xref: /freebsd/sys/vm/vm_mmap.c (revision ddd5b8e9b4d8957fce018c520657cdfa4ecffad3)
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 flags, error;
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 	/*
255 	 * Check for illegal addresses.  Watch out for address wrap... Note
256 	 * that VM_*_ADDRESS are not constants due to casts (argh).
257 	 */
258 	if (flags & MAP_FIXED) {
259 		/*
260 		 * The specified address must have the same remainder
261 		 * as the file offset taken modulo PAGE_SIZE, so it
262 		 * should be aligned after adjustment by pageoff.
263 		 */
264 		addr -= pageoff;
265 		if (addr & PAGE_MASK)
266 			return (EINVAL);
267 
268 		/* Address range must be all in user VM space. */
269 		if (addr < vm_map_min(&vms->vm_map) ||
270 		    addr + size > vm_map_max(&vms->vm_map))
271 			return (EINVAL);
272 		if (addr + size < addr)
273 			return (EINVAL);
274 	} else {
275 		/*
276 		 * XXX for non-fixed mappings where no hint is provided or
277 		 * the hint would fall in the potential heap space,
278 		 * place it after the end of the largest possible heap.
279 		 *
280 		 * There should really be a pmap call to determine a reasonable
281 		 * location.
282 		 */
283 		PROC_LOCK(td->td_proc);
284 		if (addr == 0 ||
285 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
286 		    addr < round_page((vm_offset_t)vms->vm_daddr +
287 		    lim_max(td->td_proc, RLIMIT_DATA))))
288 			addr = round_page((vm_offset_t)vms->vm_daddr +
289 			    lim_max(td->td_proc, RLIMIT_DATA));
290 		PROC_UNLOCK(td->td_proc);
291 	}
292 	if (flags & MAP_ANON) {
293 		/*
294 		 * Mapping blank space is trivial.
295 		 */
296 		handle = NULL;
297 		handle_type = OBJT_DEFAULT;
298 		maxprot = VM_PROT_ALL;
299 		cap_maxprot = VM_PROT_ALL;
300 	} else {
301 		/*
302 		 * Mapping file, get fp for validation and don't let the
303 		 * descriptor disappear on us if we block. Check capability
304 		 * rights, but also return the maximum rights to be combined
305 		 * with maxprot later.
306 		 */
307 		rights = CAP_MMAP;
308 		if (prot & PROT_READ)
309 			rights |= CAP_MMAP_R;
310 		if ((flags & MAP_SHARED) != 0) {
311 			if (prot & PROT_WRITE)
312 				rights |= CAP_MMAP_W;
313 		}
314 		if (prot & PROT_EXEC)
315 			rights |= CAP_MMAP_X;
316 		if ((error = fget_mmap(td, uap->fd, rights, &cap_maxprot,
317 		    &fp)) != 0)
318 			goto done;
319 		if (fp->f_type == DTYPE_SHM) {
320 			handle = fp->f_data;
321 			handle_type = OBJT_SWAP;
322 			maxprot = VM_PROT_NONE;
323 
324 			/* FREAD should always be set. */
325 			if (fp->f_flag & FREAD)
326 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
327 			if (fp->f_flag & FWRITE)
328 				maxprot |= VM_PROT_WRITE;
329 			goto map;
330 		}
331 		if (fp->f_type != DTYPE_VNODE) {
332 			error = ENODEV;
333 			goto done;
334 		}
335 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
336     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
337 		/*
338 		 * POSIX shared-memory objects are defined to have
339 		 * kernel persistence, and are not defined to support
340 		 * read(2)/write(2) -- or even open(2).  Thus, we can
341 		 * use MAP_ASYNC to trade on-disk coherence for speed.
342 		 * The shm_open(3) library routine turns on the FPOSIXSHM
343 		 * flag to request this behavior.
344 		 */
345 		if (fp->f_flag & FPOSIXSHM)
346 			flags |= MAP_NOSYNC;
347 #endif
348 		vp = fp->f_vnode;
349 		/*
350 		 * Ensure that file and memory protections are
351 		 * compatible.  Note that we only worry about
352 		 * writability if mapping is shared; in this case,
353 		 * current and max prot are dictated by the open file.
354 		 * XXX use the vnode instead?  Problem is: what
355 		 * credentials do we use for determination? What if
356 		 * proc does a setuid?
357 		 */
358 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
359 			maxprot = VM_PROT_NONE;
360 		else
361 			maxprot = VM_PROT_EXECUTE;
362 		if (fp->f_flag & FREAD) {
363 			maxprot |= VM_PROT_READ;
364 		} else if (prot & PROT_READ) {
365 			error = EACCES;
366 			goto done;
367 		}
368 		/*
369 		 * If we are sharing potential changes (either via
370 		 * MAP_SHARED or via the implicit sharing of character
371 		 * device mappings), and we are trying to get write
372 		 * permission although we opened it without asking
373 		 * for it, bail out.
