xref: /freebsd/sys/vm/vm_mmap.c (revision 40a8ac8f62b535d30349faf28cf47106b7041b83)
1 /*-
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35  *
36  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
37  */
38 
39 /*
40  * Mapped file (mmap) interface to VM
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capsicum.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/procctl.h>
60 #include <sys/racct.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/rwlock.h>
64 #include <sys/sysctl.h>
65 #include <sys/vnode.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/mman.h>
69 #include <sys/mount.h>
70 #include <sys/conf.h>
71 #include <sys/stat.h>
72 #include <sys/syscallsubr.h>
73 #include <sys/sysent.h>
74 #include <sys/vmmeter.h>
75 
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pager.h>
85 #include <vm/vm_pageout.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_page.h>
88 #include <vm/vnode_pager.h>
89 
90 #ifdef HWPMC_HOOKS
91 #include <sys/pmckern.h>
92 #endif
93 
94 int old_mlock = 0;
95 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
96     "Do not apply RLIMIT_MEMLOCK on mlockall");
97 
98 #ifdef MAP_32BIT
99 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
100 #endif
101 
102 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
103     int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
104 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
105     int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
106 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
107     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
108 
109 #ifndef _SYS_SYSPROTO_H_
110 struct sbrk_args {
111 	int incr;
112 };
113 #endif
114 
115 /*
116  * MPSAFE
117  */
118 /* ARGSUSED */
119 int
120 sys_sbrk(td, uap)
121 	struct thread *td;
122 	struct sbrk_args *uap;
123 {
124 	/* Not yet implemented */
125 	return (EOPNOTSUPP);
126 }
127 
128 #ifndef _SYS_SYSPROTO_H_
129 struct sstk_args {
130 	int incr;
131 };
132 #endif
133 
134 /*
135  * MPSAFE
136  */
137 /* ARGSUSED */
138 int
139 sys_sstk(td, uap)
140 	struct thread *td;
141 	struct sstk_args *uap;
142 {
143 	/* Not yet implemented */
144 	return (EOPNOTSUPP);
145 }
146 
147 #if defined(COMPAT_43)
148 #ifndef _SYS_SYSPROTO_H_
149 struct getpagesize_args {
150 	int dummy;
151 };
152 #endif
153 
154 int
155 ogetpagesize(td, uap)
156 	struct thread *td;
157 	struct getpagesize_args *uap;
158 {
159 	/* MP SAFE */
160 	td->td_retval[0] = PAGE_SIZE;
161 	return (0);
162 }
163 #endif				/* COMPAT_43 */
164 
165 
166 /*
167  * Memory Map (mmap) system call.  Note that the file offset
168  * and address are allowed to be NOT page aligned, though if
169  * the MAP_FIXED flag it set, both must have the same remainder
170  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
171  * page-aligned, the actual mapping starts at trunc_page(addr)
172  * and the return value is adjusted up by the page offset.
173  *
174  * Generally speaking, only character devices which are themselves
175  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
176  * there would be no cache coherency between a descriptor and a VM mapping
177  * both to the same character device.
178  */
179 #ifndef _SYS_SYSPROTO_H_
180 struct mmap_args {
181 	void *addr;
182 	size_t len;
183 	int prot;
184 	int flags;
185 	int fd;
186 	long pad;
187 	off_t pos;
188 };
189 #endif
190 
191 /*
192  * MPSAFE
193  */
194 int
195 sys_mmap(td, uap)
196 	struct thread *td;
197 	struct mmap_args *uap;
198 {
199 #ifdef HWPMC_HOOKS
200 	struct pmckern_map_in pkm;
201 #endif
202 	struct file *fp;
203 	struct vnode *vp;
204 	vm_offset_t addr;
205 	vm_size_t size, pageoff;
206 	vm_prot_t cap_maxprot, maxprot;
207 	void *handle;
208 	objtype_t handle_type;
209 	int align, error, flags, prot;
210 	off_t pos;
211 	struct vmspace *vms = td->td_proc->p_vmspace;
212 	cap_rights_t rights;
213 
214 	addr = (vm_offset_t) uap->addr;
215 	size = uap->len;
216 	prot = uap->prot;
217 	flags = uap->flags;
218 	pos = uap->pos;
219 
220 	fp = NULL;
221 
222 	/*
223 	 * Enforce the constraints.
224 	 * Mapping of length 0 is only allowed for old binaries.
225 	 * Anonymous mapping shall specify -1 as filedescriptor and
226 	 * zero position for new code. Be nice to ancient a.out
227 	 * binaries and correct pos for anonymous mapping, since old
228 	 * ld.so sometimes issues anonymous map requests with non-zero
229 	 * pos.
230 	 */
231 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
232 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
233 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
234 			return (EINVAL);
235 	} else {
236 		if ((flags & MAP_ANON) != 0)
237 			pos = 0;
238 	}
239 
240 	if (flags & MAP_STACK) {
241 		if ((uap->fd != -1) ||
242 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
243 			return (EINVAL);
244 		flags |= MAP_ANON;
245 		pos = 0;
246 	}
247 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_RENAME |
248 	    MAP_NORESERVE | MAP_HASSEMAPHORE | MAP_STACK | MAP_NOSYNC |
249 	    MAP_ANON | MAP_EXCL | MAP_NOCORE | MAP_PREFAULT_READ |
250 #ifdef MAP_32BIT
251 	    MAP_32BIT |
252 #endif
253 	    MAP_ALIGNMENT_MASK)) != 0)
254 		return (EINVAL);
255 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
256 		return (EINVAL);
257 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
258 		return (EINVAL);
259 	if (prot != PROT_NONE &&
260 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
261 		return (EINVAL);
262 
263 	/*
264 	 * Align the file position to a page boundary,
265 	 * and save its page offset component.
266 	 */
267 	pageoff = (pos & PAGE_MASK);
268 	pos -= pageoff;
269 
270 	/* Adjust size for rounding (on both ends). */
271 	size += pageoff;			/* low end... */
272 	size = (vm_size_t) round_page(size);	/* hi end */
273 
274 	/* Ensure alignment is at least a page and fits in a pointer. */
275 	align = flags & MAP_ALIGNMENT_MASK;
276 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
277 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
278 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
279 		return (EINVAL);
280 
281 	/*
282 	 * Check for illegal addresses.  Watch out for address wrap... Note
283 	 * that VM_*_ADDRESS are not constants due to casts (argh).
