xref: /freebsd/sys/vm/vm_mmap.c (revision 780fb4a2fa9a9aee5ac48a60b790f567c0dc13e9)
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/filedesc.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procctl.h>
62 #include <sys/racct.h>
63 #include <sys/resource.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 #include <sys/fcntl.h>
69 #include <sys/file.h>
70 #include <sys/mman.h>
71 #include <sys/mount.h>
72 #include <sys/conf.h>
73 #include <sys/stat.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysent.h>
76 #include <sys/vmmeter.h>
77 
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_extern.h>
90 #include <vm/vm_page.h>
91 #include <vm/vnode_pager.h>
92 
93 #ifdef HWPMC_HOOKS
94 #include <sys/pmckern.h>
95 #endif
96 
97 int old_mlock = 0;
98 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
99     "Do not apply RLIMIT_MEMLOCK on mlockall");
100 
101 #ifdef MAP_32BIT
102 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
103 #endif
104 
105 #ifndef _SYS_SYSPROTO_H_
106 struct sbrk_args {
107 	int incr;
108 };
109 #endif
110 
111 int
112 sys_sbrk(struct thread *td, struct sbrk_args *uap)
113 {
114 	/* Not yet implemented */
115 	return (EOPNOTSUPP);
116 }
117 
118 #ifndef _SYS_SYSPROTO_H_
119 struct sstk_args {
120 	int incr;
121 };
122 #endif
123 
124 int
125 sys_sstk(struct thread *td, struct sstk_args *uap)
126 {
127 	/* Not yet implemented */
128 	return (EOPNOTSUPP);
129 }
130 
131 #if defined(COMPAT_43)
132 #ifndef _SYS_SYSPROTO_H_
133 struct getpagesize_args {
134 	int dummy;
135 };
136 #endif
137 
138 int
139 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
140 {
141 
142 	td->td_retval[0] = PAGE_SIZE;
143 	return (0);
144 }
145 #endif				/* COMPAT_43 */
146 
147 
148 /*
149  * Memory Map (mmap) system call.  Note that the file offset
150  * and address are allowed to be NOT page aligned, though if
151  * the MAP_FIXED flag it set, both must have the same remainder
152  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
153  * page-aligned, the actual mapping starts at trunc_page(addr)
154  * and the return value is adjusted up by the page offset.
155  *
156  * Generally speaking, only character devices which are themselves
157  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
158  * there would be no cache coherency between a descriptor and a VM mapping
159  * both to the same character device.
160  */
161 #ifndef _SYS_SYSPROTO_H_
162 struct mmap_args {
163 	void *addr;
164 	size_t len;
165 	int prot;
166 	int flags;
167 	int fd;
168 	long pad;
169 	off_t pos;
170 };
171 #endif
172 
173 int
174 sys_mmap(struct thread *td, struct mmap_args *uap)
175 {
176 
177 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
178 	    uap->flags, uap->fd, uap->pos));
179 }
180 
181 int
182 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
183     int fd, off_t pos)
184 {
185 	struct vmspace *vms;
186 	struct file *fp;
187 	vm_offset_t addr;
188 	vm_size_t pageoff;
189 	vm_prot_t cap_maxprot;
190 	int align, error;
191 	cap_rights_t rights;
192 
193 	vms = td->td_proc->p_vmspace;
194 	fp = NULL;
195 	AUDIT_ARG_FD(fd);
196 	addr = addr0;
197 
198 	/*
199 	 * Ignore old flags that used to be defined but did not do anything.
200 	 */
201 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
202 
203 	/*
204 	 * Enforce the constraints.
205 	 * Mapping of length 0 is only allowed for old binaries.
206 	 * Anonymous mapping shall specify -1 as filedescriptor and
207 	 * zero position for new code. Be nice to ancient a.out
208 	 * binaries and correct pos for anonymous mapping, since old
209 	 * ld.so sometimes issues anonymous map requests with non-zero
210 	 * pos.
211 	 */
212 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
213 		if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
214 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
215 			return (EINVAL);
216 	} else {
217 		if ((flags & MAP_ANON) != 0)
218 			pos = 0;
219 	}
220 
221 	if (flags & MAP_STACK) {
222 		if ((fd != -1) ||
223 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
224 			return (EINVAL);
225 		flags |= MAP_ANON;
226 		pos = 0;
227 	}
228 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
229 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
230 	    MAP_PREFAULT_READ | MAP_GUARD |
231 #ifdef MAP_32BIT
232 	    MAP_32BIT |
233 #endif
234 	    MAP_ALIGNMENT_MASK)) != 0)
235 		return (EINVAL);
236 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
237 		return (EINVAL);
238 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
239 		return (EINVAL);
240 	if (prot != PROT_NONE &&
241 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
242 		return (EINVAL);
243 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
244 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
245 #ifdef MAP_32BIT
246 	    MAP_32BIT |
247 #endif
248 	    MAP_ALIGNMENT_MASK)) != 0))
249 		return (EINVAL);
250 
251 	/*
252 	 * Align the file position to a page boundary,
253 	 * and save its page offset component.
254 	 */
255 	pageoff = (pos & PAGE_MASK);
256 	pos -= pageoff;
257 
258 	/* Adjust size for rounding (on both ends). */
259 	size += pageoff;			/* low end... */
260 	size = (vm_size_t) round_page(size);	/* hi end */
261 
262 	/* Ensure alignment is at least a page and fits in a pointer. */
263 	align = flags & MAP_ALIGNMENT_MASK;
264 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
265 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
266 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
267 		return (EINVAL);
268 
269 	/*
270 	 * Check for illegal addresses.  Watch out for address wrap... Note
271 	 * that VM_*_ADDRESS are not constants due to casts (argh).
