xref: /freebsd/sys/vm/vm_mmap.c (revision 3fc9e2c36555140de248a0b4def91bbfa44d7c2c)
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/capability.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/racct.h>
60 #include <sys/resource.h>
61 #include <sys/resourcevar.h>
62 #include <sys/rwlock.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 #include <sys/fcntl.h>
66 #include <sys/file.h>
67 #include <sys/mman.h>
68 #include <sys/mount.h>
69 #include <sys/conf.h>
70 #include <sys/stat.h>
71 #include <sys/sysent.h>
72 #include <sys/vmmeter.h>
73 
74 #include <security/mac/mac_framework.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_pager.h>
83 #include <vm/vm_pageout.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_page.h>
86 #include <vm/vnode_pager.h>
87 
88 #ifdef HWPMC_HOOKS
89 #include <sys/pmckern.h>
90 #endif
91 
92 int old_mlock = 0;
93 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RW | CTLFLAG_TUN, &old_mlock, 0,
94     "Do not apply RLIMIT_MEMLOCK on mlockall");
95 TUNABLE_INT("vm.old_mlock", &old_mlock);
96 
97 #ifndef _SYS_SYSPROTO_H_
98 struct sbrk_args {
99 	int incr;
100 };
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 /*
111  * MPSAFE
112  */
113 /* ARGSUSED */
114 int
115 sys_sbrk(td, uap)
116 	struct thread *td;
117 	struct sbrk_args *uap;
118 {
119 	/* Not yet implemented */
120 	return (EOPNOTSUPP);
121 }
122 
123 #ifndef _SYS_SYSPROTO_H_
124 struct sstk_args {
125 	int incr;
126 };
127 #endif
128 
129 /*
130  * MPSAFE
131  */
132 /* ARGSUSED */
133 int
134 sys_sstk(td, uap)
135 	struct thread *td;
136 	struct sstk_args *uap;
137 {
138 	/* Not yet implemented */
139 	return (EOPNOTSUPP);
140 }
141 
142 #if defined(COMPAT_43)
143 #ifndef _SYS_SYSPROTO_H_
144 struct getpagesize_args {
145 	int dummy;
146 };
147 #endif
148 
149 int
150 ogetpagesize(td, uap)
151 	struct thread *td;
152 	struct getpagesize_args *uap;
153 {
154 	/* MP SAFE */
155 	td->td_retval[0] = PAGE_SIZE;
156 	return (0);
157 }
158 #endif				/* COMPAT_43 */
159 
160 
161 /*
162  * Memory Map (mmap) system call.  Note that the file offset
163  * and address are allowed to be NOT page aligned, though if
164  * the MAP_FIXED flag it set, both must have the same remainder
165  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
166  * page-aligned, the actual mapping starts at trunc_page(addr)
167  * and the return value is adjusted up by the page offset.
168  *
169  * Generally speaking, only character devices which are themselves
170  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
171  * there would be no cache coherency between a descriptor and a VM mapping
172  * both to the same character device.
173  */
174 #ifndef _SYS_SYSPROTO_H_
175 struct mmap_args {
176 	void *addr;
177 	size_t len;
178 	int prot;
179 	int flags;
180 	int fd;
181 	long pad;
182 	off_t pos;
183 };
184 #endif
185 
186 /*
187  * MPSAFE
188  */
189 int
190 sys_mmap(td, uap)
191 	struct thread *td;
192 	struct mmap_args *uap;
193 {
194 #ifdef HWPMC_HOOKS
195 	struct pmckern_map_in pkm;
196 #endif
197 	struct file *fp;
198 	struct vnode *vp;
199 	vm_offset_t addr;
200 	vm_size_t size, pageoff;
201 	vm_prot_t cap_maxprot, prot, maxprot;
202 	void *handle;
203 	objtype_t handle_type;
204 	int align, error, flags;
205 	off_t pos;
206 	struct vmspace *vms = td->td_proc->p_vmspace;
207 	cap_rights_t rights;
208 
209 	addr = (vm_offset_t) uap->addr;
210 	size = uap->len;
211 	prot = uap->prot & VM_PROT_ALL;
212 	flags = uap->flags;
213 	pos = uap->pos;
214 
215 	fp = NULL;
216 
217 	/*
218 	 * Enforce the constraints.
219 	 * Mapping of length 0 is only allowed for old binaries.
220 	 * Anonymous mapping shall specify -1 as filedescriptor and
221 	 * zero position for new code. Be nice to ancient a.out
222 	 * binaries and correct pos for anonymous mapping, since old
223 	 * ld.so sometimes issues anonymous map requests with non-zero
224 	 * pos.
225 	 */
226 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
227 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
228 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
229 			return (EINVAL);
230 	} else {
231 		if ((flags & MAP_ANON) != 0)
232 			pos = 0;
233 	}
234 
235 	if (flags & MAP_STACK) {
236 		if ((uap->fd != -1) ||
237 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
238 			return (EINVAL);
239 		flags |= MAP_ANON;
240 		pos = 0;
241 	}
242 
243 	/*
244 	 * Align the file position to a page boundary,
245 	 * and save its page offset component.
246 	 */
247 	pageoff = (pos & PAGE_MASK);
248 	pos -= pageoff;
249 
250 	/* Adjust size for rounding (on both ends). */
251 	size += pageoff;			/* low end... */
252 	size = (vm_size_t) round_page(size);	/* hi end */
253 
254 	/* Ensure alignment is at least a page and fits in a pointer. */
255 	align = flags & MAP_ALIGNMENT_MASK;
256 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
257 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
258 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
259 		return (EINVAL);
260 
261 	/*
262 	 * Check for illegal addresses.  Watch out for address wrap... Note
263 	 * that VM_*_ADDRESS are not constants due to casts (argh).
264 	 */
265 	if (flags & MAP_FIXED) {
266 		/*
267 		 * The specified address must have the same remainder
268 		 * as the file offset taken modulo PAGE_SIZE, so it
269 		 * should be aligned after adjustment by pageoff.
