xref: /freebsd/sys/vm/vm_mmap.c (revision a7623790fb345e6dc986dfd31df0ace115e6f2e4)
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 protections 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 _Static_assert(MAXPAGESIZES <= 4, "MINCORE_SUPER too narrow");
116 
117 #ifndef _SYS_SYSPROTO_H_
118 struct sbrk_args {
119 	int incr;
120 };
121 #endif
122 
123 int
124 sys_sbrk(struct thread *td, struct sbrk_args *uap)
125 {
126 	/* Not yet implemented */
127 	return (EOPNOTSUPP);
128 }
129 
130 #ifndef _SYS_SYSPROTO_H_
131 struct sstk_args {
132 	int incr;
133 };
134 #endif
135 
136 int
137 sys_sstk(struct thread *td, struct sstk_args *uap)
138 {
139 	/* Not yet implemented */
140 	return (EOPNOTSUPP);
141 }
142 
143 #if defined(COMPAT_43)
144 int
145 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
146 {
147 
148 	td->td_retval[0] = PAGE_SIZE;
149 	return (0);
150 }
151 #endif				/* COMPAT_43 */
152 
153 /*
154  * Memory Map (mmap) system call.  Note that the file offset
155  * and address are allowed to be NOT page aligned, though if
156  * the MAP_FIXED flag it set, both must have the same remainder
157  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
158  * page-aligned, the actual mapping starts at trunc_page(addr)
159  * and the return value is adjusted up by the page offset.
160  *
161  * Generally speaking, only character devices which are themselves
162  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
163  * there would be no cache coherency between a descriptor and a VM mapping
164  * both to the same character device.
165  */
166 #ifndef _SYS_SYSPROTO_H_
167 struct mmap_args {
168 	void *addr;
169 	size_t len;
170 	int prot;
171 	int flags;
172 	int fd;
173 	long pad;
174 	off_t pos;
175 };
176 #endif
177 
178 int
179 sys_mmap(struct thread *td, struct mmap_args *uap)
180 {
181 
182 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
183 	    uap->flags, uap->fd, uap->pos));
184 }
185 
186 int
187 kern_mmap_maxprot(struct proc *p, int prot)
188 {
189 
190 	if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
191 	    (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
192 		return (_PROT_ALL);
193 	if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
194 	    prot != PROT_NONE)
195 		 return (prot);
196 	return (_PROT_ALL);
197 }
198 
199 int
200 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags,
201     int fd, off_t pos)
202 {
203 	struct mmap_req mr = {
204 		.mr_hint = addr0,
205 		.mr_len = len,
206 		.mr_prot = prot,
207 		.mr_flags = flags,
208 		.mr_fd = fd,
209 		.mr_pos = pos
210 	};
211 
212 	return (kern_mmap_req(td, &mr));
213 }
214 
215 int
216 kern_mmap_req(struct thread *td, const struct mmap_req *mrp)
217 {
218 	struct vmspace *vms;
219 	struct file *fp;
220 	struct proc *p;
221 	off_t pos;
222 	vm_offset_t addr, orig_addr;
223 	vm_size_t len, pageoff, size;
224 	vm_prot_t cap_maxprot;
225 	int align, error, fd, flags, max_prot, prot;
226 	cap_rights_t rights;
227 	mmap_check_fp_fn check_fp_fn;
228 
229 	orig_addr = addr = mrp->mr_hint;
230 	len = mrp->mr_len;
231 	prot = mrp->mr_prot;
232 	flags = mrp->mr_flags;
233 	fd = mrp->mr_fd;
234 	pos = mrp->mr_pos;
235 	check_fp_fn = mrp->mr_check_fp_fn;
236 
237 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
238 		return (EINVAL);
239 	max_prot = PROT_MAX_EXTRACT(prot);
240 	prot = PROT_EXTRACT(prot);
241 	if (max_prot != 0 && (max_prot & prot) != prot)
242 		return (ENOTSUP);
243 
244 	p = td->td_proc;
245 
246 	/*
247 	 * Always honor PROT_MAX if set.  If not, default to all
248 	 * permissions unless we're implying maximum permissions.
249 	 */
250 	if (max_prot == 0)
251 		max_prot = kern_mmap_maxprot(p, prot);
252 
253 	vms = p->p_vmspace;
254 	fp = NULL;
255 	AUDIT_ARG_FD(fd);
256 
257 	/*
258 	 * Ignore old flags that used to be defined but did not do anything.
259 	 */
260 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
261 
262 	/*
263 	 * Enforce the constraints.
264 	 * Mapping of length 0 is only allowed for old binaries.
265 	 * Anonymous mapping shall specify -1 as filedescriptor and
266 	 * zero position for new code. Be nice to ancient a.out
267 	 * binaries and correct pos for anonymous mapping, since old
268 	 * ld.so sometimes issues anonymous map requests with non-zero
269 	 * pos.
270 	 */
271 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
272 		if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
273 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
274 			return (EINVAL);
275 	} else {
276 		if ((flags & MAP_ANON) != 0)
277 			pos = 0;
278 	}
279 
280 	if (flags & MAP_STACK) {
281 		if ((fd != -1) ||
282 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
283 			return (EINVAL);
284 		flags |= MAP_ANON;
285 		pos = 0;
286 	}
287 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
288 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
289 	    MAP_PREFAULT_READ | MAP_GUARD |
290 #ifdef MAP_32BIT
291 	    MAP_32BIT |
292 #endif
293 	    MAP_ALIGNMENT_MASK)) != 0)
294 		return (EINVAL);
295 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
296 		return (EINVAL);
297 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
298 		return (EINVAL);
299 	if (prot != PROT_NONE &&
300 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
301 		return (EINVAL);
302 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
303 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
304 #ifdef MAP_32BIT
305 	    MAP_32BIT |
306 #endif
307 	    MAP_ALIGNMENT_MASK)) != 0))
308 		return (EINVAL);
309 
310 	/*
311 	 * Align the file position to a page boundary,
312 	 * and save its page offset component.