374 		 */
375 		if ((flags & MAP_SHARED) != 0) {
376 			if ((fp->f_flag & FWRITE) != 0) {
377 				maxprot |= VM_PROT_WRITE;
378 			} else if ((prot & PROT_WRITE) != 0) {
379 				error = EACCES;
380 				goto done;
381 			}
382 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
383 			maxprot |= VM_PROT_WRITE;
384 			cap_maxprot |= VM_PROT_WRITE;
385 		}
386 		handle = (void *)vp;
387 		handle_type = OBJT_VNODE;
388 	}
389 map:
390 	td->td_fpop = fp;
391 	maxprot &= cap_maxprot;
392 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
393 	    flags, handle_type, handle, pos);
394 	td->td_fpop = NULL;
395 #ifdef HWPMC_HOOKS
396 	/* inform hwpmc(4) if an executable is being mapped */
397 	if (error == 0 && handle_type == OBJT_VNODE &&
398 	    (prot & PROT_EXEC)) {
399 		pkm.pm_file = handle;
400 		pkm.pm_address = (uintptr_t) addr;
401 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
402 	}
403 #endif
404 	if (error == 0)
405 		td->td_retval[0] = (register_t) (addr + pageoff);
406 done:
407 	if (fp)
408 		fdrop(fp, td);
409 
410 	return (error);
411 }
412 
413 int
414 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
415 {
416 	struct mmap_args oargs;
417 
418 	oargs.addr = uap->addr;
419 	oargs.len = uap->len;
420 	oargs.prot = uap->prot;
421 	oargs.flags = uap->flags;
422 	oargs.fd = uap->fd;
423 	oargs.pos = uap->pos;
424 	return (sys_mmap(td, &oargs));
425 }
426 
427 #ifdef COMPAT_43
428 #ifndef _SYS_SYSPROTO_H_
429 struct ommap_args {
430 	caddr_t addr;
431 	int len;
432 	int prot;
433 	int flags;
434 	int fd;
435 	long pos;
436 };
437 #endif
438 int
439 ommap(td, uap)
440 	struct thread *td;
441 	struct ommap_args *uap;
442 {
443 	struct mmap_args nargs;
444 	static const char cvtbsdprot[8] = {
445 		0,
446 		PROT_EXEC,
447 		PROT_WRITE,
448 		PROT_EXEC | PROT_WRITE,
449 		PROT_READ,
450 		PROT_EXEC | PROT_READ,
451 		PROT_WRITE | PROT_READ,
452 		PROT_EXEC | PROT_WRITE | PROT_READ,
453 	};
454 
455 #define	OMAP_ANON	0x0002
456 #define	OMAP_COPY	0x0020
457 #define	OMAP_SHARED	0x0010
458 #define	OMAP_FIXED	0x0100
459 
460 	nargs.addr = uap->addr;
461 	nargs.len = uap->len;
462 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
463 #ifdef COMPAT_FREEBSD32
464 #if defined(__amd64__) || defined(__ia64__)
465 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
466 	    nargs.prot != 0)
467 		nargs.prot |= PROT_EXEC;
468 #endif
469 #endif
470 	nargs.flags = 0;
471 	if (uap->flags & OMAP_ANON)
472 		nargs.flags |= MAP_ANON;
473 	if (uap->flags & OMAP_COPY)
474 		nargs.flags |= MAP_COPY;
475 	if (uap->flags & OMAP_SHARED)
476 		nargs.flags |= MAP_SHARED;
477 	else
478 		nargs.flags |= MAP_PRIVATE;
479 	if (uap->flags & OMAP_FIXED)
480 		nargs.flags |= MAP_FIXED;
481 	nargs.fd = uap->fd;
482 	nargs.pos = uap->pos;
483 	return (sys_mmap(td, &nargs));
484 }
485 #endif				/* COMPAT_43 */
486 
487 
488 #ifndef _SYS_SYSPROTO_H_
489 struct msync_args {
490 	void *addr;
491 	size_t len;
492 	int flags;
493 };
494 #endif
495 /*
496  * MPSAFE
497  */
498 int
499 sys_msync(td, uap)
500 	struct thread *td;
501 	struct msync_args *uap;
502 {
503 	vm_offset_t addr;
504 	vm_size_t size, pageoff;
505 	int flags;
506 	vm_map_t map;
507 	int rv;
508 
509 	addr = (vm_offset_t) uap->addr;
510 	size = uap->len;
511 	flags = uap->flags;
512 
513 	pageoff = (addr & PAGE_MASK);
514 	addr -= pageoff;
515 	size += pageoff;
516 	size = (vm_size_t) round_page(size);
517 	if (addr + size < addr)
518 		return (EINVAL);
519 
520 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
521 		return (EINVAL);
522 
523 	map = &td->td_proc->p_vmspace->vm_map;
524 
525 	/*
526 	 * Clean the pages and interpret the return value.
527 	 */
528 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
529 	    (flags & MS_INVALIDATE) != 0);
530 	switch (rv) {
531 	case KERN_SUCCESS:
532 		return (0);
533 	case KERN_INVALID_ADDRESS:
534 		return (EINVAL);	/* Sun returns ENOMEM? */
535 	case KERN_INVALID_ARGUMENT:
536 		return (EBUSY);
537 	case KERN_FAILURE:
538 		return (EIO);
539 	default:
540 		return (EINVAL);
541 	}
542 }
543 
544 #ifndef _SYS_SYSPROTO_H_
545 struct munmap_args {
546 	void *addr;
547 	size_t len;
548 };
549 #endif
550 /*
551  * MPSAFE
552  */
553 int
554 sys_munmap(td, uap)
555 	struct thread *td;
556 	struct munmap_args *uap;
557 {
558 #ifdef HWPMC_HOOKS
559 	struct pmckern_map_out pkm;
560 	vm_map_entry_t entry;
561 #endif
562 	vm_offset_t addr;
563 	vm_size_t size, pageoff;
564 	vm_map_t map;
565 
566 	addr = (vm_offset_t) uap->addr;
567 	size = uap->len;
568 	if (size == 0)
569 		return (EINVAL);
570 
571 	pageoff = (addr & PAGE_MASK);
572 	addr -= pageoff;
573 	size += pageoff;
574 	size = (vm_size_t) round_page(size);
575 	if (addr + size < addr)
576 		return (EINVAL);
577 
578 	/*
579 	 * Check for illegal addresses.  Watch out for address wrap...
