xref: /freebsd/sys/vm/vm_mmap.c (revision 95d45410b5100e07f6f98450bcd841a8945d4726)
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_RWTUN, &old_mlock, 0,
96     "Do not apply RLIMIT_MEMLOCK on mlockall");
97 
98 #ifdef MAP_32BIT
99 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
100 #endif
101 
102 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
103     int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
104 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
105     int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
106 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
107     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
108 
109 #ifndef _SYS_SYSPROTO_H_
110 struct sbrk_args {
111 	int incr;
112 };
113 #endif
114 
115 /*
116  * MPSAFE
117  */
118 /* ARGSUSED */
119 int
120 sys_sbrk(td, uap)
121 	struct thread *td;
122 	struct sbrk_args *uap;
123 {
124 	/* Not yet implemented */
125 	return (EOPNOTSUPP);
126 }
127 
128 #ifndef _SYS_SYSPROTO_H_
129 struct sstk_args {
130 	int incr;
131 };
132 #endif
133 
134 /*
135  * MPSAFE
136  */
137 /* ARGSUSED */
138 int
139 sys_sstk(td, uap)
140 	struct thread *td;
141 	struct sstk_args *uap;
142 {
143 	/* Not yet implemented */
144 	return (EOPNOTSUPP);
145 }
146 
147 #if defined(COMPAT_43)
148 #ifndef _SYS_SYSPROTO_H_
149 struct getpagesize_args {
150 	int dummy;
151 };
152 #endif
153 
154 int
155 ogetpagesize(td, uap)
156 	struct thread *td;
157 	struct getpagesize_args *uap;
158 {
159 	/* MP SAFE */
160 	td->td_retval[0] = PAGE_SIZE;
161 	return (0);
162 }
163 #endif				/* COMPAT_43 */
164 
165 
166 /*
167  * Memory Map (mmap) system call.  Note that the file offset
168  * and address are allowed to be NOT page aligned, though if
169  * the MAP_FIXED flag it set, both must have the same remainder
170  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
171  * page-aligned, the actual mapping starts at trunc_page(addr)
172  * and the return value is adjusted up by the page offset.
173  *
174  * Generally speaking, only character devices which are themselves
175  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
176  * there would be no cache coherency between a descriptor and a VM mapping
177  * both to the same character device.
178  */
179 #ifndef _SYS_SYSPROTO_H_
180 struct mmap_args {
181 	void *addr;
182 	size_t len;
183 	int prot;
184 	int flags;
185 	int fd;
186 	long pad;
187 	off_t pos;
188 };
189 #endif
190 
191 /*
192  * MPSAFE
193  */
194 int
195 sys_mmap(td, uap)
196 	struct thread *td;
197 	struct mmap_args *uap;
198 {
199 #ifdef HWPMC_HOOKS
200 	struct pmckern_map_in pkm;
201 #endif
202 	struct file *fp;
203 	struct vnode *vp;
204 	vm_offset_t addr;
205 	vm_size_t size, pageoff;
206 	vm_prot_t cap_maxprot, prot, maxprot;
207 	void *handle;
208 	objtype_t handle_type;
209 	int align, error, flags;
210 	off_t pos;
211 	struct vmspace *vms = td->td_proc->p_vmspace;
212 	cap_rights_t rights;
213 
214 	addr = (vm_offset_t) uap->addr;
215 	size = uap->len;
216 	prot = uap->prot & VM_PROT_ALL;
217 	flags = uap->flags;
218 	pos = uap->pos;
219 
220 	fp = NULL;
221 
222 	/*
223 	 * Enforce the constraints.
224 	 * Mapping of length 0 is only allowed for old binaries.
225 	 * Anonymous mapping shall specify -1 as filedescriptor and
226 	 * zero position for new code. Be nice to ancient a.out
227 	 * binaries and correct pos for anonymous mapping, since old
228 	 * ld.so sometimes issues anonymous map requests with non-zero
229 	 * pos.
230 	 */
231 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
232 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
233 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
234 			return (EINVAL);
235 	} else {
236 		if ((flags & MAP_ANON) != 0)
237 			pos = 0;
238 	}
239 
240 	if (flags & MAP_STACK) {
241 		if ((uap->fd != -1) ||
242 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
243 			return (EINVAL);
244 		flags |= MAP_ANON;
245 		pos = 0;
246 	}
247 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
248 		return (EINVAL);
249 
250 	/*
251 	 * Align the file position to a page boundary,
252 	 * and save its page offset component.
253 	 */
254 	pageoff = (pos & PAGE_MASK);
255 	pos -= pageoff;
256 
257 	/* Adjust size for rounding (on both ends). */
258 	size += pageoff;			/* low end... */
259 	size = (vm_size_t) round_page(size);	/* hi end */
260 
261 	/* Ensure alignment is at least a page and fits in a pointer. */
262 	align = flags & MAP_ALIGNMENT_MASK;
263 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
264 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
265 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
266 		return (EINVAL);
267 
268 	/*
269 	 * Check for illegal addresses.  Watch out for address wrap... Note
270 	 * that VM_*_ADDRESS are not constants due to casts (argh).
271 	 */
272 	if (flags & MAP_FIXED) {
273 		/*
274 		 * The specified address must have the same remainder
275 		 * as the file offset taken modulo PAGE_SIZE, so it
276 		 * should be aligned after adjustment by pageoff.
277 		 */
278 		addr -= pageoff;
279 		if (addr & PAGE_MASK)
280 			return (EINVAL);
281 
282 		/* Address range must be all in user VM space. */
283 		if (addr < vm_map_min(&vms->vm_map) ||
284 		    addr + size > vm_map_max(&vms->vm_map))
285 			return (EINVAL);
286 		if (addr + size < addr)
287 			return (EINVAL);
288 #ifdef MAP_32BIT
289 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
290 			return (EINVAL);
291 	} else if (flags & MAP_32BIT) {
292 		/*
293 		 * For MAP_32BIT, override the hint if it is too high and
294 		 * do not bother moving the mapping past the heap (since
295 		 * the heap is usually above 2GB).
296 		 */
297 		if (addr + size > MAP_32BIT_MAX_ADDR)
298 			addr = 0;
299 #endif
300 	} else {
301 		/*
302 		 * XXX for non-fixed mappings where no hint is provided or
303 		 * the hint would fall in the potential heap space,
304 		 * place it after the end of the largest possible heap.