270 		 */
271 		addr -= pageoff;
272 		if (addr & PAGE_MASK)
273 			return (EINVAL);
274 
275 		/* Address range must be all in user VM space. */
276 		if (addr < vm_map_min(&vms->vm_map) ||
277 		    addr + size > vm_map_max(&vms->vm_map))
278 			return (EINVAL);
279 		if (addr + size < addr)
280 			return (EINVAL);
281 	} else {
282 		/*
283 		 * XXX for non-fixed mappings where no hint is provided or
284 		 * the hint would fall in the potential heap space,
285 		 * place it after the end of the largest possible heap.
286 		 *
287 		 * There should really be a pmap call to determine a reasonable
288 		 * location.
289 		 */
290 		PROC_LOCK(td->td_proc);
291 		if (addr == 0 ||
292 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
293 		    addr < round_page((vm_offset_t)vms->vm_daddr +
294 		    lim_max(td->td_proc, RLIMIT_DATA))))
295 			addr = round_page((vm_offset_t)vms->vm_daddr +
296 			    lim_max(td->td_proc, RLIMIT_DATA));
297 		PROC_UNLOCK(td->td_proc);
298 	}
299 	if (flags & MAP_ANON) {
300 		/*
301 		 * Mapping blank space is trivial.
302 		 */
303 		handle = NULL;
304 		handle_type = OBJT_DEFAULT;
305 		maxprot = VM_PROT_ALL;
306 		cap_maxprot = VM_PROT_ALL;
307 	} else {
308 		/*
309 		 * Mapping file, get fp for validation and don't let the
310 		 * descriptor disappear on us if we block. Check capability
311 		 * rights, but also return the maximum rights to be combined
312 		 * with maxprot later.
313 		 */
314 		cap_rights_init(&rights, CAP_MMAP);
315 		if (prot & PROT_READ)
316 			cap_rights_set(&rights, CAP_MMAP_R);
317 		if ((flags & MAP_SHARED) != 0) {
318 			if (prot & PROT_WRITE)
319 				cap_rights_set(&rights, CAP_MMAP_W);
320 		}
321 		if (prot & PROT_EXEC)
322 			cap_rights_set(&rights, CAP_MMAP_X);
323 		error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
324 		if (error != 0)
325 			goto done;
326 		if (fp->f_type == DTYPE_SHM) {
327 			handle = fp->f_data;
328 			handle_type = OBJT_SWAP;
329 			maxprot = VM_PROT_NONE;
330 
331 			/* FREAD should always be set. */
332 			if (fp->f_flag & FREAD)
333 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
334 			if (fp->f_flag & FWRITE)
335 				maxprot |= VM_PROT_WRITE;
336 			goto map;
337 		}
338 		if (fp->f_type != DTYPE_VNODE) {
339 			error = ENODEV;
340 			goto done;
341 		}
342 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
343     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
344 		/*
345 		 * POSIX shared-memory objects are defined to have
346 		 * kernel persistence, and are not defined to support
347 		 * read(2)/write(2) -- or even open(2).  Thus, we can
348 		 * use MAP_ASYNC to trade on-disk coherence for speed.
349 		 * The shm_open(3) library routine turns on the FPOSIXSHM
350 		 * flag to request this behavior.
351 		 */
352 		if (fp->f_flag & FPOSIXSHM)
353 			flags |= MAP_NOSYNC;
354 #endif
355 		vp = fp->f_vnode;
356 		/*
357 		 * Ensure that file and memory protections are
358 		 * compatible.  Note that we only worry about
359 		 * writability if mapping is shared; in this case,
360 		 * current and max prot are dictated by the open file.
361 		 * XXX use the vnode instead?  Problem is: what
362 		 * credentials do we use for determination? What if
363 		 * proc does a setuid?
364 		 */
365 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
366 			maxprot = VM_PROT_NONE;
367 		else
368 			maxprot = VM_PROT_EXECUTE;
369 		if (fp->f_flag & FREAD) {
370 			maxprot |= VM_PROT_READ;
371 		} else if (prot & PROT_READ) {
372 			error = EACCES;
373 			goto done;
374 		}
375 		/*
376 		 * If we are sharing potential changes (either via
377 		 * MAP_SHARED or via the implicit sharing of character
378 		 * device mappings), and we are trying to get write
379 		 * permission although we opened it without asking
380 		 * for it, bail out.
381 		 */
382 		if ((flags & MAP_SHARED) != 0) {
383 			if ((fp->f_flag & FWRITE) != 0) {
384 				maxprot |= VM_PROT_WRITE;
385 			} else if ((prot & PROT_WRITE) != 0) {
386 				error = EACCES;
387 				goto done;
388 			}
389 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
390 			maxprot |= VM_PROT_WRITE;
391 			cap_maxprot |= VM_PROT_WRITE;
392 		}
393 		handle = (void *)vp;
394 		handle_type = OBJT_VNODE;
395 	}
396 map:
397 	td->td_fpop = fp;
398 	maxprot &= cap_maxprot;
399 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
400 	    flags, handle_type, handle, pos);
401 	td->td_fpop = NULL;
402 #ifdef HWPMC_HOOKS
