xref: /freebsd/sys/vm/vm_mmap.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1988 University of Utah.
5  * Copyright (c) 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
37  *
38  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
39  */
40 
41 /*
42  * Mapped file (mmap) interface to VM
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_hwpmc_hooks.h"
49 #include "opt_vm.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/capsicum.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/sysproto.h>
58 #include <sys/elf.h>
59 #include <sys/filedesc.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/procctl.h>
63 #include <sys/racct.h>
64 #include <sys/resource.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
69 #include <sys/fcntl.h>
70 #include <sys/file.h>
71 #include <sys/mman.h>
72 #include <sys/mount.h>
73 #include <sys/conf.h>
74 #include <sys/stat.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/vmmeter.h>
78 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
79 #include <machine/md_var.h>
80 #endif
81 
82 #include <security/audit/audit.h>
83 #include <security/mac/mac_framework.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_pageout.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vm_page.h>
95 #include <vm/vnode_pager.h>
96 
97 #ifdef HWPMC_HOOKS
98 #include <sys/pmckern.h>
99 #endif
100 
101 int old_mlock = 0;
102 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
103     "Do not apply RLIMIT_MEMLOCK on mlockall");
104 static int mincore_mapped = 1;
105 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
106     "mincore reports mappings, not residency");
107 static int imply_prot_max = 0;
108 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0,
109     "Imply maximum page permissions in mmap() when none are specified");
110 
111 #ifdef MAP_32BIT
112 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
113 #endif
114 
115 #ifndef _SYS_SYSPROTO_H_
116 struct sbrk_args {
117 	int incr;
118 };
119 #endif
120 
121 int
122 sys_sbrk(struct thread *td, 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 int
135 sys_sstk(struct thread *td, struct sstk_args *uap)
136 {
137 	/* Not yet implemented */
138 	return (EOPNOTSUPP);
139 }
140 
141 #if defined(COMPAT_43)
142 int
143 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
144 {
145 
146 	td->td_retval[0] = PAGE_SIZE;
147 	return (0);
148 }
149 #endif				/* COMPAT_43 */
150 
151 
152 /*
153  * Memory Map (mmap) system call.  Note that the file offset
154  * and address are allowed to be NOT page aligned, though if
155  * the MAP_FIXED flag it set, both must have the same remainder
156  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
157  * page-aligned, the actual mapping starts at trunc_page(addr)
158  * and the return value is adjusted up by the page offset.
159  *
160  * Generally speaking, only character devices which are themselves
161  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
162  * there would be no cache coherency between a descriptor and a VM mapping
163  * both to the same character device.
164  */
165 #ifndef _SYS_SYSPROTO_H_
166 struct mmap_args {
167 	void *addr;
168 	size_t len;
169 	int prot;
170 	int flags;
171 	int fd;
172 	long pad;
173 	off_t pos;
174 };
175 #endif
176 
177 int
178 sys_mmap(struct thread *td, struct mmap_args *uap)
179 {
180 
181 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
182 	    uap->flags, uap->fd, uap->pos));
183 }
184 
185 int
186 kern_mmap_maxprot(struct proc *p, int prot)
187 {
188 
189 	if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
190 	    (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
191 		return (_PROT_ALL);
192 	if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
193 	    prot != PROT_NONE)
194 		 return (prot);
195 	return (_PROT_ALL);
196 }
197 
198 int
199 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags,
200     int fd, off_t pos)
201 {
202 	struct vmspace *vms;
203 	struct file *fp;
204 	struct proc *p;
205 	vm_offset_t addr;
206 	vm_size_t pageoff, size;
207 	vm_prot_t cap_maxprot;
208 	int align, error, max_prot;
209 	cap_rights_t rights;
210 
211 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
212 		return (EINVAL);
213 	max_prot = PROT_MAX_EXTRACT(prot);
214 	prot = PROT_EXTRACT(prot);
215 	if (max_prot != 0 && (max_prot & prot) != prot)
216 		return (EINVAL);
217 
218 	p = td->td_proc;
219 
220 	/*
221 	 * Always honor PROT_MAX if set.  If not, default to all
222 	 * permissions unless we're implying maximum permissions.
223 	 */
224 	if (max_prot == 0)
225 		max_prot = kern_mmap_maxprot(p, prot);
226 
227 	vms = p->p_vmspace;
228 	fp = NULL;
229 	AUDIT_ARG_FD(fd);
230 	addr = addr0;
231 
232 	/*
233 	 * Ignore old flags that used to be defined but did not do anything.
234 	 */
235 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
236 
237 	/*
238 	 * Enforce the constraints.
239 	 * Mapping of length 0 is only allowed for old binaries.
240 	 * Anonymous mapping shall specify -1 as filedescriptor and
241 	 * zero position for new code. Be nice to ancient a.out
242 	 * binaries and correct pos for anonymous mapping, since old
243 	 * ld.so sometimes issues anonymous map requests with non-zero
244 	 * pos.
245 	 */
246 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
247 		if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
248 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
249 			return (EINVAL);
250 	} else {
251 		if ((flags & MAP_ANON) != 0)
252 			pos = 0;
253 	}
254 
255 	if (flags & MAP_STACK) {
256 		if ((fd != -1) ||
257 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
258 			return (EINVAL);
259 		flags |= MAP_ANON;
260 		pos = 0;
261 	}
262 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
263 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
264 	    MAP_PREFAULT_READ | MAP_GUARD |
265 #ifdef MAP_32BIT
266 	    MAP_32BIT |
267 #endif
268 	    MAP_ALIGNMENT_MASK)) != 0)
269 		return (EINVAL);
270 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
271 		return (EINVAL);
272 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
273 		return (EINVAL);
274 	if (prot != PROT_NONE &&
275 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
276 		return (EINVAL);
277 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
278 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
279 #ifdef MAP_32BIT
280 	    MAP_32BIT |
281 #endif
282 	    MAP_ALIGNMENT_MASK)) != 0))
283 		return (EINVAL);
284 
285 	/*
286 	 * Align the file position to a page boundary,
287 	 * and save its page offset component.