305 		 *
306 		 * There should really be a pmap call to determine a reasonable
307 		 * location.
308 		 */
309 		PROC_LOCK(td->td_proc);
310 		if (addr == 0 ||
311 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
312 		    addr < round_page((vm_offset_t)vms->vm_daddr +
313 		    lim_max(td->td_proc, RLIMIT_DATA))))
314 			addr = round_page((vm_offset_t)vms->vm_daddr +
315 			    lim_max(td->td_proc, RLIMIT_DATA));
316 		PROC_UNLOCK(td->td_proc);
317 	}
318 	if (flags & MAP_ANON) {
319 		/*
320 		 * Mapping blank space is trivial.
321 		 */
322 		handle = NULL;
323 		handle_type = OBJT_DEFAULT;
324 		maxprot = VM_PROT_ALL;
325 		cap_maxprot = VM_PROT_ALL;
326 	} else {
327 		/*
328 		 * Mapping file, get fp for validation and don't let the
329 		 * descriptor disappear on us if we block. Check capability
330 		 * rights, but also return the maximum rights to be combined
331 		 * with maxprot later.
332 		 */
333 		cap_rights_init(&rights, CAP_MMAP);
334 		if (prot & PROT_READ)
335 			cap_rights_set(&rights, CAP_MMAP_R);
336 		if ((flags & MAP_SHARED) != 0) {
337 			if (prot & PROT_WRITE)
338 				cap_rights_set(&rights, CAP_MMAP_W);
339 		}
340 		if (prot & PROT_EXEC)
341 			cap_rights_set(&rights, CAP_MMAP_X);
342 		error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
343 		if (error != 0)
344 			goto done;
345 		if (fp->f_type == DTYPE_SHM) {
346 			handle = fp->f_data;
347 			handle_type = OBJT_SWAP;
348 			maxprot = VM_PROT_NONE;
349 
350 			/* FREAD should always be set. */
351 			if (fp->f_flag & FREAD)
352 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
353 			if (fp->f_flag & FWRITE)
354 				maxprot |= VM_PROT_WRITE;
355 			goto map;
356 		}
357 		if (fp->f_type != DTYPE_VNODE) {
358 			error = ENODEV;
359 			goto done;
360 		}
361 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
362     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
363 		/*
364 		 * POSIX shared-memory objects are defined to have
365 		 * kernel persistence, and are not defined to support
366 		 * read(2)/write(2) -- or even open(2).  Thus, we can
367 		 * use MAP_ASYNC to trade on-disk coherence for speed.
368 		 * The shm_open(3) library routine turns on the FPOSIXSHM
369 		 * flag to request this behavior.
370 		 */
371 		if (fp->f_flag & FPOSIXSHM)
372 			flags |= MAP_NOSYNC;
373 #endif
374 		vp = fp->f_vnode;
375 		/*
376 		 * Ensure that file and memory protections are
377 		 * compatible.  Note that we only worry about
378 		 * writability if mapping is shared; in this case,
379 		 * current and max prot are dictated by the open file.
380 		 * XXX use the vnode instead?  Problem is: what
381 		 * credentials do we use for determination? What if
382 		 * proc does a setuid?
383 		 */
384 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
385 			maxprot = VM_PROT_NONE;
386 		else
387 			maxprot = VM_PROT_EXECUTE;
388 		if (fp->f_flag & FREAD) {
389 			maxprot |= VM_PROT_READ;
390 		} else if (prot & PROT_READ) {
391 			error = EACCES;
392 			goto done;
393 		}
394 		/*
395 		 * If we are sharing potential changes (either via
396 		 * MAP_SHARED or via the implicit sharing of character
397 		 * device mappings), and we are trying to get write
398 		 * permission although we opened it without asking
399 		 * for it, bail out.
400 		 */
401 		if ((flags & MAP_SHARED) != 0) {
402 			if ((fp->f_flag & FWRITE) != 0) {
403 				maxprot |= VM_PROT_WRITE;
404 			} else if ((prot & PROT_WRITE) != 0) {
405 				error = EACCES;
406 				goto done;
407 			}
408 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
409 			maxprot |= VM_PROT_WRITE;
410 			cap_maxprot |= VM_PROT_WRITE;
411 		}
412 		handle = (void *)vp;
413 		handle_type = OBJT_VNODE;
414 	}
415 map:
416 	td->td_fpop = fp;
417 	maxprot &= cap_maxprot;
418 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
419 	    flags, handle_type, handle, pos);
420 	td->td_fpop = NULL;
421 #ifdef HWPMC_HOOKS
422 	/* inform hwpmc(4) if an executable is being mapped */
423 	if (error == 0 && handle_type == OBJT_VNODE &&
424 	    (prot & PROT_EXEC)) {
425 		pkm.pm_file = handle;
426 		pkm.pm_address = (uintptr_t) addr;
427 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
428 	}
429 #endif
430 	if (error == 0)
431 		td->td_retval[0] = (register_t) (addr + pageoff);
432 done:
433 	if (fp)
434 		fdrop(fp, td);
435 
436 	return (error);
437 }
438 
439 int
440 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
441 {
442 	struct mmap_args oargs;
443 
444 	oargs.addr = uap->addr;
445 	oargs.len = uap->len;
446 	oargs.prot = uap->prot;
447 	oargs.flags = uap->flags;
448 	oargs.fd = uap->fd;
449 	oargs.pos = uap->pos;
450 	return (sys_mmap(td, &oargs));
451 }
452 
453 #ifdef COMPAT_43
454 #ifndef _SYS_SYSPROTO_H_
455 struct ommap_args {
456 	caddr_t addr;
457 	int len;
458 	int prot;
459 	int flags;
460 	int fd;
461 	long pos;
462 };
463 #endif
464 int
465 ommap(td, uap)
466 	struct thread *td;
467 	struct ommap_args *uap;
468 {
469 	struct mmap_args nargs;
470 	static const char cvtbsdprot[8] = {
471 		0,
472 		PROT_EXEC,
473 		PROT_WRITE,
474 		PROT_EXEC | PROT_WRITE,
475 		PROT_READ,
476 		PROT_EXEC | PROT_READ,
477 		PROT_WRITE | PROT_READ,
478 		PROT_EXEC | PROT_WRITE | PROT_READ,
479 	};
480 
481 #define	OMAP_ANON	0x0002
482 #define	OMAP_COPY	0x0020
483 #define	OMAP_SHARED	0x0010
484 #define	OMAP_FIXED	0x0100
485 
486 	nargs.addr = uap->addr;
487 	nargs.len = uap->len;
488 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
489 #ifdef COMPAT_FREEBSD32
490 #if defined(__amd64__)
491 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
492 	    nargs.prot != 0)