313 	 */
314 	pageoff = (pos & PAGE_MASK);
315 	pos -= pageoff;
316 
317 	/* Compute size from len by rounding (on both ends). */
318 	size = len + pageoff;			/* low end... */
319 	size = round_page(size);		/* hi end */
320 	/* Check for rounding up to zero. */
321 	if (len > size)
322 		return (ENOMEM);
323 
324 	/* Ensure alignment is at least a page and fits in a pointer. */
325 	align = flags & MAP_ALIGNMENT_MASK;
326 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
327 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
328 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
329 		return (EINVAL);
330 
331 	/*
332 	 * Check for illegal addresses.  Watch out for address wrap... Note
333 	 * that VM_*_ADDRESS are not constants due to casts (argh).
334 	 */
335 	if (flags & MAP_FIXED) {
336 		/*
337 		 * The specified address must have the same remainder
338 		 * as the file offset taken modulo PAGE_SIZE, so it
339 		 * should be aligned after adjustment by pageoff.
340 		 */
341 		addr -= pageoff;
342 		if (addr & PAGE_MASK)
343 			return (EINVAL);
344 
345 		/* Address range must be all in user VM space. */
346 		if (!vm_map_range_valid(&vms->vm_map, addr, addr + size))
347 			return (EINVAL);
348 #ifdef MAP_32BIT
349 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
350 			return (EINVAL);
351 	} else if (flags & MAP_32BIT) {
352 		/*
353 		 * For MAP_32BIT, override the hint if it is too high and
354 		 * do not bother moving the mapping past the heap (since
355 		 * the heap is usually above 2GB).
356 		 */
357 		if (addr + size > MAP_32BIT_MAX_ADDR)
358 			addr = 0;
359 #endif
360 	} else {
361 		/*
362 		 * XXX for non-fixed mappings where no hint is provided or
363 		 * the hint would fall in the potential heap space,
364 		 * place it after the end of the largest possible heap.
365 		 *
366 		 * There should really be a pmap call to determine a reasonable
367 		 * location.
368 		 */
369 		if (addr == 0 ||
370 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
371 		    addr < round_page((vm_offset_t)vms->vm_daddr +
372 		    lim_max(td, RLIMIT_DATA))))
373 			addr = round_page((vm_offset_t)vms->vm_daddr +
374 			    lim_max(td, RLIMIT_DATA));
375 	}
376 	if (len == 0) {
377 		/*
378 		 * Return success without mapping anything for old
379 		 * binaries that request a page-aligned mapping of
380 		 * length 0.  For modern binaries, this function
381 		 * returns an error earlier.
382 		 */
383 		error = 0;
384 	} else if ((flags & MAP_GUARD) != 0) {
385 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
386 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
387 	} else if ((flags & MAP_ANON) != 0) {
388 		/*
389 		 * Mapping blank space is trivial.
390 		 *
391 		 * This relies on VM_PROT_* matching PROT_*.
392 		 */
393 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
394 		    max_prot, flags, NULL, pos, FALSE, td);
395 	} else {
396 		/*
397 		 * Mapping file, get fp for validation and don't let the
398 		 * descriptor disappear on us if we block. Check capability
399 		 * rights, but also return the maximum rights to be combined
400 		 * with maxprot later.
401 		 */
402 		cap_rights_init_one(&rights, CAP_MMAP);
403 		if (prot & PROT_READ)
404 			cap_rights_set_one(&rights, CAP_MMAP_R);
405 		if ((flags & MAP_SHARED) != 0) {
406 			if (prot & PROT_WRITE)
407 				cap_rights_set_one(&rights, CAP_MMAP_W);
408 		}
409 		if (prot & PROT_EXEC)
410 			cap_rights_set_one(&rights, CAP_MMAP_X);
411 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
412 		if (error != 0)
413 			goto done;
414 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
415 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
416 			error = EINVAL;
417 			goto done;
418 		}
419 		if (check_fp_fn != NULL) {
420 			error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
421 			    flags);
422 			if (error != 0)
423 				goto done;
424 		}
425 		if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data))
426 			addr = orig_addr;
427 		/* This relies on VM_PROT_* matching PROT_*. */
428 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
429 		    max_prot & cap_maxprot, flags, pos, td);
430 	}
431 
432 	if (error == 0)
433 		td->td_retval[0] = (register_t) (addr + pageoff);
434 done:
435 	if (fp)
436 		fdrop(fp, td);
437 
438 	return (error);
439 }
440 
441 #if defined(COMPAT_FREEBSD6)
442 int
443 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
444 {
445 
446 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
447 	    uap->flags, uap->fd, uap->pos));
448 }
449 #endif
450 
451 #ifdef COMPAT_43
452 #ifndef _SYS_SYSPROTO_H_
453 struct ommap_args {
454 	caddr_t addr;
455 	int len;
456 	int prot;
457 	int flags;
458 	int fd;
459 	long pos;
460 };
461 #endif
462 int
463 ommap(struct thread *td, struct ommap_args *uap)
464 {
465 	static const char cvtbsdprot[8] = {
466 		0,
467 		PROT_EXEC,
468 		PROT_WRITE,
469 		PROT_EXEC | PROT_WRITE,
470 		PROT_READ,
471 		PROT_EXEC | PROT_READ,
472 		PROT_WRITE | PROT_READ,
473 		PROT_EXEC | PROT_WRITE | PROT_READ,
474 	};
475 	int flags, prot;
476 
477 #define	OMAP_ANON	0x0002
478 #define	OMAP_COPY	0x0020
479 #define	OMAP_SHARED	0x0010
480 #define	OMAP_FIXED	0x0100
481 
482 	prot = cvtbsdprot[uap->prot & 0x7];
483 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
484 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
485 	    prot != 0)
486 		prot |= PROT_EXEC;
487 #endif
488 	flags = 0;
489 	if (uap->flags & OMAP_ANON)
490 		flags |= MAP_ANON;
491 	if (uap->flags & OMAP_COPY)
492 		flags |= MAP_COPY;
493 	if (uap->flags & OMAP_SHARED)
494 		flags |= MAP_SHARED;
495 	else
496 		flags |= MAP_PRIVATE;
497 	if (uap->flags & OMAP_FIXED)
498 		flags |= MAP_FIXED;
499 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
500 	    uap->fd, uap->pos));
501 }
502 #endif				/* COMPAT_43 */
503 
504 #ifndef _SYS_SYSPROTO_H_
505 struct msync_args {
506 	void *addr;
507 	size_t len;
508 	int flags;
509 };
510 #endif
511 int
512 sys_msync(struct thread *td, struct msync_args *uap)
513 {
514 
515 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
516 }
517 
518 int
519 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
520 {
521 	vm_offset_t addr;
522 	vm_size_t pageoff;
523 	vm_map_t map;
524 	int rv;
525 
526 	addr = addr0;
527 	pageoff = (addr & PAGE_MASK);
528 	addr -= pageoff;
529 	size += pageoff;
530 	size = (vm_size_t) round_page(size);
531 	if (addr + size < addr)
532 		return (EINVAL);
533 
534 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
535 		return (EINVAL);
536 
537 	map = &td->td_proc->p_vmspace->vm_map;
538 
539 	/*
540 	 * Clean the pages and interpret the return value.