288 	 */
289 	pageoff = (pos & PAGE_MASK);
290 	pos -= pageoff;
291 
292 	/* Compute size from len by rounding (on both ends). */
293 	size = len + pageoff;			/* low end... */
294 	size = round_page(size);		/* hi end */
295 	/* Check for rounding up to zero. */
296 	if (len > size)
297 		return (ENOMEM);
298 
299 	/* Ensure alignment is at least a page and fits in a pointer. */
300 	align = flags & MAP_ALIGNMENT_MASK;
301 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
302 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
303 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
304 		return (EINVAL);
305 
306 	/*
307 	 * Check for illegal addresses.  Watch out for address wrap... Note
308 	 * that VM_*_ADDRESS are not constants due to casts (argh).
309 	 */
310 	if (flags & MAP_FIXED) {
311 		/*
312 		 * The specified address must have the same remainder
313 		 * as the file offset taken modulo PAGE_SIZE, so it
314 		 * should be aligned after adjustment by pageoff.
315 		 */
316 		addr -= pageoff;
317 		if (addr & PAGE_MASK)
318 			return (EINVAL);
319 
320 		/* Address range must be all in user VM space. */
321 		if (addr < vm_map_min(&vms->vm_map) ||
322 		    addr + size > vm_map_max(&vms->vm_map))
323 			return (EINVAL);
324 		if (addr + size < addr)
325 			return (EINVAL);
326 #ifdef MAP_32BIT
327 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
328 			return (EINVAL);
329 	} else if (flags & MAP_32BIT) {
330 		/*
331 		 * For MAP_32BIT, override the hint if it is too high and
332 		 * do not bother moving the mapping past the heap (since
333 		 * the heap is usually above 2GB).
334 		 */
335 		if (addr + size > MAP_32BIT_MAX_ADDR)
336 			addr = 0;
337 #endif
338 	} else {
339 		/*
340 		 * XXX for non-fixed mappings where no hint is provided or
341 		 * the hint would fall in the potential heap space,
342 		 * place it after the end of the largest possible heap.
343 		 *
344 		 * There should really be a pmap call to determine a reasonable
345 		 * location.
346 		 */
347 		if (addr == 0 ||
348 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
349 		    addr < round_page((vm_offset_t)vms->vm_daddr +
350 		    lim_max(td, RLIMIT_DATA))))
351 			addr = round_page((vm_offset_t)vms->vm_daddr +
352 			    lim_max(td, RLIMIT_DATA));
353 	}
354 	if (len == 0) {
355 		/*
356 		 * Return success without mapping anything for old
357 		 * binaries that request a page-aligned mapping of
358 		 * length 0.  For modern binaries, this function
359 		 * returns an error earlier.
360 		 */
361 		error = 0;
362 	} else if ((flags & MAP_GUARD) != 0) {
363 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
364 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
365 	} else if ((flags & MAP_ANON) != 0) {
366 		/*
367 		 * Mapping blank space is trivial.
368 		 *
369 		 * This relies on VM_PROT_* matching PROT_*.
370 		 */
371 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
372 		    max_prot, flags, NULL, pos, FALSE, td);
373 	} else {
374 		/*
375 		 * Mapping file, get fp for validation and don't let the
376 		 * descriptor disappear on us if we block. Check capability
377 		 * rights, but also return the maximum rights to be combined
378 		 * with maxprot later.
379 		 */
380 		cap_rights_init(&rights, CAP_MMAP);
381 		if (prot & PROT_READ)
382 			cap_rights_set(&rights, CAP_MMAP_R);
383 		if ((flags & MAP_SHARED) != 0) {
384 			if (prot & PROT_WRITE)
385 				cap_rights_set(&rights, CAP_MMAP_W);
386 		}
387 		if (prot & PROT_EXEC)
388 			cap_rights_set(&rights, CAP_MMAP_X);
389 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
390 		if (error != 0)
391 			goto done;
392 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
393 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
394 			error = EINVAL;
395 			goto done;
396 		}
397 
398 		/* This relies on VM_PROT_* matching PROT_*. */
399 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
400 		    max_prot & cap_maxprot, flags, pos, td);
401 	}
402 
403 	if (error == 0)
404 		td->td_retval[0] = (register_t) (addr + pageoff);
405 done:
406 	if (fp)
407 		fdrop(fp, td);
408 
409 	return (error);
410 }
411 
412 #if defined(COMPAT_FREEBSD6)
413 int
414 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
415 {
416 
417 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
418 	    uap->flags, uap->fd, uap->pos));
419 }
420 #endif
421 
422 #ifdef COMPAT_43
423 #ifndef _SYS_SYSPROTO_H_
424 struct ommap_args {
425 	caddr_t addr;
426 	int len;
427 	int prot;
428 	int flags;
429 	int fd;
430 	long pos;
431 };
432 #endif
433 int
434 ommap(struct thread *td, struct ommap_args *uap)
435 {
436 	static const char cvtbsdprot[8] = {
437 		0,
438 		PROT_EXEC,
439 		PROT_WRITE,
440 		PROT_EXEC | PROT_WRITE,
441 		PROT_READ,
442 		PROT_EXEC | PROT_READ,
443 		PROT_WRITE | PROT_READ,
444 		PROT_EXEC | PROT_WRITE | PROT_READ,
445 	};
446 	int flags, prot;
447 
448 #define	OMAP_ANON	0x0002
449 #define	OMAP_COPY	0x0020
450 #define	OMAP_SHARED	0x0010
451 #define	OMAP_FIXED	0x0100
452 
453 	prot = cvtbsdprot[uap->prot & 0x7];
454 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
455 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
456 	    prot != 0)
457 		prot |= PROT_EXEC;
458 #endif
459 	flags = 0;
460 	if (uap->flags & OMAP_ANON)
461 		flags |= MAP_ANON;
462 	if (uap->flags & OMAP_COPY)
463 		flags |= MAP_COPY;
464 	if (uap->flags & OMAP_SHARED)
465 		flags |= MAP_SHARED;
466 	else
467 		flags |= MAP_PRIVATE;
468 	if (uap->flags & OMAP_FIXED)
469 		flags |= MAP_FIXED;
470 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
471 	    uap->fd, uap->pos));
472 }
473 #endif				/* COMPAT_43 */
474 
475 
476 #ifndef _SYS_SYSPROTO_H_
477 struct msync_args {
478 	void *addr;
479 	size_t len;
480 	int flags;
481 };
482 #endif
483 int
484 sys_msync(struct thread *td, struct msync_args *uap)
485 {
486 
487 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
488 }
489 
490 int
491 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
492 {
493 	vm_offset_t addr;
494 	vm_size_t pageoff;
495 	vm_map_t map;
496 	int rv;
497 
498 	addr = addr0;
499 	pageoff = (addr & PAGE_MASK);
500 	addr -= pageoff;
501 	size += pageoff;
502 	size = (vm_size_t) round_page(size);
503 	if (addr + size < addr)
504 		return (EINVAL);
505 
506 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
507 		return (EINVAL);
508 
509 	map = &td->td_proc->p_vmspace->vm_map;
510 
511 	/*
512 	 * Clean the pages and interpret the return value.