284 	 */
285 	if (flags & MAP_FIXED) {
286 		/*
287 		 * The specified address must have the same remainder
288 		 * as the file offset taken modulo PAGE_SIZE, so it
289 		 * should be aligned after adjustment by pageoff.
290 		 */
291 		addr -= pageoff;
292 		if (addr & PAGE_MASK)
293 			return (EINVAL);
294 
295 		/* Address range must be all in user VM space. */
296 		if (addr < vm_map_min(&vms->vm_map) ||
297 		    addr + size > vm_map_max(&vms->vm_map))
298 			return (EINVAL);
299 		if (addr + size < addr)
300 			return (EINVAL);
301 #ifdef MAP_32BIT
302 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
303 			return (EINVAL);
304 	} else if (flags & MAP_32BIT) {
305 		/*
306 		 * For MAP_32BIT, override the hint if it is too high and
307 		 * do not bother moving the mapping past the heap (since
308 		 * the heap is usually above 2GB).
309 		 */
310 		if (addr + size > MAP_32BIT_MAX_ADDR)
311 			addr = 0;
312 #endif
313 	} else {
314 		/*
315 		 * XXX for non-fixed mappings where no hint is provided or
316 		 * the hint would fall in the potential heap space,
317 		 * place it after the end of the largest possible heap.
318 		 *
319 		 * There should really be a pmap call to determine a reasonable
320 		 * location.
321 		 */
322 		PROC_LOCK(td->td_proc);
323 		if (addr == 0 ||
324 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
325 		    addr < round_page((vm_offset_t)vms->vm_daddr +
326 		    lim_max(td->td_proc, RLIMIT_DATA))))
327 			addr = round_page((vm_offset_t)vms->vm_daddr +
328 			    lim_max(td->td_proc, RLIMIT_DATA));
329 		PROC_UNLOCK(td->td_proc);
330 	}
331 	if (flags & MAP_ANON) {
332 		/*
333 		 * Mapping blank space is trivial.
334 		 */
335 		handle = NULL;
336 		handle_type = OBJT_DEFAULT;
337 		maxprot = VM_PROT_ALL;
338 		cap_maxprot = VM_PROT_ALL;
339 	} else {
340 		/*
341 		 * Mapping file, get fp for validation and don't let the
342 		 * descriptor disappear on us if we block. Check capability
343 		 * rights, but also return the maximum rights to be combined
344 		 * with maxprot later.
345 		 */
346 		cap_rights_init(&rights, CAP_MMAP);
347 		if (prot & PROT_READ)
348 			cap_rights_set(&rights, CAP_MMAP_R);
349 		if ((flags & MAP_SHARED) != 0) {
350 			if (prot & PROT_WRITE)
351 				cap_rights_set(&rights, CAP_MMAP_W);
352 		}
353 		if (prot & PROT_EXEC)
354 			cap_rights_set(&rights, CAP_MMAP_X);
355 		error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
356 		if (error != 0)
357 			goto done;
358 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
359 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
360 			error = EINVAL;
361 			goto done;
362 		}
363 		if (fp->f_type == DTYPE_SHM) {
364 			handle = fp->f_data;
365 			handle_type = OBJT_SWAP;
366 			maxprot = VM_PROT_NONE;
367 
368 			/* FREAD should always be set. */
369 			if (fp->f_flag & FREAD)
370 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
371 			if (fp->f_flag & FWRITE)
372 				maxprot |= VM_PROT_WRITE;
373 			goto map;
374 		}
375 		if (fp->f_type != DTYPE_VNODE) {
376 			error = ENODEV;
377 			goto done;
378 		}
379 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
380     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
381 		/*
382 		 * POSIX shared-memory objects are defined to have
383 		 * kernel persistence, and are not defined to support
384 		 * read(2)/write(2) -- or even open(2).  Thus, we can
385 		 * use MAP_ASYNC to trade on-disk coherence for speed.
386 		 * The shm_open(3) library routine turns on the FPOSIXSHM
387 		 * flag to request this behavior.
388 		 */
389 		if (fp->f_flag & FPOSIXSHM)
390 			flags |= MAP_NOSYNC;
391 #endif
392 		vp = fp->f_vnode;
393 		/*
394 		 * Ensure that file and memory protections are
395 		 * compatible.  Note that we only worry about
396 		 * writability if mapping is shared; in this case,
397 		 * current and max prot are dictated by the open file.
398 		 * XXX use the vnode instead?  Problem is: what
399 		 * credentials do we use for determination? What if
400 		 * proc does a setuid?
401 		 */
402 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
403 			maxprot = VM_PROT_NONE;
404 		else
405 			maxprot = VM_PROT_EXECUTE;
406 		if (fp->f_flag & FREAD) {
407 			maxprot |= VM_PROT_READ;
408 		} else if (prot & PROT_READ) {
409 			error = EACCES;
410 			goto done;
411 		}
412 		/*
413 		 * If we are sharing potential changes (either via
414 		 * MAP_SHARED or via the implicit sharing of character
415 		 * device mappings), and we are trying to get write
416 		 * permission although we opened it without asking
417 		 * for it, bail out.