272 	 */
273 	if (flags & MAP_FIXED) {
274 		/*
275 		 * The specified address must have the same remainder
276 		 * as the file offset taken modulo PAGE_SIZE, so it
277 		 * should be aligned after adjustment by pageoff.
278 		 */
279 		addr -= pageoff;
280 		if (addr & PAGE_MASK)
281 			return (EINVAL);
282 
283 		/* Address range must be all in user VM space. */
284 		if (addr < vm_map_min(&vms->vm_map) ||
285 		    addr + size > vm_map_max(&vms->vm_map))
286 			return (EINVAL);
287 		if (addr + size < addr)
288 			return (EINVAL);
289 #ifdef MAP_32BIT
290 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
291 			return (EINVAL);
292 	} else if (flags & MAP_32BIT) {
293 		/*
294 		 * For MAP_32BIT, override the hint if it is too high and
295 		 * do not bother moving the mapping past the heap (since
296 		 * the heap is usually above 2GB).
297 		 */
298 		if (addr + size > MAP_32BIT_MAX_ADDR)
299 			addr = 0;
300 #endif
301 	} else {
302 		/*
303 		 * XXX for non-fixed mappings where no hint is provided or
304 		 * the hint would fall in the potential heap space,
305 		 * place it after the end of the largest possible heap.
306 		 *
307 		 * There should really be a pmap call to determine a reasonable
308 		 * location.
309 		 */
310 		if (addr == 0 ||
311 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
312 		    addr < round_page((vm_offset_t)vms->vm_daddr +
313 		    lim_max(td, RLIMIT_DATA))))
314 			addr = round_page((vm_offset_t)vms->vm_daddr +
315 			    lim_max(td, RLIMIT_DATA));
316 	}
317 	if (size == 0) {
318 		/*
319 		 * Return success without mapping anything for old
320 		 * binaries that request a page-aligned mapping of
321 		 * length 0.  For modern binaries, this function
322 		 * returns an error earlier.
323 		 */
324 		error = 0;
325 	} else if ((flags & MAP_GUARD) != 0) {
326 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
327 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
328 	} else if ((flags & MAP_ANON) != 0) {
329 		/*
330 		 * Mapping blank space is trivial.
331 		 *
332 		 * This relies on VM_PROT_* matching PROT_*.
333 		 */
334 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
335 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
336 	} else {
337 		/*
338 		 * Mapping file, get fp for validation and don't let the
339 		 * descriptor disappear on us if we block. Check capability
340 		 * rights, but also return the maximum rights to be combined
341 		 * with maxprot later.
342 		 */
343 		cap_rights_init(&rights, CAP_MMAP);
344 		if (prot & PROT_READ)
345 			cap_rights_set(&rights, CAP_MMAP_R);
346 		if ((flags & MAP_SHARED) != 0) {
347 			if (prot & PROT_WRITE)
348 				cap_rights_set(&rights, CAP_MMAP_W);
349 		}
350 		if (prot & PROT_EXEC)
351 			cap_rights_set(&rights, CAP_MMAP_X);
352 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
353 		if (error != 0)
354 			goto done;
355 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
356 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
357 			error = EINVAL;
358 			goto done;
359 		}
360 
361 		/* This relies on VM_PROT_* matching PROT_*. */
362 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
363 		    cap_maxprot, flags, pos, td);
364 	}
365 
366 	if (error == 0)
367 		td->td_retval[0] = (register_t) (addr + pageoff);
368 done:
369 	if (fp)
370 		fdrop(fp, td);
371 
372 	return (error);
373 }
374 
375 #if defined(COMPAT_FREEBSD6)
376 int
377 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
378 {
379 
380 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
381 	    uap->flags, uap->fd, uap->pos));
382 }
383 #endif
384 
385 #ifdef COMPAT_43
386 #ifndef _SYS_SYSPROTO_H_
387 struct ommap_args {
388 	caddr_t addr;
389 	int len;
390 	int prot;
391 	int flags;
392 	int fd;
393 	long pos;
394 };
395 #endif
396 int
397 ommap(struct thread *td, struct ommap_args *uap)
398 {
399 	static const char cvtbsdprot[8] = {
400 		0,
401 		PROT_EXEC,
402 		PROT_WRITE,
403 		PROT_EXEC | PROT_WRITE,
404 		PROT_READ,
405 		PROT_EXEC | PROT_READ,
406 		PROT_WRITE | PROT_READ,
407 		PROT_EXEC | PROT_WRITE | PROT_READ,
408 	};
409 	int flags, prot;
410 
411 #define	OMAP_ANON	0x0002
412 #define	OMAP_COPY	0x0020
413 #define	OMAP_SHARED	0x0010
414 #define	OMAP_FIXED	0x0100
415 
416 	prot = cvtbsdprot[uap->prot & 0x7];
417 #ifdef COMPAT_FREEBSD32
418 #if defined(__amd64__)
419 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
420 	    prot != 0)
421 		prot |= PROT_EXEC;
422 #endif
423 #endif
424 	flags = 0;
425 	if (uap->flags & OMAP_ANON)
426 		flags |= MAP_ANON;
427 	if (uap->flags & OMAP_COPY)
428 		flags |= MAP_COPY;
429 	if (uap->flags & OMAP_SHARED)
430 		flags |= MAP_SHARED;
431 	else
432 		flags |= MAP_PRIVATE;
433 	if (uap->flags & OMAP_FIXED)
434 		flags |= MAP_FIXED;
435 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
436 	    uap->fd, uap->pos));
437 }
438 #endif				/* COMPAT_43 */
439 
440 
441 #ifndef _SYS_SYSPROTO_H_
442 struct msync_args {
443 	void *addr;
444 	size_t len;
445 	int flags;
446 };
447 #endif
448 int
449 sys_msync(struct thread *td, struct msync_args *uap)
450 {
451 
452 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
453 }
454 
455 int
456 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
457 {
458 	vm_offset_t addr;
459 	vm_size_t pageoff;
460 	vm_map_t map;
461 	int rv;
462 
463 	addr = addr0;
464 	pageoff = (addr & PAGE_MASK);
465 	addr -= pageoff;
466 	size += pageoff;
467 	size = (vm_size_t) round_page(size);
468 	if (addr + size < addr)
469 		return (EINVAL);
470 
471 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
472 		return (EINVAL);
473 
474 	map = &td->td_proc->p_vmspace->vm_map;
475 
476 	/*
477 	 * Clean the pages and interpret the return value.
