xref: /freebsd/sys/vm/vm_mmap.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
1 /*-
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35  *
36  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
37  */
38 
39 /*
40  * Mapped file (mmap) interface to VM
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48 #include "opt_vm.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/capsicum.h>
53 #include <sys/kernel.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/sysproto.h>
57 #include <sys/filedesc.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procctl.h>
61 #include <sys/racct.h>
62 #include <sys/resource.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
68 #include <sys/file.h>
69 #include <sys/mman.h>
70 #include <sys/mount.h>
71 #include <sys/conf.h>
72 #include <sys/stat.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysent.h>
75 #include <sys/vmmeter.h>
76 
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_pageout.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_page.h>
90 #include <vm/vnode_pager.h>
91 
92 #ifdef HWPMC_HOOKS
93 #include <sys/pmckern.h>
94 #endif
95 
96 int old_mlock = 0;
97 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
98     "Do not apply RLIMIT_MEMLOCK on mlockall");
99 
100 #ifdef MAP_32BIT
101 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
102 #endif
103 
104 #ifndef _SYS_SYSPROTO_H_
105 struct sbrk_args {
106 	int incr;
107 };
108 #endif
109 
110 int
111 sys_sbrk(struct thread *td, struct sbrk_args *uap)
112 {
113 	/* Not yet implemented */
114 	return (EOPNOTSUPP);
115 }
116 
117 #ifndef _SYS_SYSPROTO_H_
118 struct sstk_args {
119 	int incr;
120 };
121 #endif
122 
123 int
124 sys_sstk(struct thread *td, struct sstk_args *uap)
125 {
126 	/* Not yet implemented */
127 	return (EOPNOTSUPP);
128 }
129 
130 #if defined(COMPAT_43)
131 #ifndef _SYS_SYSPROTO_H_
132 struct getpagesize_args {
133 	int dummy;
134 };
135 #endif
136 
137 int
138 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
139 {
140 
141 	td->td_retval[0] = PAGE_SIZE;
142 	return (0);
143 }
144 #endif				/* COMPAT_43 */
145 
146 
147 /*
148  * Memory Map (mmap) system call.  Note that the file offset
149  * and address are allowed to be NOT page aligned, though if
150  * the MAP_FIXED flag it set, both must have the same remainder
151  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
152  * page-aligned, the actual mapping starts at trunc_page(addr)
153  * and the return value is adjusted up by the page offset.
154  *
155  * Generally speaking, only character devices which are themselves
156  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
157  * there would be no cache coherency between a descriptor and a VM mapping
158  * both to the same character device.
159  */
160 #ifndef _SYS_SYSPROTO_H_
161 struct mmap_args {
162 	void *addr;
163 	size_t len;
164 	int prot;
165 	int flags;
166 	int fd;
167 	long pad;
168 	off_t pos;
169 };
170 #endif
171 
172 int
173 sys_mmap(struct thread *td, struct mmap_args *uap)
174 {
175 
176 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
177 	    uap->flags, uap->fd, uap->pos));
178 }
179 
180 int
181 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
182     int fd, off_t pos)
183 {
184 	struct vmspace *vms;
185 	struct file *fp;
186 	vm_offset_t addr;
187 	vm_size_t pageoff;
188 	vm_prot_t cap_maxprot;
189 	int align, error;
190 	cap_rights_t rights;
191 
192 	vms = td->td_proc->p_vmspace;
193 	fp = NULL;
194 	AUDIT_ARG_FD(fd);
195 	addr = addr0;
196 
197 	/*
198 	 * Ignore old flags that used to be defined but did not do anything.
199 	 */
200 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
201 
202 	/*
203 	 * Enforce the constraints.
204 	 * Mapping of length 0 is only allowed for old binaries.
205 	 * Anonymous mapping shall specify -1 as filedescriptor and
206 	 * zero position for new code. Be nice to ancient a.out
207 	 * binaries and correct pos for anonymous mapping, since old
208 	 * ld.so sometimes issues anonymous map requests with non-zero
209 	 * pos.