513 	 */
514 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
515 	    (flags & MS_INVALIDATE) != 0);
516 	switch (rv) {
517 	case KERN_SUCCESS:
518 		return (0);
519 	case KERN_INVALID_ADDRESS:
520 		return (ENOMEM);
521 	case KERN_INVALID_ARGUMENT:
522 		return (EBUSY);
523 	case KERN_FAILURE:
524 		return (EIO);
525 	default:
526 		return (EINVAL);
527 	}
528 }
529 
530 #ifndef _SYS_SYSPROTO_H_
531 struct munmap_args {
532 	void *addr;
533 	size_t len;
534 };
535 #endif
536 int
537 sys_munmap(struct thread *td, struct munmap_args *uap)
538 {
539 
540 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
541 }
542 
543 int
544 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
545 {
546 #ifdef HWPMC_HOOKS
547 	struct pmckern_map_out pkm;
548 	vm_map_entry_t entry;
549 	bool pmc_handled;
550 #endif
551 	vm_offset_t addr;
552 	vm_size_t pageoff;
553 	vm_map_t map;
554 
555 	if (size == 0)
556 		return (EINVAL);
557 
558 	addr = addr0;
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 	/*
567 	 * Check for illegal addresses.  Watch out for address wrap...
568 	 */
569 	map = &td->td_proc->p_vmspace->vm_map;
570 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
571 		return (EINVAL);
572 	vm_map_lock(map);
573 #ifdef HWPMC_HOOKS
574 	pmc_handled = false;
575 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
576 		pmc_handled = true;
577 		/*
578 		 * Inform hwpmc if the address range being unmapped contains
579 		 * an executable region.
580 		 */
581 		pkm.pm_address = (uintptr_t) NULL;
582 		if (vm_map_lookup_entry(map, addr, &entry)) {
583 			for (; entry->start < addr + size;
584 			    entry = entry->next) {
585 				if (vm_map_check_protection(map, entry->start,
586 					entry->end, VM_PROT_EXECUTE) == TRUE) {
587 					pkm.pm_address = (uintptr_t) addr;
588 					pkm.pm_size = (size_t) size;
589 					break;
590 				}
591 			}
592 		}
593 	}
594 #endif
595 	vm_map_delete(map, addr, addr + size);
596 
597 #ifdef HWPMC_HOOKS
598 	if (__predict_false(pmc_handled)) {
599 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
600 		vm_map_lock_downgrade(map);
601 		if (pkm.pm_address != (uintptr_t) NULL)
602 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
603 		vm_map_unlock_read(map);
604 	} else
605 #endif
606 		vm_map_unlock(map);
607 
608 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
609 	return (0);
610 }
611 
612 #ifndef _SYS_SYSPROTO_H_
613 struct mprotect_args {
614 	const void *addr;
615 	size_t len;
616 	int prot;
617 };
618 #endif
619 int
620 sys_mprotect(struct thread *td, struct mprotect_args *uap)
621 {
622 
623 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
624 }
625 
626 int
627 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
628 {
629 	vm_offset_t addr;
630 	vm_size_t pageoff;
631 	int vm_error, max_prot;
632 
633 	addr = addr0;
634 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
635 		return (EINVAL);
636 	max_prot = PROT_MAX_EXTRACT(prot);
637 	prot = PROT_EXTRACT(prot);
638 	pageoff = (addr & PAGE_MASK);
639 	addr -= pageoff;
640 	size += pageoff;
641 	size = (vm_size_t) round_page(size);
642 #ifdef COMPAT_FREEBSD32
643 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
644 		if (((addr + size) & 0xffffffff) < addr)
645 			return (EINVAL);
646 	} else
647 #endif
648 	if (addr + size < addr)
649 		return (EINVAL);
650 
651 	vm_error = KERN_SUCCESS;
652 	if (max_prot != 0) {
653 		if ((max_prot & prot) != prot)
654 			return (EINVAL);
655 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
656 		    addr, addr + size, max_prot, TRUE);
657 	}
658 	if (vm_error == KERN_SUCCESS)
659 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
660 		    addr, addr + size, prot, FALSE);
661 
662 	switch (vm_error) {
663 	case KERN_SUCCESS:
664 		return (0);
665 	case KERN_PROTECTION_FAILURE:
666 		return (EACCES);
667 	case KERN_RESOURCE_SHORTAGE:
668 		return (ENOMEM);
669 	}
670 	return (EINVAL);
671 }
672 
673 #ifndef _SYS_SYSPROTO_H_
674 struct minherit_args {
675 	void *addr;
676 	size_t len;
677 	int inherit;
678 };
679 #endif
680 int
681 sys_minherit(struct thread *td, struct minherit_args *uap)
682 {
683 	vm_offset_t addr;
684 	vm_size_t size, pageoff;
685 	vm_inherit_t inherit;
686 
687 	addr = (vm_offset_t)uap->addr;
688 	size = uap->len;
689 	inherit = uap->inherit;
690 
691 	pageoff = (addr & PAGE_MASK);
692 	addr -= pageoff;
693 	size += pageoff;
694 	size = (vm_size_t) round_page(size);
695 	if (addr + size < addr)
696 		return (EINVAL);
697 
698 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
699 	    addr + size, inherit)) {
700 	case KERN_SUCCESS:
701 		return (0);
702 	case KERN_PROTECTION_FAILURE:
703 		return (EACCES);
704 	}
705 	return (EINVAL);
706 }
707 
708 #ifndef _SYS_SYSPROTO_H_
709 struct madvise_args {
710 	void *addr;
711 	size_t len;
712 	int behav;
713 };
714 #endif
715 
716 int
717 sys_madvise(struct thread *td, struct madvise_args *uap)
718 {
719 
720 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
721 }
722 
723 int
724 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
725 {
726 	vm_map_t map;
727 	vm_offset_t addr, end, start;
728 	int flags;
729 
730 	/*
731 	 * Check for our special case, advising the swap pager we are
732 	 * "immortal."