541 	 */
542 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
543 	    (flags & MS_INVALIDATE) != 0);
544 	switch (rv) {
545 	case KERN_SUCCESS:
546 		return (0);
547 	case KERN_INVALID_ADDRESS:
548 		return (ENOMEM);
549 	case KERN_INVALID_ARGUMENT:
550 		return (EBUSY);
551 	case KERN_FAILURE:
552 		return (EIO);
553 	default:
554 		return (EINVAL);
555 	}
556 }
557 
558 #ifndef _SYS_SYSPROTO_H_
559 struct munmap_args {
560 	void *addr;
561 	size_t len;
562 };
563 #endif
564 int
565 sys_munmap(struct thread *td, struct munmap_args *uap)
566 {
567 
568 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
569 }
570 
571 int
572 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
573 {
574 #ifdef HWPMC_HOOKS
575 	struct pmckern_map_out pkm;
576 	vm_map_entry_t entry;
577 	bool pmc_handled;
578 #endif
579 	vm_offset_t addr, end;
580 	vm_size_t pageoff;
581 	vm_map_t map;
582 	int rv;
583 
584 	if (size == 0)
585 		return (EINVAL);
586 
587 	addr = addr0;
588 	pageoff = (addr & PAGE_MASK);
589 	addr -= pageoff;
590 	size += pageoff;
591 	size = (vm_size_t) round_page(size);
592 	end = addr + size;
593 	map = &td->td_proc->p_vmspace->vm_map;
594 	if (!vm_map_range_valid(map, addr, end))
595 		return (EINVAL);
596 
597 	vm_map_lock(map);
598 #ifdef HWPMC_HOOKS
599 	pmc_handled = false;
600 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
601 		pmc_handled = true;
602 		/*
603 		 * Inform hwpmc if the address range being unmapped contains
604 		 * an executable region.
605 		 */
606 		pkm.pm_address = (uintptr_t) NULL;
607 		if (vm_map_lookup_entry(map, addr, &entry)) {
608 			for (; entry->start < end;
609 			    entry = vm_map_entry_succ(entry)) {
610 				if (vm_map_check_protection(map, entry->start,
611 					entry->end, VM_PROT_EXECUTE) == TRUE) {
612 					pkm.pm_address = (uintptr_t) addr;
613 					pkm.pm_size = (size_t) size;
614 					break;
615 				}
616 			}
617 		}
618 	}
619 #endif
620 	rv = vm_map_delete(map, addr, end);
621 
622 #ifdef HWPMC_HOOKS
623 	if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) {
624 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
625 		vm_map_lock_downgrade(map);
626 		if (pkm.pm_address != (uintptr_t) NULL)
627 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
628 		vm_map_unlock_read(map);
629 	} else
630 #endif
631 		vm_map_unlock(map);
632 
633 	return (vm_mmap_to_errno(rv));
634 }
635 
636 #ifndef _SYS_SYSPROTO_H_
637 struct mprotect_args {
638 	const void *addr;
639 	size_t len;
640 	int prot;
641 };
642 #endif
643 int
644 sys_mprotect(struct thread *td, struct mprotect_args *uap)
645 {
646 
647 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
648 }
649 
650 int
651 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
652 {
653 	vm_offset_t addr;
654 	vm_size_t pageoff;
655 	int vm_error, max_prot;
656 
657 	addr = addr0;
658 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
659 		return (EINVAL);
660 	max_prot = PROT_MAX_EXTRACT(prot);
661 	prot = PROT_EXTRACT(prot);
662 	pageoff = (addr & PAGE_MASK);
663 	addr -= pageoff;
664 	size += pageoff;
665 	size = (vm_size_t) round_page(size);
666 #ifdef COMPAT_FREEBSD32
667 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
668 		if (((addr + size) & 0xffffffff) < addr)
669 			return (EINVAL);
670 	} else
671 #endif
672 	if (addr + size < addr)
673 		return (EINVAL);
674 
675 	vm_error = KERN_SUCCESS;
676 	if (max_prot != 0) {
677 		if ((max_prot & prot) != prot)
678 			return (ENOTSUP);
679 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
680 		    addr, addr + size, max_prot, TRUE);
681 	}
682 	if (vm_error == KERN_SUCCESS)
683 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
684 		    addr, addr + size, prot, FALSE);
685 
686 	switch (vm_error) {
687 	case KERN_SUCCESS:
688 		return (0);
689 	case KERN_PROTECTION_FAILURE:
690 		return (EACCES);
691 	case KERN_RESOURCE_SHORTAGE:
692 		return (ENOMEM);
693 	}
694 	return (EINVAL);
695 }
696 
697 #ifndef _SYS_SYSPROTO_H_
698 struct minherit_args {
699 	void *addr;
700 	size_t len;
701 	int inherit;
702 };
703 #endif
704 int
705 sys_minherit(struct thread *td, struct minherit_args *uap)
706 {
707 
708 	return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
709 	    uap->inherit));
710 }
711 
712 int
713 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
714 {
715 	vm_offset_t addr;
716 	vm_size_t size, pageoff;
717 	vm_inherit_t inherit;
718 
719 	addr = (vm_offset_t)addr0;
720 	size = len;
721 	inherit = inherit0;
722 
723 	pageoff = (addr & PAGE_MASK);
724 	addr -= pageoff;
725 	size += pageoff;
726 	size = (vm_size_t) round_page(size);
727 	if (addr + size < addr)
728 		return (EINVAL);
729 
730 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
731 	    addr + size, inherit)) {
732 	case KERN_SUCCESS:
733 		return (0);
734 	case KERN_PROTECTION_FAILURE:
735 		return (EACCES);
736 	}
737 	return (EINVAL);
738 }
739 
740 #ifndef _SYS_SYSPROTO_H_
741 struct madvise_args {
742 	void *addr;
743 	size_t len;
744 	int behav;
745 };
746 #endif
747 
748 int
749 sys_madvise(struct thread *td, struct madvise_args *uap)
750 {
751 
752 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
753 }
754 
755 int
756 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
757 {
758 	vm_map_t map;
759 	vm_offset_t addr, end, start;
760 	int flags;
761 
762 	/*
763 	 * Check for our special case, advising the swap pager we are
764 	 * "immortal."