210 	 */
211 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
212 		if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
213 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
214 			return (EINVAL);
215 	} else {
216 		if ((flags & MAP_ANON) != 0)
217 			pos = 0;
218 	}
219 
220 	if (flags & MAP_STACK) {
221 		if ((fd != -1) ||
222 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
223 			return (EINVAL);
224 		flags |= MAP_ANON;
225 		pos = 0;
226 	}
227 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
228 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
229 	    MAP_PREFAULT_READ | MAP_GUARD |
230 #ifdef MAP_32BIT
231 	    MAP_32BIT |
232 #endif
233 	    MAP_ALIGNMENT_MASK)) != 0)
234 		return (EINVAL);
235 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
236 		return (EINVAL);
237 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
238 		return (EINVAL);
239 	if (prot != PROT_NONE &&
240 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
241 		return (EINVAL);
242 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
243 	    pos != 0 || (flags & (MAP_SHARED | MAP_PRIVATE | MAP_PREFAULT |
244 	    MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0))
245 		return (EINVAL);
246 
247 	/*
248 	 * Align the file position to a page boundary,
249 	 * and save its page offset component.
250 	 */
251 	pageoff = (pos & PAGE_MASK);
252 	pos -= pageoff;
253 
254 	/* Adjust size for rounding (on both ends). */
255 	size += pageoff;			/* low end... */
256 	size = (vm_size_t) round_page(size);	/* hi end */
257 
258 	/* Ensure alignment is at least a page and fits in a pointer. */
259 	align = flags & MAP_ALIGNMENT_MASK;
260 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
261 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
262 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
263 		return (EINVAL);
264 
265 	/*
266 	 * Check for illegal addresses.  Watch out for address wrap... Note
267 	 * that VM_*_ADDRESS are not constants due to casts (argh).
268 	 */
269 	if (flags & MAP_FIXED) {
270 		/*
271 		 * The specified address must have the same remainder
272 		 * as the file offset taken modulo PAGE_SIZE, so it
273 		 * should be aligned after adjustment by pageoff.
274 		 */
275 		addr -= pageoff;
276 		if (addr & PAGE_MASK)
277 			return (EINVAL);
278 
279 		/* Address range must be all in user VM space. */
280 		if (addr < vm_map_min(&vms->vm_map) ||
281 		    addr + size > vm_map_max(&vms->vm_map))
282 			return (EINVAL);
283 		if (addr + size < addr)
284 			return (EINVAL);
285 #ifdef MAP_32BIT
286 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
287 			return (EINVAL);
288 	} else if (flags & MAP_32BIT) {
289 		/*
290 		 * For MAP_32BIT, override the hint if it is too high and
291 		 * do not bother moving the mapping past the heap (since
292 		 * the heap is usually above 2GB).
293 		 */
294 		if (addr + size > MAP_32BIT_MAX_ADDR)
295 			addr = 0;
296 #endif
297 	} else {
298 		/*
299 		 * XXX for non-fixed mappings where no hint is provided or
300 		 * the hint would fall in the potential heap space,
301 		 * place it after the end of the largest possible heap.
302 		 *
303 		 * There should really be a pmap call to determine a reasonable
304 		 * location.
305 		 */
306 		if (addr == 0 ||
307 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
308 		    addr < round_page((vm_offset_t)vms->vm_daddr +
309 		    lim_max(td, RLIMIT_DATA))))
310 			addr = round_page((vm_offset_t)vms->vm_daddr +
311 			    lim_max(td, RLIMIT_DATA));
312 	}
313 	if (size == 0) {
314 		/*
315 		 * Return success without mapping anything for old
316 		 * binaries that request a page-aligned mapping of
317 		 * length 0.  For modern binaries, this function
318 		 * returns an error earlier.
319 		 */
320 		error = 0;
321 	} else if ((flags & MAP_GUARD) != 0) {
322 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
323 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
324 	} else if ((flags & MAP_ANON) != 0) {
325 		/*
326 		 * Mapping blank space is trivial.
327 		 *
328 		 * This relies on VM_PROT_* matching PROT_*.
329 		 */
330 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
331 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
332 	} else {
333 		/*
334 		 * Mapping file, get fp for validation and don't let the
335 		 * descriptor disappear on us if we block. Check capability
336 		 * rights, but also return the maximum rights to be combined
337 		 * with maxprot later.