733 	 */
734 	if (behav == MADV_PROTECT) {
735 		flags = PPROT_SET;
736 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
737 		    PROC_SPROTECT, &flags));
738 	}
739 
740 	/*
741 	 * Check for illegal addresses.  Watch out for address wrap... Note
742 	 * that VM_*_ADDRESS are not constants due to casts (argh).
743 	 */
744 	map = &td->td_proc->p_vmspace->vm_map;
745 	addr = addr0;
746 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
747 		return (EINVAL);
748 	if ((addr + len) < addr)
749 		return (EINVAL);
750 
751 	/*
752 	 * Since this routine is only advisory, we default to conservative
753 	 * behavior.
754 	 */
755 	start = trunc_page(addr);
756 	end = round_page(addr + len);
757 
758 	/*
759 	 * vm_map_madvise() checks for illegal values of behav.
760 	 */
761 	return (vm_map_madvise(map, start, end, behav));
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 int
773 sys_mincore(struct thread *td, struct mincore_args *uap)
774 {
775 
776 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
777 }
778 
779 int
780 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
781 {
782 	vm_offset_t addr, first_addr;
783 	vm_offset_t end, cend;
784 	pmap_t pmap;
785 	vm_map_t map;
786 	int error = 0;
787 	int vecindex, lastvecindex;
788 	vm_map_entry_t current;
789 	vm_map_entry_t entry;
790 	vm_object_t object;
791 	vm_paddr_t locked_pa;
792 	vm_page_t m;
793 	vm_pindex_t pindex;
794 	int mincoreinfo;
795 	unsigned int timestamp;
796 	boolean_t locked;
797 
798 	/*
799 	 * Make sure that the addresses presented are valid for user
800 	 * mode.
801 	 */
802 	first_addr = addr = trunc_page(addr0);
803 	end = addr + (vm_size_t)round_page(len);
804 	map = &td->td_proc->p_vmspace->vm_map;
805 	if (end > vm_map_max(map) || end < addr)
806 		return (ENOMEM);
807 
808 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
809 
810 	vm_map_lock_read(map);
811 RestartScan:
812 	timestamp = map->timestamp;
813 
814 	if (!vm_map_lookup_entry(map, addr, &entry)) {
815 		vm_map_unlock_read(map);
816 		return (ENOMEM);
817 	}
818 
819 	/*
820 	 * Do this on a map entry basis so that if the pages are not
821 	 * in the current processes address space, we can easily look
822 	 * up the pages elsewhere.
823 	 */
824 	lastvecindex = -1;
825 	for (current = entry; current->start < end; current = current->next) {
826 
827 		/*
828 		 * check for contiguity
829 		 */
830 		if (current->end < end && current->next->start > current->end) {
831 			vm_map_unlock_read(map);
832 			return (ENOMEM);
833 		}
834 
835 		/*
836 		 * ignore submaps (for now) or null objects
837 		 */
838 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
839 			current->object.vm_object == NULL)
840 			continue;
841 
842 		/*
843 		 * limit this scan to the current map entry and the
844 		 * limits for the mincore call
845 		 */
846 		if (addr < current->start)
847 			addr = current->start;
848 		cend = current->end;
849 		if (cend > end)
850 			cend = end;
851 
852 		/*
853 		 * scan this entry one page at a time
854 		 */
855 		while (addr < cend) {
856 			/*
857 			 * Check pmap first, it is likely faster, also
858 			 * it can provide info as to whether we are the
859 			 * one referencing or modifying the page.
860 			 */
861 			object = NULL;
862 			locked_pa = 0;
863 		retry:
864 			m = NULL;
865 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
866 			if (mincore_mapped) {
867 				/*
868 				 * We only care about this pmap's
869 				 * mapping of the page, if any.
870 				 */
871 				if (locked_pa != 0) {
872 					vm_page_unlock(PHYS_TO_VM_PAGE(
873 					    locked_pa));
874 				}
875 			} else if (locked_pa != 0) {
876 				/*
877 				 * The page is mapped by this process but not
878 				 * both accessed and modified.  It is also
879 				 * managed.  Acquire the object lock so that
880 				 * other mappings might be examined.
881 				 */
882 				m = PHYS_TO_VM_PAGE(locked_pa);
883 				if (m->object != object) {
884 					if (object != NULL)
885 						VM_OBJECT_WUNLOCK(object);
886 					object = m->object;
887 					locked = VM_OBJECT_TRYWLOCK(object);
888 					vm_page_unlock(m);
889 					if (!locked) {
890 						VM_OBJECT_WLOCK(object);
891 						vm_page_lock(m);
892 						goto retry;
893 					}
894 				} else
895 					vm_page_unlock(m);
896 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
897 				    ("mincore: page %p is mapped but invalid",
898 				    m));
899 			} else if (mincoreinfo == 0) {
900 				/*
901 				 * The page is not mapped by this process.  If
902 				 * the object implements managed pages, then
903 				 * determine if the page is resident so that
904 				 * the mappings might be examined.
