xref: /freebsd/sys/vm/vm_mmap.c (revision a223d3ed90bfe313ce5987d468a25a915d7d1254)
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 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capsicum.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/procctl.h>
60 #include <sys/racct.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/rwlock.h>
64 #include <sys/sysctl.h>
65 #include <sys/vnode.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/mman.h>
69 #include <sys/mount.h>
70 #include <sys/conf.h>
71 #include <sys/stat.h>
72 #include <sys/syscallsubr.h>
73 #include <sys/sysent.h>
74 #include <sys/vmmeter.h>
75 
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pager.h>
85 #include <vm/vm_pageout.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_page.h>
88 #include <vm/vnode_pager.h>
89 
90 #ifdef HWPMC_HOOKS
91 #include <sys/pmckern.h>
92 #endif
93 
94 int old_mlock = 0;
95 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RW | CTLFLAG_TUN, &old_mlock, 0,
96     "Do not apply RLIMIT_MEMLOCK on mlockall");
97 TUNABLE_INT("vm.old_mlock", &old_mlock);
98 
99 #ifdef MAP_32BIT
100 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
101 #endif
102 
103 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
104     int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
105 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
106     int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
107 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
108     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
109 
110 #ifndef _SYS_SYSPROTO_H_
111 struct sbrk_args {
112 	int incr;
113 };
114 #endif
115 
116 /*
117  * MPSAFE
118  */
119 /* ARGSUSED */
120 int
121 sys_sbrk(td, uap)
122 	struct thread *td;
123 	struct sbrk_args *uap;
124 {
125 	/* Not yet implemented */
126 	return (EOPNOTSUPP);
127 }
128 
129 #ifndef _SYS_SYSPROTO_H_
130 struct sstk_args {
131 	int incr;
132 };
133 #endif
134 
135 /*
136  * MPSAFE
137  */
138 /* ARGSUSED */
139 int
140 sys_sstk(td, uap)
141 	struct thread *td;
142 	struct sstk_args *uap;
143 {
144 	/* Not yet implemented */
145 	return (EOPNOTSUPP);
146 }
147 
148 #if defined(COMPAT_43)
149 #ifndef _SYS_SYSPROTO_H_
150 struct getpagesize_args {
151 	int dummy;
152 };
153 #endif
154 
155 int
156 ogetpagesize(td, uap)
157 	struct thread *td;
158 	struct getpagesize_args *uap;
159 {
160 	/* MP SAFE */
161 	td->td_retval[0] = PAGE_SIZE;
162 	return (0);
163 }
164 #endif				/* COMPAT_43 */
165 
166 
167 /*
168  * Memory Map (mmap) system call.  Note that the file offset
169  * and address are allowed to be NOT page aligned, though if
170  * the MAP_FIXED flag it set, both must have the same remainder
171  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
172  * page-aligned, the actual mapping starts at trunc_page(addr)
173  * and the return value is adjusted up by the page offset.
174  *
175  * Generally speaking, only character devices which are themselves
176  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
177  * there would be no cache coherency between a descriptor and a VM mapping
178  * both to the same character device.
179  */
180 #ifndef _SYS_SYSPROTO_H_
181 struct mmap_args {
182 	void *addr;
183 	size_t len;
184 	int prot;
185 	int flags;
186 	int fd;
187 	long pad;
188 	off_t pos;
189 };
190 #endif
191 
192 /*
193  * MPSAFE
194  */
195 int
196 sys_mmap(td, uap)
197 	struct thread *td;
198 	struct mmap_args *uap;
199 {
200 #ifdef HWPMC_HOOKS
201 	struct pmckern_map_in pkm;
202 #endif
203 	struct file *fp;
204 	struct vnode *vp;
205 	vm_offset_t addr;
206 	vm_size_t size, pageoff;
207 	vm_prot_t cap_maxprot, prot, maxprot;
208 	void *handle;
209 	objtype_t handle_type;
210 	int align, error, flags;
211 	off_t pos;
212 	struct vmspace *vms = td->td_proc->p_vmspace;
213 	cap_rights_t rights;
214 
215 	addr = (vm_offset_t) uap->addr;
216 	size = uap->len;
217 	prot = uap->prot & VM_PROT_ALL;
218 	flags = uap->flags;
219 	pos = uap->pos;
220 
221 	fp = NULL;
222 
223 	/*
224 	 * Enforce the constraints.
225 	 * Mapping of length 0 is only allowed for old binaries.
226 	 * Anonymous mapping shall specify -1 as filedescriptor and
227 	 * zero position for new code. Be nice to ancient a.out
228 	 * binaries and correct pos for anonymous mapping, since old
229 	 * ld.so sometimes issues anonymous map requests with non-zero
230 	 * pos.
231 	 */
232 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
233 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
234 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
235 			return (EINVAL);
236 	} else {
237 		if ((flags & MAP_ANON) != 0)
238 			pos = 0;
239 	}
240 
241 	if (flags & MAP_STACK) {
242 		if ((uap->fd != -1) ||
243 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
244 			return (EINVAL);
245 		flags |= MAP_ANON;
246 		pos = 0;
247 	}
248 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
249 		return (EINVAL);
250 
251 	/*
252 	 * Align the file position to a page boundary,
253 	 * and save its page offset component.
254 	 */
255 	pageoff = (pos & PAGE_MASK);
256 	pos -= pageoff;
257 
258 	/* Adjust size for rounding (on both ends). */
259 	size += pageoff;			/* low end... */
260 	size = (vm_size_t) round_page(size);	/* hi end */
261 
262 	/* Ensure alignment is at least a page and fits in a pointer. */
263 	align = flags & MAP_ALIGNMENT_MASK;
264 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
265 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
266 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
267 		return (EINVAL);
268 
269 	/*
270 	 * Check for illegal addresses.  Watch out for address wrap... Note
271 	 * that VM_*_ADDRESS are not constants due to casts (argh).
272 	 */
273 	if (flags & MAP_FIXED) {
274 		/*
275 		 * The specified address must have the same remainder
276 		 * as the file offset taken modulo PAGE_SIZE, so it
277 		 * should be aligned after adjustment by pageoff.