493 		nargs.prot |= PROT_EXEC;
494 #endif
495 #endif
496 	nargs.flags = 0;
497 	if (uap->flags & OMAP_ANON)
498 		nargs.flags |= MAP_ANON;
499 	if (uap->flags & OMAP_COPY)
500 		nargs.flags |= MAP_COPY;
501 	if (uap->flags & OMAP_SHARED)
502 		nargs.flags |= MAP_SHARED;
503 	else
504 		nargs.flags |= MAP_PRIVATE;
505 	if (uap->flags & OMAP_FIXED)
506 		nargs.flags |= MAP_FIXED;
507 	nargs.fd = uap->fd;
508 	nargs.pos = uap->pos;
509 	return (sys_mmap(td, &nargs));
510 }
511 #endif				/* COMPAT_43 */
512 
513 
514 #ifndef _SYS_SYSPROTO_H_
515 struct msync_args {
516 	void *addr;
517 	size_t len;
518 	int flags;
519 };
520 #endif
521 /*
522  * MPSAFE
523  */
524 int
525 sys_msync(td, uap)
526 	struct thread *td;
527 	struct msync_args *uap;
528 {
529 	vm_offset_t addr;
530 	vm_size_t size, pageoff;
531 	int flags;
532 	vm_map_t map;
533 	int rv;
534 
535 	addr = (vm_offset_t) uap->addr;
536 	size = uap->len;
537 	flags = uap->flags;
538 
539 	pageoff = (addr & PAGE_MASK);
540 	addr -= pageoff;
541 	size += pageoff;
542 	size = (vm_size_t) round_page(size);
543 	if (addr + size < addr)
544 		return (EINVAL);
545 
546 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
547 		return (EINVAL);
548 
549 	map = &td->td_proc->p_vmspace->vm_map;
550 
551 	/*
552 	 * Clean the pages and interpret the return value.
553 	 */
554 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
555 	    (flags & MS_INVALIDATE) != 0);
556 	switch (rv) {
557 	case KERN_SUCCESS:
558 		return (0);
559 	case KERN_INVALID_ADDRESS:
560 		return (ENOMEM);
561 	case KERN_INVALID_ARGUMENT:
562 		return (EBUSY);
563 	case KERN_FAILURE:
564 		return (EIO);
565 	default:
566 		return (EINVAL);
567 	}
568 }
569 
570 #ifndef _SYS_SYSPROTO_H_
571 struct munmap_args {
572 	void *addr;
573 	size_t len;
574 };
575 #endif
576 /*
577  * MPSAFE
578  */
579 int
580 sys_munmap(td, uap)
581 	struct thread *td;
582 	struct munmap_args *uap;
583 {
584 #ifdef HWPMC_HOOKS
585 	struct pmckern_map_out pkm;
586 	vm_map_entry_t entry;
587 #endif
588 	vm_offset_t addr;
589 	vm_size_t size, pageoff;
590 	vm_map_t map;
591 
592 	addr = (vm_offset_t) uap->addr;
593 	size = uap->len;
594 	if (size == 0)
595 		return (EINVAL);
596 
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 	/*
605 	 * Check for illegal addresses.  Watch out for address wrap...
606 	 */
607 	map = &td->td_proc->p_vmspace->vm_map;
608 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
609 		return (EINVAL);
610 	vm_map_lock(map);
611 #ifdef HWPMC_HOOKS
612 	/*
613 	 * Inform hwpmc if the address range being unmapped contains
614 	 * an executable region.
615 	 */
616 	pkm.pm_address = (uintptr_t) NULL;
617 	if (vm_map_lookup_entry(map, addr, &entry)) {
618 		for (;
619 		     entry != &map->header && entry->start < addr + size;
620 		     entry = entry->next) {
621 			if (vm_map_check_protection(map, entry->start,
622 				entry->end, VM_PROT_EXECUTE) == TRUE) {
623 				pkm.pm_address = (uintptr_t) addr;
624 				pkm.pm_size = (size_t) size;
625 				break;
626 			}
627 		}
628 	}
629 #endif
630 	vm_map_delete(map, addr, addr + size);
631 
632 #ifdef HWPMC_HOOKS
633 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
634 	vm_map_lock_downgrade(map);
635 	if (pkm.pm_address != (uintptr_t) NULL)
636 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
637 	vm_map_unlock_read(map);
638 #else
639 	vm_map_unlock(map);
640 #endif
641 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
642 	return (0);
643 }
644 
645 #ifndef _SYS_SYSPROTO_H_
646 struct mprotect_args {
647 	const void *addr;
648 	size_t len;
649 	int prot;
650 };
651 #endif
652 /*
653  * MPSAFE
654  */
655 int
656 sys_mprotect(td, uap)
657 	struct thread *td;
658 	struct mprotect_args *uap;
659 {
660 	vm_offset_t addr;
661 	vm_size_t size, pageoff;
662 	vm_prot_t prot;
663 
664 	addr = (vm_offset_t) uap->addr;
665 	size = uap->len;
666 	prot = uap->prot & VM_PROT_ALL;
667 
668 	pageoff = (addr & PAGE_MASK);
669 	addr -= pageoff;
670 	size += pageoff;
671 	size = (vm_size_t) round_page(size);
672 	if (addr + size < addr)
673 		return (EINVAL);
674 
675 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
676 	    addr + size, prot, FALSE)) {
677 	case KERN_SUCCESS:
678 		return (0);
679 	case KERN_PROTECTION_FAILURE:
680 		return (EACCES);
681 	case KERN_RESOURCE_SHORTAGE:
682 		return (ENOMEM);
683 	}
684 	return (EINVAL);
685 }
686 
687 #ifndef _SYS_SYSPROTO_H_
688 struct minherit_args {
689 	void *addr;
690 	size_t len;
691 	int inherit;
692 };
693 #endif
694 /*
695  * MPSAFE
696  */
697 int
698 sys_minherit(td, uap)
699 	struct thread *td;
700 	struct minherit_args *uap;
701 {
702 	vm_offset_t addr;
703 	vm_size_t size, pageoff;
704 	vm_inherit_t inherit;
705 
706 	addr = (vm_offset_t)uap->addr;
707 	size = uap->len;
708 	inherit = uap->inherit;
709 
710 	pageoff = (addr & PAGE_MASK);
711 	addr -= pageoff;
712 	size += pageoff;
713 	size = (vm_size_t) round_page(size);
714 	if (addr + size < addr)
715 		return (EINVAL);
716 
717 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
718 	    addr + size, inherit)) {
719 	case KERN_SUCCESS:
720 		return (0);
721 	case KERN_PROTECTION_FAILURE:
722 		return (EACCES);
723 	}
724 	return (EINVAL);
725 }
726 
727 #ifndef _SYS_SYSPROTO_H_
728 struct madvise_args {
729 	void *addr;
730 	size_t len;
731 	int behav;
732 };
733 #endif
734 
735 /*
736  * MPSAFE
737  */
738 int
739 sys_madvise(td, uap)
740 	struct thread *td;
741 	struct madvise_args *uap;
742 {
743 	vm_offset_t start, end;
744 	vm_map_t map;
745 	int flags;
746 
747 	/*
748 	 * Check for our special case, advising the swap pager we are
749 	 * "immortal."
