_size_t pageoff;
516 	vm_map_t map;
517 
518 	if (size == 0)
519 		return (EINVAL);
520 
521 	addr = addr0;
522 	pageoff = (addr & PAGE_MASK);
523 	addr -= pageoff;
524 	size += pageoff;
525 	size = (vm_size_t) round_page(size);
526 	if (addr + size < addr)
527 		return (EINVAL);
528 
529 	/*
530 	 * Check for illegal addresses.  Watch out for address wrap...
531 	 */
532 	map = &td->td_proc->p_vmspace->vm_map;
533 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
534 		return (EINVAL);
535 	vm_map_lock(map);
536 #ifdef HWPMC_HOOKS
537 	pmc_handled = false;
538 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
539 		pmc_handled = true;
540 		/*
541 		 * Inform hwpmc if the address range being unmapped contains
542 		 * an executable region.
543 		 */
544 		pkm.pm_address = (uintptr_t) NULL;
545 		if (vm_map_lookup_entry(map, addr, &entry)) {
546 			for (; entry->start < addr + size;
547 			    entry = entry->next) {
548 				if (vm_map_check_protection(map, entry->start,
549 					entry->end, VM_PROT_EXECUTE) == TRUE) {
550 					pkm.pm_address = (uintptr_t) addr;
551 					pkm.pm_size = (size_t) size;
552 					break;
553 				}
554 			}
555 		}
556 	}
557 #endif
558 	vm_map_delete(map, addr, addr + size);
559 
560 #ifdef HWPMC_HOOKS
561 	if (__predict_false(pmc_handled)) {
562 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
563 		vm_map_lock_downgrade(map);
564 		if (pkm.pm_address != (uintptr_t) NULL)
565 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
566 		vm_map_unlock_read(map);
567 	} else
568 #endif
569 		vm_map_unlock(map);
570 
571 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
572 	return (0);
573 }
574 
575 #ifndef _SYS_SYSPROTO_H_
576 struct mprotect_args {
577 	const void *addr;
578 	size_t len;
579 	int prot;
580 };
581 #endif
582 int
583 sys_mprotect(struct thread *td, struct mprotect_args *uap)
584 {
585 
586 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
587 }
588 
589 int
590 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
591 {
592 	vm_offset_t addr;
593 	vm_size_t pageoff;
594 
595 	addr = addr0;
596 	prot = (prot & VM_PROT_ALL);
597 	pageoff = (addr & PAGE_MASK);
598 	addr -= pageoff;
599 	size += pageoff;
600 	size = (vm_size_t) round_page(size);
601 	if (addr + size < addr)
602 		return (EINVAL);
603 
604 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
605 	    addr + size, prot, FALSE)) {
606 	case KERN_SUCCESS:
607 		return (0);
608 	case KERN_PROTECTION_FAILURE:
609 		return (EACCES);
610 	case KERN_RESOURCE_SHORTAGE:
611 		return (ENOMEM);
612 	}
613 	return (EINVAL);
614 }
615 
616 #ifndef _SYS_SYSPROTO_H_
617 struct minherit_args {
618 	void *addr;
619 	size_t len;
620 	int inherit;
621 };
622 #endif
623 int
624 sys_minherit(struct thread *td, struct minherit_args *uap)
625 {
626 	vm_offset_t addr;
627 	vm_size_t size, pageoff;
628 	vm_inherit_t inherit;
629 
630 	addr = (vm_offset_t)uap->addr;
631 	size = uap->len;
632 	inherit = uap->inherit;
633 
634 	pageoff = (addr & PAGE_MASK);
635 	addr -= pageoff;
636 	size += pageoff;
637 	size = (vm_size_t) round_page(size);
638 	if (addr + size < addr)
639 		return (EINVAL);
640 
641 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
642 	    addr + size, inherit)) {
643 	case KERN_SUCCESS:
644 		return (0);
645 	case KERN_PROTECTION_FAILURE:
646 		return (EACCES);
647 	}
648 	return (EINVAL);
649 }
650 
651 #ifndef _SYS_SYSPROTO_H_
652 struct madvise_args {
653 	void *addr;
654 	size_t len;
655 	int behav;
656 };
657 #endif
658 
659 int
660 sys_madvise(struct thread *td, struct madvise_args *uap)
661 {
662 
663 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
664 }
665 
666 int
667 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
668 {
669 	vm_map_t map;
670 	vm_offset_t addr, end, start;
671 	int flags;
672 
673 	/*
674 	 * Check for our special case, advising the swap pager we are
675 	 * "immortal."