765 	 */
766 	if (behav == MADV_PROTECT) {
767 		flags = PPROT_SET;
768 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
769 		    PROC_SPROTECT, &flags));
770 	}
771 
772 	/*
773 	 * Check for illegal addresses.  Watch out for address wrap... Note
774 	 * that VM_*_ADDRESS are not constants due to casts (argh).
775 	 */
776 	map = &td->td_proc->p_vmspace->vm_map;
777 	addr = addr0;
778 	if (!vm_map_range_valid(map, addr, addr + len))
779 		return (EINVAL);
780 
781 	/*
782 	 * Since this routine is only advisory, we default to conservative
783 	 * behavior.
784 	 */
785 	start = trunc_page(addr);
786 	end = round_page(addr + len);
787 
788 	/*
789 	 * vm_map_madvise() checks for illegal values of behav.
790 	 */
791 	return (vm_map_madvise(map, start, end, behav));
792 }
793 
794 #ifndef _SYS_SYSPROTO_H_
795 struct mincore_args {
796 	const void *addr;
797 	size_t len;
798 	char *vec;
799 };
800 #endif
801 
802 int
803 sys_mincore(struct thread *td, struct mincore_args *uap)
804 {
805 
806 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
807 }
808 
809 int
810 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
811 {
812 	pmap_t pmap;
813 	vm_map_t map;
814 	vm_map_entry_t current, entry;
815 	vm_object_t object;
816 	vm_offset_t addr, cend, end, first_addr;
817 	vm_paddr_t pa;
818 	vm_page_t m;
819 	vm_pindex_t pindex;
820 	int error, lastvecindex, mincoreinfo, vecindex;
821 	unsigned int timestamp;
822 
823 	/*
824 	 * Make sure that the addresses presented are valid for user
825 	 * mode.
826 	 */
827 	first_addr = addr = trunc_page(addr0);
828 	end = round_page(addr0 + len);
829 	map = &td->td_proc->p_vmspace->vm_map;
830 	if (end > vm_map_max(map) || end < addr)
831 		return (ENOMEM);
832 
833 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
834 
835 	vm_map_lock_read(map);
836 RestartScan:
837 	timestamp = map->timestamp;
838 
839 	if (!vm_map_lookup_entry(map, addr, &entry)) {
840 		vm_map_unlock_read(map);
841 		return (ENOMEM);
842 	}
843 
844 	/*
845 	 * Do this on a map entry basis so that if the pages are not
846 	 * in the current processes address space, we can easily look
847 	 * up the pages elsewhere.
848 	 */
849 	lastvecindex = -1;
850 	while (entry->start < end) {
851 		/*
852 		 * check for contiguity
853 		 */
854 		current = entry;
855 		entry = vm_map_entry_succ(current);
856 		if (current->end < end &&
857 		    entry->start > current->end) {
858 			vm_map_unlock_read(map);
859 			return (ENOMEM);
860 		}
861 
862 		/*
863 		 * ignore submaps (for now) or null objects
864 		 */
865 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
866 		    current->object.vm_object == NULL)
867 			continue;
868 
869 		/*
870 		 * limit this scan to the current map entry and the
871 		 * limits for the mincore call
872 		 */
873 		if (addr < current->start)
874 			addr = current->start;
875 		cend = current->end;
876 		if (cend > end)
877 			cend = end;
878 
879 		for (; addr < cend; addr += PAGE_SIZE) {
880 			/*
881 			 * Check pmap first, it is likely faster, also
882 			 * it can provide info as to whether we are the
883 			 * one referencing or modifying the page.
884 			 */
885 			m = NULL;
886 			object = NULL;
887 retry:
888 			pa = 0;
889 			mincoreinfo = pmap_mincore(pmap, addr, &pa);
890 			if (mincore_mapped) {
891 				/*
892 				 * We only care about this pmap's
893 				 * mapping of the page, if any.
894 				 */
895 				;
896 			} else if (pa != 0) {
897 				/*
898 				 * The page is mapped by this process but not
899 				 * both accessed and modified.  It is also
900 				 * managed.  Acquire the object lock so that
901 				 * other mappings might be examined.  The page's
902 				 * identity may change at any point before its
903 				 * object lock is acquired, so re-validate if
904 				 * necessary.
905 				 */
906 				m = PHYS_TO_VM_PAGE(pa);
907 				while (object == NULL || m->object != object) {
908 					if (object != NULL)
909 						VM_OBJECT_WUNLOCK(object);
910 					object = atomic_load_ptr(&m->object);
911 					if (object == NULL)
912 						goto retry;
913 					VM_OBJECT_WLOCK(object);
914 				}
915 				if (pa != pmap_extract(pmap, addr))
916 					goto retry;
917 				KASSERT(vm_page_all_valid(m),
918 				    ("mincore: page %p is mapped but invalid",
919 				    m));
920 			} else if (mincoreinfo == 0) {
921 				/*
922 				 * The page is not mapped by this process.  If
923 				 * the object implements managed pages, then
924 				 * determine if the page is resident so that
925 				 * the mappings might be examined.