418 		 */
419 		if ((flags & MAP_SHARED) != 0) {
420 			if ((fp->f_flag & FWRITE) != 0) {
421 				maxprot |= VM_PROT_WRITE;
422 			} else if ((prot & PROT_WRITE) != 0) {
423 				error = EACCES;
424 				goto done;
425 			}
426 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
427 			maxprot |= VM_PROT_WRITE;
428 			cap_maxprot |= VM_PROT_WRITE;
429 		}
430 		handle = (void *)vp;
431 		handle_type = OBJT_VNODE;
432 	}
433 map:
434 	td->td_fpop = fp;
435 	maxprot &= cap_maxprot;
436 
437 	/* This relies on VM_PROT_* matching PROT_*. */
438 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
439 	    flags, handle_type, handle, pos);
440 	td->td_fpop = NULL;
441 #ifdef HWPMC_HOOKS
442 	/* inform hwpmc(4) if an executable is being mapped */
443 	if (error == 0 && handle_type == OBJT_VNODE &&
444 	    (prot & PROT_EXEC)) {
445 		pkm.pm_file = handle;
446 		pkm.pm_address = (uintptr_t) addr;
447 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
448 	}
449 #endif
450 	if (error == 0)
451 		td->td_retval[0] = (register_t) (addr + pageoff);
452 done:
453 	if (fp)
454 		fdrop(fp, td);
455 
456 	return (error);
457 }
458 
459 int
460 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
461 {
462 	struct mmap_args oargs;
463 
464 	oargs.addr = uap->addr;
465 	oargs.len = uap->len;
466 	oargs.prot = uap->prot;
467 	oargs.flags = uap->flags;
468 	oargs.fd = uap->fd;
469 	oargs.pos = uap->pos;
470 	return (sys_mmap(td, &oargs));
471 }
472 
473 #ifdef COMPAT_43
474 #ifndef _SYS_SYSPROTO_H_
475 struct ommap_args {
476 	caddr_t addr;
477 	int len;
478 	int prot;
479 	int flags;
480 	int fd;
481 	long pos;
482 };
483 #endif
484 int
485 ommap(td, uap)
486 	struct thread *td;
487 	struct ommap_args *uap;
488 {
489 	struct mmap_args nargs;
490 	static const char cvtbsdprot[8] = {
491 		0,
492 		PROT_EXEC,
493 		PROT_WRITE,
494 		PROT_EXEC | PROT_WRITE,
495 		PROT_READ,
496 		PROT_EXEC | PROT_READ,
497 		PROT_WRITE | PROT_READ,
498 		PROT_EXEC | PROT_WRITE | PROT_READ,
499 	};
500 
501 #define	OMAP_ANON	0x0002
502 #define	OMAP_COPY	0x0020
503 #define	OMAP_SHARED	0x0010
504 #define	OMAP_FIXED	0x0100
505 
506 	nargs.addr = uap->addr;
507 	nargs.len = uap->len;
508 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
509 #ifdef COMPAT_FREEBSD32
510 #if defined(__amd64__)
511 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
512 	    nargs.prot != 0)
513 		nargs.prot |= PROT_EXEC;
514 #endif
515 #endif
516 	nargs.flags = 0;
517 	if (uap->flags & OMAP_ANON)
518 		nargs.flags |= MAP_ANON;
519 	if (uap->flags & OMAP_COPY)
520 		nargs.flags |= MAP_COPY;
521 	if (uap->flags & OMAP_SHARED)
522 		nargs.flags |= MAP_SHARED;
523 	else
524 		nargs.flags |= MAP_PRIVATE;
525 	if (uap->flags & OMAP_FIXED)
526 		nargs.flags |= MAP_FIXED;
527 	nargs.fd = uap->fd;
528 	nargs.pos = uap->pos;
529 	return (sys_mmap(td, &nargs));
530 }
531 #endif				/* COMPAT_43 */
532 
533 
534 #ifndef _SYS_SYSPROTO_H_
535 struct msync_args {
536 	void *addr;
537 	size_t len;
538 	int flags;
539 };
540 #endif
541 /*
542  * MPSAFE
543  */
544 int
545 sys_msync(td, uap)
546 	struct thread *td;
547 	struct msync_args *uap;
548 {
549 	vm_offset_t addr;
550 	vm_size_t size, pageoff;
551 	int flags;
552 	vm_map_t map;
553 	int rv;
554 
555 	addr = (vm_offset_t) uap->addr;
556 	size = uap->len;
557 	flags = uap->flags;
558 
559 	pageoff = (addr & PAGE_MASK);
560 	addr -= pageoff;
561 	size += pageoff;
562 	size = (vm_size_t) round_page(size);
563 	if (addr + size < addr)
564 		return (EINVAL);
565 
566 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
567 		return (EINVAL);
568 
569 	map = &td->td_proc->p_vmspace->vm_map;
570 
571 	/*
572 	 * Clean the pages and interpret the return value.
573 	 */
574 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
575 	    (flags & MS_INVALIDATE) != 0);
576 	switch (rv) {
577 	case KERN_SUCCESS:
578 		return (0);
579 	case KERN_INVALID_ADDRESS:
580 		return (ENOMEM);
581 	case KERN_INVALID_ARGUMENT:
582 		return (EBUSY);
583 	case KERN_FAILURE:
584 		return (EIO);
585 	default:
586 		return (EINVAL);
587 	}
588 }
589 
590 #ifndef _SYS_SYSPROTO_H_
591 struct munmap_args {
592 	void *addr;
593 	size_t len;
594 };
595 #endif
596 /*
597  * MPSAFE
598  */
599 int
600 sys_munmap(td, uap)
601 	struct thread *td;
602 	struct munmap_args *uap;
603 {
604 #ifdef HWPMC_HOOKS
605 	struct pmckern_map_out pkm;
606 	vm_map_entry_t entry;
607 #endif
608 	vm_offset_t addr;
609 	vm_size_t size, pageoff;
610 	vm_map_t map;
611 
612 	addr = (vm_offset_t) uap->addr;
613 	size = uap->len;
614 	if (size == 0)
615 		return (EINVAL);
616 
617 	pageoff = (addr & PAGE_MASK);
618 	addr -= pageoff;
619 	size += pageoff;
620 	size = (vm_size_t) round_page(size);
621 	if (addr + size < addr)
622 		return (EINVAL);
623 
624 	/*
625 	 * Check for illegal addresses.  Watch out for address wrap...
626 	 */
627 	map = &td->td_proc->p_vmspace->vm_map;
628 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
629 		return (EINVAL);
630 	vm_map_lock(map);
631 #ifdef HWPMC_HOOKS
632 	/*
633 	 * Inform hwpmc if the address range being unmapped contains
634 	 * an executable region.