676 	 */
677 	if (behav == MADV_PROTECT) {
678 		flags = PPROT_SET;
679 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
680 		    PROC_SPROTECT, &flags));
681 	}
682 
683 	/*
684 	 * Check for illegal behavior
685 	 */
686 	if (behav < 0 || behav > MADV_CORE)
687 		return (EINVAL);
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 	if (vm_map_madvise(map, start, end, behav))
707 		return (EINVAL);
708 	return (0);
709 }
710 
711 #ifndef _SYS_SYSPROTO_H_
712 struct mincore_args {
713 	const void *addr;
714 	size_t len;
715 	char *vec;
716 };
717 #endif
718 
719 int
720 sys_mincore(struct thread *td, struct mincore_args *uap)
721 {
722 
723 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
724 }
725 
726 int
727 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
728 {
729 	vm_offset_t addr, first_addr;
730 	vm_offset_t end, cend;
731 	pmap_t pmap;
732 	vm_map_t map;
733 	int error = 0;
734 	int vecindex, lastvecindex;
735 	vm_map_entry_t current;
736 	vm_map_entry_t entry;
737 	vm_object_t object;
738 	vm_paddr_t locked_pa;
739 	vm_page_t m;
740 	vm_pindex_t pindex;
741 	int mincoreinfo;
742 	unsigned int timestamp;
743 	boolean_t locked;
744 
745 	/*
746 	 * Make sure that the addresses presented are valid for user
747 	 * mode.
748 	 */
749 	first_addr = addr = trunc_page(addr0);
750 	end = addr + (vm_size_t)round_page(len);
751 	map = &td->td_proc->p_vmspace->vm_map;
752 	if (end > vm_map_max(map) || end < addr)
753 		return (ENOMEM);
754 
755 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
756 
757 	vm_map_lock_read(map);
758 RestartScan:
759 	timestamp = map->timestamp;
760 
761 	if (!vm_map_lookup_entry(map, addr, &entry)) {
762 		vm_map_unlock_read(map);
763 		return (ENOMEM);
764 	}
765 
766 	/*
767 	 * Do this on a map entry basis so that if the pages are not
768 	 * in the current processes address space, we can easily look
769 	 * up the pages elsewhere.
770 	 */
771 	lastvecindex = -1;
772 	for (current = entry; current->start < end; current = current->next) {
773 
774 		/*
775 		 * check for contiguity
776 		 */
777 		if (current->end < end && current->next->start > current->end) {
778 			vm_map_unlock_read(map);
779 			return (ENOMEM);
780 		}
781 
782 		/*
783 		 * ignore submaps (for now) or null objects
784 		 */
785 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
786 			current->object.vm_object == NULL)
787 			continue;
788 
789 		/*
790 		 * limit this scan to the current map entry and the
791 		 * limits for the mincore call
792 		 */
793 		if (addr < current->start)
794 			addr = current->start;
795 		cend = current->end;
796 		if (cend > end)
797 			cend = end;
798 
799 		/*
800 		 * scan this entry one page at a time
801 		 */
802 		while (addr < cend) {
803 			/*
804 			 * Check pmap first, it is likely faster, also
805 			 * it can provide info as to whether we are the
806 			 * one referencing or modifying the page.
807 			 */
808 			object = NULL;
809 			locked_pa = 0;
810 		retry:
811 			m = NULL;
812 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
813 			if (locked_pa != 0) {
814 				/*
815 				 * The page is mapped by this process but not
816 				 * both accessed and modified.  It is also
817 				 * managed.  Acquire the object lock so that
818 				 * other mappings might be examined.