926 				 */
927 				if (current->object.vm_object != object) {
928 					if (object != NULL)
929 						VM_OBJECT_WUNLOCK(object);
930 					object = current->object.vm_object;
931 					VM_OBJECT_WLOCK(object);
932 				}
933 				if (object->type == OBJT_DEFAULT ||
934 				    object->type == OBJT_SWAP ||
935 				    object->type == OBJT_VNODE) {
936 					pindex = OFF_TO_IDX(current->offset +
937 					    (addr - current->start));
938 					m = vm_page_lookup(object, pindex);
939 					if (m != NULL && vm_page_none_valid(m))
940 						m = NULL;
941 					if (m != NULL)
942 						mincoreinfo = MINCORE_INCORE;
943 				}
944 			}
945 			if (m != NULL) {
946 				VM_OBJECT_ASSERT_WLOCKED(m->object);
947 
948 				/* Examine other mappings of the page. */
949 				if (m->dirty == 0 && pmap_is_modified(m))
950 					vm_page_dirty(m);
951 				if (m->dirty != 0)
952 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
953 
954 				/*
955 				 * The first test for PGA_REFERENCED is an
956 				 * optimization.  The second test is
957 				 * required because a concurrent pmap
958 				 * operation could clear the last reference
959 				 * and set PGA_REFERENCED before the call to
960 				 * pmap_is_referenced().
961 				 */
962 				if ((m->a.flags & PGA_REFERENCED) != 0 ||
963 				    pmap_is_referenced(m) ||
964 				    (m->a.flags & PGA_REFERENCED) != 0)
965 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
966 			}
967 			if (object != NULL)
968 				VM_OBJECT_WUNLOCK(object);
969 
970 			/*
971 			 * subyte may page fault.  In case it needs to modify
972 			 * the map, we release the lock.
973 			 */
974 			vm_map_unlock_read(map);
975 
976 			/*
977 			 * calculate index into user supplied byte vector
978 			 */
979 			vecindex = atop(addr - first_addr);
980 
981 			/*
982 			 * If we have skipped map entries, we need to make sure that
983 			 * the byte vector is zeroed for those skipped entries.
984 			 */
985 			while ((lastvecindex + 1) < vecindex) {
986 				++lastvecindex;
987 				error = subyte(vec + lastvecindex, 0);
988 				if (error) {
989 					error = EFAULT;
990 					goto done2;
991 				}
992 			}
993 
994 			/*
995 			 * Pass the page information to the user
996 			 */
997 			error = subyte(vec + vecindex, mincoreinfo);
998 			if (error) {
999 				error = EFAULT;
1000 				goto done2;
1001 			}
1002 
1003 			/*
1004 			 * If the map has changed, due to the subyte, the previous
1005 			 * output may be invalid.
1006 			 */
1007 			vm_map_lock_read(map);
1008 			if (timestamp != map->timestamp)
1009 				goto RestartScan;
1010 
1011 			lastvecindex = vecindex;
1012 		}
1013 	}
1014 
1015 	/*
1016 	 * subyte may page fault.  In case it needs to modify
1017 	 * the map, we release the lock.
1018 	 */
1019 	vm_map_unlock_read(map);
1020 
1021 	/*
1022 	 * Zero the last entries in the byte vector.
1023 	 */
1024 	vecindex = atop(end - first_addr);
1025 	while ((lastvecindex + 1) < vecindex) {
1026 		++lastvecindex;
1027 		error = subyte(vec + lastvecindex, 0);
1028 		if (error) {
1029 			error = EFAULT;
1030 			goto done2;
1031 		}
1032 	}
1033 
1034 	/*
1035 	 * If the map has changed, due to the subyte, the previous
1036 	 * output may be invalid.
1037 	 */
1038 	vm_map_lock_read(map);
1039 	if (timestamp != map->timestamp)
1040 		goto RestartScan;
1041 	vm_map_unlock_read(map);
1042 done2:
1043 	return (error);
1044 }
1045 
1046 #ifndef _SYS_SYSPROTO_H_
1047 struct mlock_args {
1048 	const void *addr;
1049 	size_t len;
1050 };
1051 #endif
1052 int
1053 sys_mlock(struct thread *td, struct mlock_args *uap)
1054 {
1055 
1056 	return (kern_mlock(td->td_proc, td->td_ucred,
1057 	    __DECONST(uintptr_t, uap->addr), uap->len));
1058 }
1059 
1060 int
1061 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1062 {
1063 	vm_offset_t addr, end, last, start;
1064 	vm_size_t npages, size;
1065 	vm_map_t map;
1066 	unsigned long nsize;
1067 	int error;
1068 
1069 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1070 	if (error)
1071 		return (error);
1072 	addr = addr0;
1073 	size = len;
1074 	last = addr + size;
1075 	start = trunc_page(addr);
1076 	end = round_page(last);
1077 	if (last < addr || end < addr)
1078 		return (EINVAL);
1079 	npages = atop(end - start);
1080 	if (npages > vm_page_max_user_wired)
1081 		return (ENOMEM);
1082 	map = &proc->p_vmspace->vm_map;
1083 	PROC_LOCK(proc);
1084 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1085 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1086 		PROC_UNLOCK(proc);
1087 		return (ENOMEM);
1088 	}
1089 	PROC_UNLOCK(proc);
1090 #ifdef RACCT
1091 	if (racct_enable) {
1092 		PROC_LOCK(proc);
1093 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1094 		PROC_UNLOCK(proc);
1095 		if (error != 0)
1096 			return (ENOMEM);
1097 	}
1098 #endif
1099 	error = vm_map_wire(map, start, end,
1100 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1101 #ifdef RACCT
1102 	if (racct_enable && error != KERN_SUCCESS) {
1103 		PROC_LOCK(proc);
1104 		racct_set(proc, RACCT_MEMLOCK,
1105 		    ptoa(pmap_wired_count(map->pmap)));
1106 		PROC_UNLOCK(proc);
1107 	}
1108 #endif
1109 	switch (error) {
1110 	case KERN_SUCCESS:
1111 		return (0);
1112 	case KERN_INVALID_ARGUMENT:
1113 		return (EINVAL);
1114 	default:
1115 		return (ENOMEM);
1116 	}
1117 }
1118 
1119 #ifndef _SYS_SYSPROTO_H_
1120 struct mlockall_args {
1121 	int	how;
1122 };
1123 #endif
1124 
1125 int
1126 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1127 {
1128 	vm_map_t map;
1129 	int error;
1130 
1131 	map = &td->td_proc->p_vmspace->vm_map;
1132 	error = priv_check(td, PRIV_VM_MLOCK);
1133 	if (error)
1134 		return (error);
1135 
1136 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1137 		return (EINVAL);
1138 
1139 	/*
1140 	 * If wiring all pages in the process would cause it to exceed
1141 	 * a hard resource limit, return ENOMEM.