635 	 */
636 	pkm.pm_address = (uintptr_t) NULL;
637 	if (vm_map_lookup_entry(map, addr, &entry)) {
638 		for (;
639 		     entry != &map->header && entry->start < addr + size;
640 		     entry = entry->next) {
641 			if (vm_map_check_protection(map, entry->start,
642 				entry->end, VM_PROT_EXECUTE) == TRUE) {
643 				pkm.pm_address = (uintptr_t) addr;
644 				pkm.pm_size = (size_t) size;
645 				break;
646 			}
647 		}
648 	}
649 #endif
650 	vm_map_delete(map, addr, addr + size);
651 
652 #ifdef HWPMC_HOOKS
653 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
654 	vm_map_lock_downgrade(map);
655 	if (pkm.pm_address != (uintptr_t) NULL)
656 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
657 	vm_map_unlock_read(map);
658 #else
659 	vm_map_unlock(map);
660 #endif
661 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
662 	return (0);
663 }
664 
665 #ifndef _SYS_SYSPROTO_H_
666 struct mprotect_args {
667 	const void *addr;
668 	size_t len;
669 	int prot;
670 };
671 #endif
672 /*
673  * MPSAFE
674  */
675 int
676 sys_mprotect(td, uap)
677 	struct thread *td;
678 	struct mprotect_args *uap;
679 {
680 	vm_offset_t addr;
681 	vm_size_t size, pageoff;
682 	vm_prot_t prot;
683 
684 	addr = (vm_offset_t) uap->addr;
685 	size = uap->len;
686 	prot = uap->prot & VM_PROT_ALL;
687 
688 	pageoff = (addr & PAGE_MASK);
689 	addr -= pageoff;
690 	size += pageoff;
691 	size = (vm_size_t) round_page(size);
692 	if (addr + size < addr)
693 		return (EINVAL);
694 
695 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
696 	    addr + size, prot, FALSE)) {
697 	case KERN_SUCCESS:
698 		return (0);
699 	case KERN_PROTECTION_FAILURE:
700 		return (EACCES);
701 	case KERN_RESOURCE_SHORTAGE:
702 		return (ENOMEM);
703 	}
704 	return (EINVAL);
705 }
706 
707 #ifndef _SYS_SYSPROTO_H_
708 struct minherit_args {
709 	void *addr;
710 	size_t len;
711 	int inherit;
712 };
713 #endif
714 /*
715  * MPSAFE
716  */
717 int
718 sys_minherit(td, uap)
719 	struct thread *td;
720 	struct minherit_args *uap;
721 {
722 	vm_offset_t addr;
723 	vm_size_t size, pageoff;
724 	vm_inherit_t inherit;
725 
726 	addr = (vm_offset_t)uap->addr;
727 	size = uap->len;
728 	inherit = uap->inherit;
729 
730 	pageoff = (addr & PAGE_MASK);
731 	addr -= pageoff;
732 	size += pageoff;
733 	size = (vm_size_t) round_page(size);
734 	if (addr + size < addr)
735 		return (EINVAL);
736 
737 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
738 	    addr + size, inherit)) {
739 	case KERN_SUCCESS:
740 		return (0);
741 	case KERN_PROTECTION_FAILURE:
742 		return (EACCES);
743 	}
744 	return (EINVAL);
745 }
746 
747 #ifndef _SYS_SYSPROTO_H_
748 struct madvise_args {
749 	void *addr;
750 	size_t len;
751 	int behav;
752 };
753 #endif
754 
755 /*
756  * MPSAFE
757  */
758 int
759 sys_madvise(td, uap)
760 	struct thread *td;
761 	struct madvise_args *uap;
762 {
763 	vm_offset_t start, end;
764 	vm_map_t map;
765 	int flags;
766 
767 	/*
768 	 * Check for our special case, advising the swap pager we are
769 	 * "immortal."
770 	 */
771 	if (uap->behav == MADV_PROTECT) {
772 		flags = PPROT_SET;
773 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
774 		    PROC_SPROTECT, &flags));
775 	}
776 
777 	/*
778 	 * Check for illegal behavior
779 	 */
780 	if (uap->behav < 0 || uap->behav > MADV_CORE)
781 		return (EINVAL);
782 	/*
783 	 * Check for illegal addresses.  Watch out for address wrap... Note
784 	 * that VM_*_ADDRESS are not constants due to casts (argh).
785 	 */
786 	map = &td->td_proc->p_vmspace->vm_map;
787 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
788 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
789 		return (EINVAL);
790 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
791 		return (EINVAL);
792 
793 	/*
794 	 * Since this routine is only advisory, we default to conservative
795 	 * behavior.
796 	 */
797 	start = trunc_page((vm_offset_t) uap->addr);
798 	end = round_page((vm_offset_t) uap->addr + uap->len);
799 
800 	if (vm_map_madvise(map, start, end, uap->behav))
801 		return (EINVAL);
802 	return (0);
803 }
804 
805 #ifndef _SYS_SYSPROTO_H_
806 struct mincore_args {
807 	const void *addr;
808 	size_t len;
809 	char *vec;
810 };
811 #endif
812 
813 /*
814  * MPSAFE
815  */
816 int
817 sys_mincore(td, uap)
818 	struct thread *td;
819 	struct mincore_args *uap;
820 {
821 	vm_offset_t addr, first_addr;
822 	vm_offset_t end, cend;
823 	pmap_t pmap;
824 	vm_map_t map;
825 	char *vec;
826 	int error = 0;
827 	int vecindex, lastvecindex;
828 	vm_map_entry_t current;
829 	vm_map_entry_t entry;
830 	vm_object_t object;
831 	vm_paddr_t locked_pa;
832 	vm_page_t m;
833 	vm_pindex_t pindex;
834 	int mincoreinfo;
835 	unsigned int timestamp;
836 	boolean_t locked;
837 
838 	/*
839 	 * Make sure that the addresses presented are valid for user
840 	 * mode.
841 	 */
842 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
843 	end = addr + (vm_size_t)round_page(uap->len);
844 	map = &td->td_proc->p_vmspace->vm_map;
845 	if (end > vm_map_max(map) || end < addr)
846 		return (ENOMEM);
847 
848 	/*
849 	 * Address of byte vector
850 	 */
851 	vec = uap->vec;
852 
853 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
854 
855 	vm_map_lock_read(map);
856 RestartScan:
857 	timestamp = map->timestamp;
858 
859 	if (!vm_map_lookup_entry(map, addr, &entry)) {
860 		vm_map_unlock_read(map);
861 		return (ENOMEM);
862 	}
863 
864 	/*
865 	 * Do this on a map entry basis so that if the pages are not
866 	 * in the current processes address space, we can easily look
867 	 * up the pages elsewhere.
868 	 */
869 	lastvecindex = -1;
870 	for (current = entry;
871 	    (current != &map->header) && (current->start < end);
872 	    current = current->next) {
873 
874 		/*
875 		 * check for contiguity
876 		 */
877 		if (current->end < end &&
878 		    (entry->next == &map->header ||
879 		     current->next->start > current->end)) {
880 			vm_map_unlock_read(map);
881 			return (ENOMEM);
882 		}
883 
884 		/*
885 		 * ignore submaps (for now) or null objects
886 		 */
887 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
888 			current->object.vm_object == NULL)
889 			continue;
890 
891 		/*
892 		 * limit this scan to the current map entry and the
893 		 * limits for the mincore call
894 		 */
895 		if (addr < current->start)
896 			addr = current->start;
897 		cend = current->end;
898 		if (cend > end)
899 			cend = end;
900 
901 		/*
902 		 * scan this entry one page at a time
903 		 */
904 		while (addr < cend) {
905 			/*
906 			 * Check pmap first, it is likely faster, also
907 			 * it can provide info as to whether we are the
908 			 * one referencing or modifying the page.