819 				 */
820 				m = PHYS_TO_VM_PAGE(locked_pa);
821 				if (m->object != object) {
822 					if (object != NULL)
823 						VM_OBJECT_WUNLOCK(object);
824 					object = m->object;
825 					locked = VM_OBJECT_TRYWLOCK(object);
826 					vm_page_unlock(m);
827 					if (!locked) {
828 						VM_OBJECT_WLOCK(object);
829 						vm_page_lock(m);
830 						goto retry;
831 					}
832 				} else
833 					vm_page_unlock(m);
834 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
835 				    ("mincore: page %p is mapped but invalid",
836 				    m));
837 			} else if (mincoreinfo == 0) {
838 				/*
839 				 * The page is not mapped by this process.  If
840 				 * the object implements managed pages, then
841 				 * determine if the page is resident so that
842 				 * the mappings might be examined.
843 				 */
844 				if (current->object.vm_object != object) {
845 					if (object != NULL)
846 						VM_OBJECT_WUNLOCK(object);
847 					object = current->object.vm_object;
848 					VM_OBJECT_WLOCK(object);
849 				}
850 				if (object->type == OBJT_DEFAULT ||
851 				    object->type == OBJT_SWAP ||
852 				    object->type == OBJT_VNODE) {
853 					pindex = OFF_TO_IDX(current->offset +
854 					    (addr - current->start));
855 					m = vm_page_lookup(object, pindex);
856 					if (m != NULL && m->valid == 0)
857 						m = NULL;
858 					if (m != NULL)
859 						mincoreinfo = MINCORE_INCORE;
860 				}
861 			}
862 			if (m != NULL) {
863 				/* Examine other mappings to the page. */
864 				if (m->dirty == 0 && pmap_is_modified(m))
865 					vm_page_dirty(m);
866 				if (m->dirty != 0)
867 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
868 				/*
869 				 * The first test for PGA_REFERENCED is an
870 				 * optimization.  The second test is
871 				 * required because a concurrent pmap
872 				 * operation could clear the last reference
873 				 * and set PGA_REFERENCED before the call to
874 				 * pmap_is_referenced().
875 				 */
876 				if ((m->aflags & PGA_REFERENCED) != 0 ||
877 				    pmap_is_referenced(m) ||
878 				    (m->aflags & PGA_REFERENCED) != 0)
879 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
880 			}
881 			if (object != NULL)
882 				VM_OBJECT_WUNLOCK(object);
883 
884 			/*
885 			 * subyte may page fault.  In case it needs to modify
886 			 * the map, we release the lock.
887 			 */
888 			vm_map_unlock_read(map);
889 
890 			/*
891 			 * calculate index into user supplied byte vector
892 			 */
893 			vecindex = atop(addr - first_addr);
894 
895 			/*
896 			 * If we have skipped map entries, we need to make sure that
897 			 * the byte vector is zeroed for those skipped entries.
898 			 */
899 			while ((lastvecindex + 1) < vecindex) {
900 				++lastvecindex;
901 				error = subyte(vec + lastvecindex, 0);
902 				if (error) {
903 					error = EFAULT;
904 					goto done2;
905 				}
906 			}
907 
908 			/*
909 			 * Pass the page information to the user
910 			 */
911 			error = subyte(vec + vecindex, mincoreinfo);
912 			if (error) {
913 				error = EFAULT;
914 				goto done2;
915 			}
916 
917 			/*
918 			 * If the map has changed, due to the subyte, the previous
919 			 * output may be invalid.
920 			 */
921 			vm_map_lock_read(map);
922 			if (timestamp != map->timestamp)
923 				goto RestartScan;
924 
925 			lastvecindex = vecindex;
926 			addr += PAGE_SIZE;
927 		}
928 	}
929 
930 	/*
931 	 * subyte may page fault.  In case it needs to modify
932 	 * the map, we release the lock.
933 	 */
934 	vm_map_unlock_read(map);
935 
936 	/*
937 	 * Zero the last entries in the byte vector.