580 	 */
581 	map = &td->td_proc->p_vmspace->vm_map;
582 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
583 		return (EINVAL);
584 	vm_map_lock(map);
585 #ifdef HWPMC_HOOKS
586 	/*
587 	 * Inform hwpmc if the address range being unmapped contains
588 	 * an executable region.
589 	 */
590 	pkm.pm_address = (uintptr_t) NULL;
591 	if (vm_map_lookup_entry(map, addr, &entry)) {
592 		for (;
593 		     entry != &map->header && entry->start < addr + size;
594 		     entry = entry->next) {
595 			if (vm_map_check_protection(map, entry->start,
596 				entry->end, VM_PROT_EXECUTE) == TRUE) {
597 				pkm.pm_address = (uintptr_t) addr;
598 				pkm.pm_size = (size_t) size;
599 				break;
600 			}
601 		}
602 	}
603 #endif
604 	vm_map_delete(map, addr, addr + size);
605 
606 #ifdef HWPMC_HOOKS
607 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
608 	vm_map_lock_downgrade(map);
609 	if (pkm.pm_address != (uintptr_t) NULL)
610 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
611 	vm_map_unlock_read(map);
612 #else
613 	vm_map_unlock(map);
614 #endif
615 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
616 	return (0);
617 }
618 
619 #ifndef _SYS_SYSPROTO_H_
620 struct mprotect_args {
621 	const void *addr;
622 	size_t len;
623 	int prot;
624 };
625 #endif
626 /*
627  * MPSAFE
628  */
629 int
630 sys_mprotect(td, uap)
631 	struct thread *td;
632 	struct mprotect_args *uap;
633 {
634 	vm_offset_t addr;
635 	vm_size_t size, pageoff;
636 	vm_prot_t prot;
637 
638 	addr = (vm_offset_t) uap->addr;
639 	size = uap->len;
640 	prot = uap->prot & VM_PROT_ALL;
641 
642 	pageoff = (addr & PAGE_MASK);
643 	addr -= pageoff;
644 	size += pageoff;
645 	size = (vm_size_t) round_page(size);
646 	if (addr + size < addr)
647 		return (EINVAL);
648 
649 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
650 	    addr + size, prot, FALSE)) {
651 	case KERN_SUCCESS:
652 		return (0);
653 	case KERN_PROTECTION_FAILURE:
654 		return (EACCES);
655 	case KERN_RESOURCE_SHORTAGE:
656 		return (ENOMEM);
657 	}
658 	return (EINVAL);
659 }
660 
661 #ifndef _SYS_SYSPROTO_H_
662 struct minherit_args {
663 	void *addr;
664 	size_t len;
665 	int inherit;
666 };
667 #endif
668 /*
669  * MPSAFE
670  */
671 int
672 sys_minherit(td, uap)
673 	struct thread *td;
674 	struct minherit_args *uap;
675 {
676 	vm_offset_t addr;
677 	vm_size_t size, pageoff;
678 	vm_inherit_t inherit;
679 
680 	addr = (vm_offset_t)uap->addr;
681 	size = uap->len;
682 	inherit = uap->inherit;
683 
684 	pageoff = (addr & PAGE_MASK);
685 	addr -= pageoff;
686 	size += pageoff;
687 	size = (vm_size_t) round_page(size);
688 	if (addr + size < addr)
689 		return (EINVAL);
690 
691 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
692 	    addr + size, inherit)) {
693 	case KERN_SUCCESS:
694 		return (0);
695 	case KERN_PROTECTION_FAILURE:
696 		return (EACCES);
697 	}
698 	return (EINVAL);
699 }
700 
701 #ifndef _SYS_SYSPROTO_H_
702 struct madvise_args {
703 	void *addr;
704 	size_t len;
705 	int behav;
706 };
707 #endif
708 
709 /*
710  * MPSAFE
711  */
712 int
713 sys_madvise(td, uap)
714 	struct thread *td;
715 	struct madvise_args *uap;
716 {
717 	vm_offset_t start, end;
718 	vm_map_t map;
719 	struct proc *p;
720 	int error;
721 
722 	/*
723 	 * Check for our special case, advising the swap pager we are
724 	 * "immortal."
725 	 */
726 	if (uap->behav == MADV_PROTECT) {
727 		error = priv_check(td, PRIV_VM_MADV_PROTECT);
728 		if (error == 0) {
729 			p = td->td_proc;
730 			PROC_LOCK(p);
731 			p->p_flag |= P_PROTECTED;
732 			PROC_UNLOCK(p);
733 		}
734 		return (error);
735 	}
736 	/*
737 	 * Check for illegal behavior
738 	 */
739 	if (uap->behav < 0 || uap->behav > MADV_CORE)
740 		return (EINVAL);
741 	/*
742 	 * Check for illegal addresses.  Watch out for address wrap... Note
743 	 * that VM_*_ADDRESS are not constants due to casts (argh).
744 	 */
745 	map = &td->td_proc->p_vmspace->vm_map;
746 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
747 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
748 		return (EINVAL);
749 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
750 		return (EINVAL);
751 
752 	/*
753 	 * Since this routine is only advisory, we default to conservative
754 	 * behavior.