403 	/* inform hwpmc(4) if an executable is being mapped */
404 	if (error == 0 && handle_type == OBJT_VNODE &&
405 	    (prot & PROT_EXEC)) {
406 		pkm.pm_file = handle;
407 		pkm.pm_address = (uintptr_t) addr;
408 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
409 	}
410 #endif
411 	if (error == 0)
412 		td->td_retval[0] = (register_t) (addr + pageoff);
413 done:
414 	if (fp)
415 		fdrop(fp, td);
416 
417 	return (error);
418 }
419 
420 int
421 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
422 {
423 	struct mmap_args oargs;
424 
425 	oargs.addr = uap->addr;
426 	oargs.len = uap->len;
427 	oargs.prot = uap->prot;
428 	oargs.flags = uap->flags;
429 	oargs.fd = uap->fd;
430 	oargs.pos = uap->pos;
431 	return (sys_mmap(td, &oargs));
432 }
433 
434 #ifdef COMPAT_43
435 #ifndef _SYS_SYSPROTO_H_
436 struct ommap_args {
437 	caddr_t addr;
438 	int len;
439 	int prot;
440 	int flags;
441 	int fd;
442 	long pos;
443 };
444 #endif
445 int
446 ommap(td, uap)
447 	struct thread *td;
448 	struct ommap_args *uap;
449 {
450 	struct mmap_args nargs;
451 	static const char cvtbsdprot[8] = {
452 		0,
453 		PROT_EXEC,
454 		PROT_WRITE,
455 		PROT_EXEC | PROT_WRITE,
456 		PROT_READ,
457 		PROT_EXEC | PROT_READ,
458 		PROT_WRITE | PROT_READ,
459 		PROT_EXEC | PROT_WRITE | PROT_READ,
460 	};
461 
462 #define	OMAP_ANON	0x0002
463 #define	OMAP_COPY	0x0020
464 #define	OMAP_SHARED	0x0010
465 #define	OMAP_FIXED	0x0100
466 
467 	nargs.addr = uap->addr;
468 	nargs.len = uap->len;
469 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
470 #ifdef COMPAT_FREEBSD32
471 #if defined(__amd64__) || defined(__ia64__)
472 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
473 	    nargs.prot != 0)
474 		nargs.prot |= PROT_EXEC;
475 #endif
476 #endif
477 	nargs.flags = 0;
478 	if (uap->flags & OMAP_ANON)
479 		nargs.flags |= MAP_ANON;
480 	if (uap->flags & OMAP_COPY)
481 		nargs.flags |= MAP_COPY;
482 	if (uap->flags & OMAP_SHARED)
483 		nargs.flags |= MAP_SHARED;
484 	else
485 		nargs.flags |= MAP_PRIVATE;
486 	if (uap->flags & OMAP_FIXED)
487 		nargs.flags |= MAP_FIXED;
488 	nargs.fd = uap->fd;
489 	nargs.pos = uap->pos;
490 	return (sys_mmap(td, &nargs));
491 }
492 #endif				/* COMPAT_43 */
493 
494 
495 #ifndef _SYS_SYSPROTO_H_
496 struct msync_args {
497 	void *addr;
498 	size_t len;
499 	int flags;
500 };
501 #endif
502 /*
503  * MPSAFE
504  */
505 int
506 sys_msync(td, uap)
507 	struct thread *td;
508 	struct msync_args *uap;
509 {
510 	vm_offset_t addr;
511 	vm_size_t size, pageoff;
512 	int flags;
513 	vm_map_t map;
514 	int rv;
515 
516 	addr = (vm_offset_t) uap->addr;
517 	size = uap->len;
518 	flags = uap->flags;
519 
520 	pageoff = (addr & PAGE_MASK);
521 	addr -= pageoff;
522 	size += pageoff;
523 	size = (vm_size_t) round_page(size);
524 	if (addr + size < addr)
525 		return (EINVAL);
526 
527 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
528 		return (EINVAL);
529 
530 	map = &td->td_proc->p_vmspace->vm_map;
531 
532 	/*
533 	 * Clean the pages and interpret the return value.
534 	 */
535 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
536 	    (flags & MS_INVALIDATE) != 0);
537 	switch (rv) {
538 	case KERN_SUCCESS:
539 		return (0);
540 	case KERN_INVALID_ADDRESS:
541 		return (EINVAL);	/* Sun returns ENOMEM? */
542 	case KERN_INVALID_ARGUMENT:
543 		return (EBUSY);
544 	case KERN_FAILURE:
545 		return (EIO);
546 	default:
547 		return (EINVAL);
548 	}
549 }
550 
551 #ifndef _SYS_SYSPROTO_H_
552 struct munmap_args {
553 	void *addr;
554 	size_t len;
555 };
556 #endif
557 /*
558  * MPSAFE
559  */
560 int
561 sys_munmap(td, uap)
562 	struct thread *td;
563 	struct munmap_args *uap;
564 {
565 #ifdef HWPMC_HOOKS
566 	struct pmckern_map_out pkm;
567 	vm_map_entry_t entry;
568 #endif
569 	vm_offset_t addr;
570 	vm_size_t size, pageoff;
571 	vm_map_t map;
572 
573 	addr = (vm_offset_t) uap->addr;
574 	size = uap->len;
575 	if (size == 0)
576 		return (EINVAL);
577 
578 	pageoff = (addr & PAGE_MASK);
579 	addr -= pageoff;
580 	size += pageoff;
581 	size = (vm_size_t) round_page(size);
582 	if (addr + size < addr)
583 		return (EINVAL);
584 
585 	/*
586 	 * Check for illegal addresses.  Watch out for address wrap...
587 	 */
588 	map = &td->td_proc->p_vmspace->vm_map;
589 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
590 		return (EINVAL);
591 	vm_map_lock(map);
592 #ifdef HWPMC_HOOKS
593 	/*
594 	 * Inform hwpmc if the address range being unmapped contains
595 	 * an executable region.