278 		 */
279 		addr -= pageoff;
280 		if (addr & PAGE_MASK)
281 			return (EINVAL);
282 
283 		/* Address range must be all in user VM space. */
284 		if (addr < vm_map_min(&vms->vm_map) ||
285 		    addr + size > vm_map_max(&vms->vm_map))
286 			return (EINVAL);
287 		if (addr + size < addr)
288 			return (EINVAL);
289 #ifdef MAP_32BIT
290 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
291 			return (EINVAL);
292 	} else if (flags & MAP_32BIT) {
293 		/*
294 		 * For MAP_32BIT, override the hint if it is too high and
295 		 * do not bother moving the mapping past the heap (since
296 		 * the heap is usually above 2GB).
297 		 */
298 		if (addr + size > MAP_32BIT_MAX_ADDR)
299 			addr = 0;
300 #endif
301 	} else {
302 		/*
303 		 * XXX for non-fixed mappings where no hint is provided or
304 		 * the hint would fall in the potential heap space,
305 		 * place it after the end of the largest possible heap.
306 		 *
307 		 * There should really be a pmap call to determine a reasonable
308 		 * location.
309 		 */
310 		PROC_LOCK(td->td_proc);
311 		if (addr == 0 ||
312 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
313 		    addr < round_page((vm_offset_t)vms->vm_daddr +
314 		    lim_max(td->td_proc, RLIMIT_DATA))))
315 			addr = round_page((vm_offset_t)vms->vm_daddr +
316 			    lim_max(td->td_proc, RLIMIT_DATA));
317 		PROC_UNLOCK(td->td_proc);
318 	}
319 	if (flags & MAP_ANON) {
320 		/*
321 		 * Mapping blank space is trivial.
322 		 */
323 		handle = NULL;
324 		handle_type = OBJT_DEFAULT;
325 		maxprot = VM_PROT_ALL;
326 		cap_maxprot = VM_PROT_ALL;
327 	} else {
328 		/*
329 		 * Mapping file, get fp for validation and don't let the
330 		 * descriptor disappear on us if we block. Check capability
331 		 * rights, but also return the maximum rights to be combined
332 		 * with maxprot later.
333 		 */
334 		cap_rights_init(&rights, CAP_MMAP);
335 		if (prot & PROT_READ)
336 			cap_rights_set(&rights, CAP_MMAP_R);
337 		if ((flags & MAP_SHARED) != 0) {
338 			if (prot & PROT_WRITE)
339 				cap_rights_set(&rights, CAP_MMAP_W);
340 		}
341 		if (prot & PROT_EXEC)
342 			cap_rights_set(&rights, CAP_MMAP_X);
343 		error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
344 		if (error != 0)
345 			goto done;
346 		if (fp->f_type == DTYPE_SHM) {
347 			handle = fp->f_data;
348 			handle_type = OBJT_SWAP;
349 			maxprot = VM_PROT_NONE;
350 
351 			/* FREAD should always be set. */
352 			if (fp->f_flag & FREAD)
353 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
354 			if (fp->f_flag & FWRITE)
355 				maxprot |= VM_PROT_WRITE;
356 			goto map;
357 		}
358 		if (fp->f_type != DTYPE_VNODE) {
359 			error = ENODEV;
360 			goto done;
361 		}
362 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
363     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
364 		/*
365 		 * POSIX shared-memory objects are defined to have
366 		 * kernel persistence, and are not defined to support
367 		 * read(2)/write(2) -- or even open(2).  Thus, we can
368 		 * use MAP_ASYNC to trade on-disk coherence for speed.
369 		 * The shm_open(3) library routine turns on the FPOSIXSHM
370 		 * flag to request this behavior.
371 		 */
372 		if (fp->f_flag & FPOSIXSHM)
373 			flags |= MAP_NOSYNC;
374 #endif
375 		vp = fp->f_vnode;
376 		/*
377 		 * Ensure that file and memory protections are
378 		 * compatible.  Note that we only worry about
379 		 * writability if mapping is shared; in this case,
380 		 * current and max prot are dictated by the open file.
381 		 * XXX use the vnode instead?  Problem is: what
382 		 * credentials do we use for determination? What if
383 		 * proc does a setuid?
384 		 */
385 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
386 			maxprot = VM_PROT_NONE;
387 		else
388 			maxprot = VM_PROT_EXECUTE;
389 		if (fp->f_flag & FREAD) {
390 			maxprot |= VM_PROT_READ;
391 		} else if (prot & PROT_READ) {
392 			error = EACCES;
393 			goto done;
394 		}
395 		/*
396 		 * If we are sharing potential changes (either via
397 		 * MAP_SHARED or via the implicit sharing of character
398 		 * device mappings), and we are trying to get write
399 		 * permission although we opened it without asking
400 		 * for it, bail out.