909 			 */
910 			object = NULL;
911 			locked_pa = 0;
912 		retry:
913 			m = NULL;
914 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
915 			if (locked_pa != 0) {
916 				/*
917 				 * The page is mapped by this process but not
918 				 * both accessed and modified.  It is also
919 				 * managed.  Acquire the object lock so that
920 				 * other mappings might be examined.
921 				 */
922 				m = PHYS_TO_VM_PAGE(locked_pa);
923 				if (m->object != object) {
924 					if (object != NULL)
925 						VM_OBJECT_WUNLOCK(object);
926 					object = m->object;
927 					locked = VM_OBJECT_TRYWLOCK(object);
928 					vm_page_unlock(m);
929 					if (!locked) {
930 						VM_OBJECT_WLOCK(object);
931 						vm_page_lock(m);
932 						goto retry;
933 					}
934 				} else
935 					vm_page_unlock(m);
936 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
937 				    ("mincore: page %p is mapped but invalid",
938 				    m));
939 			} else if (mincoreinfo == 0) {
940 				/*
941 				 * The page is not mapped by this process.  If
942 				 * the object implements managed pages, then
943 				 * determine if the page is resident so that
944 				 * the mappings might be examined.
945 				 */
946 				if (current->object.vm_object != object) {
947 					if (object != NULL)
948 						VM_OBJECT_WUNLOCK(object);
949 					object = current->object.vm_object;
950 					VM_OBJECT_WLOCK(object);
951 				}
952 				if (object->type == OBJT_DEFAULT ||
953 				    object->type == OBJT_SWAP ||
954 				    object->type == OBJT_VNODE) {
955 					pindex = OFF_TO_IDX(current->offset +
956 					    (addr - current->start));
957 					m = vm_page_lookup(object, pindex);
958 					if (m == NULL &&
959 					    vm_page_is_cached(object, pindex))
960 						mincoreinfo = MINCORE_INCORE;
961 					if (m != NULL && m->valid == 0)
962 						m = NULL;
963 					if (m != NULL)
964 						mincoreinfo = MINCORE_INCORE;
965 				}
966 			}
967 			if (m != NULL) {
968 				/* Examine other mappings to the page. */
969 				if (m->dirty == 0 && pmap_is_modified(m))
970 					vm_page_dirty(m);
971 				if (m->dirty != 0)
972 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
973 				/*
974 				 * The first test for PGA_REFERENCED is an
975 				 * optimization.  The second test is
976 				 * required because a concurrent pmap
977 				 * operation could clear the last reference
978 				 * and set PGA_REFERENCED before the call to
979 				 * pmap_is_referenced().
980 				 */
981 				if ((m->aflags & PGA_REFERENCED) != 0 ||
982 				    pmap_is_referenced(m) ||
983 				    (m->aflags & PGA_REFERENCED) != 0)
984 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
985 			}
986 			if (object != NULL)
987 				VM_OBJECT_WUNLOCK(object);
988 
989 			/*
990 			 * subyte may page fault.  In case it needs to modify
991 			 * the map, we release the lock.
992 			 */
993 			vm_map_unlock_read(map);
994 
995 			/*
996 			 * calculate index into user supplied byte vector
997 			 */
998 			vecindex = OFF_TO_IDX(addr - first_addr);
999 
1000 			/*
1001 			 * If we have skipped map entries, we need to make sure that
1002 			 * the byte vector is zeroed for those skipped entries.
1003 			 */
1004 			while ((lastvecindex + 1) < vecindex) {
1005 				++lastvecindex;
1006 				error = subyte(vec + lastvecindex, 0);
1007 				if (error) {
1008 					error = EFAULT;
1009 					goto done2;
1010 				}
1011 			}
1012 
1013 			/*
1014 			 * Pass the page information to the user
1015 			 */
1016 			error = subyte(vec + vecindex, mincoreinfo);
1017 			if (error) {
1018 				error = EFAULT;
1019 				goto done2;
1020 			}
1021 
1022 			/*
1023 			 * If the map has changed, due to the subyte, the previous
1024 			 * output may be invalid.
1025 			 */
1026 			vm_map_lock_read(map);
1027 			if (timestamp != map->timestamp)
1028 				goto RestartScan;
1029 
1030 			lastvecindex = vecindex;
1031 			addr += PAGE_SIZE;
1032 		}
1033 	}
1034 
1035 	/*
1036 	 * subyte may page fault.  In case it needs to modify
1037 	 * the map, we release the lock.
1038 	 */
1039 	vm_map_unlock_read(map);
1040 
1041 	/*
1042 	 * Zero the last entries in the byte vector.
1043 	 */
1044 	vecindex = OFF_TO_IDX(end - first_addr);
1045 	while ((lastvecindex + 1) < vecindex) {
1046 		++lastvecindex;
1047 		error = subyte(vec + lastvecindex, 0);
1048 		if (error) {
1049 			error = EFAULT;
1050 			goto done2;
1051 		}
1052 	}
1053 
1054 	/*
1055 	 * If the map has changed, due to the subyte, the previous
1056 	 * output may be invalid.