938 	 */
939 	vecindex = atop(end - first_addr);
940 	while ((lastvecindex + 1) < vecindex) {
941 		++lastvecindex;
942 		error = subyte(vec + lastvecindex, 0);
943 		if (error) {
944 			error = EFAULT;
945 			goto done2;
946 		}
947 	}
948 
949 	/*
950 	 * If the map has changed, due to the subyte, the previous
951 	 * output may be invalid.
952 	 */
953 	vm_map_lock_read(map);
954 	if (timestamp != map->timestamp)
955 		goto RestartScan;
956 	vm_map_unlock_read(map);
957 done2:
958 	return (error);
959 }
960 
961 #ifndef _SYS_SYSPROTO_H_
962 struct mlock_args {
963 	const void *addr;
964 	size_t len;
965 };
966 #endif
967 int
968 sys_mlock(struct thread *td, struct mlock_args *uap)
969 {
970 
971 	return (kern_mlock(td->td_proc, td->td_ucred,
972 	    __DECONST(uintptr_t, uap->addr), uap->len));
973 }
974 
975 int
976 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
977 {
978 	vm_offset_t addr, end, last, start;
979 	vm_size_t npages, size;
980 	vm_map_t map;
981 	unsigned long nsize;
982 	int error;
983 
984 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
985 	if (error)
986 		return (error);
987 	addr = addr0;
988 	size = len;
989 	last = addr + size;
990 	start = trunc_page(addr);
991 	end = round_page(last);
992 	if (last < addr || end < addr)
993 		return (EINVAL);
994 	npages = atop(end - start);
995 	if (npages > vm_page_max_wired)
996 		return (ENOMEM);
997 	map = &proc->p_vmspace->vm_map;
998 	PROC_LOCK(proc);
999 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1000 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1001 		PROC_UNLOCK(proc);
1002 		return (ENOMEM);
1003 	}
1004 	PROC_UNLOCK(proc);
1005 	if (npages + vm_wire_count() > vm_page_max_wired)
1006 		return (EAGAIN);
1007 #ifdef RACCT
1008 	if (racct_enable) {
1009 		PROC_LOCK(proc);
1010 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1011 		PROC_UNLOCK(proc);
1012 		if (error != 0)
1013 			return (ENOMEM);
1014 	}
1015 #endif
1016 	error = vm_map_wire(map, start, end,
1017 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1018 #ifdef RACCT
1019 	if (racct_enable && error != KERN_SUCCESS) {
1020 		PROC_LOCK(proc);
1021 		racct_set(proc, RACCT_MEMLOCK,
1022 		    ptoa(pmap_wired_count(map->pmap)));
1023 		PROC_UNLOCK(proc);
1024 	}
1025 #endif
1026 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1027 }
1028 
1029 #ifndef _SYS_SYSPROTO_H_
1030 struct mlockall_args {
1031 	int	how;
1032 };
1033 #endif
1034 
1035 int
1036 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1037 {
1038 	vm_map_t map;
1039 	int error;
1040 
1041 	map = &td->td_proc->p_vmspace->vm_map;
1042 	error = priv_check(td, PRIV_VM_MLOCK);
1043 	if (error)
1044 		return (error);
1045 
1046 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1047 		return (EINVAL);
1048 
1049 	/*
1050 	 * If wiring all pages in the process would cause it to exceed
1051 	 * a hard resource limit, return ENOMEM.
1052 	 */
1053 	if (!old_mlock && uap->how & MCL_CURRENT) {
1054 		PROC_LOCK(td->td_proc);
1055 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1056 			PROC_UNLOCK(td->td_proc);
1057 			return (ENOMEM);
1058 		}
1059 		PROC_UNLOCK(td->td_proc);
1060 	}
1061 #ifdef RACCT
1062 	if (racct_enable) {
1063 		PROC_LOCK(td->td_proc);
1064 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1065 		PROC_UNLOCK(td->td_proc);
1066 		if (error != 0)
1067 			return (ENOMEM);
1068 	}
1069 #endif
1070 
1071 	if (uap->how & MCL_FUTURE) {
1072 		vm_map_lock(map);
1073 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1074 		vm_map_unlock(map);
1075 		error = 0;
1076 	}
1077 
1078 	if (uap->how & MCL_CURRENT) {
1079 		/*
1080 		 * P1003.1-2001 mandates that all currently mapped pages
1081 		 * will be memory resident and locked (wired) upon return
1082 		 * from mlockall(). vm_map_wire() will wire pages, by
1083 		 * calling vm_fault_wire() for each page in the region.