401 		 */
402 		if ((flags & MAP_SHARED) != 0) {
403 			if ((fp->f_flag & FWRITE) != 0) {
404 				maxprot |= VM_PROT_WRITE;
405 			} else if ((prot & PROT_WRITE) != 0) {
406 				error = EACCES;
407 				goto done;
408 			}
409 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
410 			maxprot |= VM_PROT_WRITE;
411 			cap_maxprot |= VM_PROT_WRITE;
412 		}
413 		handle = (void *)vp;
414 		handle_type = OBJT_VNODE;
415 	}
416 map:
417 	td->td_fpop = fp;
418 	maxprot &= cap_maxprot;
419 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
420 	    flags, handle_type, handle, pos);
421 	td->td_fpop = NULL;
422 #ifdef HWPMC_HOOKS
423 	/* inform hwpmc(4) if an executable is being mapped */
424 	if (error == 0 && handle_type == OBJT_VNODE &&
425 	    (prot & PROT_EXEC)) {
426 		pkm.pm_file = handle;
427 		pkm.pm_address = (uintptr_t) addr;
428 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
429 	}
430 #endif
431 	if (error == 0)
432 		td->td_retval[0] = (register_t) (addr + pageoff);
433 done:
434 	if (fp)
435 		fdrop(fp, td);
436 
437 	return (error);
438 }
439 
440 int
441 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
442 {
443 	struct mmap_args oargs;
444 
445 	oargs.addr = uap->addr;
446 	oargs.len = uap->len;
447 	oargs.prot = uap->prot;
448 	oargs.flags = uap->flags;
449 	oargs.fd = uap->fd;
450 	oargs.pos = uap->pos;
451 	return (sys_mmap(td, &oargs));
452 }
453 
454 #ifdef COMPAT_43
455 #ifndef _SYS_SYSPROTO_H_
456 struct ommap_args {
457 	caddr_t addr;
458 	int len;
459 	int prot;
460 	int flags;
461 	int fd;
462 	long pos;
463 };
464 #endif
465 int
466 ommap(td, uap)
467 	struct thread *td;
468 	struct ommap_args *uap;
469 {
470 	struct mmap_args nargs;
471 	static const char cvtbsdprot[8] = {
472 		0,
473 		PROT_EXEC,
474 		PROT_WRITE,
475 		PROT_EXEC | PROT_WRITE,
476 		PROT_READ,
477 		PROT_EXEC | PROT_READ,
478 		PROT_WRITE | PROT_READ,
479 		PROT_EXEC | PROT_WRITE | PROT_READ,
480 	};
481 
482 #define	OMAP_ANON	0x0002
483 #define	OMAP_COPY	0x0020
484 #define	OMAP_SHARED	0x0010
485 #define	OMAP_FIXED	0x0100
486 
487 	nargs.addr = uap->addr;
488 	nargs.len = uap->len;
489 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
490 #ifdef COMPAT_FREEBSD32
491 #if defined(__amd64__) || defined(__ia64__)
492 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
493 	    nargs.prot != 0)
494 		nargs.prot |= PROT_EXEC;
495 #endif
496 #endif
497 	nargs.flags = 0;
498 	if (uap->flags & OMAP_ANON)
499 		nargs.flags |= MAP_ANON;
500 	if (uap->flags & OMAP_COPY)
501 		nargs.flags |= MAP_COPY;
502 	if (uap->flags & OMAP_SHARED)
503 		nargs.flags |= MAP_SHARED;
504 	else
505 		nargs.flags |= MAP_PRIVATE;
506 	if (uap->flags & OMAP_FIXED)
507 		nargs.flags |= MAP_FIXED;
508 	nargs.fd = uap->fd;
509 	nargs.pos = uap->pos;
510 	return (sys_mmap(td, &nargs));
511 }
512 #endif				/* COMPAT_43 */
513 
514 
515 #ifndef _SYS_SYSPROTO_H_
516 struct msync_args {
517 	void *addr;
518 	size_t len;
519 	int flags;
520 };
521 #endif
522 /*
523  * MPSAFE
524  */
525 int
526 sys_msync(td, uap)
527 	struct thread *td;
528 	struct msync_args *uap;
529 {
530 	vm_offset_t addr;
531 	vm_size_t size, pageoff;
532 	int flags;
533 	vm_map_t map;
534 	int rv;
535 
536 	addr = (vm_offset_t) uap->addr;
537 	size = uap->len;
538 	flags = uap->flags;
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 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
548 		return (EINVAL);
549 
550 	map = &td->td_proc->p_vmspace->vm_map;
551 
552 	/*
553 	 * Clean the pages and interpret the return value.
554 	 */
555 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
556 	    (flags & MS_INVALIDATE) != 0);
557 	switch (rv) {
558 	case KERN_SUCCESS:
559 		return (0);
560 	case KERN_INVALID_ADDRESS:
561 		return (ENOMEM);
562 	case KERN_INVALID_ARGUMENT:
563 		return (EBUSY);
564 	case KERN_FAILURE:
565 		return (EIO);
566 	default:
567 		return (EINVAL);
568 	}
569 }
570 
571 #ifndef _SYS_SYSPROTO_H_
572 struct munmap_args {
573 	void *addr;
574 	size_t len;
575 };
576 #endif
577 /*
578  * MPSAFE
579  */
580 int
581 sys_munmap(td, uap)
582 	struct thread *td;
583 	struct munmap_args *uap;
584 {
585 #ifdef HWPMC_HOOKS
586 	struct pmckern_map_out pkm;
587 	vm_map_entry_t entry;
588 #endif
589 	vm_offset_t addr;
590 	vm_size_t size, pageoff;
591 	vm_map_t map;
592 
593 	addr = (vm_offset_t) uap->addr;
594 	size = uap->len;
595 	if (size == 0)
596 		return (EINVAL);
597 
598 	pageoff = (addr & PAGE_MASK);
599 	addr -= pageoff;
600 	size += pageoff;
601 	size = (vm_size_t) round_page(size);
602 	if (addr + size < addr)
603 		return (EINVAL);
604 
605 	/*
606 	 * Check for illegal addresses.  Watch out for address wrap...
607 	 */
608 	map = &td->td_proc->p_vmspace->vm_map;
609 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
610 		return (EINVAL);
611 	vm_map_lock(map);
612 #ifdef HWPMC_HOOKS
613 	/*
614 	 * Inform hwpmc if the address range being unmapped contains
615 	 * an executable region.