596 	 */
597 	pkm.pm_address = (uintptr_t) NULL;
598 	if (vm_map_lookup_entry(map, addr, &entry)) {
599 		for (;
600 		     entry != &map->header && entry->start < addr + size;
601 		     entry = entry->next) {
602 			if (vm_map_check_protection(map, entry->start,
603 				entry->end, VM_PROT_EXECUTE) == TRUE) {
604 				pkm.pm_address = (uintptr_t) addr;
605 				pkm.pm_size = (size_t) size;
606 				break;
607 			}
608 		}
609 	}
610 #endif
611 	vm_map_delete(map, addr, addr + size);
612 
613 #ifdef HWPMC_HOOKS
614 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
615 	vm_map_lock_downgrade(map);
616 	if (pkm.pm_address != (uintptr_t) NULL)
617 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
618 	vm_map_unlock_read(map);
619 #else
620 	vm_map_unlock(map);
621 #endif
622 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
623 	return (0);
624 }
625 
626 #ifndef _SYS_SYSPROTO_H_
627 struct mprotect_args {
628 	const void *addr;
629 	size_t len;
630 	int prot;
631 };
632 #endif
633 /*
634  * MPSAFE
635  */
636 int
637 sys_mprotect(td, uap)
638 	struct thread *td;
639 	struct mprotect_args *uap;
640 {
641 	vm_offset_t addr;
642 	vm_size_t size, pageoff;
643 	vm_prot_t prot;
644 
645 	addr = (vm_offset_t) uap->addr;
646 	size = uap->len;
647 	prot = uap->prot & VM_PROT_ALL;
648 
649 	pageoff = (addr & PAGE_MASK);
650 	addr -= pageoff;
651 	size += pageoff;
652 	size = (vm_size_t) round_page(size);
653 	if (addr + size < addr)
654 		return (EINVAL);
655 
656 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
657 	    addr + size, prot, FALSE)) {
658 	case KERN_SUCCESS:
659 		return (0);
660 	case KERN_PROTECTION_FAILURE:
661 		return (EACCES);
662 	case KERN_RESOURCE_SHORTAGE:
663 		return (ENOMEM);
664 	}
665 	return (EINVAL);
666 }
667 
668 #ifndef _SYS_SYSPROTO_H_
669 struct minherit_args {
670 	void *addr;
671 	size_t len;
672 	int inherit;
673 };
674 #endif
675 /*
676  * MPSAFE
677  */
678 int
679 sys_minherit(td, uap)
680 	struct thread *td;
681 	struct minherit_args *uap;
682 {
683 	vm_offset_t addr;
684 	vm_size_t size, pageoff;
685 	vm_inherit_t inherit;
686 
687 	addr = (vm_offset_t)uap->addr;
688 	size = uap->len;
689 	inherit = uap->inherit;
690 
691 	pageoff = (addr & PAGE_MASK);
692 	addr -= pageoff;
693 	size += pageoff;
694 	size = (vm_size_t) round_page(size);
695 	if (addr + size < addr)
696 		return (EINVAL);
697 
698 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
699 	    addr + size, inherit)) {
700 	case KERN_SUCCESS:
701 		return (0);
702 	case KERN_PROTECTION_FAILURE:
703 		return (EACCES);
704 	}
705 	return (EINVAL);
706 }
707 
708 #ifndef _SYS_SYSPROTO_H_
709 struct madvise_args {
710 	void *addr;
711 	size_t len;
712 	int behav;
713 };
714 #endif
715 
716 /*
717  * MPSAFE
718  */
719 int
720 sys_madvise(td, uap)
721 	struct thread *td;
722 	struct madvise_args *uap;
723 {
724 	vm_offset_t start, end;
725 	vm_map_t map;
726 	struct proc *p;
727 	int error;
728 
729 	/*
730 	 * Check for our special case, advising the swap pager we are
731 	 * "immortal."
732 	 */
733 	if (uap->behav == MADV_PROTECT) {
734 		error = priv_check(td, PRIV_VM_MADV_PROTECT);
735 		if (error == 0) {
736 			p = td->td_proc;
737 			PROC_LOCK(p);
738 			p->p_flag |= P_PROTECTED;
739 			PROC_UNLOCK(p);
740 		}
741 		return (error);
742 	}
743 	/*
744 	 * Check for illegal behavior
745 	 */
746 	if (uap->behav < 0 || uap->behav > MADV_CORE)
747 		return (EINVAL);
748 	/*
749 	 * Check for illegal addresses.  Watch out for address wrap... Note
750 	 * that VM_*_ADDRESS are not constants due to casts (argh).
751 	 */
752 	map = &td->td_proc->p_vmspace->vm_map;
753 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
754 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
755 		return (EINVAL);
756 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
757 		return (EINVAL);
758 
759 	/*
760 	 * Since this routine is only advisory, we default to conservative
761 	 * behavior.
762 	 */
763 	start = trunc_page((vm_offset_t) uap->addr);
764 	end = round_page((vm_offset_t) uap->addr + uap->len);
765 
766 	if (vm_map_madvise(map, start, end, uap->behav))
767 		return (EINVAL);
768 	return (0);
769 }
770 
771 #ifndef _SYS_SYSPROTO_H_
772 struct mincore_args {
773 	const void *addr;
774 	size_t len;
775 	char *vec;
776 };
777 #endif
778 
779 /*
780  * MPSAFE
781  */
782 int
783 sys_mincore(td, uap)
784 	struct thread *td;
785 	struct mincore_args *uap;
786 {
787 	vm_offset_t addr, first_addr;
788 	vm_offset_t end, cend;
789 	pmap_t pmap;
790 	vm_map_t map;
791 	char *vec;
792 	int error = 0;
793 	int vecindex, lastvecindex;
794 	vm_map_entry_t current;
795 	vm_map_entry_t entry;
796 	vm_object_t object;
797 	vm_paddr_t locked_pa;
798 	vm_page_t m;
799 	vm_pindex_t pindex;
800 	int mincoreinfo;
801 	unsigned int timestamp;
802 	boolean_t locked;
803 
804 	/*
805 	 * Make sure that the addresses presented are valid for user
806 	 * mode.
807 	 */
808 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
809 	end = addr + (vm_size_t)round_page(uap->len);
810 	map = &td->td_proc->p_vmspace->vm_map;
811 	if (end > vm_map_max(map) || end < addr)
812 		return (ENOMEM);
813 
814 	/*
815 	 * Address of byte vector
816 	 */
817 	vec = uap->vec;
818 
819 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
820 
821 	vm_map_lock_read(map);
822 RestartScan:
823 	timestamp = map->timestamp;
824 
825 	if (!vm_map_lookup_entry(map, addr, &entry)) {
826 		vm_map_unlock_read(map);
827 		return (ENOMEM);
828 	}
829 
830 	/*
831 	 * Do this on a map entry basis so that if the pages are not
832 	 * in the current processes address space, we can easily look
833 	 * up the pages elsewhere.
834 	 */
835 	lastvecindex = -1;
836 	for (current = entry;
837 	    (current != &map->header) && (current->start < end);
838 	    current = current->next) {
839 
840 		/*
841 		 * check for contiguity
842 		 */
843 		if (current->end < end &&
844 		    (entry->next == &map->header ||
845 		     current->next->start > current->end)) {
846 			vm_map_unlock_read(map);
847 			return (ENOMEM);
848 		}
849 
850 		/*
851 		 * ignore submaps (for now) or null objects
852 		 */
853 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
854 			current->object.vm_object == NULL)
855 			continue;
856 
857 		/*
858 		 * limit this scan to the current map entry and the
859 		 * limits for the mincore call
860 		 */
861 		if (addr < current->start)
862 			addr = current->start;
863 		cend = current->end;
864 		if (cend > end)
865 			cend = end;
866 
867 		/*
868 		 * scan this entry one page at a time
869 		 */
870 		while (addr < cend) {
871 			/*
872 			 * Check pmap first, it is likely faster, also
873 			 * it can provide info as to whether we are the
874 			 * one referencing or modifying the page.