1142 	 */
1143 	if (!old_mlock && uap->how & MCL_CURRENT) {
1144 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1145 			return (ENOMEM);
1146 	}
1147 #ifdef RACCT
1148 	if (racct_enable) {
1149 		PROC_LOCK(td->td_proc);
1150 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1151 		PROC_UNLOCK(td->td_proc);
1152 		if (error != 0)
1153 			return (ENOMEM);
1154 	}
1155 #endif
1156 
1157 	if (uap->how & MCL_FUTURE) {
1158 		vm_map_lock(map);
1159 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1160 		vm_map_unlock(map);
1161 		error = 0;
1162 	}
1163 
1164 	if (uap->how & MCL_CURRENT) {
1165 		/*
1166 		 * P1003.1-2001 mandates that all currently mapped pages
1167 		 * will be memory resident and locked (wired) upon return
1168 		 * from mlockall(). vm_map_wire() will wire pages, by
1169 		 * calling vm_fault_wire() for each page in the region.
1170 		 */
1171 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1172 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1173 		if (error == KERN_SUCCESS)
1174 			error = 0;
1175 		else if (error == KERN_RESOURCE_SHORTAGE)
1176 			error = ENOMEM;
1177 		else
1178 			error = EAGAIN;
1179 	}
1180 #ifdef RACCT
1181 	if (racct_enable && error != KERN_SUCCESS) {
1182 		PROC_LOCK(td->td_proc);
1183 		racct_set(td->td_proc, RACCT_MEMLOCK,
1184 		    ptoa(pmap_wired_count(map->pmap)));
1185 		PROC_UNLOCK(td->td_proc);
1186 	}
1187 #endif
1188 
1189 	return (error);
1190 }
1191 
1192 #ifndef _SYS_SYSPROTO_H_
1193 struct munlockall_args {
1194 	register_t dummy;
1195 };
1196 #endif
1197 
1198 int
1199 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1200 {
1201 	vm_map_t map;
1202 	int error;
1203 
1204 	map = &td->td_proc->p_vmspace->vm_map;
1205 	error = priv_check(td, PRIV_VM_MUNLOCK);
1206 	if (error)
1207 		return (error);
1208 
1209 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1210 	vm_map_lock(map);
1211 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1212 	vm_map_unlock(map);
1213 
1214 	/* Forcibly unwire all pages. */
1215 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1216 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1217 #ifdef RACCT
1218 	if (racct_enable && error == KERN_SUCCESS) {
1219 		PROC_LOCK(td->td_proc);
1220 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1221 		PROC_UNLOCK(td->td_proc);
1222 	}
1223 #endif
1224 
1225 	return (error);
1226 }
1227 
1228 #ifndef _SYS_SYSPROTO_H_
1229 struct munlock_args {
1230 	const void *addr;
1231 	size_t len;
1232 };
1233 #endif
1234 int
1235 sys_munlock(struct thread *td, struct munlock_args *uap)
1236 {
1237 
1238 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1239 }
1240 
1241 int
1242 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1243 {
1244 	vm_offset_t addr, end, last, start;
1245 #ifdef RACCT
1246 	vm_map_t map;
1247 #endif
1248 	int error;
1249 
1250 	error = priv_check(td, PRIV_VM_MUNLOCK);
1251 	if (error)
1252 		return (error);
1253 	addr = addr0;
1254 	last = addr + size;
1255 	start = trunc_page(addr);
1256 	end = round_page(last);
1257 	if (last < addr || end < addr)
1258 		return (EINVAL);
1259 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1260 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1261 #ifdef RACCT
1262 	if (racct_enable && error == KERN_SUCCESS) {
1263 		PROC_LOCK(td->td_proc);
1264 		map = &td->td_proc->p_vmspace->vm_map;
1265 		racct_set(td->td_proc, RACCT_MEMLOCK,
1266 		    ptoa(pmap_wired_count(map->pmap)));
1267 		PROC_UNLOCK(td->td_proc);
1268 	}
1269 #endif
1270 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1271 }
1272 
1273 /*
1274  * vm_mmap_vnode()
1275  *
1276  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1277  * operations on vnodes.
1278  */
1279 int
1280 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1281     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1282     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1283     boolean_t *writecounted)
1284 {
1285 	struct vattr va;
1286 	vm_object_t obj;
1287 	vm_ooffset_t foff;
1288 	struct ucred *cred;
1289 	int error, flags;
1290 	bool writex;
1291 
1292 	cred = td->td_ucred;
1293 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1294 	    (*flagsp & MAP_SHARED) != 0;
1295 	if ((error = vget(vp, LK_SHARED)) != 0)
1296 		return (error);
1297 	AUDIT_ARG_VNODE1(vp);
1298 	foff = *foffp;
1299 	flags = *flagsp;
1300 	obj = vp->v_object;
1301 	if (vp->v_type == VREG) {
1302 		/*
1303 		 * Get the proper underlying object
1304 		 */
1305 		if (obj == NULL) {
1306 			error = EINVAL;
1307 			goto done;
1308 		}
1309 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1310 			vput(vp);
1311 			vp = (struct vnode *)obj->handle;
1312 			/*
1313 			 * Bypass filesystems obey the mpsafety of the
1314 			 * underlying fs.  Tmpfs never bypasses.