616 	 */
617 	pkm.pm_address = (uintptr_t) NULL;
618 	if (vm_map_lookup_entry(map, addr, &entry)) {
619 		for (;
620 		     entry != &map->header && entry->start < addr + size;
621 		     entry = entry->next) {
622 			if (vm_map_check_protection(map, entry->start,
623 				entry->end, VM_PROT_EXECUTE) == TRUE) {
624 				pkm.pm_address = (uintptr_t) addr;
625 				pkm.pm_size = (size_t) size;
626 				break;
627 			}
628 		}
629 	}
630 #endif
631 	vm_map_delete(map, addr, addr + size);
632 
633 #ifdef HWPMC_HOOKS
634 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
635 	vm_map_lock_downgrade(map);
636 	if (pkm.pm_address != (uintptr_t) NULL)
637 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
638 	vm_map_unlock_read(map);
639 #else
640 	vm_map_unlock(map);
641 #endif
642 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
643 	return (0);
644 }
645 
646 #ifndef _SYS_SYSPROTO_H_
647 struct mprotect_args {
648 	const void *addr;
649 	size_t len;
650 	int prot;
651 };
652 #endif
653 /*
654  * MPSAFE
655  */
656 int
657 sys_mprotect(td, uap)
658 	struct thread *td;
659 	struct mprotect_args *uap;
660 {
661 	vm_offset_t addr;
662 	vm_size_t size, pageoff;
663 	vm_prot_t prot;
664 
665 	addr = (vm_offset_t) uap->addr;
666 	size = uap->len;
667 	prot = uap->prot & VM_PROT_ALL;
668 
669 	pageoff = (addr & PAGE_MASK);
670 	addr -= pageoff;
671 	size += pageoff;
672 	size = (vm_size_t) round_page(size);
673 	if (addr + size < addr)
674 		return (EINVAL);
675 
676 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
677 	    addr + size, prot, FALSE)) {
678 	case KERN_SUCCESS:
679 		return (0);
680 	case KERN_PROTECTION_FAILURE:
681 		return (EACCES);
682 	case KERN_RESOURCE_SHORTAGE:
683 		return (ENOMEM);
684 	}
685 	return (EINVAL);
686 }
687 
688 #ifndef _SYS_SYSPROTO_H_
689 struct minherit_args {
690 	void *addr;
691 	size_t len;
692 	int inherit;
693 };
694 #endif
695 /*
696  * MPSAFE
697  */
698 int
699 sys_minherit(td, uap)
700 	struct thread *td;
701 	struct minherit_args *uap;
702 {
703 	vm_offset_t addr;
704 	vm_size_t size, pageoff;
705 	vm_inherit_t inherit;
706 
707 	addr = (vm_offset_t)uap->addr;
708 	size = uap->len;
709 	inherit = uap->inherit;
710 
711 	pageoff = (addr & PAGE_MASK);
712 	addr -= pageoff;
713 	size += pageoff;
714 	size = (vm_size_t) round_page(size);
715 	if (addr + size < addr)
716 		return (EINVAL);
717 
718 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
719 	    addr + size, inherit)) {
720 	case KERN_SUCCESS:
721 		return (0);
722 	case KERN_PROTECTION_FAILURE:
723 		return (EACCES);
724 	}
725 	return (EINVAL);
726 }
727 
728 #ifndef _SYS_SYSPROTO_H_
729 struct madvise_args {
730 	void *addr;
731 	size_t len;
732 	int behav;
733 };
734 #endif
735 
736 /*
737  * MPSAFE
738  */
739 int
740 sys_madvise(td, uap)
741 	struct thread *td;
742 	struct madvise_args *uap;
743 {
744 	vm_offset_t start, end;
745 	vm_map_t map;
746 	int flags;
747 
748 	/*
749 	 * Check for our special case, advising the swap pager we are
750 	 * "immortal."
751 	 */
752 	if (uap->behav == MADV_PROTECT) {
753 		flags = PPROT_SET;
754 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
755 		    PROC_SPROTECT, &flags));
756 	}
757 
758 	/*
759 	 * Check for illegal behavior
760 	 */
761 	if (uap->behav < 0 || uap->behav > MADV_CORE)
762 		return (EINVAL);
763 	/*
764 	 * Check for illegal addresses.  Watch out for address wrap... Note
765 	 * that VM_*_ADDRESS are not constants due to casts (argh).
766 	 */
767 	map = &td->td_proc->p_vmspace->vm_map;
768 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
769 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
770 		return (EINVAL);
771 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
772 		return (EINVAL);
773 
774 	/*
775 	 * Since this routine is only advisory, we default to conservative
776 	 * behavior.
777 	 */
778 	start = trunc_page((vm_offset_t) uap->addr);
779 	end = round_page((vm_offset_t) uap->addr + uap->len);
780 
781 	if (vm_map_madvise(map, start, end, uap->behav))
782 		return (EINVAL);
783 	return (0);
784 }
785 
786 #ifndef _SYS_SYSPROTO_H_
787 struct mincore_args {
788 	const void *addr;
789 	size_t len;
790 	char *vec;
791 };
792 #endif
793 
794 /*
795  * MPSAFE
796  */
797 int
798 sys_mincore(td, uap)
799 	struct thread *td;
800 	struct mincore_args *uap;
801 {
802 	vm_offset_t addr, first_addr;
803 	vm_offset_t end, cend;
804 	pmap_t pmap;
805 	vm_map_t map;
806 	char *vec;
807 	int error = 0;
808 	int vecindex, lastvecindex;
809 	vm_map_entry_t current;
810 	vm_map_entry_t entry;
811 	vm_object_t object;
812 	vm_paddr_t locked_pa;
813 	vm_page_t m;
814 	vm_pindex_t pindex;
815 	int mincoreinfo;
816 	unsigned int timestamp;
817 	boolean_t locked;
818 
819 	/*
820 	 * Make sure that the addresses presented are valid for user
821 	 * mode.
822 	 */
823 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
824 	end = addr + (vm_size_t)round_page(uap->len);
825 	map = &td->td_proc->p_vmspace->vm_map;
826 	if (end > vm_map_max(map) || end < addr)
827 		return (ENOMEM);
828 
829 	/*
830 	 * Address of byte vector
831 	 */
832 	vec = uap->vec;
833 
834 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
835 
836 	vm_map_lock_read(map);
837 RestartScan:
838 	timestamp = map->timestamp;
839 
840 	if (!vm_map_lookup_entry(map, addr, &entry)) {
841 		vm_map_unlock_read(map);
842 		return (ENOMEM);
843 	}
844 
845 	/*
846 	 * Do this on a map entry basis so that if the pages are not
847 	 * in the current processes address space, we can easily look
848 	 * up the pages elsewhere.
849 	 */
850 	lastvecindex = -1;
851 	for (current = entry;
852 	    (current != &map->header) && (current->start < end);
853 	    current = current->next) {
854 
855 		/*
856 		 * check for contiguity
857 		 */
858 		if (current->end < end &&
859 		    (entry->next == &map->header ||
860 		     current->next->start > current->end)) {
861 			vm_map_unlock_read(map);
862 			return (ENOMEM);
863 		}
864 
865 		/*
866 		 * ignore submaps (for now) or null objects
867 		 */
868 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
869 			current->object.vm_object == NULL)
870 			continue;
871 
872 		/*
873 		 * limit this scan to the current map entry and the
874 		 * limits for the mincore call
875 		 */
876 		if (addr < current->start)
877 			addr = current->start;
878 		cend = current->end;
879 		if (cend > end)
880 			cend = end;
881 
882 		/*
883 		 * scan this entry one page at a time
884 		 */
885 		while (addr < cend) {
886 			/*
887 			 * Check pmap first, it is likely faster, also
888 			 * it can provide info as to whether we are the
889 			 * one referencing or modifying the page.