755 	 */
756 	start = trunc_page((vm_offset_t) uap->addr);
757 	end = round_page((vm_offset_t) uap->addr + uap->len);
758 
759 	if (vm_map_madvise(map, start, end, uap->behav))
760 		return (EINVAL);
761 	return (0);
762 }
763 
764 #ifndef _SYS_SYSPROTO_H_
765 struct mincore_args {
766 	const void *addr;
767 	size_t len;
768 	char *vec;
769 };
770 #endif
771 
772 /*
773  * MPSAFE
774  */
775 int
776 sys_mincore(td, uap)
777 	struct thread *td;
778 	struct mincore_args *uap;
779 {
780 	vm_offset_t addr, first_addr;
781 	vm_offset_t end, cend;
782 	pmap_t pmap;
783 	vm_map_t map;
784 	char *vec;
785 	int error = 0;
786 	int vecindex, lastvecindex;
787 	vm_map_entry_t current;
788 	vm_map_entry_t entry;
789 	vm_object_t object;
790 	vm_paddr_t locked_pa;
791 	vm_page_t m;
792 	vm_pindex_t pindex;
793 	int mincoreinfo;
794 	unsigned int timestamp;
795 	boolean_t locked;
796 
797 	/*
798 	 * Make sure that the addresses presented are valid for user
799 	 * mode.
800 	 */
801 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
802 	end = addr + (vm_size_t)round_page(uap->len);
803 	map = &td->td_proc->p_vmspace->vm_map;
804 	if (end > vm_map_max(map) || end < addr)
805 		return (ENOMEM);
806 
807 	/*
808 	 * Address of byte vector
809 	 */
810 	vec = uap->vec;
811 
812 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
813 
814 	vm_map_lock_read(map);
815 RestartScan:
816 	timestamp = map->timestamp;
817 
818 	if (!vm_map_lookup_entry(map, addr, &entry)) {
819 		vm_map_unlock_read(map);
820 		return (ENOMEM);
821 	}
822 
823 	/*
824 	 * Do this on a map entry basis so that if the pages are not
825 	 * in the current processes address space, we can easily look
826 	 * up the pages elsewhere.
827 	 */
828 	lastvecindex = -1;
829 	for (current = entry;
830 	    (current != &map->header) && (current->start < end);
831 	    current = current->next) {
832 
833 		/*
834 		 * check for contiguity
835 		 */
836 		if (current->end < end &&
837 		    (entry->next == &map->header ||
838 		     current->next->start > current->end)) {
839 			vm_map_unlock_read(map);
840 			return (ENOMEM);
841 		}
842 
843 		/*
844 		 * ignore submaps (for now) or null objects
845 		 */
846 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
847 			current->object.vm_object == NULL)
848 			continue;
849 
850 		/*
851 		 * limit this scan to the current map entry and the
852 		 * limits for the mincore call
853 		 */
854 		if (addr < current->start)
855 			addr = current->start;
856 		cend = current->end;
857 		if (cend > end)
858 			cend = end;
859 
860 		/*
861 		 * scan this entry one page at a time
862 		 */
863 		while (addr < cend) {
864 			/*
865 			 * Check pmap first, it is likely faster, also
866 			 * it can provide info as to whether we are the
867 			 * one referencing or modifying the page.
868 			 */
869 			object = NULL;
870 			locked_pa = 0;
871 		retry:
872 			m = NULL;
873 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
874 			if (locked_pa != 0) {
875 				/*
876 				 * The page is mapped by this process but not
877 				 * both accessed and modified.  It is also
878 				 * managed.  Acquire the object lock so that
879 				 * other mappings might be examined.
880 				 */
881 				m = PHYS_TO_VM_PAGE(locked_pa);
882 				if (m->object != object) {
883 					if (object != NULL)
884 						VM_OBJECT_WUNLOCK(object);
885 					object = m->object;
886 					locked = VM_OBJECT_TRYWLOCK(object);
887 					vm_page_unlock(m);
888 					if (!locked) {
889 						VM_OBJECT_WLOCK(object);
890 						vm_page_lock(m);
891 						goto retry;
892 					}
893 				} else
894 					vm_page_unlock(m);
895 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
896 				    ("mincore: page %p is mapped but invalid",
897 				    m));
898 			} else if (mincoreinfo == 0) {
899 				/*
900 				 * The page is not mapped by this process.  If
901 				 * the object implements managed pages, then
902 				 * determine if the page is resident so that
903 				 * the mappings might be examined.
904 				 */
905 				if (current->object.vm_object != object) {
906 					if (object != NULL)
907 						VM_OBJECT_WUNLOCK(object);
908 					object = current->object.vm_object;
909 					VM_OBJECT_WLOCK(object);
910 				}
911 				if (object->type == OBJT_DEFAULT ||
912 				    object->type == OBJT_SWAP ||
913 				    object->type == OBJT_VNODE) {
914 					pindex = OFF_TO_IDX(current->offset +
915 					    (addr - current->start));
916 					m = vm_page_lookup(object, pindex);
917 					if (m == NULL &&
918 					    vm_page_is_cached(object, pindex))
919 						mincoreinfo = MINCORE_INCORE;
920 					if (m != NULL && m->valid == 0)
921 						m = NULL;
922 					if (m != NULL)
923 						mincoreinfo = MINCORE_INCORE;
924 				}
925 			}
926 			if (m != NULL) {
927 				/* Examine other mappings to the page. */
928 				if (m->dirty == 0 && pmap_is_modified(m))
929 					vm_page_dirty(m);
930 				if (m->dirty != 0)
931 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
932 				/*
933 				 * The first test for PGA_REFERENCED is an
934 				 * optimization.  The second test is
935 				 * required because a concurrent pmap
936 				 * operation could clear the last reference
937 				 * and set PGA_REFERENCED before the call to
938 				 * pmap_is_referenced().