875 			 */
876 			object = NULL;
877 			locked_pa = 0;
878 		retry:
879 			m = NULL;
880 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
881 			if (locked_pa != 0) {
882 				/*
883 				 * The page is mapped by this process but not
884 				 * both accessed and modified.  It is also
885 				 * managed.  Acquire the object lock so that
886 				 * other mappings might be examined.
887 				 */
888 				m = PHYS_TO_VM_PAGE(locked_pa);
889 				if (m->object != object) {
890 					if (object != NULL)
891 						VM_OBJECT_WUNLOCK(object);
892 					object = m->object;
893 					locked = VM_OBJECT_TRYWLOCK(object);
894 					vm_page_unlock(m);
895 					if (!locked) {
896 						VM_OBJECT_WLOCK(object);
897 						vm_page_lock(m);
898 						goto retry;
899 					}
900 				} else
901 					vm_page_unlock(m);
902 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
903 				    ("mincore: page %p is mapped but invalid",
904 				    m));
905 			} else if (mincoreinfo == 0) {
906 				/*
907 				 * The page is not mapped by this process.  If
908 				 * the object implements managed pages, then
909 				 * determine if the page is resident so that
910 				 * the mappings might be examined.
911 				 */
912 				if (current->object.vm_object != object) {
913 					if (object != NULL)
914 						VM_OBJECT_WUNLOCK(object);
915 					object = current->object.vm_object;
916 					VM_OBJECT_WLOCK(object);
917 				}
918 				if (object->type == OBJT_DEFAULT ||
919 				    object->type == OBJT_SWAP ||
920 				    object->type == OBJT_VNODE) {
921 					pindex = OFF_TO_IDX(current->offset +
922 					    (addr - current->start));
923 					m = vm_page_lookup(object, pindex);
924 					if (m == NULL &&
925 					    vm_page_is_cached(object, pindex))
926 						mincoreinfo = MINCORE_INCORE;
927 					if (m != NULL && m->valid == 0)
928 						m = NULL;
929 					if (m != NULL)
930 						mincoreinfo = MINCORE_INCORE;
931 				}
932 			}
933 			if (m != NULL) {
934 				/* Examine other mappings to the page. */
935 				if (m->dirty == 0 && pmap_is_modified(m))
936 					vm_page_dirty(m);
937 				if (m->dirty != 0)
938 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
939 				/*
940 				 * The first test for PGA_REFERENCED is an
941 				 * optimization.  The second test is
942 				 * required because a concurrent pmap
943 				 * operation could clear the last reference
944 				 * and set PGA_REFERENCED before the call to
945 				 * pmap_is_referenced().
946 				 */
947 				if ((m->aflags & PGA_REFERENCED) != 0 ||
948 				    pmap_is_referenced(m) ||
949 				    (m->aflags & PGA_REFERENCED) != 0)
950 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
951 			}
952 			if (object != NULL)
953 				VM_OBJECT_WUNLOCK(object);
954 
955 			/*
956 			 * subyte may page fault.  In case it needs to modify
957 			 * the map, we release the lock.
958 			 */
959 			vm_map_unlock_read(map);
960 
961 			/*
962 			 * calculate index into user supplied byte vector
963 			 */
964 			vecindex = OFF_TO_IDX(addr - first_addr);
965 
966 			/*
967 			 * If we have skipped map entries, we need to make sure that
968 			 * the byte vector is zeroed for those skipped entries.
969 			 */
970 			while ((lastvecindex + 1) < vecindex) {
971 				error = subyte(vec + lastvecindex, 0);
972 				if (error) {
973 					error = EFAULT;
974 					goto done2;
975 				}
976 				++lastvecindex;
977 			}
978 
979 			/*
980 			 * Pass the page information to the user
981 			 */
982 			error = subyte(vec + vecindex, mincoreinfo);
983 			if (error) {
984 				error = EFAULT;
985 				goto done2;
986 			}
987 
988 			/*
989 			 * If the map has changed, due to the subyte, the previous
990 			 * output may be invalid.
991 			 */
992 			vm_map_lock_read(map);
993 			if (timestamp != map->timestamp)
994 				goto RestartScan;
995 
996 			lastvecindex = vecindex;
997 			addr += PAGE_SIZE;
998 		}
999 	}
1000 
1001 	/*
1002 	 * subyte may page fault.  In case it needs to modify
1003 	 * the map, we release the lock.
1004 	 */
1005 	vm_map_unlock_read(map);
1006 
1007 	/*
1008 	 * Zero the last entries in the byte vector.
1009 	 */
1010 	vecindex = OFF_TO_IDX(end - first_addr);
1011 	while ((lastvecindex + 1) < vecindex) {
1012 		error = subyte(vec + lastvecindex, 0);
1013 		if (error) {
1014 			error = EFAULT;
1015 			goto done2;
1016 		}
1017 		++lastvecindex;
1018 	}
1019 
1020 	/*
1021 	 * If the map has changed, due to the subyte, the previous
1022 	 * output may be invalid.