1315 			 */
1316 			error = vget(vp, LK_SHARED);
1317 			if (error != 0)
1318 				return (error);
1319 		}
1320 		if (writex) {
1321 			*writecounted = TRUE;
1322 			vm_pager_update_writecount(obj, 0, objsize);
1323 		}
1324 	} else {
1325 		error = EINVAL;
1326 		goto done;
1327 	}
1328 	if ((error = VOP_GETATTR(vp, &va, cred)))
1329 		goto done;
1330 #ifdef MAC
1331 	/* This relies on VM_PROT_* matching PROT_*. */
1332 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1333 	if (error != 0)
1334 		goto done;
1335 #endif
1336 	if ((flags & MAP_SHARED) != 0) {
1337 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1338 			if (prot & VM_PROT_WRITE) {
1339 				error = EPERM;
1340 				goto done;
1341 			}
1342 			*maxprotp &= ~VM_PROT_WRITE;
1343 		}
1344 	}
1345 	/*
1346 	 * If it is a regular file without any references
1347 	 * we do not need to sync it.
1348 	 * Adjust object size to be the size of actual file.
1349 	 */
1350 	objsize = round_page(va.va_size);
1351 	if (va.va_nlink == 0)
1352 		flags |= MAP_NOSYNC;
1353 	if (obj->type == OBJT_VNODE) {
1354 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1355 		    cred);
1356 		if (obj == NULL) {
1357 			error = ENOMEM;
1358 			goto done;
1359 		}
1360 	} else {
1361 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1362 		    ("wrong object type"));
1363 		vm_object_reference(obj);
1364 #if VM_NRESERVLEVEL > 0
1365 		if ((obj->flags & OBJ_COLORED) == 0) {
1366 			VM_OBJECT_WLOCK(obj);
1367 			vm_object_color(obj, 0);
1368 			VM_OBJECT_WUNLOCK(obj);
1369 		}
1370 #endif
1371 	}
1372 	*objp = obj;
1373 	*flagsp = flags;
1374 
1375 	VOP_MMAPPED(vp);
1376 
1377 done:
1378 	if (error != 0 && *writecounted) {
1379 		*writecounted = FALSE;
1380 		vm_pager_update_writecount(obj, objsize, 0);
1381 	}
1382 	vput(vp);
1383 	return (error);
1384 }
1385 
1386 /*
1387  * vm_mmap_cdev()
1388  *
1389  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1390  * operations on cdevs.
1391  */
1392 int
1393 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1394     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1395     vm_ooffset_t *foff, vm_object_t *objp)
1396 {
1397 	vm_object_t obj;
1398 	int error, flags;
1399 
1400 	flags = *flagsp;
1401 
1402 	if (dsw->d_flags & D_MMAP_ANON) {
1403 		*objp = NULL;
1404 		*foff = 0;
1405 		*maxprotp = VM_PROT_ALL;
1406 		*flagsp |= MAP_ANON;
1407 		return (0);
1408 	}
1409 	/*
1410 	 * cdevs do not provide private mappings of any kind.
1411 	 */
1412 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1413 	    (prot & VM_PROT_WRITE) != 0)
1414 		return (EACCES);
1415 	if (flags & (MAP_PRIVATE|MAP_COPY))
1416 		return (EINVAL);
1417 	/*
1418 	 * Force device mappings to be shared.
1419 	 */
1420 	flags |= MAP_SHARED;
1421 #ifdef MAC_XXX
1422 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1423 	if (error != 0)
1424 		return (error);
1425 #endif
1426 	/*
1427 	 * First, try d_mmap_single().  If that is not implemented
1428 	 * (returns ENODEV), fall back to using the device pager.
1429 	 * Note that d_mmap_single() must return a reference to the
1430 	 * object (it needs to bump the reference count of the object
1431 	 * it returns somehow).
1432 	 *
1433 	 * XXX assumes VM_PROT_* == PROT_*
1434 	 */
1435 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1436 	if (error != ENODEV)
1437 		return (error);
1438 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1439 	    td->td_ucred);
1440 	if (obj == NULL)
1441 		return (EINVAL);
1442 	*objp = obj;
1443 	*flagsp = flags;
1444 	return (0);
1445 }
1446 
1447 /*
1448  * vm_mmap()
1449  *
1450  * Internal version of mmap used by exec, sys5 shared memory, and
1451  * various device drivers.  Handle is either a vnode pointer, a
1452  * character device, or NULL for MAP_ANON.
1453  */
1454 int
1455 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1456 	vm_prot_t maxprot, int flags,
1457 	objtype_t handle_type, void *handle,
1458 	vm_ooffset_t foff)
1459 {
1460 	vm_object_t object;
1461 	struct thread *td = curthread;
1462 	int error;
1463 	boolean_t writecounted;
1464 
1465 	if (size == 0)
1466 		return (EINVAL);
1467 
1468 	size = round_page(size);
1469 	object = NULL;
1470 	writecounted = FALSE;
1471 
1472 	/*
1473 	 * Lookup/allocate object.
1474 	 */
1475 	switch (handle_type) {
1476 	case OBJT_DEVICE: {
1477 		struct cdevsw *dsw;
1478 		struct cdev *cdev;
1479 		int ref;
1480 
1481 		cdev = handle;
1482 		dsw = dev_refthread(cdev, &ref);
1483 		if (dsw == NULL)
1484 			return (ENXIO);
1485 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1486 		    dsw, &foff, &object);
1487 		dev_relthread(cdev, ref);
1488 		break;
1489 	}
1490 	case OBJT_VNODE:
1491 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1492 		    handle, &foff, &object, &writecounted);
1493 		break;
1494 	case OBJT_DEFAULT:
1495 		if (handle == NULL) {
1496 			error = 0;
1497 			break;
1498 		}
1499 		/* FALLTHROUGH */
1500 	default:
1501 		error = EINVAL;
1502 		break;
1503 	}
1504 	if (error)
1505 		return (error);
1506 
1507 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1508 	    foff, writecounted, td);
1509 	if (error != 0 && object != NULL) {
1510 		/*
1511 		 * If this mapping was accounted for in the vnode's
1512 		 * writecount, then undo that now.