1084 		 */
1085 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1086 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1087 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1088 	}
1089 #ifdef RACCT
1090 	if (racct_enable && error != KERN_SUCCESS) {
1091 		PROC_LOCK(td->td_proc);
1092 		racct_set(td->td_proc, RACCT_MEMLOCK,
1093 		    ptoa(pmap_wired_count(map->pmap)));
1094 		PROC_UNLOCK(td->td_proc);
1095 	}
1096 #endif
1097 
1098 	return (error);
1099 }
1100 
1101 #ifndef _SYS_SYSPROTO_H_
1102 struct munlockall_args {
1103 	register_t dummy;
1104 };
1105 #endif
1106 
1107 int
1108 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1109 {
1110 	vm_map_t map;
1111 	int error;
1112 
1113 	map = &td->td_proc->p_vmspace->vm_map;
1114 	error = priv_check(td, PRIV_VM_MUNLOCK);
1115 	if (error)
1116 		return (error);
1117 
1118 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1119 	vm_map_lock(map);
1120 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1121 	vm_map_unlock(map);
1122 
1123 	/* Forcibly unwire all pages. */
1124 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1125 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1126 #ifdef RACCT
1127 	if (racct_enable && error == KERN_SUCCESS) {
1128 		PROC_LOCK(td->td_proc);
1129 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1130 		PROC_UNLOCK(td->td_proc);
1131 	}
1132 #endif
1133 
1134 	return (error);
1135 }
1136 
1137 #ifndef _SYS_SYSPROTO_H_
1138 struct munlock_args {
1139 	const void *addr;
1140 	size_t len;
1141 };
1142 #endif
1143 int
1144 sys_munlock(struct thread *td, struct munlock_args *uap)
1145 {
1146 
1147 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1148 }
1149 
1150 int
1151 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1152 {
1153 	vm_offset_t addr, end, last, start;
1154 #ifdef RACCT
1155 	vm_map_t map;
1156 #endif
1157 	int error;
1158 
1159 	error = priv_check(td, PRIV_VM_MUNLOCK);
1160 	if (error)
1161 		return (error);
1162 	addr = addr0;
1163 	last = addr + size;
1164 	start = trunc_page(addr);
1165 	end = round_page(last);
1166 	if (last < addr || end < addr)
1167 		return (EINVAL);
1168 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1169 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1170 #ifdef RACCT
1171 	if (racct_enable && error == KERN_SUCCESS) {
1172 		PROC_LOCK(td->td_proc);
1173 		map = &td->td_proc->p_vmspace->vm_map;
1174 		racct_set(td->td_proc, RACCT_MEMLOCK,
1175 		    ptoa(pmap_wired_count(map->pmap)));
1176 		PROC_UNLOCK(td->td_proc);
1177 	}
1178 #endif
1179 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1180 }
1181 
1182 /*
1183  * vm_mmap_vnode()
1184  *
1185  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1186  * operations on vnodes.
1187  */
1188 int
1189 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1190     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1191     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1192     boolean_t *writecounted)
1193 {
1194 	struct vattr va;
1195 	vm_object_t obj;
1196 	vm_ooffset_t foff;
1197 	struct ucred *cred;
1198 	int error, flags, locktype;
1199 
1200 	cred = td->td_ucred;
1201 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1202 		locktype = LK_EXCLUSIVE;
1203 	else
1204 		locktype = LK_SHARED;
1205 	if ((error = vget(vp, locktype, td)) != 0)
1206 		return (error);
1207 	AUDIT_ARG_VNODE1(vp);
1208 	foff = *foffp;
1209 	flags = *flagsp;
1210 	obj = vp->v_object;
1211 	if (vp->v_type == VREG) {
1212 		/*
1213 		 * Get the proper underlying object
1214 		 */
1215 		if (obj == NULL) {
1216 			error = EINVAL;
1217 			goto done;
1218 		}
1219 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1220 			vput(vp);
1221 			vp = (struct vnode *)obj->handle;
1222 			/*
1223 			 * Bypass filesystems obey the mpsafety of the
1224 			 * underlying fs.  Tmpfs never bypasses.