1023 	 */
1024 	vm_map_lock_read(map);
1025 	if (timestamp != map->timestamp)
1026 		goto RestartScan;
1027 	vm_map_unlock_read(map);
1028 done2:
1029 	return (error);
1030 }
1031 
1032 #ifndef _SYS_SYSPROTO_H_
1033 struct mlock_args {
1034 	const void *addr;
1035 	size_t len;
1036 };
1037 #endif
1038 /*
1039  * MPSAFE
1040  */
1041 int
1042 sys_mlock(td, uap)
1043 	struct thread *td;
1044 	struct mlock_args *uap;
1045 {
1046 
1047 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1048 }
1049 
1050 int
1051 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1052 {
1053 	vm_offset_t addr, end, last, start;
1054 	vm_size_t npages, size;
1055 	vm_map_t map;
1056 	unsigned long nsize;
1057 	int error;
1058 
1059 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1060 	if (error)
1061 		return (error);
1062 	addr = (vm_offset_t)addr0;
1063 	size = len;
1064 	last = addr + size;
1065 	start = trunc_page(addr);
1066 	end = round_page(last);
1067 	if (last < addr || end < addr)
1068 		return (EINVAL);
1069 	npages = atop(end - start);
1070 	if (npages > vm_page_max_wired)
1071 		return (ENOMEM);
1072 	map = &proc->p_vmspace->vm_map;
1073 	PROC_LOCK(proc);
1074 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1075 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1076 		PROC_UNLOCK(proc);
1077 		return (ENOMEM);
1078 	}
1079 	PROC_UNLOCK(proc);
1080 	if (npages + cnt.v_wire_count > vm_page_max_wired)
1081 		return (EAGAIN);
1082 #ifdef RACCT
1083 	PROC_LOCK(proc);
1084 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1085 	PROC_UNLOCK(proc);
1086 	if (error != 0)
1087 		return (ENOMEM);
1088 #endif
1089 	error = vm_map_wire(map, start, end,
1090 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1091 #ifdef RACCT
1092 	if (error != KERN_SUCCESS) {
1093 		PROC_LOCK(proc);
1094 		racct_set(proc, RACCT_MEMLOCK,
1095 		    ptoa(pmap_wired_count(map->pmap)));
1096 		PROC_UNLOCK(proc);
1097 	}
1098 #endif
1099 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1100 }
1101 
1102 #ifndef _SYS_SYSPROTO_H_
1103 struct mlockall_args {
1104 	int	how;
1105 };
1106 #endif
1107 
1108 /*
1109  * MPSAFE
1110  */
1111 int
1112 sys_mlockall(td, uap)
1113 	struct thread *td;
1114 	struct mlockall_args *uap;
1115 {
1116 	vm_map_t map;
1117 	int error;
1118 
1119 	map = &td->td_proc->p_vmspace->vm_map;
1120 	error = priv_check(td, PRIV_VM_MLOCK);
1121 	if (error)
1122 		return (error);
1123 
1124 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1125 		return (EINVAL);
1126 
1127 	/*
1128 	 * If wiring all pages in the process would cause it to exceed
1129 	 * a hard resource limit, return ENOMEM.
1130 	 */
1131 	if (!old_mlock && uap->how & MCL_CURRENT) {
1132 		PROC_LOCK(td->td_proc);
1133 		if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1134 			PROC_UNLOCK(td->td_proc);
1135 			return (ENOMEM);
1136 		}
1137 		PROC_UNLOCK(td->td_proc);
1138 	}
1139 #ifdef RACCT
1140 	PROC_LOCK(td->td_proc);
1141 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1142 	PROC_UNLOCK(td->td_proc);
1143 	if (error != 0)
1144 		return (ENOMEM);
1145 #endif
1146 
1147 	if (uap->how & MCL_FUTURE) {
1148 		vm_map_lock(map);
1149 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1150 		vm_map_unlock(map);
1151 		error = 0;
1152 	}
1153 
1154 	if (uap->how & MCL_CURRENT) {
1155 		/*
1156 		 * P1003.1-2001 mandates that all currently mapped pages
1157 		 * will be memory resident and locked (wired) upon return
1158 		 * from mlockall(). vm_map_wire() will wire pages, by
1159 		 * calling vm_fault_wire() for each page in the region.
1160 		 */
1161 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1162 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1163 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1164 	}
1165 #ifdef RACCT
1166 	if (error != KERN_SUCCESS) {
1167 		PROC_LOCK(td->td_proc);
1168 		racct_set(td->td_proc, RACCT_MEMLOCK,
1169 		    ptoa(pmap_wired_count(map->pmap)));
1170 		PROC_UNLOCK(td->td_proc);
1171 	}
1172 #endif
1173 
1174 	return (error);
1175 }
1176 
1177 #ifndef _SYS_SYSPROTO_H_
1178 struct munlockall_args {
1179 	register_t dummy;
1180 };
1181 #endif
1182 
1183 /*
1184  * MPSAFE
1185  */
1186 int
1187 sys_munlockall(td, uap)
1188 	struct thread *td;
1189 	struct munlockall_args *uap;
1190 {
1191 	vm_map_t map;
1192 	int error;
1193 
1194 	map = &td->td_proc->p_vmspace->vm_map;
1195 	error = priv_check(td, PRIV_VM_MUNLOCK);
1196 	if (error)
1197 		return (error);
1198 
1199 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1200 	vm_map_lock(map);
1201 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1202 	vm_map_unlock(map);
1203 
1204 	/* Forcibly unwire all pages. */
1205 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1206 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1207 #ifdef RACCT
1208 	if (error == KERN_SUCCESS) {
1209 		PROC_LOCK(td->td_proc);
1210 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1211 		PROC_UNLOCK(td->td_proc);
1212 	}
1213 #endif
1214 
1215 	return (error);
1216 }
1217 
1218 #ifndef _SYS_SYSPROTO_H_
1219 struct munlock_args {
1220 	const void *addr;
1221 	size_t len;
1222 };
1223 #endif
1224 /*
1225  * MPSAFE
1226  */
1227 int
1228 sys_munlock(td, uap)
1229 	struct thread *td;
1230 	struct munlock_args *uap;
1231 {
1232 	vm_offset_t addr, end, last, start;
1233 	vm_size_t size;
1234 #ifdef RACCT
1235 	vm_map_t map;
1236 #endif
1237 	int error;
1238 
1239 	error = priv_check(td, PRIV_VM_MUNLOCK);
1240 	if (error)
1241 		return (error);
1242 	addr = (vm_offset_t)uap->addr;
1243 	size = uap->len;
1244 	last = addr + size;
1245 	start = trunc_page(addr);
1246 	end = round_page(last);
1247 	if (last < addr || end < addr)
1248 		return (EINVAL);
1249 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1250 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1251 #ifdef RACCT
1252 	if (error == KERN_SUCCESS) {
1253 		PROC_LOCK(td->td_proc);
1254 		map = &td->td_proc->p_vmspace->vm_map;
1255 		racct_set(td->td_proc, RACCT_MEMLOCK,
1256 		    ptoa(pmap_wired_count(map->pmap)));
1257 		PROC_UNLOCK(td->td_proc);
1258 	}
1259 #endif
1260 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1261 }
1262 
1263 /*
1264  * vm_mmap_vnode()
1265  *
1266  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1267  * operations on vnodes.
1268  *
1269  * For VCHR vnodes, the vnode lock is held over the call to
1270  * vm_mmap_cdev() to keep vp->v_rdev valid.