478 	 */
479 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
480 	    (flags & MS_INVALIDATE) != 0);
481 	switch (rv) {
482 	case KERN_SUCCESS:
483 		return (0);
484 	case KERN_INVALID_ADDRESS:
485 		return (ENOMEM);
486 	case KERN_INVALID_ARGUMENT:
487 		return (EBUSY);
488 	case KERN_FAILURE:
489 		return (EIO);
490 	default:
491 		return (EINVAL);
492 	}
493 }
494 
495 #ifndef _SYS_SYSPROTO_H_
496 struct munmap_args {
497 	void *addr;
498 	size_t len;
499 };
500 #endif
501 int
502 sys_munmap(struct thread *td, struct munmap_args *uap)
503 {
504 
505 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
506 }
507 
508 int
509 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
510 {
511 #ifdef HWPMC_HOOKS
512 	struct pmckern_map_out pkm;
513 	vm_map_entry_t entry;
514 	bool pmc_handled;
515 #endif
516 	vm_offset_t addr;
517 	vm_size_t pageoff;
518 	vm_map_t map;
519 
520 	if (size == 0)
521 		return (EINVAL);
522 
523 	addr = addr0;
524 	pageoff = (addr & PAGE_MASK);
525 	addr -= pageoff;
526 	size += pageoff;
527 	size = (vm_size_t) round_page(size);
528 	if (addr + size < addr)
529 		return (EINVAL);
530 
531 	/*
532 	 * Check for illegal addresses.  Watch out for address wrap...
533 	 */
534 	map = &td->td_proc->p_vmspace->vm_map;
535 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
536 		return (EINVAL);
537 	vm_map_lock(map);
538 #ifdef HWPMC_HOOKS
539 	pmc_handled = false;
540 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
541 		pmc_handled = true;
542 		/*
543 		 * Inform hwpmc if the address range being unmapped contains
544 		 * an executable region.
545 		 */
546 		pkm.pm_address = (uintptr_t) NULL;
547 		if (vm_map_lookup_entry(map, addr, &entry)) {
548 			for (; entry->start < addr + size;
549 			    entry = entry->next) {
550 				if (vm_map_check_protection(map, entry->start,
551 					entry->end, VM_PROT_EXECUTE) == TRUE) {
552 					pkm.pm_address = (uintptr_t) addr;
553 					pkm.pm_size = (size_t) size;
554 					break;
555 				}
556 			}
557 		}
558 	}
559 #endif
560 	vm_map_delete(map, addr, addr + size);
561 
562 #ifdef HWPMC_HOOKS
563 	if (__predict_false(pmc_handled)) {
564 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
565 		vm_map_lock_downgrade(map);
566 		if (pkm.pm_address != (uintptr_t) NULL)
567 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
568 		vm_map_unlock_read(map);
569 	} else
570 #endif
571 		vm_map_unlock(map);
572 
573 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
574 	return (0);
575 }
576 
577 #ifndef _SYS_SYSPROTO_H_
578 struct mprotect_args {
579 	const void *addr;
580 	size_t len;
581 	int prot;
582 };
583 #endif
584 int
585 sys_mprotect(struct thread *td, struct mprotect_args *uap)
586 {
587 
588 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
589 }
590 
591 int
592 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
593 {
594 	vm_offset_t addr;
595 	vm_size_t pageoff;
596 
597 	addr = addr0;
598 	prot = (prot & VM_PROT_ALL);
599 	pageoff = (addr & PAGE_MASK);
600 	addr -= pageoff;
601 	size += pageoff;
602 	size = (vm_size_t) round_page(size);
603 	if (addr + size < addr)
604 		return (EINVAL);
605 
606 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
607 	    addr + size, prot, FALSE)) {
608 	case KERN_SUCCESS:
609 		return (0);
610 	case KERN_PROTECTION_FAILURE:
611 		return (EACCES);
612 	case KERN_RESOURCE_SHORTAGE:
613 		return (ENOMEM);
614 	}
615 	return (EINVAL);
616 }
617 
618 #ifndef _SYS_SYSPROTO_H_
619 struct minherit_args {
620 	void *addr;
621 	size_t len;
622 	int inherit;
623 };
624 #endif
625 int
626 sys_minherit(struct thread *td, struct minherit_args *uap)
627 {
628 	vm_offset_t addr;
629 	vm_size_t size, pageoff;
630 	vm_inherit_t inherit;
631 
632 	addr = (vm_offset_t)uap->addr;
633 	size = uap->len;
634 	inherit = uap->inherit;
635 
636 	pageoff = (addr & PAGE_MASK);
637 	addr -= pageoff;
638 	size += pageoff;
639 	size = (vm_size_t) round_page(size);
640 	if (addr + size < addr)
641 		return (EINVAL);
642 
643 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
644 	    addr + size, inherit)) {
645 	case KERN_SUCCESS:
646 		return (0);
647 	case KERN_PROTECTION_FAILURE:
648 		return (EACCES);
649 	}
650 	return (EINVAL);
651 }
652 
653 #ifndef _SYS_SYSPROTO_H_
654 struct madvise_args {
655 	void *addr;
656 	size_t len;
657 	int behav;
658 };
659 #endif
660 
661 int
662 sys_madvise(struct thread *td, struct madvise_args *uap)
663 {
664 
665 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
666 }
667 
668 int
669 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
670 {
671 	vm_map_t map;
672 	vm_offset_t addr, end, start;
673 	int flags;
674 
675 	/*
676 	 * Check for our special case, advising the swap pager we are
677 	 * "immortal."