1225 			 */
1226 			error = vget(vp, locktype, td);
1227 			if (error != 0)
1228 				return (error);
1229 		}
1230 		if (locktype == LK_EXCLUSIVE) {
1231 			*writecounted = TRUE;
1232 			vnode_pager_update_writecount(obj, 0, objsize);
1233 		}
1234 	} else {
1235 		error = EINVAL;
1236 		goto done;
1237 	}
1238 	if ((error = VOP_GETATTR(vp, &va, cred)))
1239 		goto done;
1240 #ifdef MAC
1241 	/* This relies on VM_PROT_* matching PROT_*. */
1242 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1243 	if (error != 0)
1244 		goto done;
1245 #endif
1246 	if ((flags & MAP_SHARED) != 0) {
1247 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1248 			if (prot & VM_PROT_WRITE) {
1249 				error = EPERM;
1250 				goto done;
1251 			}
1252 			*maxprotp &= ~VM_PROT_WRITE;
1253 		}
1254 	}
1255 	/*
1256 	 * If it is a regular file without any references
1257 	 * we do not need to sync it.
1258 	 * Adjust object size to be the size of actual file.
1259 	 */
1260 	objsize = round_page(va.va_size);
1261 	if (va.va_nlink == 0)
1262 		flags |= MAP_NOSYNC;
1263 	if (obj->type == OBJT_VNODE) {
1264 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1265 		    cred);
1266 		if (obj == NULL) {
1267 			error = ENOMEM;
1268 			goto done;
1269 		}
1270 	} else {
1271 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1272 		    ("wrong object type"));
1273 		VM_OBJECT_WLOCK(obj);
1274 		vm_object_reference_locked(obj);
1275 #if VM_NRESERVLEVEL > 0
1276 		vm_object_color(obj, 0);
1277 #endif
1278 		VM_OBJECT_WUNLOCK(obj);
1279 	}
1280 	*objp = obj;
1281 	*flagsp = flags;
1282 
1283 	vfs_mark_atime(vp, cred);
1284 
1285 done:
1286 	if (error != 0 && *writecounted) {
1287 		*writecounted = FALSE;
1288 		vnode_pager_update_writecount(obj, objsize, 0);
1289 	}
1290 	vput(vp);
1291 	return (error);
1292 }
1293 
1294 /*
1295  * vm_mmap_cdev()
1296  *
1297  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1298  * operations on cdevs.
1299  */
1300 int
1301 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1302     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1303     vm_ooffset_t *foff, vm_object_t *objp)
1304 {
1305 	vm_object_t obj;
1306 	int error, flags;
1307 
1308 	flags = *flagsp;
1309 
1310 	if (dsw->d_flags & D_MMAP_ANON) {
1311 		*objp = NULL;
1312 		*foff = 0;
1313 		*maxprotp = VM_PROT_ALL;
1314 		*flagsp |= MAP_ANON;
1315 		return (0);
1316 	}
1317 	/*
1318 	 * cdevs do not provide private mappings of any kind.
1319 	 */
1320 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1321 	    (prot & VM_PROT_WRITE) != 0)
1322 		return (EACCES);
1323 	if (flags & (MAP_PRIVATE|MAP_COPY))
1324 		return (EINVAL);
1325 	/*
1326 	 * Force device mappings to be shared.
1327 	 */
1328 	flags |= MAP_SHARED;
1329 #ifdef MAC_XXX
1330 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1331 	if (error != 0)
1332 		return (error);
1333 #endif
1334 	/*
1335 	 * First, try d_mmap_single().  If that is not implemented
1336 	 * (returns ENODEV), fall back to using the device pager.
1337 	 * Note that d_mmap_single() must return a reference to the
1338 	 * object (it needs to bump the reference count of the object
1339 	 * it returns somehow).