1271  */
1272 int
1273 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1274     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1275     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1276     boolean_t *writecounted)
1277 {
1278 	struct vattr va;
1279 	vm_object_t obj;
1280 	vm_offset_t foff;
1281 	struct mount *mp;
1282 	struct ucred *cred;
1283 	int error, flags, locktype;
1284 
1285 	mp = vp->v_mount;
1286 	cred = td->td_ucred;
1287 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1288 		locktype = LK_EXCLUSIVE;
1289 	else
1290 		locktype = LK_SHARED;
1291 	if ((error = vget(vp, locktype, td)) != 0)
1292 		return (error);
1293 	foff = *foffp;
1294 	flags = *flagsp;
1295 	obj = vp->v_object;
1296 	if (vp->v_type == VREG) {
1297 		/*
1298 		 * Get the proper underlying object
1299 		 */
1300 		if (obj == NULL) {
1301 			error = EINVAL;
1302 			goto done;
1303 		}
1304 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1305 			vput(vp);
1306 			vp = (struct vnode *)obj->handle;
1307 			/*
1308 			 * Bypass filesystems obey the mpsafety of the
1309 			 * underlying fs.  Tmpfs never bypasses.
1310 			 */
1311 			error = vget(vp, locktype, td);
1312 			if (error != 0)
1313 				return (error);
1314 		}
1315 		if (locktype == LK_EXCLUSIVE) {
1316 			*writecounted = TRUE;
1317 			vnode_pager_update_writecount(obj, 0, objsize);
1318 		}
1319 	} else if (vp->v_type == VCHR) {
1320 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1321 		    vp->v_rdev, foffp, objp);
1322 		if (error == 0)
1323 			goto mark_atime;
1324 		goto done;
1325 	} else {
1326 		error = EINVAL;
1327 		goto done;
1328 	}
1329 	if ((error = VOP_GETATTR(vp, &va, cred)))
1330 		goto done;
1331 #ifdef MAC
1332 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1333 	if (error != 0)
1334 		goto done;
1335 #endif
1336 	if ((flags & MAP_SHARED) != 0) {
1337 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1338 			if (prot & PROT_WRITE) {
1339 				error = EPERM;
1340 				goto done;
1341 			}
1342 			*maxprotp &= ~VM_PROT_WRITE;
1343 		}
1344 	}
1345 	/*
1346 	 * If it is a regular file without any references
1347 	 * we do not need to sync it.
1348 	 * Adjust object size to be the size of actual file.
1349 	 */
1350 	objsize = round_page(va.va_size);
1351 	if (va.va_nlink == 0)
1352 		flags |= MAP_NOSYNC;
1353 	if (obj->type == OBJT_VNODE)
1354 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1355 		    cred);
1356 	else {
1357 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1358 		    ("wrong object type"));
1359 		vm_object_reference(obj);
1360 	}
1361 	if (obj == NULL) {
1362 		error = ENOMEM;
1363 		goto done;
1364 	}
1365 	*objp = obj;
1366 	*flagsp = flags;
1367 
1368 mark_atime:
1369 	vfs_mark_atime(vp, cred);
1370 
1371 done:
1372 	if (error != 0 && *writecounted) {
1373 		*writecounted = FALSE;
1374 		vnode_pager_update_writecount(obj, objsize, 0);
1375 	}
1376 	vput(vp);
1377 	return (error);
1378 }
1379 
1380 /*
1381  * vm_mmap_cdev()
1382  *
1383  * MPSAFE
1384  *
1385  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1386  * operations on cdevs.
1387  */
1388 int
1389 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1390     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1391     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1392 {
1393 	vm_object_t obj;
1394 	struct cdevsw *dsw;
1395 	int error, flags, ref;
1396 
1397 	flags = *flagsp;
1398 
1399 	dsw = dev_refthread(cdev, &ref);
1400 	if (dsw == NULL)
1401 		return (ENXIO);
1402 	if (dsw->d_flags & D_MMAP_ANON) {
1403 		dev_relthread(cdev, ref);
1404 		*maxprotp = VM_PROT_ALL;
1405 		*flagsp |= MAP_ANON;
1406 		return (0);
1407 	}
1408 	/*
1409 	 * cdevs do not provide private mappings of any kind.
1410 	 */
1411 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1412 	    (prot & PROT_WRITE) != 0) {
1413 		dev_relthread(cdev, ref);
1414 		return (EACCES);
1415 	}
1416 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1417 		dev_relthread(cdev, ref);
1418 		return (EINVAL);
1419 	}
1420 	/*
1421 	 * Force device mappings to be shared.
1422 	 */
1423 	flags |= MAP_SHARED;
1424 #ifdef MAC_XXX
1425 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1426 	if (error != 0) {
1427 		dev_relthread(cdev, ref);
1428 		return (error);
1429 	}
1430 #endif
1431 	/*
1432 	 * First, try d_mmap_single().  If that is not implemented
1433 	 * (returns ENODEV), fall back to using the device pager.
1434 	 * Note that d_mmap_single() must return a reference to the
1435 	 * object (it needs to bump the reference count of the object
1436 	 * it returns somehow).
1437 	 *
1438 	 * XXX assumes VM_PROT_* == PROT_*
1439 	 */
1440 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1441 	dev_relthread(cdev, ref);
1442 	if (error != ENODEV)
1443 		return (error);
1444 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1445 	    td->td_ucred);
1446 	if (obj == NULL)
1447 		return (EINVAL);
1448 	*objp = obj;
1449 	*flagsp = flags;
1450 	return (0);
1451 }
1452 
1453 /*
1454  * vm_mmap_shm()
1455  *
1456  * MPSAFE
1457  *
1458  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1459  * operations on shm file descriptors.