890 			 */
891 			object = NULL;
892 			locked_pa = 0;
893 		retry:
894 			m = NULL;
895 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
896 			if (locked_pa != 0) {
897 				/*
898 				 * The page is mapped by this process but not
899 				 * both accessed and modified.  It is also
900 				 * managed.  Acquire the object lock so that
901 				 * other mappings might be examined.
902 				 */
903 				m = PHYS_TO_VM_PAGE(locked_pa);
904 				if (m->object != object) {
905 					if (object != NULL)
906 						VM_OBJECT_WUNLOCK(object);
907 					object = m->object;
908 					locked = VM_OBJECT_TRYWLOCK(object);
909 					vm_page_unlock(m);
910 					if (!locked) {
911 						VM_OBJECT_WLOCK(object);
912 						vm_page_lock(m);
913 						goto retry;
914 					}
915 				} else
916 					vm_page_unlock(m);
917 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
918 				    ("mincore: page %p is mapped but invalid",
919 				    m));
920 			} else if (mincoreinfo == 0) {
921 				/*
922 				 * The page is not mapped by this process.  If
923 				 * the object implements managed pages, then
924 				 * determine if the page is resident so that
925 				 * the mappings might be examined.
926 				 */
927 				if (current->object.vm_object != object) {
928 					if (object != NULL)
929 						VM_OBJECT_WUNLOCK(object);
930 					object = current->object.vm_object;
931 					VM_OBJECT_WLOCK(object);
932 				}
933 				if (object->type == OBJT_DEFAULT ||
934 				    object->type == OBJT_SWAP ||
935 				    object->type == OBJT_VNODE) {
936 					pindex = OFF_TO_IDX(current->offset +
937 					    (addr - current->start));
938 					m = vm_page_lookup(object, pindex);
939 					if (m == NULL &&
940 					    vm_page_is_cached(object, pindex))
941 						mincoreinfo = MINCORE_INCORE;
942 					if (m != NULL && m->valid == 0)
943 						m = NULL;
944 					if (m != NULL)
945 						mincoreinfo = MINCORE_INCORE;
946 				}
947 			}
948 			if (m != NULL) {
949 				/* Examine other mappings to the page. */
950 				if (m->dirty == 0 && pmap_is_modified(m))
951 					vm_page_dirty(m);
952 				if (m->dirty != 0)
953 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
954 				/*
955 				 * The first test for PGA_REFERENCED is an
956 				 * optimization.  The second test is
957 				 * required because a concurrent pmap
958 				 * operation could clear the last reference
959 				 * and set PGA_REFERENCED before the call to
960 				 * pmap_is_referenced().
961 				 */
962 				if ((m->aflags & PGA_REFERENCED) != 0 ||
963 				    pmap_is_referenced(m) ||
964 				    (m->aflags & PGA_REFERENCED) != 0)
965 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
966 			}
967 			if (object != NULL)
968 				VM_OBJECT_WUNLOCK(object);
969 
970 			/*
971 			 * subyte may page fault.  In case it needs to modify
972 			 * the map, we release the lock.
973 			 */
974 			vm_map_unlock_read(map);
975 
976 			/*
977 			 * calculate index into user supplied byte vector
978 			 */
979 			vecindex = OFF_TO_IDX(addr - first_addr);
980 
981 			/*
982 			 * If we have skipped map entries, we need to make sure that
983 			 * the byte vector is zeroed for those skipped entries.
984 			 */
985 			while ((lastvecindex + 1) < vecindex) {
986 				++lastvecindex;
987 				error = subyte(vec + lastvecindex, 0);
988 				if (error) {
989 					error = EFAULT;
990 					goto done2;
991 				}
992 			}
993 
994 			/*
995 			 * Pass the page information to the user
996 			 */
997 			error = subyte(vec + vecindex, mincoreinfo);
998 			if (error) {
999 				error = EFAULT;
1000 				goto done2;
1001 			}
1002 
1003 			/*
1004 			 * If the map has changed, due to the subyte, the previous
1005 			 * output may be invalid.
1006 			 */
1007 			vm_map_lock_read(map);
1008 			if (timestamp != map->timestamp)
1009 				goto RestartScan;
1010 
1011 			lastvecindex = vecindex;
1012 			addr += PAGE_SIZE;
1013 		}
1014 	}
1015 
1016 	/*
1017 	 * subyte may page fault.  In case it needs to modify
1018 	 * the map, we release the lock.
1019 	 */
1020 	vm_map_unlock_read(map);
1021 
1022 	/*
1023 	 * Zero the last entries in the byte vector.
1024 	 */
1025 	vecindex = OFF_TO_IDX(end - first_addr);
1026 	while ((lastvecindex + 1) < vecindex) {
1027 		++lastvecindex;
1028 		error = subyte(vec + lastvecindex, 0);
1029 		if (error) {
1030 			error = EFAULT;
1031 			goto done2;
1032 		}
1033 	}
1034 
1035 	/*
1036 	 * If the map has changed, due to the subyte, the previous
1037 	 * output may be invalid.