939 				 */
940 				if ((m->aflags & PGA_REFERENCED) != 0 ||
941 				    pmap_is_referenced(m) ||
942 				    (m->aflags & PGA_REFERENCED) != 0)
943 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
944 			}
945 			if (object != NULL)
946 				VM_OBJECT_WUNLOCK(object);
947 
948 			/*
949 			 * subyte may page fault.  In case it needs to modify
950 			 * the map, we release the lock.
951 			 */
952 			vm_map_unlock_read(map);
953 
954 			/*
955 			 * calculate index into user supplied byte vector
956 			 */
957 			vecindex = OFF_TO_IDX(addr - first_addr);
958 
959 			/*
960 			 * If we have skipped map entries, we need to make sure that
961 			 * the byte vector is zeroed for those skipped entries.
962 			 */
963 			while ((lastvecindex + 1) < vecindex) {
964 				error = subyte(vec + lastvecindex, 0);
965 				if (error) {
966 					error = EFAULT;
967 					goto done2;
968 				}
969 				++lastvecindex;
970 			}
971 
972 			/*
973 			 * Pass the page information to the user
974 			 */
975 			error = subyte(vec + vecindex, mincoreinfo);
976 			if (error) {
977 				error = EFAULT;
978 				goto done2;
979 			}
980 
981 			/*
982 			 * If the map has changed, due to the subyte, the previous
983 			 * output may be invalid.
984 			 */
985 			vm_map_lock_read(map);
986 			if (timestamp != map->timestamp)
987 				goto RestartScan;
988 
989 			lastvecindex = vecindex;
990 			addr += PAGE_SIZE;
991 		}
992 	}
993 
994 	/*
995 	 * subyte may page fault.  In case it needs to modify
996 	 * the map, we release the lock.
997 	 */
998 	vm_map_unlock_read(map);
999 
1000 	/*
1001 	 * Zero the last entries in the byte vector.
1002 	 */
1003 	vecindex = OFF_TO_IDX(end - first_addr);
1004 	while ((lastvecindex + 1) < vecindex) {
1005 		error = subyte(vec + lastvecindex, 0);
1006 		if (error) {
1007 			error = EFAULT;
1008 			goto done2;
1009 		}
1010 		++lastvecindex;
1011 	}
1012 
1013 	/*
1014 	 * If the map has changed, due to the subyte, the previous
1015 	 * output may be invalid.
1016 	 */
1017 	vm_map_lock_read(map);
1018 	if (timestamp != map->timestamp)
1019 		goto RestartScan;
1020 	vm_map_unlock_read(map);
1021 done2:
1022 	return (error);
1023 }
1024 
1025 #ifndef _SYS_SYSPROTO_H_
1026 struct mlock_args {
1027 	const void *addr;
1028 	size_t len;
1029 };
1030 #endif
1031 /*
1032  * MPSAFE
1033  */
1034 int
1035 sys_mlock(td, uap)
1036 	struct thread *td;
1037 	struct mlock_args *uap;
1038 {
1039 	struct proc *proc;
1040 	vm_offset_t addr, end, last, start;
1041 	vm_size_t npages, size;
1042 	vm_map_t map;
1043 	unsigned long nsize;
1044 	int error;
1045 
1046 	error = priv_check(td, PRIV_VM_MLOCK);
1047 	if (error)
1048 		return (error);
1049 	addr = (vm_offset_t)uap->addr;
1050 	size = uap->len;
1051 	last = addr + size;
1052 	start = trunc_page(addr);
1053 	end = round_page(last);
1054 	if (last < addr || end < addr)
1055 		return (EINVAL);
1056 	npages = atop(end - start);
1057 	if (npages > vm_page_max_wired)
1058 		return (ENOMEM);
1059 	proc = td->td_proc;
1060 	map = &proc->p_vmspace->vm_map;
1061 	PROC_LOCK(proc);
1062 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1063 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1064 		PROC_UNLOCK(proc);
1065 		return (ENOMEM);
1066 	}
1067 	PROC_UNLOCK(proc);
1068 	if (npages + cnt.v_wire_count > vm_page_max_wired)
1069 		return (EAGAIN);
1070 #ifdef RACCT
1071 	PROC_LOCK(proc);
1072 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1073 	PROC_UNLOCK(proc);
1074 	if (error != 0)
1075 		return (ENOMEM);
1076 #endif
1077 	error = vm_map_wire(map, start, end,
1078 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1079 #ifdef RACCT
1080 	if (error != KERN_SUCCESS) {
1081 		PROC_LOCK(proc);
1082 		racct_set(proc, RACCT_MEMLOCK,
1083 		    ptoa(pmap_wired_count(map->pmap)));
1084 		PROC_UNLOCK(proc);
1085 	}
1086 #endif
1087 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1088 }
1089 
1090 #ifndef _SYS_SYSPROTO_H_
1091 struct mlockall_args {
1092 	int	how;
1093 };
1094 #endif
1095 
1096 /*
1097  * MPSAFE
1098  */
1099 int
1100 sys_mlockall(td, uap)
1101 	struct thread *td;
1102 	struct mlockall_args *uap;
1103 {
1104 	vm_map_t map;
1105 	int error;
1106 
1107 	map = &td->td_proc->p_vmspace->vm_map;
1108 	error = priv_check(td, PRIV_VM_MLOCK);
1109 	if (error)
1110 		return (error);
1111 
1112 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1113 		return (EINVAL);
1114 
1115 	/*
1116 	 * If wiring all pages in the process would cause it to exceed
1117 	 * a hard resource limit, return ENOMEM.