1460  */
1461 int
1462 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1463     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1464     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1465 {
1466 	int error;
1467 
1468 	if ((*flagsp & MAP_SHARED) != 0 &&
1469 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1470 	    (prot & PROT_WRITE) != 0)
1471 		return (EACCES);
1472 #ifdef MAC
1473 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1474 	if (error != 0)
1475 		return (error);
1476 #endif
1477 	error = shm_mmap(shmfd, objsize, foff, objp);
1478 	if (error)
1479 		return (error);
1480 	return (0);
1481 }
1482 
1483 /*
1484  * vm_mmap()
1485  *
1486  * MPSAFE
1487  *
1488  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1489  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1490  */
1491 int
1492 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1493 	vm_prot_t maxprot, int flags,
1494 	objtype_t handle_type, void *handle,
1495 	vm_ooffset_t foff)
1496 {
1497 	boolean_t fitit;
1498 	vm_object_t object = NULL;
1499 	struct thread *td = curthread;
1500 	int docow, error, findspace, rv;
1501 	boolean_t writecounted;
1502 
1503 	if (size == 0)
1504 		return (0);
1505 
1506 	size = round_page(size);
1507 
1508 	if (map == &td->td_proc->p_vmspace->vm_map) {
1509 		PROC_LOCK(td->td_proc);
1510 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1511 			PROC_UNLOCK(td->td_proc);
1512 			return (ENOMEM);
1513 		}
1514 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1515 			PROC_UNLOCK(td->td_proc);
1516 			return (ENOMEM);
1517 		}
1518 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1519 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1520 			    lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1521 				racct_set_force(td->td_proc, RACCT_VMEM,
1522 				    map->size);
1523 				PROC_UNLOCK(td->td_proc);
1524 				return (ENOMEM);
1525 			}
1526 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1527 			    ptoa(pmap_wired_count(map->pmap)) + size);
1528 			if (error != 0) {
1529 				racct_set_force(td->td_proc, RACCT_VMEM,
1530 				    map->size);
1531 				PROC_UNLOCK(td->td_proc);
1532 				return (error);
1533 			}
1534 		}
1535 		PROC_UNLOCK(td->td_proc);
1536 	}
1537 
1538 	/*
1539 	 * We currently can only deal with page aligned file offsets.
1540 	 * The check is here rather than in the syscall because the
1541 	 * kernel calls this function internally for other mmaping
1542 	 * operations (such as in exec) and non-aligned offsets will
1543 	 * cause pmap inconsistencies...so we want to be sure to
1544 	 * disallow this in all cases.
1545 	 */
1546 	if (foff & PAGE_MASK)
1547 		return (EINVAL);
1548 
1549 	if ((flags & MAP_FIXED) == 0) {
1550 		fitit = TRUE;
1551 		*addr = round_page(*addr);
1552 	} else {
1553 		if (*addr != trunc_page(*addr))
1554 			return (EINVAL);
1555 		fitit = FALSE;
1556 	}
1557 	writecounted = FALSE;
1558 
1559 	/*
1560 	 * Lookup/allocate object.
1561 	 */
1562 	switch (handle_type) {
1563 	case OBJT_DEVICE:
1564 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1565 		    handle, &foff, &object);
1566 		break;
1567 	case OBJT_VNODE:
1568 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1569 		    handle, &foff, &object, &writecounted);
1570 		break;
1571 	case OBJT_SWAP:
1572 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1573 		    handle, foff, &object);
1574 		break;
1575 	case OBJT_DEFAULT:
1576 		if (handle == NULL) {
1577 			error = 0;
1578 			break;
1579 		}
1580 		/* FALLTHROUGH */
1581 	default:
1582 		error = EINVAL;
1583 		break;
1584 	}
1585 	if (error)
1586 		return (error);
1587 	if (flags & MAP_ANON) {
1588 		object = NULL;
1589 		docow = 0;
1590 		/*
1591 		 * Unnamed anonymous regions always start at 0.
1592 		 */
1593 		if (handle == 0)
1594 			foff = 0;
1595 	} else if (flags & MAP_PREFAULT_READ)
1596 		docow = MAP_PREFAULT;
1597 	else
1598 		docow = MAP_PREFAULT_PARTIAL;
1599 
1600 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1601 		docow |= MAP_COPY_ON_WRITE;
1602 	if (flags & MAP_NOSYNC)
1603 		docow |= MAP_DISABLE_SYNCER;
1604 	if (flags & MAP_NOCORE)
1605 		docow |= MAP_DISABLE_COREDUMP;
1606 	/* Shared memory is also shared with children. */
1607 	if (flags & MAP_SHARED)
1608 		docow |= MAP_INHERIT_SHARE;
1609 	if (writecounted)
1610 		docow |= MAP_VN_WRITECOUNT;
1611 
1612 	if (flags & MAP_STACK)
1613 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1614 		    docow | MAP_STACK_GROWS_DOWN);
1615 	else if (fitit) {
1616 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1617 			findspace = VMFS_SUPER_SPACE;
1618 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1619 			findspace = VMFS_ALIGNED_SPACE(flags >>
1620 			    MAP_ALIGNMENT_SHIFT);
1621 		else
1622 			findspace = VMFS_OPTIMAL_SPACE;
1623 		rv = vm_map_find(map, object, foff, addr, size, findspace,
1624 		    prot, maxprot, docow);
1625 	} else
1626 		rv = vm_map_fixed(map, object, foff, *addr, size,
1627 				 prot, maxprot, docow);
1628 
1629 	if (rv == KERN_SUCCESS) {
1630 		/*
1631 		 * If the process has requested that all future mappings
1632 		 * be wired, then heed this.
1633 		 */
1634 		if (map->flags & MAP_WIREFUTURE) {
1635 			vm_map_wire(map, *addr, *addr + size,
1636 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1637 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1638 		}
1639 	} else {
1640 		/*
1641 		 * If this mapping was accounted for in the vnode's
1642 		 * writecount, then undo that now.
1643 		 */
1644 		if (writecounted)
1645 			vnode_pager_release_writecount(object, 0, size);
1646 		/*
1647 		 * Lose the object reference.  Will destroy the
1648 		 * object if it's an unnamed anonymous mapping
1649 		 * or named anonymous without other references.
1650 		 */
1651 		vm_object_deallocate(object);
1652 	}
1653 	return (vm_mmap_to_errno(rv));
1654 }
1655 
1656 /*
1657  * Translate a Mach VM return code to zero on success or the appropriate errno
1658  * on failure.
1659  */
1660 int
1661 vm_mmap_to_errno(int rv)
1662 {
1663 
1664 	switch (rv) {
1665 	case KERN_SUCCESS:
1666 		return (0);
1667 	case KERN_INVALID_ADDRESS:
1668 	case KERN_NO_SPACE:
1669 		return (ENOMEM);
1670 	case KERN_PROTECTION_FAILURE:
1671 		return (EACCES);
1672 	default:
1673 		return (EINVAL);
1674 	}
1675 }
1676