338 		 */
339 		cap_rights_init(&rights, CAP_MMAP);
340 		if (prot & PROT_READ)
341 			cap_rights_set(&rights, CAP_MMAP_R);
342 		if ((flags & MAP_SHARED) != 0) {
343 			if (prot & PROT_WRITE)
344 				cap_rights_set(&rights, CAP_MMAP_W);
345 		}
346 		if (prot & PROT_EXEC)
347 			cap_rights_set(&rights, CAP_MMAP_X);
348 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
349 		if (error != 0)
350 			goto done;
351 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
352 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
353 			error = EINVAL;
354 			goto done;
355 		}
356 
357 		/* This relies on VM_PROT_* matching PROT_*. */
358 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
359 		    cap_maxprot, flags, pos, td);
360 	}
361 
362 	if (error == 0)
363 		td->td_retval[0] = (register_t) (addr + pageoff);
364 done:
365 	if (fp)
366 		fdrop(fp, td);
367 
368 	return (error);
369 }
370 
371 #if defined(COMPAT_FREEBSD6)
372 int
373 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
374 {
375 
376 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
377 	    uap->flags, uap->fd, uap->pos));
378 }
379 #endif
380 
381 #ifdef COMPAT_43
382 #ifndef _SYS_SYSPROTO_H_
383 struct ommap_args {
384 	caddr_t addr;
385 	int len;
386 	int prot;
387 	int flags;
388 	int fd;
389 	long pos;
390 };
391 #endif
392 int
393 ommap(struct thread *td, struct ommap_args *uap)
394 {
395 	static const char cvtbsdprot[8] = {
396 		0,
397 		PROT_EXEC,
398 		PROT_WRITE,
399 		PROT_EXEC | PROT_WRITE,
400 		PROT_READ,
401 		PROT_EXEC | PROT_READ,
402 		PROT_WRITE | PROT_READ,
403 		PROT_EXEC | PROT_WRITE | PROT_READ,
404 	};
405 	int flags, prot;
406 
407 #define	OMAP_ANON	0x0002
408 #define	OMAP_COPY	0x0020
409 #define	OMAP_SHARED	0x0010
410 #define	OMAP_FIXED	0x0100
411 
412 	prot = cvtbsdprot[uap->prot & 0x7];
413 #ifdef COMPAT_FREEBSD32
414 #if defined(__amd64__)
415 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
416 	    prot != 0)
417 		prot |= PROT_EXEC;
418 #endif
419 #endif
420 	flags = 0;
421 	if (uap->flags & OMAP_ANON)
422 		flags |= MAP_ANON;
423 	if (uap->flags & OMAP_COPY)
424 		flags |= MAP_COPY;
425 	if (uap->flags & OMAP_SHARED)
426 		flags |= MAP_SHARED;
427 	else
428 		flags |= MAP_PRIVATE;
429 	if (uap->flags & OMAP_FIXED)
430 		flags |= MAP_FIXED;
431 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
432 	    uap->fd, uap->pos));
433 }
434 #endif				/* COMPAT_43 */
435 
436 
437 #ifndef _SYS_SYSPROTO_H_
438 struct msync_args {
439 	void *addr;
440 	size_t len;
441 	int flags;
442 };
443 #endif
444 int
445 sys_msync(struct thread *td, struct msync_args *uap)
446 {
447 
448 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
449 }
450 
451 int
452 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
453 {
454 	vm_offset_t addr;
455 	vm_size_t pageoff;
456 	vm_map_t map;
457 	int rv;
458 
459 	addr = addr0;
460 	pageoff = (addr & PAGE_MASK);
461 	addr -= pageoff;
462 	size += pageoff;
463 	size = (vm_size_t) round_page(size);
464 	if (addr + size < addr)
465 		return (EINVAL);
466 
467 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
468 		return (EINVAL);
469 
470 	map = &td->td_proc->p_vmspace->vm_map;
471 
472 	/*
473 	 * Clean the pages and interpret the return value.
474 	 */
475 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
476 	    (flags & MS_INVALIDATE) != 0);
477 	switch (rv) {
478 	case KERN_SUCCESS:
479 		return (0);
480 	case KERN_INVALID_ADDRESS:
481 		return (ENOMEM);
482 	case KERN_INVALID_ARGUMENT:
483 		return (EBUSY);
484 	case KERN_FAILURE:
485 		return (EIO);
486 	default:
487 		return (EINVAL);
488 	}
489 }
490 
491 #ifndef _SYS_SYSPROTO_H_
492 struct munmap_args {
493 	void *addr;
494 	size_t len;
495 };
496 #endif
497 int
498 sys_munmap(struct thread *td, struct munmap_args *uap)
499 {
500 
501 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
502 }
503 
504 int
505 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
506 {
507 #ifdef HWPMC_HOOKS
508 	struct pmckern_map_out pkm;
509 	vm_map_entry_t entry;
510 	bool pmc_handled;
511 #endif
512 	vm_offset_t addr;
513 	vm_size_t pageoff;
514 	vm_map_t map;
515 
516 	if (size == 0)
517 		return (EINVAL);
518 
519 	addr = addr0;
520 	pageoff = (addr & PAGE_MASK);
521 	addr -= pageoff;
522 	size += pageoff;
523 	size = (vm_size_t) round_page(size);
524 	if (addr + size < addr)
525 		return (EINVAL);
526 
527 	/*
528 	 * Check for illegal addresses.  Watch out for address wrap...