750 	 */
751 	if (uap->behav == MADV_PROTECT) {
752 		flags = PPROT_SET;
753 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
754 		    PROC_SPROTECT, &flags));
755 	}
756 
757 	/*
758 	 * Check for illegal behavior
759 	 */
760 	if (uap->behav < 0 || uap->behav > MADV_CORE)
761 		return (EINVAL);
762 	/*
763 	 * Check for illegal addresses.  Watch out for address wrap... Note
764 	 * that VM_*_ADDRESS are not constants due to casts (argh).
765 	 */
766 	map = &td->td_proc->p_vmspace->vm_map;
767 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
768 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
769 		return (EINVAL);
770 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
771 		return (EINVAL);
772 
773 	/*
774 	 * Since this routine is only advisory, we default to conservative
775 	 * behavior.
776 	 */
777 	start = trunc_page((vm_offset_t) uap->addr);
778 	end = round_page((vm_offset_t) uap->addr + uap->len);
779 
780 	if (vm_map_madvise(map, start, end, uap->behav))
781 		return (EINVAL);
782 	return (0);
783 }
784 
785 #ifndef _SYS_SYSPROTO_H_
786 struct mincore_args {
787 	const void *addr;
788 	size_t len;
789 	char *vec;
790 };
791 #endif
792 
793 /*
794  * MPSAFE
795  */
796 int
797 sys_mincore(td, uap)
798 	struct thread *td;
799 	struct mincore_args *uap;
800 {
801 	vm_offset_t addr, first_addr;
802 	vm_offset_t end, cend;
803 	pmap_t pmap;
804 	vm_map_t map;
805 	char *vec;
806 	int error = 0;
807 	int vecindex, lastvecindex;
808 	vm_map_entry_t current;
809 	vm_map_entry_t entry;
810 	vm_object_t object;
811 	vm_paddr_t locked_pa;
812 	vm_page_t m;
813 	vm_pindex_t pindex;
814 	int mincoreinfo;
815 	unsigned int timestamp;
816 	boolean_t locked;
817 
818 	/*
819 	 * Make sure that the addresses presented are valid for user
820 	 * mode.
821 	 */
822 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
823 	end = addr + (vm_size_t)round_page(uap->len);
824 	map = &td->td_proc->p_vmspace->vm_map;
825 	if (end > vm_map_max(map) || end < addr)
826 		return (ENOMEM);
827 
828 	/*
829 	 * Address of byte vector
830 	 */
831 	vec = uap->vec;
832 
833 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
834 
835 	vm_map_lock_read(map);
836 RestartScan:
837 	timestamp = map->timestamp;
838 
839 	if (!vm_map_lookup_entry(map, addr, &entry)) {
840 		vm_map_unlock_read(map);
841 		return (ENOMEM);
842 	}
843 
844 	/*
845 	 * Do this on a map entry basis so that if the pages are not
846 	 * in the current processes address space, we can easily look
847 	 * up the pages elsewhere.
848 	 */
849 	lastvecindex = -1;
850 	for (current = entry;
851 	    (current != &map->header) && (current->start < end);
852 	    current = current->next) {
853 
854 		/*
855 		 * check for contiguity
856 		 */
857 		if (current->end < end &&
858 		    (entry->next == &map->header ||
859 		     current->next->start > current->end)) {
860 			vm_map_unlock_read(map);
861 			return (ENOMEM);
862 		}
863 
864 		/*
865 		 * ignore submaps (for now) or null objects
866 		 */
867 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
868 			current->object.vm_object == NULL)
869 			continue;
870 
871 		/*
872 		 * limit this scan to the current map entry and the
873 		 * limits for the mincore call
874 		 */
875 		if (addr < current->start)
876 			addr = current->start;
877 		cend = current->end;
878 		if (cend > end)
879 			cend = end;
880 
881 		/*
882 		 * scan this entry one page at a time
883 		 */
884 		while (addr < cend) {
885 			/*
886 			 * Check pmap first, it is likely faster, also
887 			 * it can provide info as to whether we are the
888 			 * one referencing or modifying the page.
889 			 */
890 			object = NULL;
891 			locked_pa = 0;
892 		retry:
893 			m = NULL;
894 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
895 			if (locked_pa != 0) {
896 				/*
897 				 * The page is mapped by this process but not
898 				 * both accessed and modified.  It is also
899 				 * managed.  Acquire the object lock so that
900 				 * other mappings might be examined.