678 	 */
679 	if (behav == MADV_PROTECT) {
680 		flags = PPROT_SET;
681 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
682 		    PROC_SPROTECT, &flags));
683 	}
684 
685 	/*
686 	 * Check for illegal addresses.  Watch out for address wrap... Note
687 	 * that VM_*_ADDRESS are not constants due to casts (argh).
688 	 */
689 	map = &td->td_proc->p_vmspace->vm_map;
690 	addr = addr0;
691 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
692 		return (EINVAL);
693 	if ((addr + len) < addr)
694 		return (EINVAL);
695 
696 	/*
697 	 * Since this routine is only advisory, we default to conservative
698 	 * behavior.
699 	 */
700 	start = trunc_page(addr);
701 	end = round_page(addr + len);
702 
703 	/*
704 	 * vm_map_madvise() checks for illegal values of behav.
705 	 */
706 	return (vm_map_madvise(map, start, end, behav));
707 }
708 
709 #ifndef _SYS_SYSPROTO_H_
710 struct mincore_args {
711 	const void *addr;
712 	size_t len;
713 	char *vec;
714 };
715 #endif
716 
717 int
718 sys_mincore(struct thread *td, struct mincore_args *uap)
719 {
720 
721 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
722 }
723 
724 int
725 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
726 {
727 	vm_offset_t addr, first_addr;
728 	vm_offset_t end, cend;
729 	pmap_t pmap;
730 	vm_map_t map;
731 	int error = 0;
732 	int vecindex, lastvecindex;
733 	vm_map_entry_t current;
734 	vm_map_entry_t entry;
735 	vm_object_t object;
736 	vm_paddr_t locked_pa;
737 	vm_page_t m;
738 	vm_pindex_t pindex;
739 	int mincoreinfo;
740 	unsigned int timestamp;
741 	boolean_t locked;
742 
743 	/*
744 	 * Make sure that the addresses presented are valid for user
745 	 * mode.
746 	 */
747 	first_addr = addr = trunc_page(addr0);
748 	end = addr + (vm_size_t)round_page(len);
749 	map = &td->td_proc->p_vmspace->vm_map;
750 	if (end > vm_map_max(map) || end < addr)
751 		return (ENOMEM);
752 
753 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
754 
755 	vm_map_lock_read(map);
756 RestartScan:
757 	timestamp = map->timestamp;
758 
759 	if (!vm_map_lookup_entry(map, addr, &entry)) {
760 		vm_map_unlock_read(map);
761 		return (ENOMEM);
762 	}
763 
764 	/*
765 	 * Do this on a map entry basis so that if the pages are not
766 	 * in the current processes address space, we can easily look
767 	 * up the pages elsewhere.
768 	 */
769 	lastvecindex = -1;
770 	for (current = entry; current->start < end; current = current->next) {
771 
772 		/*
773 		 * check for contiguity
774 		 */
775 		if (current->end < end && current->next->start > current->end) {
776 			vm_map_unlock_read(map);
777 			return (ENOMEM);
778 		}
779 
780 		/*
781 		 * ignore submaps (for now) or null objects
782 		 */
783 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
784 			current->object.vm_object == NULL)
785 			continue;
786 
787 		/*
788 		 * limit this scan to the current map entry and the
789 		 * limits for the mincore call
790 		 */
791 		if (addr < current->start)
792 			addr = current->start;
793 		cend = current->end;
794 		if (cend > end)
795 			cend = end;
796 
797 		/*
798 		 * scan this entry one page at a time
799 		 */
800 		while (addr < cend) {
801 			/*
802 			 * Check pmap first, it is likely faster, also
803 			 * it can provide info as to whether we are the
804 			 * one referencing or modifying the page.
805 			 */
806 			object = NULL;
807 			locked_pa = 0;
808 		retry:
809 			m = NULL;
810 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
811 			if (locked_pa != 0) {
812 				/*
813 				 * The page is mapped by this process but not
814 				 * both accessed and modified.  It is also
815 				 * managed.  Acquire the object lock so that
816 				 * other mappings might be examined.
817 				 */
818 				m = PHYS_TO_VM_PAGE(locked_pa);
819 				if (m->object != object) {
820 					if (object != NULL)
821 						VM_OBJECT_WUNLOCK(object);
822 					object = m->object;
823 					locked = VM_OBJECT_TRYWLOCK(object);
824 					vm_page_unlock(m);
825 					if (!locked) {
826 						VM_OBJECT_WLOCK(object);
827 						vm_page_lock(m);
828 						goto retry;
829 					}
830 				} else
831 					vm_page_unlock(m);
832 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
833 				    ("mincore: page %p is mapped but invalid",
834 				    m));
835 			} else if (mincoreinfo == 0) {
836 				/*
837 				 * The page is not mapped by this process.  If
838 				 * the object implements managed pages, then
839 				 * determine if the page is resident so that
840 				 * the mappings might be examined.