1038 	 */
1039 	vm_map_lock_read(map);
1040 	if (timestamp != map->timestamp)
1041 		goto RestartScan;
1042 	vm_map_unlock_read(map);
1043 done2:
1044 	return (error);
1045 }
1046 
1047 #ifndef _SYS_SYSPROTO_H_
1048 struct mlock_args {
1049 	const void *addr;
1050 	size_t len;
1051 };
1052 #endif
1053 /*
1054  * MPSAFE
1055  */
1056 int
1057 sys_mlock(td, uap)
1058 	struct thread *td;
1059 	struct mlock_args *uap;
1060 {
1061 
1062 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1063 }
1064 
1065 int
1066 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1067 {
1068 	vm_offset_t addr, end, last, start;
1069 	vm_size_t npages, size;
1070 	vm_map_t map;
1071 	unsigned long nsize;
1072 	int error;
1073 
1074 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1075 	if (error)
1076 		return (error);
1077 	addr = (vm_offset_t)addr0;
1078 	size = len;
1079 	last = addr + size;
1080 	start = trunc_page(addr);
1081 	end = round_page(last);
1082 	if (last < addr || end < addr)
1083 		return (EINVAL);
1084 	npages = atop(end - start);
1085 	if (npages > vm_page_max_wired)
1086 		return (ENOMEM);
1087 	map = &proc->p_vmspace->vm_map;
1088 	PROC_LOCK(proc);
1089 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1090 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1091 		PROC_UNLOCK(proc);
1092 		return (ENOMEM);
1093 	}
1094 	PROC_UNLOCK(proc);
1095 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1096 		return (EAGAIN);
1097 #ifdef RACCT
1098 	PROC_LOCK(proc);
1099 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1100 	PROC_UNLOCK(proc);
1101 	if (error != 0)
1102 		return (ENOMEM);
1103 #endif
1104 	error = vm_map_wire(map, start, end,
1105 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1106 #ifdef RACCT
1107 	if (error != KERN_SUCCESS) {
1108 		PROC_LOCK(proc);
1109 		racct_set(proc, RACCT_MEMLOCK,
1110 		    ptoa(pmap_wired_count(map->pmap)));
1111 		PROC_UNLOCK(proc);
1112 	}
1113 #endif
1114 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1115 }
1116 
1117 #ifndef _SYS_SYSPROTO_H_
1118 struct mlockall_args {
1119 	int	how;
1120 };
1121 #endif
1122 
1123 /*
1124  * MPSAFE
1125  */
1126 int
1127 sys_mlockall(td, uap)
1128 	struct thread *td;
1129 	struct mlockall_args *uap;
1130 {
1131 	vm_map_t map;
1132 	int error;
1133 
1134 	map = &td->td_proc->p_vmspace->vm_map;
1135 	error = priv_check(td, PRIV_VM_MLOCK);
1136 	if (error)
1137 		return (error);
1138 
1139 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1140 		return (EINVAL);
1141 
1142 	/*
1143 	 * If wiring all pages in the process would cause it to exceed
1144 	 * a hard resource limit, return ENOMEM.
1145 	 */
1146 	if (!old_mlock && uap->how & MCL_CURRENT) {
1147 		PROC_LOCK(td->td_proc);
1148 		if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1149 			PROC_UNLOCK(td->td_proc);
1150 			return (ENOMEM);
1151 		}
1152 		PROC_UNLOCK(td->td_proc);
1153 	}
1154 #ifdef RACCT
1155 	PROC_LOCK(td->td_proc);
1156 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1157 	PROC_UNLOCK(td->td_proc);
1158 	if (error != 0)
1159 		return (ENOMEM);
1160 #endif
1161 
1162 	if (uap->how & MCL_FUTURE) {
1163 		vm_map_lock(map);
1164 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1165 		vm_map_unlock(map);
1166 		error = 0;
1167 	}
1168 
1169 	if (uap->how & MCL_CURRENT) {
1170 		/*
1171 		 * P1003.1-2001 mandates that all currently mapped pages
1172 		 * will be memory resident and locked (wired) upon return
1173 		 * from mlockall(). vm_map_wire() will wire pages, by
1174 		 * calling vm_fault_wire() for each page in the region.
1175 		 */
1176 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1177 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1178 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1179 	}
1180 #ifdef RACCT
1181 	if (error != KERN_SUCCESS) {
1182 		PROC_LOCK(td->td_proc);
1183 		racct_set(td->td_proc, RACCT_MEMLOCK,
1184 		    ptoa(pmap_wired_count(map->pmap)));
1185 		PROC_UNLOCK(td->td_proc);
1186 	}
1187 #endif
1188 
1189 	return (error);
1190 }
1191 
1192 #ifndef _SYS_SYSPROTO_H_
1193 struct munlockall_args {
1194 	register_t dummy;
1195 };
1196 #endif
1197 
1198 /*
1199  * MPSAFE
1200  */
1201 int
1202 sys_munlockall(td, uap)
1203 	struct thread *td;
1204 	struct munlockall_args *uap;
1205 {
1206 	vm_map_t map;
1207 	int error;
1208 
1209 	map = &td->td_proc->p_vmspace->vm_map;
1210 	error = priv_check(td, PRIV_VM_MUNLOCK);
1211 	if (error)
1212 		return (error);
1213 
1214 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1215 	vm_map_lock(map);
1216 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1217 	vm_map_unlock(map);
1218 
1219 	/* Forcibly unwire all pages. */
1220 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1221 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1222 #ifdef RACCT
1223 	if (error == KERN_SUCCESS) {
1224 		PROC_LOCK(td->td_proc);
1225 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1226 		PROC_UNLOCK(td->td_proc);
1227 	}
1228 #endif
1229 
1230 	return (error);
1231 }
1232 
1233 #ifndef _SYS_SYSPROTO_H_
1234 struct munlock_args {
1235 	const void *addr;
1236 	size_t len;
1237 };
1238 #endif
1239 /*
1240  * MPSAFE
1241  */
1242 int
1243 sys_munlock(td, uap)
1244 	struct thread *td;
1245 	struct munlock_args *uap;
1246 {
1247 	vm_offset_t addr, end, last, start;
1248 	vm_size_t size;
1249 #ifdef RACCT
1250 	vm_map_t map;
1251 #endif
1252 	int error;
1253 
1254 	error = priv_check(td, PRIV_VM_MUNLOCK);
1255 	if (error)
1256 		return (error);
1257 	addr = (vm_offset_t)uap->addr;
1258 	size = uap->len;
1259 	last = addr + size;
1260 	start = trunc_page(addr);
1261 	end = round_page(last);
1262 	if (last < addr || end < addr)
1263 		return (EINVAL);
1264 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1265 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1266 #ifdef RACCT
1267 	if (error == KERN_SUCCESS) {
1268 		PROC_LOCK(td->td_proc);
1269 		map = &td->td_proc->p_vmspace->vm_map;
1270 		racct_set(td->td_proc, RACCT_MEMLOCK,
1271 		    ptoa(pmap_wired_count(map->pmap)));
1272 		PROC_UNLOCK(td->td_proc);
1273 	}
1274 #endif
1275 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1276 }
1277 
1278 /*
1279  * vm_mmap_vnode()
1280  *
1281  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1282  * operations on vnodes.
1283  *
1284  * For VCHR vnodes, the vnode lock is held over the call to
1285  * vm_mmap_cdev() to keep vp->v_rdev valid.