901 				 */
902 				m = PHYS_TO_VM_PAGE(locked_pa);
903 				if (m->object != object) {
904 					if (object != NULL)
905 						VM_OBJECT_WUNLOCK(object);
906 					object = m->object;
907 					locked = VM_OBJECT_TRYWLOCK(object);
908 					vm_page_unlock(m);
909 					if (!locked) {
910 						VM_OBJECT_WLOCK(object);
911 						vm_page_lock(m);
912 						goto retry;
913 					}
914 				} else
915 					vm_page_unlock(m);
916 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
917 				    ("mincore: page %p is mapped but invalid",
918 				    m));
919 			} else if (mincoreinfo == 0) {
920 				/*
921 				 * The page is not mapped by this process.  If
922 				 * the object implements managed pages, then
923 				 * determine if the page is resident so that
924 				 * the mappings might be examined.
925 				 */
926 				if (current->object.vm_object != object) {
927 					if (object != NULL)
928 						VM_OBJECT_WUNLOCK(object);
929 					object = current->object.vm_object;
930 					VM_OBJECT_WLOCK(object);
931 				}
932 				if (object->type == OBJT_DEFAULT ||
933 				    object->type == OBJT_SWAP ||
934 				    object->type == OBJT_VNODE) {
935 					pindex = OFF_TO_IDX(current->offset +
936 					    (addr - current->start));
937 					m = vm_page_lookup(object, pindex);
938 					if (m == NULL &&
939 					    vm_page_is_cached(object, pindex))
940 						mincoreinfo = MINCORE_INCORE;
941 					if (m != NULL && m->valid == 0)
942 						m = NULL;
943 					if (m != NULL)
944 						mincoreinfo = MINCORE_INCORE;
945 				}
946 			}
947 			if (m != NULL) {
948 				/* Examine other mappings to the page. */
949 				if (m->dirty == 0 && pmap_is_modified(m))
950 					vm_page_dirty(m);
951 				if (m->dirty != 0)
952 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
953 				/*
954 				 * The first test for PGA_REFERENCED is an
955 				 * optimization.  The second test is
956 				 * required because a concurrent pmap
957 				 * operation could clear the last reference
958 				 * and set PGA_REFERENCED before the call to
959 				 * pmap_is_referenced().
960 				 */
961 				if ((m->aflags & PGA_REFERENCED) != 0 ||
962 				    pmap_is_referenced(m) ||
963 				    (m->aflags & PGA_REFERENCED) != 0)
964 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
965 			}
966 			if (object != NULL)
967 				VM_OBJECT_WUNLOCK(object);
968 
969 			/*
970 			 * subyte may page fault.  In case it needs to modify
971 			 * the map, we release the lock.
972 			 */
973 			vm_map_unlock_read(map);
974 
975 			/*
976 			 * calculate index into user supplied byte vector
977 			 */
978 			vecindex = OFF_TO_IDX(addr - first_addr);
979 
980 			/*
981 			 * If we have skipped map entries, we need to make sure that
982 			 * the byte vector is zeroed for those skipped entries.
983 			 */
984 			while ((lastvecindex + 1) < vecindex) {
985 				++lastvecindex;
986 				error = subyte(vec + lastvecindex, 0);
987 				if (error) {
988 					error = EFAULT;
989 					goto done2;
990 				}
991 			}
992 
993 			/*
994 			 * Pass the page information to the user
995 			 */
996 			error = subyte(vec + vecindex, mincoreinfo);
997 			if (error) {
998 				error = EFAULT;
999 				goto done2;
1000 			}
1001 
1002 			/*
1003 			 * If the map has changed, due to the subyte, the previous
1004 			 * output may be invalid.
1005 			 */
1006 			vm_map_lock_read(map);
1007 			if (timestamp != map->timestamp)
1008 				goto RestartScan;
1009 
1010 			lastvecindex = vecindex;
1011 			addr += PAGE_SIZE;
1012 		}
1013 	}
1014 
1015 	/*
1016 	 * subyte may page fault.  In case it needs to modify
1017 	 * the map, we release the lock.
1018 	 */
1019 	vm_map_unlock_read(map);
1020 
1021 	/*
1022 	 * Zero the last entries in the byte vector.
1023 	 */
1024 	vecindex = OFF_TO_IDX(end - first_addr);
1025 	while ((lastvecindex + 1) < vecindex) {
1026 		++lastvecindex;
1027 		error = subyte(vec + lastvecindex, 0);
1028 		if (error) {
1029 			error = EFAULT;
1030 			goto done2;
1031 		}
1032 	}
1033 
1034 	/*
1035 	 * If the map has changed, due to the subyte, the previous
1036 	 * output may be invalid.
1037 	 */
1038 	vm_map_lock_read(map);
1039 	if (timestamp != map->timestamp)
1040 		goto RestartScan;
1041 	vm_map_unlock_read(map);
1042 done2:
1043 	return (error);
1044 }
1045 
1046 #ifndef _SYS_SYSPROTO_H_
1047 struct mlock_args {
1048 	const void *addr;
1049 	size_t len;
1050 };
1051 #endif
1052 /*
1053  * MPSAFE
1054  */
1055 int
1056 sys_mlock(td, uap)
1057 	struct thread *td;
1058 	struct mlock_args *uap;
1059 {
1060 
1061 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1062 }
1063 
1064 int
1065 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1066 {
1067 	vm_offset_t addr, end, last, start;
1068 	vm_size_t npages, size;
1069 	vm_map_t map;
1070 	unsigned long nsize;
1071 	int error;
1072 
1073 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1074 	if (error)
1075 		return (error);
1076 	addr = (vm_offset_t)addr0;
1077 	size = len;
1078 	last = addr + size;
1079 	start = trunc_page(addr);
1080 	end = round_page(last);
1081 	if (last < addr || end < addr)
1082 		return (EINVAL);
1083 	npages = atop(end - start);
1084 	if (npages > vm_page_max_wired)
1085 		return (ENOMEM);
1086 	map = &proc->p_vmspace->vm_map;
1087 	PROC_LOCK(proc);
1088 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1089 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1090 		PROC_UNLOCK(proc);
1091 		return (ENOMEM);
1092 	}
1093 	PROC_UNLOCK(proc);
1094 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1095 		return (EAGAIN);
1096 #ifdef RACCT
1097 	PROC_LOCK(proc);
1098 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1099 	PROC_UNLOCK(proc);
1100 	if (error != 0)
1101 		return (ENOMEM);
1102 #endif
1103 	error = vm_map_wire(map, start, end,
1104 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1105 #ifdef RACCT
1106 	if (error != KERN_SUCCESS) {
1107 		PROC_LOCK(proc);
1108 		racct_set(proc, RACCT_MEMLOCK,
1109 		    ptoa(pmap_wired_count(map->pmap)));
1110 		PROC_UNLOCK(proc);
1111 	}
1112 #endif
1113 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1114 }
1115 
1116 #ifndef _SYS_SYSPROTO_H_
1117 struct mlockall_args {
1118 	int	how;
1119 };
1120 #endif
1121 
1122 /*
1123  * MPSAFE
1124  */
1125 int
1126 sys_mlockall(td, uap)
1127 	struct thread *td;
1128 	struct mlockall_args *uap;
1129 {
1130 	vm_map_t map;
1131 	int error;
1132 
1133 	map = &td->td_proc->p_vmspace->vm_map;
1134 	error = priv_check(td, PRIV_VM_MLOCK);
1135 	if (error)
1136 		return (error);
1137 
1138 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1139 		return (EINVAL);
1140 
1141 	/*
1142 	 * If wiring all pages in the process would cause it to exceed
1143 	 * a hard resource limit, return ENOMEM.