841 				 */
842 				if (current->object.vm_object != object) {
843 					if (object != NULL)
844 						VM_OBJECT_WUNLOCK(object);
845 					object = current->object.vm_object;
846 					VM_OBJECT_WLOCK(object);
847 				}
848 				if (object->type == OBJT_DEFAULT ||
849 				    object->type == OBJT_SWAP ||
850 				    object->type == OBJT_VNODE) {
851 					pindex = OFF_TO_IDX(current->offset +
852 					    (addr - current->start));
853 					m = vm_page_lookup(object, pindex);
854 					if (m != NULL && m->valid == 0)
855 						m = NULL;
856 					if (m != NULL)
857 						mincoreinfo = MINCORE_INCORE;
858 				}
859 			}
860 			if (m != NULL) {
861 				/* Examine other mappings to the page. */
862 				if (m->dirty == 0 && pmap_is_modified(m))
863 					vm_page_dirty(m);
864 				if (m->dirty != 0)
865 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
866 				/*
867 				 * The first test for PGA_REFERENCED is an
868 				 * optimization.  The second test is
869 				 * required because a concurrent pmap
870 				 * operation could clear the last reference
871 				 * and set PGA_REFERENCED before the call to
872 				 * pmap_is_referenced().
873 				 */
874 				if ((m->aflags & PGA_REFERENCED) != 0 ||
875 				    pmap_is_referenced(m) ||
876 				    (m->aflags & PGA_REFERENCED) != 0)
877 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
878 			}
879 			if (object != NULL)
880 				VM_OBJECT_WUNLOCK(object);
881 
882 			/*
883 			 * subyte may page fault.  In case it needs to modify
884 			 * the map, we release the lock.
885 			 */
886 			vm_map_unlock_read(map);
887 
888 			/*
889 			 * calculate index into user supplied byte vector
890 			 */
891 			vecindex = atop(addr - first_addr);
892 
893 			/*
894 			 * If we have skipped map entries, we need to make sure that
895 			 * the byte vector is zeroed for those skipped entries.
896 			 */
897 			while ((lastvecindex + 1) < vecindex) {
898 				++lastvecindex;
899 				error = subyte(vec + lastvecindex, 0);
900 				if (error) {
901 					error = EFAULT;
902 					goto done2;
903 				}
904 			}
905 
906 			/*
907 			 * Pass the page information to the user
908 			 */
909 			error = subyte(vec + vecindex, mincoreinfo);
910 			if (error) {
911 				error = EFAULT;
912 				goto done2;
913 			}
914 
915 			/*
916 			 * If the map has changed, due to the subyte, the previous
917 			 * output may be invalid.
918 			 */
919 			vm_map_lock_read(map);
920 			if (timestamp != map->timestamp)
921 				goto RestartScan;
922 
923 			lastvecindex = vecindex;
924 			addr += PAGE_SIZE;
925 		}
926 	}
927 
928 	/*
929 	 * subyte may page fault.  In case it needs to modify
930 	 * the map, we release the lock.
931 	 */
932 	vm_map_unlock_read(map);
933 
934 	/*
935 	 * Zero the last entries in the byte vector.
936 	 */
937 	vecindex = atop(end - first_addr);
938 	while ((lastvecindex + 1) < vecindex) {
939 		++lastvecindex;
940 		error = subyte(vec + lastvecindex, 0);
941 		if (error) {
942 			error = EFAULT;
943 			goto done2;
944 		}
945 	}
946 
947 	/*
948 	 * If the map has changed, due to the subyte, the previous
949 	 * output may be invalid.
950 	 */
951 	vm_map_lock_read(map);
952 	if (timestamp != map->timestamp)
953 		goto RestartScan;
954 	vm_map_unlock_read(map);
955 done2:
956 	return (error);
957 }
958 
959 #ifndef _SYS_SYSPROTO_H_
960 struct mlock_args {
961 	const void *addr;
962 	size_t len;
963 };
964 #endif
965 int
966 sys_mlock(struct thread *td, struct mlock_args *uap)
967 {
968 
969 	return (kern_mlock(td->td_proc, td->td_ucred,
970 	    __DECONST(uintptr_t, uap->addr), uap->len));
971 }
972 
973 int
974 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
975 {
976 	vm_offset_t addr, end, last, start;
977 	vm_size_t npages, size;
978 	vm_map_t map;
979 	unsigned long nsize;
980 	int error;
981 
982 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
983 	if (error)
984 		return (error);
985 	addr = addr0;
986 	size = len;
987 	last = addr + size;
988 	start = trunc_page(addr);
989 	end = round_page(last);
990 	if (last < addr || end < addr)
991 		return (EINVAL);
992 	npages = atop(end - start);
993 	if (npages > vm_page_max_wired)
994 		return (ENOMEM);
995 	map = &proc->p_vmspace->vm_map;
996 	PROC_LOCK(proc);
997 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
998 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
999 		PROC_UNLOCK(proc);
1000 		return (ENOMEM);
1001 	}
1002 	PROC_UNLOCK(proc);
1003 	if (npages + vm_wire_count() > vm_page_max_wired)
1004 		return (EAGAIN);
1005 #ifdef RACCT
1006 	if (racct_enable) {
1007 		PROC_LOCK(proc);
1008 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1009 		PROC_UNLOCK(proc);
1010 		if (error != 0)
1011 			return (ENOMEM);
1012 	}
1013 #endif
1014 	error = vm_map_wire(map, start, end,
1015 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1016 #ifdef RACCT
1017 	if (racct_enable && error != KERN_SUCCESS) {
1018 		PROC_LOCK(proc);
1019 		racct_set(proc, RACCT_MEMLOCK,
1020 		    ptoa(pmap_wired_count(map->pmap)));
1021 		PROC_UNLOCK(proc);
1022 	}
1023 #endif
1024 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1025 }
1026 
1027 #ifndef _SYS_SYSPROTO_H_
1028 struct mlockall_args {
1029 	int	how;
1030 };
1031 #endif
1032 
1033 int
1034 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1035 {
1036 	vm_map_t map;
1037 	int error;
1038 
1039 	map = &td->td_proc->p_vmspace->vm_map;
1040 	error = priv_check(td, PRIV_VM_MLOCK);
1041 	if (error)
1042 		return (error);
1043 
1044 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1045 		return (EINVAL);
1046 
1047 	/*
1048 	 * If wiring all pages in the process would cause it to exceed
1049 	 * a hard resource limit, return ENOMEM.