1340 	 *
1341 	 * XXX assumes VM_PROT_* == PROT_*
1342 	 */
1343 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1344 	if (error != ENODEV)
1345 		return (error);
1346 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1347 	    td->td_ucred);
1348 	if (obj == NULL)
1349 		return (EINVAL);
1350 	*objp = obj;
1351 	*flagsp = flags;
1352 	return (0);
1353 }
1354 
1355 /*
1356  * vm_mmap()
1357  *
1358  * Internal version of mmap used by exec, sys5 shared memory, and
1359  * various device drivers.  Handle is either a vnode pointer, a
1360  * character device, or NULL for MAP_ANON.
1361  */
1362 int
1363 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1364 	vm_prot_t maxprot, int flags,
1365 	objtype_t handle_type, void *handle,
1366 	vm_ooffset_t foff)
1367 {
1368 	vm_object_t object;
1369 	struct thread *td = curthread;
1370 	int error;
1371 	boolean_t writecounted;
1372 
1373 	if (size == 0)
1374 		return (EINVAL);
1375 
1376 	size = round_page(size);
1377 	object = NULL;
1378 	writecounted = FALSE;
1379 
1380 	/*
1381 	 * Lookup/allocate object.
1382 	 */
1383 	switch (handle_type) {
1384 	case OBJT_DEVICE: {
1385 		struct cdevsw *dsw;
1386 		struct cdev *cdev;
1387 		int ref;
1388 
1389 		cdev = handle;
1390 		dsw = dev_refthread(cdev, &ref);
1391 		if (dsw == NULL)
1392 			return (ENXIO);
1393 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1394 		    dsw, &foff, &object);
1395 		dev_relthread(cdev, ref);
1396 		break;
1397 	}
1398 	case OBJT_VNODE:
1399 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1400 		    handle, &foff, &object, &writecounted);
1401 		break;
1402 	case OBJT_DEFAULT:
1403 		if (handle == NULL) {
1404 			error = 0;
1405 			break;
1406 		}
1407 		/* FALLTHROUGH */
1408 	default:
1409 		error = EINVAL;
1410 		break;
1411 	}
1412 	if (error)
1413 		return (error);
1414 
1415 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1416 	    foff, writecounted, td);
1417 	if (error != 0 && object != NULL) {
1418 		/*
1419 		 * If this mapping was accounted for in the vnode's
1420 		 * writecount, then undo that now.
1421 		 */
1422 		if (writecounted)
1423 			vnode_pager_release_writecount(object, 0, size);
1424 		vm_object_deallocate(object);
1425 	}
1426 	return (error);
1427 }
1428 
1429 /*
1430  * Internal version of mmap that maps a specific VM object into an
1431  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1432  */
1433 int
1434 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1435     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1436     boolean_t writecounted, struct thread *td)
1437 {
1438 	boolean_t curmap, fitit;
1439 	vm_offset_t max_addr;
1440 	int docow, error, findspace, rv;
1441 
1442 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1443 	if (curmap) {
1444 		PROC_LOCK(td->td_proc);
1445 		if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1446 			PROC_UNLOCK(td->td_proc);
1447 			return (ENOMEM);
1448 		}
1449 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1450 			PROC_UNLOCK(td->td_proc);
1451 			return (ENOMEM);
1452 		}
1453 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1454 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1455 			    lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1456 				racct_set_force(td->td_proc, RACCT_VMEM,
1457 				    map->size);
1458 				PROC_UNLOCK(td->td_proc);
1459 				return (ENOMEM);
1460 			}
1461 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1462 			    ptoa(pmap_wired_count(map->pmap)) + size);
1463 			if (error != 0) {
1464 				racct_set_force(td->td_proc, RACCT_VMEM,
1465 				    map->size);
1466 				PROC_UNLOCK(td->td_proc);
1467 				return (error);
1468 			}
1469 		}
1470 		PROC_UNLOCK(td->td_proc);
1471 	}
1472 
1473 	/*
1474 	 * We currently can only deal with page aligned file offsets.