1118 	 */
1119 	if (!old_mlock && uap->how & MCL_CURRENT) {
1120 		PROC_LOCK(td->td_proc);
1121 		if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1122 			PROC_UNLOCK(td->td_proc);
1123 			return (ENOMEM);
1124 		}
1125 		PROC_UNLOCK(td->td_proc);
1126 	}
1127 #ifdef RACCT
1128 	PROC_LOCK(td->td_proc);
1129 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1130 	PROC_UNLOCK(td->td_proc);
1131 	if (error != 0)
1132 		return (ENOMEM);
1133 #endif
1134 
1135 	if (uap->how & MCL_FUTURE) {
1136 		vm_map_lock(map);
1137 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1138 		vm_map_unlock(map);
1139 		error = 0;
1140 	}
1141 
1142 	if (uap->how & MCL_CURRENT) {
1143 		/*
1144 		 * P1003.1-2001 mandates that all currently mapped pages
1145 		 * will be memory resident and locked (wired) upon return
1146 		 * from mlockall(). vm_map_wire() will wire pages, by
1147 		 * calling vm_fault_wire() for each page in the region.
1148 		 */
1149 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1150 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1151 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1152 	}
1153 #ifdef RACCT
1154 	if (error != KERN_SUCCESS) {
1155 		PROC_LOCK(td->td_proc);
1156 		racct_set(td->td_proc, RACCT_MEMLOCK,
1157 		    ptoa(pmap_wired_count(map->pmap)));
1158 		PROC_UNLOCK(td->td_proc);
1159 	}
1160 #endif
1161 
1162 	return (error);
1163 }
1164 
1165 #ifndef _SYS_SYSPROTO_H_
1166 struct munlockall_args {
1167 	register_t dummy;
1168 };
1169 #endif
1170 
1171 /*
1172  * MPSAFE
1173  */
1174 int
1175 sys_munlockall(td, uap)
1176 	struct thread *td;
1177 	struct munlockall_args *uap;
1178 {
1179 	vm_map_t map;
1180 	int error;
1181 
1182 	map = &td->td_proc->p_vmspace->vm_map;
1183 	error = priv_check(td, PRIV_VM_MUNLOCK);
1184 	if (error)
1185 		return (error);
1186 
1187 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1188 	vm_map_lock(map);
1189 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1190 	vm_map_unlock(map);
1191 
1192 	/* Forcibly unwire all pages. */
1193 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1194 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1195 #ifdef RACCT
1196 	if (error == KERN_SUCCESS) {
1197 		PROC_LOCK(td->td_proc);
1198 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1199 		PROC_UNLOCK(td->td_proc);
1200 	}
1201 #endif
1202 
1203 	return (error);
1204 }
1205 
1206 #ifndef _SYS_SYSPROTO_H_
1207 struct munlock_args {
1208 	const void *addr;
1209 	size_t len;
1210 };
1211 #endif
1212 /*
1213  * MPSAFE
1214  */
1215 int
1216 sys_munlock(td, uap)
1217 	struct thread *td;
1218 	struct munlock_args *uap;
1219 {
1220 	vm_offset_t addr, end, last, start;
1221 	vm_size_t size;
1222 	int error;
1223 
1224 	error = priv_check(td, PRIV_VM_MUNLOCK);
1225 	if (error)
1226 		return (error);
1227 	addr = (vm_offset_t)uap->addr;
1228 	size = uap->len;
1229 	last = addr + size;
1230 	start = trunc_page(addr);
1231 	end = round_page(last);
1232 	if (last < addr || end < addr)
1233 		return (EINVAL);
1234 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1235 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1236 #ifdef RACCT
1237 	if (error == KERN_SUCCESS) {
1238 		PROC_LOCK(td->td_proc);
1239 		racct_sub(td->td_proc, RACCT_MEMLOCK, ptoa(end - start));
1240 		PROC_UNLOCK(td->td_proc);
1241 	}
1242 #endif
1243 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1244 }
1245 
1246 /*
1247  * vm_mmap_vnode()
1248  *
1249  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1250  * operations on vnodes.
1251  *
1252  * For VCHR vnodes, the vnode lock is held over the call to
1253  * vm_mmap_cdev() to keep vp->v_rdev valid.
1254  */
1255 int
1256 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1257     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1258     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1259     boolean_t *writecounted)
1260 {
1261 	struct vattr va;
1262 	vm_object_t obj;
1263 	vm_offset_t foff;
1264 	struct mount *mp;
1265 	struct ucred *cred;
1266 	int error, flags, locktype;
1267 
1268 	mp = vp->v_mount;
1269 	cred = td->td_ucred;
1270 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1271 		locktype = LK_EXCLUSIVE;
1272 	else
1273 		locktype = LK_SHARED;
1274 	if ((error = vget(vp, locktype, td)) != 0)
1275 		return (error);
1276 	foff = *foffp;
1277 	flags = *flagsp;
1278 	obj = vp->v_object;
1279 	if (vp->v_type == VREG) {
1280 		/*
1281 		 * Get the proper underlying object
1282 		 */
1283 		if (obj == NULL) {
1284 			error = EINVAL;
1285 			goto done;
1286 		}
1287 		if (obj->handle != vp) {
1288 			vput(vp);
1289 			vp = (struct vnode *)obj->handle;
1290 			/*
1291 			 * Bypass filesystems obey the mpsafety of the
1292 			 * underlying fs.