1050 	 */
1051 	if (!old_mlock && uap->how & MCL_CURRENT) {
1052 		PROC_LOCK(td->td_proc);
1053 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1054 			PROC_UNLOCK(td->td_proc);
1055 			return (ENOMEM);
1056 		}
1057 		PROC_UNLOCK(td->td_proc);
1058 	}
1059 #ifdef RACCT
1060 	if (racct_enable) {
1061 		PROC_LOCK(td->td_proc);
1062 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1063 		PROC_UNLOCK(td->td_proc);
1064 		if (error != 0)
1065 			return (ENOMEM);
1066 	}
1067 #endif
1068 
1069 	if (uap->how & MCL_FUTURE) {
1070 		vm_map_lock(map);
1071 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1072 		vm_map_unlock(map);
1073 		error = 0;
1074 	}
1075 
1076 	if (uap->how & MCL_CURRENT) {
1077 		/*
1078 		 * P1003.1-2001 mandates that all currently mapped pages
1079 		 * will be memory resident and locked (wired) upon return
1080 		 * from mlockall(). vm_map_wire() will wire pages, by
1081 		 * calling vm_fault_wire() for each page in the region.
1082 		 */
1083 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1084 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1085 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1086 	}
1087 #ifdef RACCT
1088 	if (racct_enable && error != KERN_SUCCESS) {
1089 		PROC_LOCK(td->td_proc);
1090 		racct_set(td->td_proc, RACCT_MEMLOCK,
1091 		    ptoa(pmap_wired_count(map->pmap)));
1092 		PROC_UNLOCK(td->td_proc);
1093 	}
1094 #endif
1095 
1096 	return (error);
1097 }
1098 
1099 #ifndef _SYS_SYSPROTO_H_
1100 struct munlockall_args {
1101 	register_t dummy;
1102 };
1103 #endif
1104 
1105 int
1106 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1107 {
1108 	vm_map_t map;
1109 	int error;
1110 
1111 	map = &td->td_proc->p_vmspace->vm_map;
1112 	error = priv_check(td, PRIV_VM_MUNLOCK);
1113 	if (error)
1114 		return (error);
1115 
1116 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1117 	vm_map_lock(map);
1118 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1119 	vm_map_unlock(map);
1120 
1121 	/* Forcibly unwire all pages. */
1122 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1123 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1124 #ifdef RACCT
1125 	if (racct_enable && error == KERN_SUCCESS) {
1126 		PROC_LOCK(td->td_proc);
1127 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1128 		PROC_UNLOCK(td->td_proc);
1129 	}
1130 #endif
1131 
1132 	return (error);
1133 }
1134 
1135 #ifndef _SYS_SYSPROTO_H_
1136 struct munlock_args {
1137 	const void *addr;
1138 	size_t len;
1139 };
1140 #endif
1141 int
1142 sys_munlock(struct thread *td, struct munlock_args *uap)
1143 {
1144 
1145 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1146 }
1147 
1148 int
1149 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1150 {
1151 	vm_offset_t addr, end, last, start;
1152 #ifdef RACCT
1153 	vm_map_t map;
1154 #endif
1155 	int error;
1156 
1157 	error = priv_check(td, PRIV_VM_MUNLOCK);
1158 	if (error)
1159 		return (error);
1160 	addr = addr0;
1161 	last = addr + size;
1162 	start = trunc_page(addr);
1163 	end = round_page(last);
1164 	if (last < addr || end < addr)
1165 		return (EINVAL);
1166 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1167 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1168 #ifdef RACCT
1169 	if (racct_enable && error == KERN_SUCCESS) {
1170 		PROC_LOCK(td->td_proc);
1171 		map = &td->td_proc->p_vmspace->vm_map;
1172 		racct_set(td->td_proc, RACCT_MEMLOCK,
1173 		    ptoa(pmap_wired_count(map->pmap)));
1174 		PROC_UNLOCK(td->td_proc);
1175 	}
1176 #endif
1177 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1178 }
1179 
1180 /*
1181  * vm_mmap_vnode()
1182  *
1183  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1184  * operations on vnodes.
1185  */
1186 int
1187 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1188     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1189     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1190     boolean_t *writecounted)
1191 {
1192 	struct vattr va;
1193 	vm_object_t obj;
1194 	vm_ooffset_t foff;
1195 	struct ucred *cred;
1196 	int error, flags, locktype;
1197 
1198 	cred = td->td_ucred;
1199 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1200 		locktype = LK_EXCLUSIVE;
1201 	else
1202 		locktype = LK_SHARED;
1203 	if ((error = vget(vp, locktype, td)) != 0)
1204 		return (error);
1205 	AUDIT_ARG_VNODE1(vp);
1206 	foff = *foffp;
1207 	flags = *flagsp;
1208 	obj = vp->v_object;
1209 	if (vp->v_type == VREG) {
1210 		/*
1211 		 * Get the proper underlying object
1212 		 */
1213 		if (obj == NULL) {
1214 			error = EINVAL;
1215 			goto done;
1216 		}
1217 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1218 			vput(vp);
1219 			vp = (struct vnode *)obj->handle;
1220 			/*
1221 			 * Bypass filesystems obey the mpsafety of the
1222 			 * underlying fs.  Tmpfs never bypasses.