1057 	 */
1058 	vm_map_lock_read(map);
1059 	if (timestamp != map->timestamp)
1060 		goto RestartScan;
1061 	vm_map_unlock_read(map);
1062 done2:
1063 	return (error);
1064 }
1065 
1066 #ifndef _SYS_SYSPROTO_H_
1067 struct mlock_args {
1068 	const void *addr;
1069 	size_t len;
1070 };
1071 #endif
1072 /*
1073  * MPSAFE
1074  */
1075 int
1076 sys_mlock(td, uap)
1077 	struct thread *td;
1078 	struct mlock_args *uap;
1079 {
1080 
1081 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1082 }
1083 
1084 int
1085 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1086 {
1087 	vm_offset_t addr, end, last, start;
1088 	vm_size_t npages, size;
1089 	vm_map_t map;
1090 	unsigned long nsize;
1091 	int error;
1092 
1093 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1094 	if (error)
1095 		return (error);
1096 	addr = (vm_offset_t)addr0;
1097 	size = len;
1098 	last = addr + size;
1099 	start = trunc_page(addr);
1100 	end = round_page(last);
1101 	if (last < addr || end < addr)
1102 		return (EINVAL);
1103 	npages = atop(end - start);
1104 	if (npages > vm_page_max_wired)
1105 		return (ENOMEM);
1106 	map = &proc->p_vmspace->vm_map;
1107 	PROC_LOCK(proc);
1108 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1109 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1110 		PROC_UNLOCK(proc);
1111 		return (ENOMEM);
1112 	}
1113 	PROC_UNLOCK(proc);
1114 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1115 		return (EAGAIN);
1116 #ifdef RACCT
1117 	PROC_LOCK(proc);
1118 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1119 	PROC_UNLOCK(proc);
1120 	if (error != 0)
1121 		return (ENOMEM);
1122 #endif
1123 	error = vm_map_wire(map, start, end,
1124 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1125 #ifdef RACCT
1126 	if (error != KERN_SUCCESS) {
1127 		PROC_LOCK(proc);
1128 		racct_set(proc, RACCT_MEMLOCK,
1129 		    ptoa(pmap_wired_count(map->pmap)));
1130 		PROC_UNLOCK(proc);
1131 	}
1132 #endif
1133 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1134 }
1135 
1136 #ifndef _SYS_SYSPROTO_H_
1137 struct mlockall_args {
1138 	int	how;
1139 };
1140 #endif
1141 
1142 /*
1143  * MPSAFE
1144  */
1145 int
1146 sys_mlockall(td, uap)
1147 	struct thread *td;
1148 	struct mlockall_args *uap;
1149 {
1150 	vm_map_t map;
1151 	int error;
1152 
1153 	map = &td->td_proc->p_vmspace->vm_map;
1154 	error = priv_check(td, PRIV_VM_MLOCK);
1155 	if (error)
1156 		return (error);
1157 
1158 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1159 		return (EINVAL);
1160 
1161 	/*
1162 	 * If wiring all pages in the process would cause it to exceed
1163 	 * a hard resource limit, return ENOMEM.
1164 	 */
1165 	if (!old_mlock && uap->how & MCL_CURRENT) {
1166 		PROC_LOCK(td->td_proc);
1167 		if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1168 			PROC_UNLOCK(td->td_proc);
1169 			return (ENOMEM);
1170 		}
1171 		PROC_UNLOCK(td->td_proc);
1172 	}
1173 #ifdef RACCT
1174 	PROC_LOCK(td->td_proc);
1175 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1176 	PROC_UNLOCK(td->td_proc);
1177 	if (error != 0)
1178 		return (ENOMEM);
1179 #endif
1180 
1181 	if (uap->how & MCL_FUTURE) {
1182 		vm_map_lock(map);
1183 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1184 		vm_map_unlock(map);
1185 		error = 0;
1186 	}
1187 
1188 	if (uap->how & MCL_CURRENT) {
1189 		/*
1190 		 * P1003.1-2001 mandates that all currently mapped pages
1191 		 * will be memory resident and locked (wired) upon return
1192 		 * from mlockall(). vm_map_wire() will wire pages, by
1193 		 * calling vm_fault_wire() for each page in the region.
1194 		 */
1195 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1196 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1197 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1198 	}
1199 #ifdef RACCT
1200 	if (error != KERN_SUCCESS) {
1201 		PROC_LOCK(td->td_proc);
1202 		racct_set(td->td_proc, RACCT_MEMLOCK,
1203 		    ptoa(pmap_wired_count(map->pmap)));
1204 		PROC_UNLOCK(td->td_proc);
1205 	}
1206 #endif
1207 
1208 	return (error);
1209 }
1210 
1211 #ifndef _SYS_SYSPROTO_H_
1212 struct munlockall_args {
1213 	register_t dummy;
1214 };
1215 #endif
1216 
1217 /*
1218  * MPSAFE
1219  */
1220 int
1221 sys_munlockall(td, uap)
1222 	struct thread *td;
1223 	struct munlockall_args *uap;
1224 {
1225 	vm_map_t map;
1226 	int error;
1227 
1228 	map = &td->td_proc->p_vmspace->vm_map;
1229 	error = priv_check(td, PRIV_VM_MUNLOCK);
1230 	if (error)
1231 		return (error);
1232 
1233 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1234 	vm_map_lock(map);
1235 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1236 	vm_map_unlock(map);
1237 
1238 	/* Forcibly unwire all pages. */
1239 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1240 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1241 #ifdef RACCT
1242 	if (error == KERN_SUCCESS) {
1243 		PROC_LOCK(td->td_proc);
1244 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1245 		PROC_UNLOCK(td->td_proc);
1246 	}
1247 #endif
1248 
1249 	return (error);
1250 }
1251 
1252 #ifndef _SYS_SYSPROTO_H_
1253 struct munlock_args {
1254 	const void *addr;
1255 	size_t len;
1256 };
1257 #endif
1258 /*
1259  * MPSAFE
1260  */
1261 int
1262 sys_munlock(td, uap)
1263 	struct thread *td;
1264 	struct munlock_args *uap;
1265 {
1266 	vm_offset_t addr, end, last, start;
1267 	vm_size_t size;
1268 #ifdef RACCT
1269 	vm_map_t map;
1270 #endif
1271 	int error;
1272 
1273 	error = priv_check(td, PRIV_VM_MUNLOCK);
1274 	if (error)
1275 		return (error);
1276 	addr = (vm_offset_t)uap->addr;
1277 	size = uap->len;
1278 	last = addr + size;
1279 	start = trunc_page(addr);
1280 	end = round_page(last);
1281 	if (last < addr || end < addr)
1282 		return (EINVAL);
1283 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1284 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1285 #ifdef RACCT
1286 	if (error == KERN_SUCCESS) {
1287 		PROC_LOCK(td->td_proc);
1288 		map = &td->td_proc->p_vmspace->vm_map;
1289 		racct_set(td->td_proc, RACCT_MEMLOCK,
1290 		    ptoa(pmap_wired_count(map->pmap)));
1291 		PROC_UNLOCK(td->td_proc);
1292 	}
1293 #endif
1294 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1295 }
1296 
1297 /*
1298  * vm_mmap_vnode()
1299  *
1300  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1301  * operations on vnodes.
1302  *
1303  * For VCHR vnodes, the vnode lock is held over the call to
1304  * vm_mmap_cdev() to keep vp->v_rdev valid.