1144 	 */
1145 	if (!old_mlock && uap->how & MCL_CURRENT) {
1146 		PROC_LOCK(td->td_proc);
1147 		if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1148 			PROC_UNLOCK(td->td_proc);
1149 			return (ENOMEM);
1150 		}
1151 		PROC_UNLOCK(td->td_proc);
1152 	}
1153 #ifdef RACCT
1154 	PROC_LOCK(td->td_proc);
1155 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1156 	PROC_UNLOCK(td->td_proc);
1157 	if (error != 0)
1158 		return (ENOMEM);
1159 #endif
1160 
1161 	if (uap->how & MCL_FUTURE) {
1162 		vm_map_lock(map);
1163 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1164 		vm_map_unlock(map);
1165 		error = 0;
1166 	}
1167 
1168 	if (uap->how & MCL_CURRENT) {
1169 		/*
1170 		 * P1003.1-2001 mandates that all currently mapped pages
1171 		 * will be memory resident and locked (wired) upon return
1172 		 * from mlockall(). vm_map_wire() will wire pages, by
1173 		 * calling vm_fault_wire() for each page in the region.
1174 		 */
1175 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1176 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1177 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1178 	}
1179 #ifdef RACCT
1180 	if (error != KERN_SUCCESS) {
1181 		PROC_LOCK(td->td_proc);
1182 		racct_set(td->td_proc, RACCT_MEMLOCK,
1183 		    ptoa(pmap_wired_count(map->pmap)));
1184 		PROC_UNLOCK(td->td_proc);
1185 	}
1186 #endif
1187 
1188 	return (error);
1189 }
1190 
1191 #ifndef _SYS_SYSPROTO_H_
1192 struct munlockall_args {
1193 	register_t dummy;
1194 };
1195 #endif
1196 
1197 /*
1198  * MPSAFE
1199  */
1200 int
1201 sys_munlockall(td, uap)
1202 	struct thread *td;
1203 	struct munlockall_args *uap;
1204 {
1205 	vm_map_t map;
1206 	int error;
1207 
1208 	map = &td->td_proc->p_vmspace->vm_map;
1209 	error = priv_check(td, PRIV_VM_MUNLOCK);
1210 	if (error)
1211 		return (error);
1212 
1213 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1214 	vm_map_lock(map);
1215 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1216 	vm_map_unlock(map);
1217 
1218 	/* Forcibly unwire all pages. */
1219 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1220 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1221 #ifdef RACCT
1222 	if (error == KERN_SUCCESS) {
1223 		PROC_LOCK(td->td_proc);
1224 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1225 		PROC_UNLOCK(td->td_proc);
1226 	}
1227 #endif
1228 
1229 	return (error);
1230 }
1231 
1232 #ifndef _SYS_SYSPROTO_H_
1233 struct munlock_args {
1234 	const void *addr;
1235 	size_t len;
1236 };
1237 #endif
1238 /*
1239  * MPSAFE
1240  */
1241 int
1242 sys_munlock(td, uap)
1243 	struct thread *td;
1244 	struct munlock_args *uap;
1245 {
1246 	vm_offset_t addr, end, last, start;
1247 	vm_size_t size;
1248 #ifdef RACCT
1249 	vm_map_t map;
1250 #endif
1251 	int error;
1252 
1253 	error = priv_check(td, PRIV_VM_MUNLOCK);
1254 	if (error)
1255 		return (error);
1256 	addr = (vm_offset_t)uap->addr;
1257 	size = uap->len;
1258 	last = addr + size;
1259 	start = trunc_page(addr);
1260 	end = round_page(last);
1261 	if (last < addr || end < addr)
1262 		return (EINVAL);
1263 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1264 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1265 #ifdef RACCT
1266 	if (error == KERN_SUCCESS) {
1267 		PROC_LOCK(td->td_proc);
1268 		map = &td->td_proc->p_vmspace->vm_map;
1269 		racct_set(td->td_proc, RACCT_MEMLOCK,
1270 		    ptoa(pmap_wired_count(map->pmap)));
1271 		PROC_UNLOCK(td->td_proc);
1272 	}
1273 #endif
1274 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1275 }
1276 
1277 /*
1278  * vm_mmap_vnode()
1279  *
1280  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1281  * operations on vnodes.
1282  *
1283  * For VCHR vnodes, the vnode lock is held over the call to
1284  * vm_mmap_cdev() to keep vp->v_rdev valid.