905 				 */
906 				if (current->object.vm_object != object) {
907 					if (object != NULL)
908 						VM_OBJECT_WUNLOCK(object);
909 					object = current->object.vm_object;
910 					VM_OBJECT_WLOCK(object);
911 				}
912 				if (object->type == OBJT_DEFAULT ||
913 				    object->type == OBJT_SWAP ||
914 				    object->type == OBJT_VNODE) {
915 					pindex = OFF_TO_IDX(current->offset +
916 					    (addr - current->start));
917 					m = vm_page_lookup(object, pindex);
918 					if (m != NULL && m->valid == 0)
919 						m = NULL;
920 					if (m != NULL)
921 						mincoreinfo = MINCORE_INCORE;
922 				}
923 			}
924 			if (m != NULL) {
925 				/* Examine other mappings to the page. */
926 				if (m->dirty == 0 && pmap_is_modified(m))
927 					vm_page_dirty(m);
928 				if (m->dirty != 0)
929 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
930 				/*
931 				 * The first test for PGA_REFERENCED is an
932 				 * optimization.  The second test is
933 				 * required because a concurrent pmap
934 				 * operation could clear the last reference
935 				 * and set PGA_REFERENCED before the call to
936 				 * pmap_is_referenced().
937 				 */
938 				if ((m->aflags & PGA_REFERENCED) != 0 ||
939 				    pmap_is_referenced(m) ||
940 				    (m->aflags & PGA_REFERENCED) != 0)
941 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
942 			}
943 			if (object != NULL)
944 				VM_OBJECT_WUNLOCK(object);
945 
946 			/*
947 			 * subyte may page fault.  In case it needs to modify
948 			 * the map, we release the lock.
949 			 */
950 			vm_map_unlock_read(map);
951 
952 			/*
953 			 * calculate index into user supplied byte vector
954 			 */
955 			vecindex = atop(addr - first_addr);
956 
957 			/*
958 			 * If we have skipped map entries, we need to make sure that
959 			 * the byte vector is zeroed for those skipped entries.
960 			 */
961 			while ((lastvecindex + 1) < vecindex) {
962 				++lastvecindex;
963 				error = subyte(vec + lastvecindex, 0);
964 				if (error) {
965 					error = EFAULT;
966 					goto done2;
967 				}
968 			}
969 
970 			/*
971 			 * Pass the page information to the user
972 			 */
973 			error = subyte(vec + vecindex, mincoreinfo);
974 			if (error) {
975 				error = EFAULT;
976 				goto done2;
977 			}
978 
979 			/*
980 			 * If the map has changed, due to the subyte, the previous
981 			 * output may be invalid.
982 			 */
983 			vm_map_lock_read(map);
984 			if (timestamp != map->timestamp)
985 				goto RestartScan;
986 
987 			lastvecindex = vecindex;
988 			addr += PAGE_SIZE;
989 		}
990 	}
991 
992 	/*
993 	 * subyte may page fault.  In case it needs to modify
994 	 * the map, we release the lock.
995 	 */
996 	vm_map_unlock_read(map);
997 
998 	/*
999 	 * Zero the last entries in the byte vector.
1000 	 */
1001 	vecindex = atop(end - first_addr);
1002 	while ((lastvecindex + 1) < vecindex) {
1003 		++lastvecindex;
1004 		error = subyte(vec + lastvecindex, 0);
1005 		if (error) {
1006 			error = EFAULT;
1007 			goto done2;
1008 		}
1009 	}
1010 
1011 	/*
1012 	 * If the map has changed, due to the subyte, the previous
1013 	 * output may be invalid.
1014 	 */
1015 	vm_map_lock_read(map);
1016 	if (timestamp != map->timestamp)
1017 		goto RestartScan;
1018 	vm_map_unlock_read(map);
1019 done2:
1020 	return (error);
1021 }
1022 
1023 #ifndef _SYS_SYSPROTO_H_
1024 struct mlock_args {
1025 	const void *addr;
1026 	size_t len;
1027 };
1028 #endif
1029 int
1030 sys_mlock(struct thread *td, struct mlock_args *uap)
1031 {
1032 
1033 	return (kern_mlock(td->td_proc, td->td_ucred,
1034 	    __DECONST(uintptr_t, uap->addr), uap->len));
1035 }
1036 
1037 int
1038 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1039 {
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_cred(cred, PRIV_VM_MLOCK);
1047 	if (error)
1048 		return (error);
1049 	addr = addr0;
1050 	size = 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_user_wired)
1058 		return (ENOMEM);
1059 	map = &proc->p_vmspace->vm_map;
1060 	PROC_LOCK(proc);
1061 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1062 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1063 		PROC_UNLOCK(proc);
1064 		return (ENOMEM);
1065 	}
1066 	PROC_UNLOCK(proc);
1067 #ifdef RACCT
1068 	if (racct_enable) {
1069 		PROC_LOCK(proc);
1070 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1071 		PROC_UNLOCK(proc);
1072 		if (error != 0)
1073 			return (ENOMEM);
1074 	}
1075 #endif
1076 	error = vm_map_wire(map, start, end,
1077 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1078 #ifdef RACCT
1079 	if (racct_enable && error != KERN_SUCCESS) {
1080 		PROC_LOCK(proc);
1081 		racct_set(proc, RACCT_MEMLOCK,
1082 		    ptoa(pmap_wired_count(map->pmap)));
1083 		PROC_UNLOCK(proc);
1084 	}
1085 #endif
1086 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1087 }
1088 
1089 #ifndef _SYS_SYSPROTO_H_
1090 struct mlockall_args {
1091 	int	how;
1092 };
1093 #endif
1094 
1095 int
1096 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1097 {
1098 	vm_map_t map;
1099 	int error;
1100 
1101 	map = &td->td_proc->p_vmspace->vm_map;
1102 	error = priv_check(td, PRIV_VM_MLOCK);
1103 	if (error)
1104 		return (error);
1105 
1106 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1107 		return (EINVAL);
1108 
1109 	/*
1110 	 * If wiring all pages in the process would cause it to exceed
1111 	 * a hard resource limit, return ENOMEM.