1293 			 */
1294 			error = vget(vp, locktype, td);
1295 			if (error != 0)
1296 				return (error);
1297 		}
1298 		if (locktype == LK_EXCLUSIVE) {
1299 			*writecounted = TRUE;
1300 			vnode_pager_update_writecount(obj, 0, objsize);
1301 		}
1302 	} else if (vp->v_type == VCHR) {
1303 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1304 		    vp->v_rdev, foffp, objp);
1305 		if (error == 0)
1306 			goto mark_atime;
1307 		goto done;
1308 	} else {
1309 		error = EINVAL;
1310 		goto done;
1311 	}
1312 	if ((error = VOP_GETATTR(vp, &va, cred)))
1313 		goto done;
1314 #ifdef MAC
1315 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1316 	if (error != 0)
1317 		goto done;
1318 #endif
1319 	if ((flags & MAP_SHARED) != 0) {
1320 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1321 			if (prot & PROT_WRITE) {
1322 				error = EPERM;
1323 				goto done;
1324 			}
1325 			*maxprotp &= ~VM_PROT_WRITE;
1326 		}
1327 	}
1328 	/*
1329 	 * If it is a regular file without any references
1330 	 * we do not need to sync it.
1331 	 * Adjust object size to be the size of actual file.
1332 	 */
1333 	objsize = round_page(va.va_size);
1334 	if (va.va_nlink == 0)
1335 		flags |= MAP_NOSYNC;
1336 	obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, cred);
1337 	if (obj == NULL) {
1338 		error = ENOMEM;
1339 		goto done;
1340 	}
1341 	*objp = obj;
1342 	*flagsp = flags;
1343 
1344 mark_atime:
1345 	vfs_mark_atime(vp, cred);
1346 
1347 done:
1348 	if (error != 0 && *writecounted) {
1349 		*writecounted = FALSE;
1350 		vnode_pager_update_writecount(obj, objsize, 0);
1351 	}
1352 	vput(vp);
1353 	return (error);
1354 }
1355 
1356 /*
1357  * vm_mmap_cdev()
1358  *
1359  * MPSAFE
1360  *
1361  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1362  * operations on cdevs.
1363  */
1364 int
1365 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1366     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1367     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1368 {
1369 	vm_object_t obj;
1370 	struct cdevsw *dsw;
1371 	int error, flags, ref;
1372 
1373 	flags = *flagsp;
1374 
1375 	dsw = dev_refthread(cdev, &ref);
1376 	if (dsw == NULL)
1377 		return (ENXIO);
1378 	if (dsw->d_flags & D_MMAP_ANON) {
1379 		dev_relthread(cdev, ref);
1380 		*maxprotp = VM_PROT_ALL;
1381 		*flagsp |= MAP_ANON;
1382 		return (0);
1383 	}
1384 	/*
1385 	 * cdevs do not provide private mappings of any kind.
1386 	 */
1387 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1388 	    (prot & PROT_WRITE) != 0) {
1389 		dev_relthread(cdev, ref);
1390 		return (EACCES);
1391 	}
1392 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1393 		dev_relthread(cdev, ref);
1394 		return (EINVAL);
1395 	}
1396 	/*
1397 	 * Force device mappings to be shared.
1398 	 */
1399 	flags |= MAP_SHARED;
1400 #ifdef MAC_XXX
1401 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1402 	if (error != 0) {
1403 		dev_relthread(cdev, ref);
1404 		return (error);
1405 	}
1406 #endif
1407 	/*
1408 	 * First, try d_mmap_single().  If that is not implemented
1409 	 * (returns ENODEV), fall back to using the device pager.
1410 	 * Note that d_mmap_single() must return a reference to the
1411 	 * object (it needs to bump the reference count of the object
1412 	 * it returns somehow).
1413 	 *
1414 	 * XXX assumes VM_PROT_* == PROT_*
1415 	 */
1416 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1417 	dev_relthread(cdev, ref);
1418 	if (error != ENODEV)
1419 		return (error);
1420 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1421 	    td->td_ucred);
1422 	if (obj == NULL)
1423 		return (EINVAL);
1424 	*objp = obj;
1425 	*flagsp = flags;
1426 	return (0);
1427 }
1428 
1429 /*
1430  * vm_mmap_shm()
1431  *
1432  * MPSAFE
1433  *
1434  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1435  * operations on shm file descriptors.
1436  */
1437 int
1438 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1439     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1440     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1441 {
1442 	int error;
1443 
1444 	if ((*flagsp & MAP_SHARED) != 0 &&
1445 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1446 	    (prot & PROT_WRITE) != 0)
1447 		return (EACCES);
1448 #ifdef MAC
1449 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1450 	if (error != 0)
1451 		return (error);
1452 #endif
1453 	error = shm_mmap(shmfd, objsize, foff, objp);
1454 	if (error)
1455 		return (error);
1456 	return (0);
1457 }
1458 
1459 /*
1460  * vm_mmap()
1461  *
1462  * MPSAFE
1463  *
1464  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1465  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1466  */
1467 int
1468 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1469 	vm_prot_t maxprot, int flags,
1470 	objtype_t handle_type, void *handle,
1471 	vm_ooffset_t foff)
1472 {
1473 	boolean_t fitit;
1474 	vm_object_t object = NULL;
1475 	struct thread *td = curthread;
1476 	int docow, error, rv;
1477 	boolean_t writecounted;
1478 
1479 	if (size == 0)
1480 		return (0);
1481 
1482 	size = round_page(size);
1483 
1484 	if (map == &td->td_proc->p_vmspace->vm_map) {
1485 		PROC_LOCK(td->td_proc);
1486 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1487 			PROC_UNLOCK(td->td_proc);
1488 			return (ENOMEM);
1489 		}
1490 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1491 			PROC_UNLOCK(td->td_proc);
1492 			return (ENOMEM);
1493 		}
1494 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1495 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1496 			    lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1497 				racct_set_force(td->td_proc, RACCT_VMEM,
1498 				    map->size);
1499 				PROC_UNLOCK(td->td_proc);
1500 				return (ENOMEM);
1501 			}
1502 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1503 			    ptoa(pmap_wired_count(map->pmap)) + size);
1504 			if (error != 0) {
1505 				racct_set_force(td->td_proc, RACCT_VMEM,
1506 				    map->size);
1507 				PROC_UNLOCK(td->td_proc);
1508 				return (error);
1509 			}
1510 		}
1511 		PROC_UNLOCK(td->td_proc);
1512 	}
1513 
1514 	/*
1515 	 * We currently can only deal with page aligned file offsets.