1285  */
1286 int
1287 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1288     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1289     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1290     boolean_t *writecounted)
1291 {
1292 	struct vattr va;
1293 	vm_object_t obj;
1294 	vm_offset_t foff;
1295 	struct mount *mp;
1296 	struct ucred *cred;
1297 	int error, flags, locktype;
1298 
1299 	mp = vp->v_mount;
1300 	cred = td->td_ucred;
1301 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1302 		locktype = LK_EXCLUSIVE;
1303 	else
1304 		locktype = LK_SHARED;
1305 	if ((error = vget(vp, locktype, td)) != 0)
1306 		return (error);
1307 	foff = *foffp;
1308 	flags = *flagsp;
1309 	obj = vp->v_object;
1310 	if (vp->v_type == VREG) {
1311 		/*
1312 		 * Get the proper underlying object
1313 		 */
1314 		if (obj == NULL) {
1315 			error = EINVAL;
1316 			goto done;
1317 		}
1318 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1319 			vput(vp);
1320 			vp = (struct vnode *)obj->handle;
1321 			/*
1322 			 * Bypass filesystems obey the mpsafety of the
1323 			 * underlying fs.  Tmpfs never bypasses.
1324 			 */
1325 			error = vget(vp, locktype, td);
1326 			if (error != 0)
1327 				return (error);
1328 		}
1329 		if (locktype == LK_EXCLUSIVE) {
1330 			*writecounted = TRUE;
1331 			vnode_pager_update_writecount(obj, 0, objsize);
1332 		}
1333 	} else if (vp->v_type == VCHR) {
1334 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1335 		    vp->v_rdev, foffp, objp);
1336 		if (error == 0)
1337 			goto mark_atime;
1338 		goto done;
1339 	} else {
1340 		error = EINVAL;
1341 		goto done;
1342 	}
1343 	if ((error = VOP_GETATTR(vp, &va, cred)))
1344 		goto done;
1345 #ifdef MAC
1346 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1347 	if (error != 0)
1348 		goto done;
1349 #endif
1350 	if ((flags & MAP_SHARED) != 0) {
1351 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1352 			if (prot & PROT_WRITE) {
1353 				error = EPERM;
1354 				goto done;
1355 			}
1356 			*maxprotp &= ~VM_PROT_WRITE;
1357 		}
1358 	}
1359 	/*
1360 	 * If it is a regular file without any references
1361 	 * we do not need to sync it.
1362 	 * Adjust object size to be the size of actual file.
1363 	 */
1364 	objsize = round_page(va.va_size);
1365 	if (va.va_nlink == 0)
1366 		flags |= MAP_NOSYNC;
1367 	if (obj->type == OBJT_VNODE)
1368 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1369 		    cred);
1370 	else {
1371 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1372 		    ("wrong object type"));
1373 		vm_object_reference(obj);
1374 	}
1375 	if (obj == NULL) {
1376 		error = ENOMEM;
1377 		goto done;
1378 	}
1379 	*objp = obj;
1380 	*flagsp = flags;
1381 
1382 mark_atime:
1383 	vfs_mark_atime(vp, cred);
1384 
1385 done:
1386 	if (error != 0 && *writecounted) {
1387 		*writecounted = FALSE;
1388 		vnode_pager_update_writecount(obj, objsize, 0);
1389 	}
1390 	vput(vp);
1391 	return (error);
1392 }
1393 
1394 /*
1395  * vm_mmap_cdev()
1396  *
1397  * MPSAFE
1398  *
1399  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1400  * operations on cdevs.
1401  */
1402 int
1403 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1404     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1405     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1406 {
1407 	vm_object_t obj;
1408 	struct cdevsw *dsw;
1409 	int error, flags, ref;
1410 
1411 	flags = *flagsp;
1412 
1413 	dsw = dev_refthread(cdev, &ref);
1414 	if (dsw == NULL)
1415 		return (ENXIO);
1416 	if (dsw->d_flags & D_MMAP_ANON) {
1417 		dev_relthread(cdev, ref);
1418 		*maxprotp = VM_PROT_ALL;
1419 		*flagsp |= MAP_ANON;
1420 		return (0);
1421 	}
1422 	/*
1423 	 * cdevs do not provide private mappings of any kind.
1424 	 */
1425 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1426 	    (prot & PROT_WRITE) != 0) {
1427 		dev_relthread(cdev, ref);
1428 		return (EACCES);
1429 	}
1430 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1431 		dev_relthread(cdev, ref);
1432 		return (EINVAL);
1433 	}
1434 	/*
1435 	 * Force device mappings to be shared.
1436 	 */
1437 	flags |= MAP_SHARED;
1438 #ifdef MAC_XXX
1439 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1440 	if (error != 0) {
1441 		dev_relthread(cdev, ref);
1442 		return (error);
1443 	}
1444 #endif
1445 	/*
1446 	 * First, try d_mmap_single().  If that is not implemented
1447 	 * (returns ENODEV), fall back to using the device pager.
1448 	 * Note that d_mmap_single() must return a reference to the
1449 	 * object (it needs to bump the reference count of the object
1450 	 * it returns somehow).
1451 	 *
1452 	 * XXX assumes VM_PROT_* == PROT_*
1453 	 */
1454 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1455 	dev_relthread(cdev, ref);
1456 	if (error != ENODEV)
1457 		return (error);
1458 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1459 	    td->td_ucred);
1460 	if (obj == NULL)
1461 		return (EINVAL);
1462 	*objp = obj;
1463 	*flagsp = flags;
1464 	return (0);
1465 }
1466 
1467 /*
1468  * vm_mmap_shm()
1469  *
1470  * MPSAFE
1471  *
1472  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1473  * operations on shm file descriptors.