1112 	 */
1113 	if (!old_mlock && uap->how & MCL_CURRENT) {
1114 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1115 			return (ENOMEM);
1116 	}
1117 #ifdef RACCT
1118 	if (racct_enable) {
1119 		PROC_LOCK(td->td_proc);
1120 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1121 		PROC_UNLOCK(td->td_proc);
1122 		if (error != 0)
1123 			return (ENOMEM);
1124 	}
1125 #endif
1126 
1127 	if (uap->how & MCL_FUTURE) {
1128 		vm_map_lock(map);
1129 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1130 		vm_map_unlock(map);
1131 		error = 0;
1132 	}
1133 
1134 	if (uap->how & MCL_CURRENT) {
1135 		/*
1136 		 * P1003.1-2001 mandates that all currently mapped pages
1137 		 * will be memory resident and locked (wired) upon return
1138 		 * from mlockall(). vm_map_wire() will wire pages, by
1139 		 * calling vm_fault_wire() for each page in the region.
1140 		 */
1141 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1142 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1143 		if (error == KERN_SUCCESS)
1144 			error = 0;
1145 		else if (error == KERN_RESOURCE_SHORTAGE)
1146 			error = ENOMEM;
1147 		else
1148 			error = EAGAIN;
1149 	}
1150 #ifdef RACCT
1151 	if (racct_enable && error != KERN_SUCCESS) {
1152 		PROC_LOCK(td->td_proc);
1153 		racct_set(td->td_proc, RACCT_MEMLOCK,
1154 		    ptoa(pmap_wired_count(map->pmap)));
1155 		PROC_UNLOCK(td->td_proc);
1156 	}
1157 #endif
1158 
1159 	return (error);
1160 }
1161 
1162 #ifndef _SYS_SYSPROTO_H_
1163 struct munlockall_args {
1164 	register_t dummy;
1165 };
1166 #endif
1167 
1168 int
1169 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1170 {
1171 	vm_map_t map;
1172 	int error;
1173 
1174 	map = &td->td_proc->p_vmspace->vm_map;
1175 	error = priv_check(td, PRIV_VM_MUNLOCK);
1176 	if (error)
1177 		return (error);
1178 
1179 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1180 	vm_map_lock(map);
1181 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1182 	vm_map_unlock(map);
1183 
1184 	/* Forcibly unwire all pages. */
1185 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1186 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1187 #ifdef RACCT
1188 	if (racct_enable && error == KERN_SUCCESS) {
1189 		PROC_LOCK(td->td_proc);
1190 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1191 		PROC_UNLOCK(td->td_proc);
1192 	}
1193 #endif
1194 
1195 	return (error);
1196 }
1197 
1198 #ifndef _SYS_SYSPROTO_H_
1199 struct munlock_args {
1200 	const void *addr;
1201 	size_t len;
1202 };
1203 #endif
1204 int
1205 sys_munlock(struct thread *td, struct munlock_args *uap)
1206 {
1207 
1208 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1209 }
1210 
1211 int
1212 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1213 {
1214 	vm_offset_t addr, end, last, start;
1215 #ifdef RACCT
1216 	vm_map_t map;
1217 #endif
1218 	int error;
1219 
1220 	error = priv_check(td, PRIV_VM_MUNLOCK);
1221 	if (error)
1222 		return (error);
1223 	addr = addr0;
1224 	last = addr + size;
1225 	start = trunc_page(addr);
1226 	end = round_page(last);
1227 	if (last < addr || end < addr)
1228 		return (EINVAL);
1229 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1230 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1231 #ifdef RACCT
1232 	if (racct_enable && error == KERN_SUCCESS) {
1233 		PROC_LOCK(td->td_proc);
1234 		map = &td->td_proc->p_vmspace->vm_map;
1235 		racct_set(td->td_proc, RACCT_MEMLOCK,
1236 		    ptoa(pmap_wired_count(map->pmap)));
1237 		PROC_UNLOCK(td->td_proc);
1238 	}
1239 #endif
1240 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1241 }
1242 
1243 /*
1244  * vm_mmap_vnode()
1245  *
1246  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1247  * operations on vnodes.
1248  */
1249 int
1250 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1251     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1252     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1253     boolean_t *writecounted)
1254 {
1255 	struct vattr va;
1256 	vm_object_t obj;
1257 	vm_ooffset_t foff;
1258 	struct ucred *cred;
1259 	int error, flags;
1260 	bool writex;
1261 
1262 	cred = td->td_ucred;
1263 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1264 	    (*flagsp & MAP_SHARED) != 0;
1265 	if ((error = vget(vp, LK_SHARED, td)) != 0)
1266 		return (error);
1267 	AUDIT_ARG_VNODE1(vp);
1268 	foff = *foffp;
1269 	flags = *flagsp;
1270 	obj = vp->v_object;
1271 	if (vp->v_type == VREG) {
1272 		/*
1273 		 * Get the proper underlying object
1274 		 */
1275 		if (obj == NULL) {
1276 			error = EINVAL;
1277 			goto done;
1278 		}
1279 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1280 			vput(vp);
1281 			vp = (struct vnode *)obj->handle;
1282 			/*
1283 			 * Bypass filesystems obey the mpsafety of the
1284 			 * underlying fs.  Tmpfs never bypasses.