1305  */
1306 int
1307 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1308     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1309     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1310     boolean_t *writecounted)
1311 {
1312 	struct vattr va;
1313 	vm_object_t obj;
1314 	vm_offset_t foff;
1315 	struct mount *mp;
1316 	struct ucred *cred;
1317 	int error, flags, locktype;
1318 
1319 	mp = vp->v_mount;
1320 	cred = td->td_ucred;
1321 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1322 		locktype = LK_EXCLUSIVE;
1323 	else
1324 		locktype = LK_SHARED;
1325 	if ((error = vget(vp, locktype, td)) != 0)
1326 		return (error);
1327 	foff = *foffp;
1328 	flags = *flagsp;
1329 	obj = vp->v_object;
1330 	if (vp->v_type == VREG) {
1331 		/*
1332 		 * Get the proper underlying object
1333 		 */
1334 		if (obj == NULL) {
1335 			error = EINVAL;
1336 			goto done;
1337 		}
1338 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1339 			vput(vp);
1340 			vp = (struct vnode *)obj->handle;
1341 			/*
1342 			 * Bypass filesystems obey the mpsafety of the
1343 			 * underlying fs.  Tmpfs never bypasses.
1344 			 */
1345 			error = vget(vp, locktype, td);
1346 			if (error != 0)
1347 				return (error);
1348 		}
1349 		if (locktype == LK_EXCLUSIVE) {
1350 			*writecounted = TRUE;
1351 			vnode_pager_update_writecount(obj, 0, objsize);
1352 		}
1353 	} else if (vp->v_type == VCHR) {
1354 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1355 		    vp->v_rdev, foffp, objp);
1356 		if (error == 0)
1357 			goto mark_atime;
1358 		goto done;
1359 	} else {
1360 		error = EINVAL;
1361 		goto done;
1362 	}
1363 	if ((error = VOP_GETATTR(vp, &va, cred)))
1364 		goto done;
1365 #ifdef MAC
1366 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1367 	if (error != 0)
1368 		goto done;
1369 #endif
1370 	if ((flags & MAP_SHARED) != 0) {
1371 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1372 			if (prot & PROT_WRITE) {
1373 				error = EPERM;
1374 				goto done;
1375 			}
1376 			*maxprotp &= ~VM_PROT_WRITE;
1377 		}
1378 	}
1379 	/*
1380 	 * If it is a regular file without any references
1381 	 * we do not need to sync it.
1382 	 * Adjust object size to be the size of actual file.
1383 	 */
1384 	objsize = round_page(va.va_size);
1385 	if (va.va_nlink == 0)
1386 		flags |= MAP_NOSYNC;
1387 	if (obj->type == OBJT_VNODE)
1388 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1389 		    cred);
1390 	else {
1391 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1392 		    ("wrong object type"));
1393 		vm_object_reference(obj);
1394 	}
1395 	if (obj == NULL) {
1396 		error = ENOMEM;
1397 		goto done;
1398 	}
1399 	*objp = obj;
1400 	*flagsp = flags;
1401 
1402 mark_atime:
1403 	vfs_mark_atime(vp, cred);
1404 
1405 done:
1406 	if (error != 0 && *writecounted) {
1407 		*writecounted = FALSE;
1408 		vnode_pager_update_writecount(obj, objsize, 0);
1409 	}
1410 	vput(vp);
1411 	return (error);
1412 }
1413 
1414 /*
1415  * vm_mmap_cdev()
1416  *
1417  * MPSAFE
1418  *
1419  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1420  * operations on cdevs.
1421  */
1422 int
1423 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1424     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1425     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1426 {
1427 	vm_object_t obj;
1428 	struct cdevsw *dsw;
1429 	int error, flags, ref;
1430 
1431 	flags = *flagsp;
1432 
1433 	dsw = dev_refthread(cdev, &ref);
1434 	if (dsw == NULL)
1435 		return (ENXIO);
1436 	if (dsw->d_flags & D_MMAP_ANON) {
1437 		dev_relthread(cdev, ref);
1438 		*maxprotp = VM_PROT_ALL;
1439 		*flagsp |= MAP_ANON;
1440 		return (0);
1441 	}
1442 	/*
1443 	 * cdevs do not provide private mappings of any kind.
1444 	 */
1445 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1446 	    (prot & PROT_WRITE) != 0) {
1447 		dev_relthread(cdev, ref);
1448 		return (EACCES);
1449 	}
1450 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1451 		dev_relthread(cdev, ref);
1452 		return (EINVAL);
1453 	}
1454 	/*
1455 	 * Force device mappings to be shared.
1456 	 */
1457 	flags |= MAP_SHARED;
1458 #ifdef MAC_XXX
1459 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1460 	if (error != 0) {
1461 		dev_relthread(cdev, ref);
1462 		return (error);
1463 	}
1464 #endif
1465 	/*
1466 	 * First, try d_mmap_single().  If that is not implemented
1467 	 * (returns ENODEV), fall back to using the device pager.
1468 	 * Note that d_mmap_single() must return a reference to the
1469 	 * object (it needs to bump the reference count of the object
1470 	 * it returns somehow).
1471 	 *
1472 	 * XXX assumes VM_PROT_* == PROT_*
1473 	 */
1474 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1475 	dev_relthread(cdev, ref);
1476 	if (error != ENODEV)
1477 		return (error);
1478 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1479 	    td->td_ucred);
1480 	if (obj == NULL)
1481 		return (EINVAL);
1482 	*objp = obj;
1483 	*flagsp = flags;
1484 	return (0);
1485 }
1486 
1487 /*
1488  * vm_mmap_shm()
1489  *
1490  * MPSAFE
1491  *
1492  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1493  * operations on shm file descriptors.