1223 			 */
1224 			error = vget(vp, locktype, td);
1225 			if (error != 0)
1226 				return (error);
1227 		}
1228 		if (locktype == LK_EXCLUSIVE) {
1229 			*writecounted = TRUE;
1230 			vnode_pager_update_writecount(obj, 0, objsize);
1231 		}
1232 	} else {
1233 		error = EINVAL;
1234 		goto done;
1235 	}
1236 	if ((error = VOP_GETATTR(vp, &va, cred)))
1237 		goto done;
1238 #ifdef MAC
1239 	/* This relies on VM_PROT_* matching PROT_*. */
1240 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1241 	if (error != 0)
1242 		goto done;
1243 #endif
1244 	if ((flags & MAP_SHARED) != 0) {
1245 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1246 			if (prot & VM_PROT_WRITE) {
1247 				error = EPERM;
1248 				goto done;
1249 			}
1250 			*maxprotp &= ~VM_PROT_WRITE;
1251 		}
1252 	}
1253 	/*
1254 	 * If it is a regular file without any references
1255 	 * we do not need to sync it.
1256 	 * Adjust object size to be the size of actual file.
1257 	 */
1258 	objsize = round_page(va.va_size);
1259 	if (va.va_nlink == 0)
1260 		flags |= MAP_NOSYNC;
1261 	if (obj->type == OBJT_VNODE) {
1262 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1263 		    cred);
1264 		if (obj == NULL) {
1265 			error = ENOMEM;
1266 			goto done;
1267 		}
1268 	} else {
1269 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1270 		    ("wrong object type"));
1271 		VM_OBJECT_WLOCK(obj);
1272 		vm_object_reference_locked(obj);
1273 #if VM_NRESERVLEVEL > 0
1274 		vm_object_color(obj, 0);
1275 #endif
1276 		VM_OBJECT_WUNLOCK(obj);
1277 	}
1278 	*objp = obj;
1279 	*flagsp = flags;
1280 
1281 	vfs_mark_atime(vp, cred);
1282 
1283 done:
1284 	if (error != 0 && *writecounted) {
1285 		*writecounted = FALSE;
1286 		vnode_pager_update_writecount(obj, objsize, 0);
1287 	}
1288 	vput(vp);
1289 	return (error);
1290 }
1291 
1292 /*
1293  * vm_mmap_cdev()
1294  *
1295  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1296  * operations on cdevs.
1297  */
1298 int
1299 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1300     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1301     vm_ooffset_t *foff, vm_object_t *objp)
1302 {
1303 	vm_object_t obj;
1304 	int error, flags;
1305 
1306 	flags = *flagsp;
1307 
1308 	if (dsw->d_flags & D_MMAP_ANON) {
1309 		*objp = NULL;
1310 		*foff = 0;
1311 		*maxprotp = VM_PROT_ALL;
1312 		*flagsp |= MAP_ANON;
1313 		return (0);
1314 	}
1315 	/*
1316 	 * cdevs do not provide private mappings of any kind.
1317 	 */
1318 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1319 	    (prot & VM_PROT_WRITE) != 0)
1320 		return (EACCES);
1321 	if (flags & (MAP_PRIVATE|MAP_COPY))
1322 		return (EINVAL);
1323 	/*
1324 	 * Force device mappings to be shared.
1325 	 */
1326 	flags |= MAP_SHARED;
1327 #ifdef MAC_XXX
1328 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1329 	if (error != 0)
1330 		return (error);
1331 #endif
1332 	/*
1333 	 * First, try d_mmap_single().  If that is not implemented
1334 	 * (returns ENODEV), fall back to using the device pager.
1335 	 * Note that d_mmap_single() must return a reference to the
1336 	 * object (it needs to bump the reference count of the object
1337 	 * it returns somehow).
1338 	 *
1339 	 * XXX assumes VM_PROT_* == PROT_*
1340 	 */
1341 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1342 	if (error != ENODEV)
1343 		return (error);
1344 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1345 	    td->td_ucred);
1346 	if (obj == NULL)
1347 		return (EINVAL);
1348 	*objp = obj;
1349 	*flagsp = flags;
1350 	return (0);
1351 }
1352 
1353 /*
1354  * vm_mmap()
1355  *
1356  * Internal version of mmap used by exec, sys5 shared memory, and
1357  * various device drivers.  Handle is either a vnode pointer, a
1358  * character device, or NULL for MAP_ANON.
1359  */
1360 int
1361 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1362 	vm_prot_t maxprot, int flags,
1363 	objtype_t handle_type, void *handle,
1364 	vm_ooffset_t foff)
1365 {
1366 	vm_object_t object;
1367 	struct thread *td = curthread;
1368 	int error;
1369 	boolean_t writecounted;
1370 
1371 	if (size == 0)
1372 		return (EINVAL);
1373 
1374 	size = round_page(size);
1375 	object = NULL;
1376 	writecounted = FALSE;
1377 
1378 	/*
1379 	 * Lookup/allocate object.
1380 	 */
1381 	switch (handle_type) {
1382 	case OBJT_DEVICE: {
1383 		struct cdevsw *dsw;
1384 		struct cdev *cdev;
1385 		int ref;
1386 
1387 		cdev = handle;
1388 		dsw = dev_refthread(cdev, &ref);
1389 		if (dsw == NULL)
1390 			return (ENXIO);
1391 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1392 		    dsw, &foff, &object);
1393 		dev_relthread(cdev, ref);
1394 		break;
1395 	}
1396 	case OBJT_VNODE:
1397 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1398 		    handle, &foff, &object, &writecounted);
1399 		break;
1400 	case OBJT_DEFAULT:
1401 		if (handle == NULL) {
1402 			error = 0;
1403 			break;
1404 		}
1405 		/* FALLTHROUGH */
1406 	default:
1407 		error = EINVAL;
1408 		break;
1409 	}
1410 	if (error)
1411 		return (error);
1412 
1413 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1414 	    foff, writecounted, td);
1415 	if (error != 0 && object != NULL) {
1416 		/*
1417 		 * If this mapping was accounted for in the vnode's
1418 		 * writecount, then undo that now.