1285 			 */
1286 			error = vget(vp, LK_SHARED, td);
1287 			if (error != 0)
1288 				return (error);
1289 		}
1290 		if (writex) {
1291 			*writecounted = TRUE;
1292 			vnode_pager_update_writecount(obj, 0, objsize);
1293 		}
1294 	} else {
1295 		error = EINVAL;
1296 		goto done;
1297 	}
1298 	if ((error = VOP_GETATTR(vp, &va, cred)))
1299 		goto done;
1300 #ifdef MAC
1301 	/* This relies on VM_PROT_* matching PROT_*. */
1302 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1303 	if (error != 0)
1304 		goto done;
1305 #endif
1306 	if ((flags & MAP_SHARED) != 0) {
1307 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1308 			if (prot & VM_PROT_WRITE) {
1309 				error = EPERM;
1310 				goto done;
1311 			}
1312 			*maxprotp &= ~VM_PROT_WRITE;
1313 		}
1314 	}
1315 	/*
1316 	 * If it is a regular file without any references
1317 	 * we do not need to sync it.
1318 	 * Adjust object size to be the size of actual file.
1319 	 */
1320 	objsize = round_page(va.va_size);
1321 	if (va.va_nlink == 0)
1322 		flags |= MAP_NOSYNC;
1323 	if (obj->type == OBJT_VNODE) {
1324 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1325 		    cred);
1326 		if (obj == NULL) {
1327 			error = ENOMEM;
1328 			goto done;
1329 		}
1330 	} else {
1331 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1332 		    ("wrong object type"));
1333 		VM_OBJECT_WLOCK(obj);
1334 		vm_object_reference_locked(obj);
1335 #if VM_NRESERVLEVEL > 0
1336 		vm_object_color(obj, 0);
1337 #endif
1338 		VM_OBJECT_WUNLOCK(obj);
1339 	}
1340 	*objp = obj;
1341 	*flagsp = flags;
1342 
1343 	vfs_mark_atime(vp, cred);
1344 
1345 done:
1346 	if (error != 0 && *writecounted) {
1347 		*writecounted = FALSE;
1348 		vnode_pager_update_writecount(obj, objsize, 0);
1349 	}
1350 	vput(vp);
1351 	return (error);
1352 }
1353 
1354 /*
1355  * vm_mmap_cdev()
1356  *
1357  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1358  * operations on cdevs.
1359  */
1360 int
1361 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1362     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1363     vm_ooffset_t *foff, vm_object_t *objp)
1364 {
1365 	vm_object_t obj;
1366 	int error, flags;
1367 
1368 	flags = *flagsp;
1369 
1370 	if (dsw->d_flags & D_MMAP_ANON) {
1371 		*objp = NULL;
1372 		*foff = 0;
1373 		*maxprotp = VM_PROT_ALL;
1374 		*flagsp |= MAP_ANON;
1375 		return (0);
1376 	}
1377 	/*
1378 	 * cdevs do not provide private mappings of any kind.
1379 	 */
1380 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1381 	    (prot & VM_PROT_WRITE) != 0)
1382 		return (EACCES);
1383 	if (flags & (MAP_PRIVATE|MAP_COPY))
1384 		return (EINVAL);
1385 	/*
1386 	 * Force device mappings to be shared.
1387 	 */
1388 	flags |= MAP_SHARED;
1389 #ifdef MAC_XXX
1390 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1391 	if (error != 0)
1392 		return (error);
1393 #endif
1394 	/*
1395 	 * First, try d_mmap_single().  If that is not implemented
1396 	 * (returns ENODEV), fall back to using the device pager.
1397 	 * Note that d_mmap_single() must return a reference to the
1398 	 * object (it needs to bump the reference count of the object
1399 	 * it returns somehow).
1400 	 *
1401 	 * XXX assumes VM_PROT_* == PROT_*
1402 	 */
1403 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1404 	if (error != ENODEV)
1405 		return (error);
1406 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1407 	    td->td_ucred);
1408 	if (obj == NULL)
1409 		return (EINVAL);
1410 	*objp = obj;
1411 	*flagsp = flags;
1412 	return (0);
1413 }
1414 
1415 /*
1416  * vm_mmap()
1417  *
1418  * Internal version of mmap used by exec, sys5 shared memory, and
1419  * various device drivers.  Handle is either a vnode pointer, a
1420  * character device, or NULL for MAP_ANON.
1421  */
1422 int
1423 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1424 	vm_prot_t maxprot, int flags,
1425 	objtype_t handle_type, void *handle,
1426 	vm_ooffset_t foff)
1427 {
1428 	vm_object_t object;
1429 	struct thread *td = curthread;
1430 	int error;
1431 	boolean_t writecounted;
1432 
1433 	if (size == 0)
1434 		return (EINVAL);
1435 
1436 	size = round_page(size);
1437 	object = NULL;
1438 	writecounted = FALSE;
1439 
1440 	/*
1441 	 * Lookup/allocate object.
1442 	 */
1443 	switch (handle_type) {
1444 	case OBJT_DEVICE: {
1445 		struct cdevsw *dsw;
1446 		struct cdev *cdev;
1447 		int ref;
1448 
1449 		cdev = handle;
1450 		dsw = dev_refthread(cdev, &ref);
1451 		if (dsw == NULL)
1452 			return (ENXIO);
1453 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1454 		    dsw, &foff, &object);
1455 		dev_relthread(cdev, ref);
1456 		break;
1457 	}
1458 	case OBJT_VNODE:
1459 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1460 		    handle, &foff, &object, &writecounted);
1461 		break;
1462 	case OBJT_DEFAULT:
1463 		if (handle == NULL) {
1464 			error = 0;
1465 			break;
1466 		}
1467 		/* FALLTHROUGH */
1468 	default:
1469 		error = EINVAL;
1470 		break;
1471 	}
1472 	if (error)
1473 		return (error);
1474 
1475 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1476 	    foff, writecounted, td);
1477 	if (error != 0 && object != NULL) {
1478 		/*
1479 		 * If this mapping was accounted for in the vnode's
1480 		 * writecount, then undo that now.