1475 	 * The mmap() system call already enforces this by subtracting
1476 	 * the page offset from the file offset, but checking here
1477 	 * catches errors in device drivers (e.g. d_single_mmap()
1478 	 * callbacks) and other internal mapping requests (such as in
1479 	 * exec).
1480 	 */
1481 	if (foff & PAGE_MASK)
1482 		return (EINVAL);
1483 
1484 	if ((flags & MAP_FIXED) == 0) {
1485 		fitit = TRUE;
1486 		*addr = round_page(*addr);
1487 	} else {
1488 		if (*addr != trunc_page(*addr))
1489 			return (EINVAL);
1490 		fitit = FALSE;
1491 	}
1492 
1493 	if (flags & MAP_ANON) {
1494 		if (object != NULL || foff != 0)
1495 			return (EINVAL);
1496 		docow = 0;
1497 	} else if (flags & MAP_PREFAULT_READ)
1498 		docow = MAP_PREFAULT;
1499 	else
1500 		docow = MAP_PREFAULT_PARTIAL;
1501 
1502 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1503 		docow |= MAP_COPY_ON_WRITE;
1504 	if (flags & MAP_NOSYNC)
1505 		docow |= MAP_DISABLE_SYNCER;
1506 	if (flags & MAP_NOCORE)
1507 		docow |= MAP_DISABLE_COREDUMP;
1508 	/* Shared memory is also shared with children. */
1509 	if (flags & MAP_SHARED)
1510 		docow |= MAP_INHERIT_SHARE;
1511 	if (writecounted)
1512 		docow |= MAP_VN_WRITECOUNT;
1513 	if (flags & MAP_STACK) {
1514 		if (object != NULL)
1515 			return (EINVAL);
1516 		docow |= MAP_STACK_GROWS_DOWN;
1517 	}
1518 	if ((flags & MAP_EXCL) != 0)
1519 		docow |= MAP_CHECK_EXCL;
1520 	if ((flags & MAP_GUARD) != 0)
1521 		docow |= MAP_CREATE_GUARD;
1522 
1523 	if (fitit) {
1524 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1525 			findspace = VMFS_SUPER_SPACE;
1526 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1527 			findspace = VMFS_ALIGNED_SPACE(flags >>
1528 			    MAP_ALIGNMENT_SHIFT);
1529 		else
1530 			findspace = VMFS_OPTIMAL_SPACE;
1531 		max_addr = 0;
1532 #ifdef MAP_32BIT
1533 		if ((flags & MAP_32BIT) != 0)
1534 			max_addr = MAP_32BIT_MAX_ADDR;
1535 #endif
1536 		if (curmap) {
1537 			rv = vm_map_find_min(map, object, foff, addr, size,
1538 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1539 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1540 			    findspace, prot, maxprot, docow);
1541 		} else {
1542 			rv = vm_map_find(map, object, foff, addr, size,
1543 			    max_addr, findspace, prot, maxprot, docow);
1544 		}
1545 	} else {
1546 		rv = vm_map_fixed(map, object, foff, *addr, size,
1547 		    prot, maxprot, docow);
1548 	}
1549 
1550 	if (rv == KERN_SUCCESS) {
1551 		/*
1552 		 * If the process has requested that all future mappings
1553 		 * be wired, then heed this.
1554 		 */
1555 		if (map->flags & MAP_WIREFUTURE) {
1556 			vm_map_wire(map, *addr, *addr + size,
1557 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1558 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1559 		}
1560 	}
1561 	return (vm_mmap_to_errno(rv));
1562 }
1563 
1564 /*
1565  * Translate a Mach VM return code to zero on success or the appropriate errno
1566  * on failure.
1567  */
1568 int
1569 vm_mmap_to_errno(int rv)
1570 {
1571 
1572 	switch (rv) {
1573 	case KERN_SUCCESS:
1574 		return (0);
1575 	case KERN_INVALID_ADDRESS:
1576 	case KERN_NO_SPACE:
1577 		return (ENOMEM);
1578 	case KERN_PROTECTION_FAILURE:
1579 		return (EACCES);
1580 	default:
1581 		return (EINVAL);
1582 	}
1583 }
1584