1286  */
1287 int
1288 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1289     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1290     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1291     boolean_t *writecounted)
1292 {
1293 	struct vattr va;
1294 	vm_object_t obj;
1295 	vm_offset_t foff;
1296 	struct mount *mp;
1297 	struct ucred *cred;
1298 	int error, flags, locktype;
1299 
1300 	mp = vp->v_mount;
1301 	cred = td->td_ucred;
1302 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1303 		locktype = LK_EXCLUSIVE;
1304 	else
1305 		locktype = LK_SHARED;
1306 	if ((error = vget(vp, locktype, td)) != 0)
1307 		return (error);
1308 	foff = *foffp;
1309 	flags = *flagsp;
1310 	obj = vp->v_object;
1311 	if (vp->v_type == VREG) {
1312 		/*
1313 		 * Get the proper underlying object
1314 		 */
1315 		if (obj == NULL) {
1316 			error = EINVAL;
1317 			goto done;
1318 		}
1319 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1320 			vput(vp);
1321 			vp = (struct vnode *)obj->handle;
1322 			/*
1323 			 * Bypass filesystems obey the mpsafety of the
1324 			 * underlying fs.  Tmpfs never bypasses.
1325 			 */
1326 			error = vget(vp, locktype, td);
1327 			if (error != 0)
1328 				return (error);
1329 		}
1330 		if (locktype == LK_EXCLUSIVE) {
1331 			*writecounted = TRUE;
1332 			vnode_pager_update_writecount(obj, 0, objsize);
1333 		}
1334 	} else if (vp->v_type == VCHR) {
1335 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1336 		    vp->v_rdev, foffp, objp);
1337 		if (error == 0)
1338 			goto mark_atime;
1339 		goto done;
1340 	} else {
1341 		error = EINVAL;
1342 		goto done;
1343 	}
1344 	if ((error = VOP_GETATTR(vp, &va, cred)))
1345 		goto done;
1346 #ifdef MAC
1347 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1348 	if (error != 0)
1349 		goto done;
1350 #endif
1351 	if ((flags & MAP_SHARED) != 0) {
1352 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1353 			if (prot & PROT_WRITE) {
1354 				error = EPERM;
1355 				goto done;
1356 			}
1357 			*maxprotp &= ~VM_PROT_WRITE;
1358 		}
1359 	}
1360 	/*
1361 	 * If it is a regular file without any references
1362 	 * we do not need to sync it.
1363 	 * Adjust object size to be the size of actual file.
1364 	 */
1365 	objsize = round_page(va.va_size);
1366 	if (va.va_nlink == 0)
1367 		flags |= MAP_NOSYNC;
1368 	if (obj->type == OBJT_VNODE)
1369 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1370 		    cred);
1371 	else {
1372 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1373 		    ("wrong object type"));
1374 		vm_object_reference(obj);
1375 	}
1376 	if (obj == NULL) {
1377 		error = ENOMEM;
1378 		goto done;
1379 	}
1380 	*objp = obj;
1381 	*flagsp = flags;
1382 
1383 mark_atime:
1384 	vfs_mark_atime(vp, cred);
1385 
1386 done:
1387 	if (error != 0 && *writecounted) {
1388 		*writecounted = FALSE;
1389 		vnode_pager_update_writecount(obj, objsize, 0);
1390 	}
1391 	vput(vp);
1392 	return (error);
1393 }
1394 
1395 /*
1396  * vm_mmap_cdev()
1397  *
1398  * MPSAFE
1399  *
1400  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1401  * operations on cdevs.
1402  */
1403 int
1404 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1405     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1406     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1407 {
1408 	vm_object_t obj;
1409 	struct cdevsw *dsw;
1410 	int error, flags, ref;
1411 
1412 	flags = *flagsp;
1413 
1414 	dsw = dev_refthread(cdev, &ref);
1415 	if (dsw == NULL)
1416 		return (ENXIO);
1417 	if (dsw->d_flags & D_MMAP_ANON) {
1418 		dev_relthread(cdev, ref);
1419 		*maxprotp = VM_PROT_ALL;
1420 		*flagsp |= MAP_ANON;
1421 		return (0);
1422 	}
1423 	/*
1424 	 * cdevs do not provide private mappings of any kind.
1425 	 */
1426 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1427 	    (prot & PROT_WRITE) != 0) {
1428 		dev_relthread(cdev, ref);
1429 		return (EACCES);
1430 	}
1431 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1432 		dev_relthread(cdev, ref);
1433 		return (EINVAL);
1434 	}
1435 	/*
1436 	 * Force device mappings to be shared.
1437 	 */
1438 	flags |= MAP_SHARED;
1439 #ifdef MAC_XXX
1440 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1441 	if (error != 0) {
1442 		dev_relthread(cdev, ref);
1443 		return (error);
1444 	}
1445 #endif
1446 	/*
1447 	 * First, try d_mmap_single().  If that is not implemented
1448 	 * (returns ENODEV), fall back to using the device pager.
1449 	 * Note that d_mmap_single() must return a reference to the
1450 	 * object (it needs to bump the reference count of the object
1451 	 * it returns somehow).
1452 	 *
1453 	 * XXX assumes VM_PROT_* == PROT_*
1454 	 */
1455 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1456 	dev_relthread(cdev, ref);
1457 	if (error != ENODEV)
1458 		return (error);
1459 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1460 	    td->td_ucred);
1461 	if (obj == NULL)
1462 		return (EINVAL);
1463 	*objp = obj;
1464 	*flagsp = flags;
1465 	return (0);
1466 }
1467 
1468 /*
1469  * vm_mmap_shm()
1470  *
1471  * MPSAFE
1472  *
1473  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1474  * operations on shm file descriptors.