1481 		 */
1482 		if (writecounted)
1483 			vnode_pager_release_writecount(object, 0, size);
1484 		vm_object_deallocate(object);
1485 	}
1486 	return (error);
1487 }
1488 
1489 /*
1490  * Internal version of mmap that maps a specific VM object into an
1491  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1492  */
1493 int
1494 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1495     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1496     boolean_t writecounted, struct thread *td)
1497 {
1498 	boolean_t curmap, fitit;
1499 	vm_offset_t max_addr;
1500 	int docow, error, findspace, rv;
1501 
1502 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1503 	if (curmap) {
1504 		RACCT_PROC_LOCK(td->td_proc);
1505 		if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1506 			RACCT_PROC_UNLOCK(td->td_proc);
1507 			return (ENOMEM);
1508 		}
1509 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1510 			RACCT_PROC_UNLOCK(td->td_proc);
1511 			return (ENOMEM);
1512 		}
1513 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1514 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1515 			    lim_cur(td, RLIMIT_MEMLOCK)) {
1516 				racct_set_force(td->td_proc, RACCT_VMEM,
1517 				    map->size);
1518 				RACCT_PROC_UNLOCK(td->td_proc);
1519 				return (ENOMEM);
1520 			}
1521 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1522 			    ptoa(pmap_wired_count(map->pmap)) + size);
1523 			if (error != 0) {
1524 				racct_set_force(td->td_proc, RACCT_VMEM,
1525 				    map->size);
1526 				RACCT_PROC_UNLOCK(td->td_proc);
1527 				return (error);
1528 			}
1529 		}
1530 		RACCT_PROC_UNLOCK(td->td_proc);
1531 	}
1532 
1533 	/*
1534 	 * We currently can only deal with page aligned file offsets.
1535 	 * The mmap() system call already enforces this by subtracting
1536 	 * the page offset from the file offset, but checking here
1537 	 * catches errors in device drivers (e.g. d_single_mmap()
1538 	 * callbacks) and other internal mapping requests (such as in
1539 	 * exec).
1540 	 */
1541 	if (foff & PAGE_MASK)
1542 		return (EINVAL);
1543 
1544 	if ((flags & MAP_FIXED) == 0) {
1545 		fitit = TRUE;
1546 		*addr = round_page(*addr);
1547 	} else {
1548 		if (*addr != trunc_page(*addr))
1549 			return (EINVAL);
1550 		fitit = FALSE;
1551 	}
1552 
1553 	if (flags & MAP_ANON) {
1554 		if (object != NULL || foff != 0)
1555 			return (EINVAL);
1556 		docow = 0;
1557 	} else if (flags & MAP_PREFAULT_READ)
1558 		docow = MAP_PREFAULT;
1559 	else
1560 		docow = MAP_PREFAULT_PARTIAL;
1561 
1562 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1563 		docow |= MAP_COPY_ON_WRITE;
1564 	if (flags & MAP_NOSYNC)
1565 		docow |= MAP_DISABLE_SYNCER;
1566 	if (flags & MAP_NOCORE)
1567 		docow |= MAP_DISABLE_COREDUMP;
1568 	/* Shared memory is also shared with children. */
1569 	if (flags & MAP_SHARED)
1570 		docow |= MAP_INHERIT_SHARE;
1571 	if (writecounted)
1572 		docow |= MAP_VN_WRITECOUNT;
1573 	if (flags & MAP_STACK) {
1574 		if (object != NULL)
1575 			return (EINVAL);
1576 		docow |= MAP_STACK_GROWS_DOWN;
1577 	}
1578 	if ((flags & MAP_EXCL) != 0)
1579 		docow |= MAP_CHECK_EXCL;
1580 	if ((flags & MAP_GUARD) != 0)
1581 		docow |= MAP_CREATE_GUARD;
1582 
1583 	if (fitit) {
1584 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1585 			findspace = VMFS_SUPER_SPACE;
1586 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1587 			findspace = VMFS_ALIGNED_SPACE(flags >>
1588 			    MAP_ALIGNMENT_SHIFT);
1589 		else
1590 			findspace = VMFS_OPTIMAL_SPACE;
1591 		max_addr = 0;
1592 #ifdef MAP_32BIT
1593 		if ((flags & MAP_32BIT) != 0)
1594 			max_addr = MAP_32BIT_MAX_ADDR;
1595 #endif
1596 		if (curmap) {
1597 			rv = vm_map_find_min(map, object, foff, addr, size,
1598 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1599 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1600 			    findspace, prot, maxprot, docow);
1601 		} else {
1602 			rv = vm_map_find(map, object, foff, addr, size,
1603 			    max_addr, findspace, prot, maxprot, docow);
1604 		}
1605 	} else {
1606 		rv = vm_map_fixed(map, object, foff, *addr, size,
1607 		    prot, maxprot, docow);
1608 	}
1609 
1610 	if (rv == KERN_SUCCESS) {
1611 		/*
1612 		 * If the process has requested that all future mappings
1613 		 * be wired, then heed this.
1614 		 */
1615 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1616 			vm_map_lock(map);
1617 			if ((map->flags & MAP_WIREFUTURE) != 0)
1618 				(void)vm_map_wire_locked(map, *addr,
1619 				    *addr + size, VM_MAP_WIRE_USER |
1620 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1621 				    VM_MAP_WIRE_NOHOLES));
1622 			vm_map_unlock(map);
1623 		}
1624 	}
1625 	return (vm_mmap_to_errno(rv));
1626 }
1627 
1628 /*
1629  * Translate a Mach VM return code to zero on success or the appropriate errno
1630  * on failure.
1631  */
1632 int
1633 vm_mmap_to_errno(int rv)
1634 {
1635 
1636 	switch (rv) {
1637 	case KERN_SUCCESS:
1638 		return (0);
1639 	case KERN_INVALID_ADDRESS:
1640 	case KERN_NO_SPACE:
1641 		return (ENOMEM);
1642 	case KERN_PROTECTION_FAILURE:
1643 		return (EACCES);
1644 	default:
1645 		return (EINVAL);
1646 	}
1647 }
1648