1419 		 */
1420 		if (writecounted)
1421 			vnode_pager_release_writecount(object, 0, size);
1422 		vm_object_deallocate(object);
1423 	}
1424 	return (error);
1425 }
1426 
1427 /*
1428  * Internal version of mmap that maps a specific VM object into an
1429  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1430  */
1431 int
1432 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1433     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1434     boolean_t writecounted, struct thread *td)
1435 {
1436 	boolean_t curmap, fitit;
1437 	vm_offset_t max_addr;
1438 	int docow, error, findspace, rv;
1439 
1440 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1441 	if (curmap) {
1442 		PROC_LOCK(td->td_proc);
1443 		if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1444 			PROC_UNLOCK(td->td_proc);
1445 			return (ENOMEM);
1446 		}
1447 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1448 			PROC_UNLOCK(td->td_proc);
1449 			return (ENOMEM);
1450 		}
1451 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1452 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1453 			    lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1454 				racct_set_force(td->td_proc, RACCT_VMEM,
1455 				    map->size);
1456 				PROC_UNLOCK(td->td_proc);
1457 				return (ENOMEM);
1458 			}
1459 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1460 			    ptoa(pmap_wired_count(map->pmap)) + size);
1461 			if (error != 0) {
1462 				racct_set_force(td->td_proc, RACCT_VMEM,
1463 				    map->size);
1464 				PROC_UNLOCK(td->td_proc);
1465 				return (error);
1466 			}
1467 		}
1468 		PROC_UNLOCK(td->td_proc);
1469 	}
1470 
1471 	/*
1472 	 * We currently can only deal with page aligned file offsets.
1473 	 * The mmap() system call already enforces this by subtracting
1474 	 * the page offset from the file offset, but checking here
1475 	 * catches errors in device drivers (e.g. d_single_mmap()
1476 	 * callbacks) and other internal mapping requests (such as in
1477 	 * exec).
1478 	 */
1479 	if (foff & PAGE_MASK)
1480 		return (EINVAL);
1481 
1482 	if ((flags & MAP_FIXED) == 0) {
1483 		fitit = TRUE;
1484 		*addr = round_page(*addr);
1485 	} else {
1486 		if (*addr != trunc_page(*addr))
1487 			return (EINVAL);
1488 		fitit = FALSE;
1489 	}
1490 
1491 	if (flags & MAP_ANON) {
1492 		if (object != NULL || foff != 0)
1493 			return (EINVAL);
1494 		docow = 0;
1495 	} else if (flags & MAP_PREFAULT_READ)
1496 		docow = MAP_PREFAULT;
1497 	else
1498 		docow = MAP_PREFAULT_PARTIAL;
1499 
1500 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1501 		docow |= MAP_COPY_ON_WRITE;
1502 	if (flags & MAP_NOSYNC)
1503 		docow |= MAP_DISABLE_SYNCER;
1504 	if (flags & MAP_NOCORE)
1505 		docow |= MAP_DISABLE_COREDUMP;
1506 	/* Shared memory is also shared with children. */
1507 	if (flags & MAP_SHARED)
1508 		docow |= MAP_INHERIT_SHARE;
1509 	if (writecounted)
1510 		docow |= MAP_VN_WRITECOUNT;
1511 	if (flags & MAP_STACK) {
1512 		if (object != NULL)
1513 			return (EINVAL);
1514 		docow |= MAP_STACK_GROWS_DOWN;
1515 	}
1516 	if ((flags & MAP_EXCL) != 0)
1517 		docow |= MAP_CHECK_EXCL;
1518 	if ((flags & MAP_GUARD) != 0)
1519 		docow |= MAP_CREATE_GUARD;
1520 
1521 	if (fitit) {
1522 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1523 			findspace = VMFS_SUPER_SPACE;
1524 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1525 			findspace = VMFS_ALIGNED_SPACE(flags >>
1526 			    MAP_ALIGNMENT_SHIFT);
1527 		else
1528 			findspace = VMFS_OPTIMAL_SPACE;
1529 		max_addr = 0;
1530 #ifdef MAP_32BIT
1531 		if ((flags & MAP_32BIT) != 0)
1532 			max_addr = MAP_32BIT_MAX_ADDR;
1533 #endif
1534 		if (curmap) {
1535 			rv = vm_map_find_min(map, object, foff, addr, size,
1536 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1537 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1538 			    findspace, prot, maxprot, docow);
1539 		} else {
1540 			rv = vm_map_find(map, object, foff, addr, size,
1541 			    max_addr, findspace, prot, maxprot, docow);
1542 		}
1543 	} else {
1544 		rv = vm_map_fixed(map, object, foff, *addr, size,
1545 		    prot, maxprot, docow);
1546 	}
1547 
1548 	if (rv == KERN_SUCCESS) {
1549 		/*
1550 		 * If the process has requested that all future mappings
1551 		 * be wired, then heed this.
1552 		 */
1553 		if (map->flags & MAP_WIREFUTURE) {
1554 			vm_map_wire(map, *addr, *addr + size,
1555 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1556 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1557 		}
1558 	}
1559 	return (vm_mmap_to_errno(rv));
1560 }
1561 
1562 /*
1563  * Translate a Mach VM return code to zero on success or the appropriate errno
1564  * on failure.
1565  */
1566 int
1567 vm_mmap_to_errno(int rv)
1568 {
1569 
1570 	switch (rv) {
1571 	case KERN_SUCCESS:
1572 		return (0);
1573 	case KERN_INVALID_ADDRESS:
1574 	case KERN_NO_SPACE:
1575 		return (ENOMEM);
1576 	case KERN_PROTECTION_FAILURE:
1577 		return (EACCES);
1578 	default:
1579 		return (EINVAL);
1580 	}
1581 }
1582