1475  */
1476 int
1477 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1478     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1479     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1480 {
1481 	int error;
1482 
1483 	if ((*flagsp & MAP_SHARED) != 0 &&
1484 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1485 	    (prot & PROT_WRITE) != 0)
1486 		return (EACCES);
1487 #ifdef MAC
1488 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1489 	if (error != 0)
1490 		return (error);
1491 #endif
1492 	error = shm_mmap(shmfd, objsize, foff, objp);
1493 	if (error)
1494 		return (error);
1495 	return (0);
1496 }
1497 
1498 /*
1499  * vm_mmap()
1500  *
1501  * MPSAFE
1502  *
1503  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1504  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1505  */
1506 int
1507 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1508 	vm_prot_t maxprot, int flags,
1509 	objtype_t handle_type, void *handle,
1510 	vm_ooffset_t foff)
1511 {
1512 	boolean_t fitit;
1513 	vm_object_t object = NULL;
1514 	struct thread *td = curthread;
1515 	int docow, error, findspace, rv;
1516 	boolean_t writecounted;
1517 
1518 	if (size == 0)
1519 		return (0);
1520 
1521 	size = round_page(size);
1522 
1523 	if (map == &td->td_proc->p_vmspace->vm_map) {
1524 		PROC_LOCK(td->td_proc);
1525 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1526 			PROC_UNLOCK(td->td_proc);
1527 			return (ENOMEM);
1528 		}
1529 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1530 			PROC_UNLOCK(td->td_proc);
1531 			return (ENOMEM);
1532 		}
1533 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1534 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1535 			    lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1536 				racct_set_force(td->td_proc, RACCT_VMEM,
1537 				    map->size);
1538 				PROC_UNLOCK(td->td_proc);
1539 				return (ENOMEM);
1540 			}
1541 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1542 			    ptoa(pmap_wired_count(map->pmap)) + size);
1543 			if (error != 0) {
1544 				racct_set_force(td->td_proc, RACCT_VMEM,
1545 				    map->size);
1546 				PROC_UNLOCK(td->td_proc);
1547 				return (error);
1548 			}
1549 		}
1550 		PROC_UNLOCK(td->td_proc);
1551 	}
1552 
1553 	/*
1554 	 * We currently can only deal with page aligned file offsets.
1555 	 * The check is here rather than in the syscall because the
1556 	 * kernel calls this function internally for other mmaping
1557 	 * operations (such as in exec) and non-aligned offsets will
1558 	 * cause pmap inconsistencies...so we want to be sure to
1559 	 * disallow this in all cases.
1560 	 */
1561 	if (foff & PAGE_MASK)
1562 		return (EINVAL);
1563 
1564 	if ((flags & MAP_FIXED) == 0) {
1565 		fitit = TRUE;
1566 		*addr = round_page(*addr);
1567 	} else {
1568 		if (*addr != trunc_page(*addr))
1569 			return (EINVAL);
1570 		fitit = FALSE;
1571 	}
1572 	writecounted = FALSE;
1573 
1574 	/*
1575 	 * Lookup/allocate object.
1576 	 */
1577 	switch (handle_type) {
1578 	case OBJT_DEVICE:
1579 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1580 		    handle, &foff, &object);
1581 		break;
1582 	case OBJT_VNODE:
1583 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1584 		    handle, &foff, &object, &writecounted);
1585 		break;
1586 	case OBJT_SWAP:
1587 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1588 		    handle, foff, &object);
1589 		break;
1590 	case OBJT_DEFAULT:
1591 		if (handle == NULL) {
1592 			error = 0;
1593 			break;
1594 		}
1595 		/* FALLTHROUGH */
1596 	default:
1597 		error = EINVAL;
1598 		break;
1599 	}
1600 	if (error)
1601 		return (error);
1602 	if (flags & MAP_ANON) {
1603 		object = NULL;
1604 		docow = 0;
1605 		/*
1606 		 * Unnamed anonymous regions always start at 0.
1607 		 */
1608 		if (handle == 0)
1609 			foff = 0;
1610 	} else if (flags & MAP_PREFAULT_READ)
1611 		docow = MAP_PREFAULT;
1612 	else
1613 		docow = MAP_PREFAULT_PARTIAL;
1614 
1615 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1616 		docow |= MAP_COPY_ON_WRITE;
1617 	if (flags & MAP_NOSYNC)
1618 		docow |= MAP_DISABLE_SYNCER;
1619 	if (flags & MAP_NOCORE)
1620 		docow |= MAP_DISABLE_COREDUMP;
1621 	/* Shared memory is also shared with children. */
1622 	if (flags & MAP_SHARED)
1623 		docow |= MAP_INHERIT_SHARE;
1624 	if (writecounted)
1625 		docow |= MAP_VN_WRITECOUNT;
1626 	if (flags & MAP_STACK) {
1627 		if (object != NULL)
1628 			return (EINVAL);
1629 		docow |= MAP_STACK_GROWS_DOWN;
1630 	}
1631 	if ((flags & MAP_EXCL) != 0)
1632 		docow |= MAP_CHECK_EXCL;
1633 
1634 	if (fitit) {
1635 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1636 			findspace = VMFS_SUPER_SPACE;
1637 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1638 			findspace = VMFS_ALIGNED_SPACE(flags >>
1639 			    MAP_ALIGNMENT_SHIFT);
1640 		else
1641 			findspace = VMFS_OPTIMAL_SPACE;
1642 		rv = vm_map_find(map, object, foff, addr, size,
1643 #ifdef MAP_32BIT
1644 		    flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1645 #endif
1646 		    0, findspace, prot, maxprot, docow);
1647 	} else {
1648 		rv = vm_map_fixed(map, object, foff, *addr, size,
1649 		    prot, maxprot, docow);
1650 	}
1651 
1652 	if (rv == KERN_SUCCESS) {
1653 		/*
1654 		 * If the process has requested that all future mappings
1655 		 * be wired, then heed this.
1656 		 */
1657 		if (map->flags & MAP_WIREFUTURE) {
1658 			vm_map_wire(map, *addr, *addr + size,
1659 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1660 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1661 		}
1662 	} else {
1663 		/*
1664 		 * If this mapping was accounted for in the vnode's
1665 		 * writecount, then undo that now.
1666 		 */
1667 		if (writecounted)
1668 			vnode_pager_release_writecount(object, 0, size);
1669 		/*
1670 		 * Lose the object reference.  Will destroy the
1671 		 * object if it's an unnamed anonymous mapping
1672 		 * or named anonymous without other references.
1673 		 */
1674 		vm_object_deallocate(object);
1675 	}
1676 	return (vm_mmap_to_errno(rv));
1677 }
1678 
1679 /*
1680  * Translate a Mach VM return code to zero on success or the appropriate errno
1681  * on failure.
1682  */
1683 int
1684 vm_mmap_to_errno(int rv)
1685 {
1686 
1687 	switch (rv) {
1688 	case KERN_SUCCESS:
1689 		return (0);
1690 	case KERN_INVALID_ADDRESS:
1691 	case KERN_NO_SPACE:
1692 		return (ENOMEM);
1693 	case KERN_PROTECTION_FAILURE:
1694 		return (EACCES);
1695 	default:
1696 		return (EINVAL);
1697 	}
1698 }
1699