1474  */
1475 int
1476 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1477     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1478     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1479 {
1480 	int error;
1481 
1482 	if ((*flagsp & MAP_SHARED) != 0 &&
1483 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1484 	    (prot & PROT_WRITE) != 0)
1485 		return (EACCES);
1486 #ifdef MAC
1487 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1488 	if (error != 0)
1489 		return (error);
1490 #endif
1491 	error = shm_mmap(shmfd, objsize, foff, objp);
1492 	if (error)
1493 		return (error);
1494 	return (0);
1495 }
1496 
1497 /*
1498  * vm_mmap()
1499  *
1500  * MPSAFE
1501  *
1502  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1503  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1504  */
1505 int
1506 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1507 	vm_prot_t maxprot, int flags,
1508 	objtype_t handle_type, void *handle,
1509 	vm_ooffset_t foff)
1510 {
1511 	boolean_t fitit;
1512 	vm_object_t object = NULL;
1513 	struct thread *td = curthread;
1514 	int docow, error, findspace, rv;
1515 	boolean_t writecounted;
1516 
1517 	if (size == 0)
1518 		return (0);
1519 
1520 	size = round_page(size);
1521 
1522 	if (map == &td->td_proc->p_vmspace->vm_map) {
1523 		PROC_LOCK(td->td_proc);
1524 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1525 			PROC_UNLOCK(td->td_proc);
1526 			return (ENOMEM);
1527 		}
1528 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1529 			PROC_UNLOCK(td->td_proc);
1530 			return (ENOMEM);
1531 		}
1532 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1533 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1534 			    lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1535 				racct_set_force(td->td_proc, RACCT_VMEM,
1536 				    map->size);
1537 				PROC_UNLOCK(td->td_proc);
1538 				return (ENOMEM);
1539 			}
1540 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1541 			    ptoa(pmap_wired_count(map->pmap)) + size);
1542 			if (error != 0) {
1543 				racct_set_force(td->td_proc, RACCT_VMEM,
1544 				    map->size);
1545 				PROC_UNLOCK(td->td_proc);
1546 				return (error);
1547 			}
1548 		}
1549 		PROC_UNLOCK(td->td_proc);
1550 	}
1551 
1552 	/*
1553 	 * We currently can only deal with page aligned file offsets.
1554 	 * The check is here rather than in the syscall because the
1555 	 * kernel calls this function internally for other mmaping
1556 	 * operations (such as in exec) and non-aligned offsets will
1557 	 * cause pmap inconsistencies...so we want to be sure to
1558 	 * disallow this in all cases.
1559 	 */
1560 	if (foff & PAGE_MASK)
1561 		return (EINVAL);
1562 
1563 	if ((flags & MAP_FIXED) == 0) {
1564 		fitit = TRUE;
1565 		*addr = round_page(*addr);
1566 	} else {
1567 		if (*addr != trunc_page(*addr))
1568 			return (EINVAL);
1569 		fitit = FALSE;
1570 	}
1571 	writecounted = FALSE;
1572 
1573 	/*
1574 	 * Lookup/allocate object.
1575 	 */
1576 	switch (handle_type) {
1577 	case OBJT_DEVICE:
1578 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1579 		    handle, &foff, &object);
1580 		break;
1581 	case OBJT_VNODE:
1582 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1583 		    handle, &foff, &object, &writecounted);
1584 		break;
1585 	case OBJT_SWAP:
1586 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1587 		    handle, foff, &object);
1588 		break;
1589 	case OBJT_DEFAULT:
1590 		if (handle == NULL) {
1591 			error = 0;
1592 			break;
1593 		}
1594 		/* FALLTHROUGH */
1595 	default:
1596 		error = EINVAL;
1597 		break;
1598 	}
1599 	if (error)
1600 		return (error);
1601 	if (flags & MAP_ANON) {
1602 		object = NULL;
1603 		docow = 0;
1604 		/*
1605 		 * Unnamed anonymous regions always start at 0.
1606 		 */
1607 		if (handle == 0)
1608 			foff = 0;
1609 	} else if (flags & MAP_PREFAULT_READ)
1610 		docow = MAP_PREFAULT;
1611 	else
1612 		docow = MAP_PREFAULT_PARTIAL;
1613 
1614 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1615 		docow |= MAP_COPY_ON_WRITE;
1616 	if (flags & MAP_NOSYNC)
1617 		docow |= MAP_DISABLE_SYNCER;
1618 	if (flags & MAP_NOCORE)
1619 		docow |= MAP_DISABLE_COREDUMP;
1620 	/* Shared memory is also shared with children. */
1621 	if (flags & MAP_SHARED)
1622 		docow |= MAP_INHERIT_SHARE;
1623 	if (writecounted)
1624 		docow |= MAP_VN_WRITECOUNT;
1625 	if (flags & MAP_STACK) {
1626 		if (object != NULL)
1627 			return (EINVAL);
1628 		docow |= MAP_STACK_GROWS_DOWN;
1629 	}
1630 	if ((flags & MAP_EXCL) != 0)
1631 		docow |= MAP_CHECK_EXCL;
1632 
1633 	if (fitit) {
1634 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1635 			findspace = VMFS_SUPER_SPACE;
1636 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1637 			findspace = VMFS_ALIGNED_SPACE(flags >>
1638 			    MAP_ALIGNMENT_SHIFT);
1639 		else
1640 			findspace = VMFS_OPTIMAL_SPACE;
1641 		rv = vm_map_find(map, object, foff, addr, size,
1642 #ifdef MAP_32BIT
1643 		    flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1644 #endif
1645 		    0, findspace, prot, maxprot, docow);
1646 	} else {
1647 		rv = vm_map_fixed(map, object, foff, *addr, size,
1648 		    prot, maxprot, docow);
1649 	}
1650 
1651 	if (rv == KERN_SUCCESS) {
1652 		/*
1653 		 * If the process has requested that all future mappings
1654 		 * be wired, then heed this.
1655 		 */
1656 		if (map->flags & MAP_WIREFUTURE) {
1657 			vm_map_wire(map, *addr, *addr + size,
1658 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1659 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1660 		}
1661 	} else {
1662 		/*
1663 		 * If this mapping was accounted for in the vnode's
1664 		 * writecount, then undo that now.
1665 		 */
1666 		if (writecounted)
1667 			vnode_pager_release_writecount(object, 0, size);
1668 		/*
1669 		 * Lose the object reference.  Will destroy the
1670 		 * object if it's an unnamed anonymous mapping
1671 		 * or named anonymous without other references.
1672 		 */
1673 		vm_object_deallocate(object);
1674 	}
1675 	return (vm_mmap_to_errno(rv));
1676 }
1677 
1678 /*
1679  * Translate a Mach VM return code to zero on success or the appropriate errno
1680  * on failure.
1681  */
1682 int
1683 vm_mmap_to_errno(int rv)
1684 {
1685 
1686 	switch (rv) {
1687 	case KERN_SUCCESS:
1688 		return (0);
1689 	case KERN_INVALID_ADDRESS:
1690 	case KERN_NO_SPACE:
1691 		return (ENOMEM);
1692 	case KERN_PROTECTION_FAILURE:
1693 		return (EACCES);
1694 	default:
1695 		return (EINVAL);
1696 	}
1697 }
1698