529 	 */
530 	map = &td->td_proc->p_vmspace->vm_map;
531 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
532 		return (EINVAL);
533 	vm_map_lock(map);
534 #ifdef HWPMC_HOOKS
535 	pmc_handled = false;
536 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
537 		pmc_handled = true;
538 		/*
539 		 * Inform hwpmc if the address range being unmapped contains
540 		 * an executable region.
541 		 */
542 		pkm.pm_address = (uintptr_t) NULL;
543 		if (vm_map_lookup_entry(map, addr, &entry)) {
544 			for (;
545 			    entry != &map->header && entry->start < addr + size;
546 			    entry = entry->next) {
547 				if (vm_map_check_protection(map, entry->start,
548 					entry->end, VM_PROT_EXECUTE) == TRUE) {
549 					pkm.pm_address = (uintptr_t) addr;
550 					pkm.pm_size = (size_t) size;
551 					break;
552 				}
553 			}
554 		}
555 	}
556 #endif
557 	vm_map_delete(map, addr, addr + size);
558 
559 #ifdef HWPMC_HOOKS
560 	if (__predict_false(pmc_handled)) {
561 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
562 		vm_map_lock_downgrade(map);
563 		if (pkm.pm_address != (uintptr_t) NULL)
564 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
565 		vm_map_unlock_read(map);
566 	} else
567 #endif
568 		vm_map_unlock(map);
569 
570 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
571 	return (0);
572 }
573 
574 #ifndef _SYS_SYSPROTO_H_
575 struct mprotect_args {
576 	const void *addr;
577 	size_t len;
578 	int prot;
579 };
580 #endif
581 int
582 sys_mprotect(struct thread *td, struct mprotect_args *uap)
583 {
584 
585 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
586 }
587 
588 int
589 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
590 {
591 	vm_offset_t addr;
592 	vm_size_t pageoff;
593 
594 	addr = addr0;
595 	prot = (prot & VM_PROT_ALL);
596 	pageoff = (addr & PAGE_MASK);
597 	addr -= pageoff;
598 	size += pageoff;
599 	size = (vm_size_t) round_page(size);
600 	if (addr + size < addr)
601 		return (EINVAL);
602 
603 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
604 	    addr + size, prot, FALSE)) {
605 	case KERN_SUCCESS:
606 		return (0);
607 	case KERN_PROTECTION_FAILURE:
608 		return (EACCES);
609 	case KERN_RESOURCE_SHORTAGE:
610 		return (ENOMEM);
611 	}
612 	return (EINVAL);
613 }
614 
615 #ifndef _SYS_SYSPROTO_H_
616 struct minherit_args {
617 	void *addr;
618 	size_t len;
619 	int inherit;
620 };
621 #endif
622 int
623 sys_minherit(struct thread *td, struct minherit_args *uap)
624 {
625 	vm_offset_t addr;
626 	vm_size_t size, pageoff;
627 	vm_inherit_t inherit;
628 
629 	addr = (vm_offset_t)uap->addr;
630 	size = uap->len;
631 	inherit = uap->inherit;
632 
633 	pageoff = (addr & PAGE_MASK);
634 	addr -= pageoff;
635 	size += pageoff;
636 	size = (vm_size_t) round_page(size);
637 	if (addr + size < addr)
638 		return (EINVAL);
639 
640 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
641 	    addr + size, inherit)) {
642 	case KERN_SUCCESS:
643 		return (0);
644 	case KERN_PROTECTION_FAILURE:
645 		return (EACCES);
646 	}
647 	return (EINVAL);
648 }
649 
650 #ifndef _SYS_SYSPROTO_H_
651 struct madvise_args {
652 	void *addr;
653 	size_t len;
654 	int behav;
655 };
656 #endif
657 
658 int
659 sys_madvise(struct thread *td, struct madvise_args *uap)
660 {
661 
662 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
663 }
664 
665 int
666 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
667 {
668 	vm_map_t map;
669 	vm_offset_t addr, end, start;
670 	int flags;
671 
672 	/*
673 	 * Check for our special case, advising the swap pager we are
674 	 * "immortal."