1494  */
1495 int
1496 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1497     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1498     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1499 {
1500 	int error;
1501 
1502 	if ((*flagsp & MAP_SHARED) != 0 &&
1503 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1504 	    (prot & PROT_WRITE) != 0)
1505 		return (EACCES);
1506 #ifdef MAC
1507 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1508 	if (error != 0)
1509 		return (error);
1510 #endif
1511 	error = shm_mmap(shmfd, objsize, foff, objp);
1512 	if (error)
1513 		return (error);
1514 	return (0);
1515 }
1516 
1517 /*
1518  * vm_mmap()
1519  *
1520  * MPSAFE
1521  *
1522  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1523  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1524  */
1525 int
1526 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1527 	vm_prot_t maxprot, int flags,
1528 	objtype_t handle_type, void *handle,
1529 	vm_ooffset_t foff)
1530 {
1531 	boolean_t fitit;
1532 	vm_object_t object = NULL;
1533 	struct thread *td = curthread;
1534 	int docow, error, findspace, rv;
1535 	boolean_t writecounted;
1536 
1537 	if (size == 0)
1538 		return (0);
1539 
1540 	size = round_page(size);
1541 
1542 	if (map == &td->td_proc->p_vmspace->vm_map) {
1543 		PROC_LOCK(td->td_proc);
1544 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1545 			PROC_UNLOCK(td->td_proc);
1546 			return (ENOMEM);
1547 		}
1548 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1549 			PROC_UNLOCK(td->td_proc);
1550 			return (ENOMEM);
1551 		}
1552 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1553 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1554 			    lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1555 				racct_set_force(td->td_proc, RACCT_VMEM,
1556 				    map->size);
1557 				PROC_UNLOCK(td->td_proc);
1558 				return (ENOMEM);
1559 			}
1560 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1561 			    ptoa(pmap_wired_count(map->pmap)) + size);
1562 			if (error != 0) {
1563 				racct_set_force(td->td_proc, RACCT_VMEM,
1564 				    map->size);
1565 				PROC_UNLOCK(td->td_proc);
1566 				return (error);
1567 			}
1568 		}
1569 		PROC_UNLOCK(td->td_proc);
1570 	}
1571 
1572 	/*
1573 	 * We currently can only deal with page aligned file offsets.
1574 	 * The check is here rather than in the syscall because the
1575 	 * kernel calls this function internally for other mmaping
1576 	 * operations (such as in exec) and non-aligned offsets will
1577 	 * cause pmap inconsistencies...so we want to be sure to
1578 	 * disallow this in all cases.
1579 	 */
1580 	if (foff & PAGE_MASK)
1581 		return (EINVAL);
1582 
1583 	if ((flags & MAP_FIXED) == 0) {
1584 		fitit = TRUE;
1585 		*addr = round_page(*addr);
1586 	} else {
1587 		if (*addr != trunc_page(*addr))
1588 			return (EINVAL);
1589 		fitit = FALSE;
1590 	}
1591 	writecounted = FALSE;
1592 
1593 	/*
1594 	 * Lookup/allocate object.
1595 	 */
1596 	switch (handle_type) {
1597 	case OBJT_DEVICE:
1598 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1599 		    handle, &foff, &object);
1600 		break;
1601 	case OBJT_VNODE:
1602 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1603 		    handle, &foff, &object, &writecounted);
1604 		break;
1605 	case OBJT_SWAP:
1606 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1607 		    handle, foff, &object);
1608 		break;
1609 	case OBJT_DEFAULT:
1610 		if (handle == NULL) {
1611 			error = 0;
1612 			break;
1613 		}
1614 		/* FALLTHROUGH */
1615 	default:
1616 		error = EINVAL;
1617 		break;
1618 	}
1619 	if (error)
1620 		return (error);
1621 	if (flags & MAP_ANON) {
1622 		object = NULL;
1623 		docow = 0;
1624 		/*
1625 		 * Unnamed anonymous regions always start at 0.
1626 		 */
1627 		if (handle == 0)
1628 			foff = 0;
1629 	} else if (flags & MAP_PREFAULT_READ)
1630 		docow = MAP_PREFAULT;
1631 	else
1632 		docow = MAP_PREFAULT_PARTIAL;
1633 
1634 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1635 		docow |= MAP_COPY_ON_WRITE;
1636 	if (flags & MAP_NOSYNC)
1637 		docow |= MAP_DISABLE_SYNCER;
1638 	if (flags & MAP_NOCORE)
1639 		docow |= MAP_DISABLE_COREDUMP;
1640 	/* Shared memory is also shared with children. */
1641 	if (flags & MAP_SHARED)
1642 		docow |= MAP_INHERIT_SHARE;
1643 	if (writecounted)
1644 		docow |= MAP_VN_WRITECOUNT;
1645 	if (flags & MAP_STACK) {
1646 		if (object != NULL)
1647 			return (EINVAL);
1648 		docow |= MAP_STACK_GROWS_DOWN;
1649 	}
1650 	if ((flags & MAP_EXCL) != 0)
1651 		docow |= MAP_CHECK_EXCL;
1652 
1653 	if (fitit) {
1654 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1655 			findspace = VMFS_SUPER_SPACE;
1656 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1657 			findspace = VMFS_ALIGNED_SPACE(flags >>
1658 			    MAP_ALIGNMENT_SHIFT);
1659 		else
1660 			findspace = VMFS_OPTIMAL_SPACE;
1661 		rv = vm_map_find(map, object, foff, addr, size,
1662 #ifdef MAP_32BIT
1663 		    flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1664 #endif
1665 		    0, findspace, prot, maxprot, docow);
1666 	} else {
1667 		rv = vm_map_fixed(map, object, foff, *addr, size,
1668 		    prot, maxprot, docow);
1669 	}
1670 
1671 	if (rv == KERN_SUCCESS) {
1672 		/*
1673 		 * If the process has requested that all future mappings
1674 		 * be wired, then heed this.
1675 		 */
1676 		if (map->flags & MAP_WIREFUTURE) {
1677 			vm_map_wire(map, *addr, *addr + size,
1678 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1679 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1680 		}
1681 	} else {
1682 		/*
1683 		 * If this mapping was accounted for in the vnode's
1684 		 * writecount, then undo that now.
1685 		 */
1686 		if (writecounted)
1687 			vnode_pager_release_writecount(object, 0, size);
1688 		/*
1689 		 * Lose the object reference.  Will destroy the
1690 		 * object if it's an unnamed anonymous mapping
1691 		 * or named anonymous without other references.
1692 		 */
1693 		vm_object_deallocate(object);
1694 	}
1695 	return (vm_mmap_to_errno(rv));
1696 }
1697 
1698 /*
1699  * Translate a Mach VM return code to zero on success or the appropriate errno
1700  * on failure.
1701  */
1702 int
1703 vm_mmap_to_errno(int rv)
1704 {
1705 
1706 	switch (rv) {
1707 	case KERN_SUCCESS:
1708 		return (0);
1709 	case KERN_INVALID_ADDRESS:
1710 	case KERN_NO_SPACE:
1711 		return (ENOMEM);
1712 	case KERN_PROTECTION_FAILURE:
1713 		return (EACCES);
1714 	default:
1715 		return (EINVAL);
1716 	}
1717 }
1718