1516 	 * The check is here rather than in the syscall because the
1517 	 * kernel calls this function internally for other mmaping
1518 	 * operations (such as in exec) and non-aligned offsets will
1519 	 * cause pmap inconsistencies...so we want to be sure to
1520 	 * disallow this in all cases.
1521 	 */
1522 	if (foff & PAGE_MASK)
1523 		return (EINVAL);
1524 
1525 	if ((flags & MAP_FIXED) == 0) {
1526 		fitit = TRUE;
1527 		*addr = round_page(*addr);
1528 	} else {
1529 		if (*addr != trunc_page(*addr))
1530 			return (EINVAL);
1531 		fitit = FALSE;
1532 	}
1533 	writecounted = FALSE;
1534 
1535 	/*
1536 	 * Lookup/allocate object.
1537 	 */
1538 	switch (handle_type) {
1539 	case OBJT_DEVICE:
1540 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1541 		    handle, &foff, &object);
1542 		break;
1543 	case OBJT_VNODE:
1544 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1545 		    handle, &foff, &object, &writecounted);
1546 		break;
1547 	case OBJT_SWAP:
1548 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1549 		    handle, foff, &object);
1550 		break;
1551 	case OBJT_DEFAULT:
1552 		if (handle == NULL) {
1553 			error = 0;
1554 			break;
1555 		}
1556 		/* FALLTHROUGH */
1557 	default:
1558 		error = EINVAL;
1559 		break;
1560 	}
1561 	if (error)
1562 		return (error);
1563 	if (flags & MAP_ANON) {
1564 		object = NULL;
1565 		docow = 0;
1566 		/*
1567 		 * Unnamed anonymous regions always start at 0.
1568 		 */
1569 		if (handle == 0)
1570 			foff = 0;
1571 	} else if (flags & MAP_PREFAULT_READ)
1572 		docow = MAP_PREFAULT;
1573 	else
1574 		docow = MAP_PREFAULT_PARTIAL;
1575 
1576 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1577 		docow |= MAP_COPY_ON_WRITE;
1578 	if (flags & MAP_NOSYNC)
1579 		docow |= MAP_DISABLE_SYNCER;
1580 	if (flags & MAP_NOCORE)
1581 		docow |= MAP_DISABLE_COREDUMP;
1582 	/* Shared memory is also shared with children. */
1583 	if (flags & MAP_SHARED)
1584 		docow |= MAP_INHERIT_SHARE;
1585 	if (writecounted)
1586 		docow |= MAP_VN_WRITECOUNT;
1587 
1588 	if (flags & MAP_STACK)
1589 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1590 		    docow | MAP_STACK_GROWS_DOWN);
1591 	else if (fitit)
1592 		rv = vm_map_find(map, object, foff, addr, size,
1593 		    object != NULL && object->type == OBJT_DEVICE ?
1594 		    VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1595 	else
1596 		rv = vm_map_fixed(map, object, foff, *addr, size,
1597 				 prot, maxprot, docow);
1598 
1599 	if (rv == KERN_SUCCESS) {
1600 		/*
1601 		 * If the process has requested that all future mappings
1602 		 * be wired, then heed this.
1603 		 */
1604 		if (map->flags & MAP_WIREFUTURE) {
1605 			vm_map_wire(map, *addr, *addr + size,
1606 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1607 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1608 		}
1609 	} else {
1610 		/*
1611 		 * If this mapping was accounted for in the vnode's
1612 		 * writecount, then undo that now.
1613 		 */
1614 		if (writecounted)
1615 			vnode_pager_release_writecount(object, 0, size);
1616 		/*
1617 		 * Lose the object reference.  Will destroy the
1618 		 * object if it's an unnamed anonymous mapping
1619 		 * or named anonymous without other references.
1620 		 */
1621 		vm_object_deallocate(object);
1622 	}
1623 	return (vm_mmap_to_errno(rv));
1624 }
1625 
1626 /*
1627  * Translate a Mach VM return code to zero on success or the appropriate errno
1628  * on failure.
1629  */
1630 int
1631 vm_mmap_to_errno(int rv)
1632 {
1633 
1634 	switch (rv) {
1635 	case KERN_SUCCESS:
1636 		return (0);
1637 	case KERN_INVALID_ADDRESS:
1638 	case KERN_NO_SPACE:
1639 		return (ENOMEM);
1640 	case KERN_PROTECTION_FAILURE:
1641 		return (EACCES);
1642 	default:
1643 		return (EINVAL);
1644 	}
1645 }
1646