1513 		 */
1514 		if (writecounted)
1515 			vm_pager_release_writecount(object, 0, size);
1516 		vm_object_deallocate(object);
1517 	}
1518 	return (error);
1519 }
1520 
1521 int
1522 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size)
1523 {
1524 	int error;
1525 
1526 	RACCT_PROC_LOCK(td->td_proc);
1527 	if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1528 		RACCT_PROC_UNLOCK(td->td_proc);
1529 		return (ENOMEM);
1530 	}
1531 	if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1532 		RACCT_PROC_UNLOCK(td->td_proc);
1533 		return (ENOMEM);
1534 	}
1535 	if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1536 		if (ptoa(pmap_wired_count(map->pmap)) + size >
1537 		    lim_cur(td, RLIMIT_MEMLOCK)) {
1538 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1539 			RACCT_PROC_UNLOCK(td->td_proc);
1540 			return (ENOMEM);
1541 		}
1542 		error = racct_set(td->td_proc, RACCT_MEMLOCK,
1543 		    ptoa(pmap_wired_count(map->pmap)) + size);
1544 		if (error != 0) {
1545 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1546 			RACCT_PROC_UNLOCK(td->td_proc);
1547 			return (error);
1548 		}
1549 	}
1550 	RACCT_PROC_UNLOCK(td->td_proc);
1551 	return (0);
1552 }
1553 
1554 /*
1555  * Internal version of mmap that maps a specific VM object into an
1556  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1557  */
1558 int
1559 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1560     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1561     boolean_t writecounted, struct thread *td)
1562 {
1563 	vm_offset_t max_addr;
1564 	int docow, error, findspace, rv;
1565 	bool curmap, fitit;
1566 
1567 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1568 	if (curmap) {
1569 		error = kern_mmap_racct_check(td, map, size);
1570 		if (error != 0)
1571 			return (error);
1572 	}
1573 
1574 	/*
1575 	 * We currently can only deal with page aligned file offsets.
1576 	 * The mmap() system call already enforces this by subtracting
1577 	 * the page offset from the file offset, but checking here
1578 	 * catches errors in device drivers (e.g. d_single_mmap()
1579 	 * callbacks) and other internal mapping requests (such as in
1580 	 * exec).
1581 	 */
1582 	if (foff & PAGE_MASK)
1583 		return (EINVAL);
1584 
1585 	if ((flags & MAP_FIXED) == 0) {
1586 		fitit = TRUE;
1587 		*addr = round_page(*addr);
1588 	} else {
1589 		if (*addr != trunc_page(*addr))
1590 			return (EINVAL);
1591 		fitit = FALSE;
1592 	}
1593 
1594 	if (flags & MAP_ANON) {
1595 		if (object != NULL || foff != 0)
1596 			return (EINVAL);
1597 		docow = 0;
1598 	} else if (flags & MAP_PREFAULT_READ)
1599 		docow = MAP_PREFAULT;
1600 	else
1601 		docow = MAP_PREFAULT_PARTIAL;
1602 
1603 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1604 		docow |= MAP_COPY_ON_WRITE;
1605 	if (flags & MAP_NOSYNC)
1606 		docow |= MAP_DISABLE_SYNCER;
1607 	if (flags & MAP_NOCORE)
1608 		docow |= MAP_DISABLE_COREDUMP;
1609 	/* Shared memory is also shared with children. */
1610 	if (flags & MAP_SHARED)
1611 		docow |= MAP_INHERIT_SHARE;
1612 	if (writecounted)
1613 		docow |= MAP_WRITECOUNT;
1614 	if (flags & MAP_STACK) {
1615 		if (object != NULL)
1616 			return (EINVAL);
1617 		docow |= MAP_STACK_GROWS_DOWN;
1618 	}
1619 	if ((flags & MAP_EXCL) != 0)
1620 		docow |= MAP_CHECK_EXCL;
1621 	if ((flags & MAP_GUARD) != 0)
1622 		docow |= MAP_CREATE_GUARD;
1623 
1624 	if (fitit) {
1625 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1626 			findspace = VMFS_SUPER_SPACE;
1627 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1628 			findspace = VMFS_ALIGNED_SPACE(flags >>
1629 			    MAP_ALIGNMENT_SHIFT);
1630 		else
1631 			findspace = VMFS_OPTIMAL_SPACE;
1632 		max_addr = 0;
1633 #ifdef MAP_32BIT
1634 		if ((flags & MAP_32BIT) != 0)
1635 			max_addr = MAP_32BIT_MAX_ADDR;
1636 #endif
1637 		if (curmap) {
1638 			rv = vm_map_find_min(map, object, foff, addr, size,
1639 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1640 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1641 			    findspace, prot, maxprot, docow);
1642 		} else {
1643 			rv = vm_map_find(map, object, foff, addr, size,
1644 			    max_addr, findspace, prot, maxprot, docow);
1645 		}
1646 	} else {
1647 		rv = vm_map_fixed(map, object, foff, *addr, size,
1648 		    prot, maxprot, docow);
1649 	}
1650 
1651 	if (rv == KERN_SUCCESS) {
1652 		/*
1653 		 * If the process has requested that all future mappings
1654 		 * be wired, then heed this.
1655 		 */
1656 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1657 			vm_map_lock(map);
1658 			if ((map->flags & MAP_WIREFUTURE) != 0)
1659 				(void)vm_map_wire_locked(map, *addr,
1660 				    *addr + size, VM_MAP_WIRE_USER |
1661 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1662 				    VM_MAP_WIRE_NOHOLES));
1663 			vm_map_unlock(map);
1664 		}
1665 	}
1666 	return (vm_mmap_to_errno(rv));
1667 }
1668 
1669 /*
1670  * Translate a Mach VM return code to zero on success or the appropriate errno
1671  * on failure.
1672  */
1673 int
1674 vm_mmap_to_errno(int rv)
1675 {
1676 
1677 	switch (rv) {
1678 	case KERN_SUCCESS:
1679 		return (0);
1680 	case KERN_INVALID_ADDRESS:
1681 	case KERN_NO_SPACE:
1682 		return (ENOMEM);
1683 	case KERN_PROTECTION_FAILURE:
1684 		return (EACCES);
1685 	default:
1686 		return (EINVAL);
1687 	}
1688 }
1689