675 	 */
676 	if (behav == MADV_PROTECT) {
677 		flags = PPROT_SET;
678 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
679 		    PROC_SPROTECT, &flags));
680 	}
681 
682 	/*
683 	 * Check for illegal behavior
684 	 */
685 	if (behav < 0 || behav > MADV_CORE)
686 		return (EINVAL);
687 	/*
688 	 * Check for illegal addresses.  Watch out for address wrap... Note
689 	 * that VM_*_ADDRESS are not constants due to casts (argh).
690 	 */
691 	map = &td->td_proc->p_vmspace->vm_map;
692 	addr = addr0;
693 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
694 		return (EINVAL);
695 	if ((addr + len) < addr)
696 		return (EINVAL);
697 
698 	/*
699 	 * Since this routine is only advisory, we default to conservative
700 	 * behavior.
701 	 */
702 	start = trunc_page(addr);
703 	end = round_page(addr + len);
704 
705 	if (vm_map_madvise(map, start, end, behav))
706 		return (EINVAL);
707 	return (0);
708 }
709 
710 #ifndef _SYS_SYSPROTO_H_
711 struct mincore_args {
712 	const void *addr;
713 	size_t len;
714 	char *vec;
715 };
716 #endif
717 
718 int
719 sys_mincore(struct thread *td, struct mincore_args *uap)
720 {
721 
722 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
723 }
724 
725 int
726 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
727 {
728 	vm_offset_t addr, first_addr;
729 	vm_offset_t end, cend;
730 	pmap_t pmap;
731 	vm_map_t map;
732 	int error = 0;
733 	int vecindex, lastvecindex;
734 	vm_map_entry_t current;
735 	vm_map_entry_t entry;
736 	vm_object_t object;
737 	vm_paddr_t locked_pa;
738 	vm_page_t m;
739 	vm_pindex_t pindex;
740 	int mincoreinfo;
741 	unsigned int timestamp;
742 	boolean_t locked;
743 
744 	/*
745 	 * Make sure that the addresses presented are valid for user
746 	 * mode.
747 	 */
748 	first_addr = addr = trunc_page(addr0);
749 	end = addr + (vm_size_t)round_page(len);
750 	map = &td->td_proc->p_vmspace->vm_map;
751 	if (end > vm_map_max(map) || end < addr)
752 		return (ENOMEM);
753 
754 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
755 
756 	vm_map_lock_read(map);
757 RestartScan:
758 	timestamp = map->timestamp;
759 
760 	if (!vm_map_lookup_entry(map, addr, &entry)) {
761 		vm_map_unlock_read(map);
762 		return (ENOMEM);
763 	}
764 
765 	/*
766 	 * Do this on a map entry basis so that if the pages are not
767 	 * in the current processes address space, we can easily look
768 	 * up the pages elsewhere.
769 	 */
770 	lastvecindex = -1;
771 	for (current = entry;
772 	    (current != &map->header) && (current->start < end);
773 	    current = current->next) {
774 
775 		/*
776 		 * check for contiguity
777 		 */
778 		if (current->end < end &&
779 		    (entry->next == &map->header ||
780 		     current->next->start > current->end)) {
781 			vm_map_unlock_read(map);
782 			return (ENOMEM);
783 		}
784 
785 		/*
786 		 * ignore submaps (for now) or null objects
787 		 */
788 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
789 			current->object.vm_object == NULL)
790 			continue;
791 
792 		/*
793 		 * limit this scan to the current map entry and the
794 		 * limits for the mincore call
795 		 */
796 		if (addr < current->start)
797 			addr = current->start;
798 		cend = current->end;
799 		if (cend > end)
800 			cend = end;
801 
802 		/*
803 		 * scan this entry one page at a time
804 		 */
805 		while (addr < cend) {
806 			/*
807 			 * Check pmap first, it is likely faster, also
808 			 * it can provide info as to whether we are the
809 			 * one referencing or modifying the page.
810 			 */
811 			object = NULL;
812 			locked_pa = 0;
813 		retry:
814 			m = NULL;
815 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
816 			if (locked_pa != 0) {
817 				/*
818 				 * The page is mapped by this process but not
819 				 * both accessed and modified.  It is also
820 				 * managed.  Acquire the object lock so that
821 				 * other mappings might be examined.
822 				 */
823 				m = PHYS_TO_VM_PAGE(locked_pa);
824 				if (m->object != object) {
825 					if (object != NULL)
826 						VM_OBJECT_WUNLOCK(object);
827 					object = m->object;
828 					locked = VM_OBJECT_TRYWLOCK(object);
829 					vm_page_unlock(m);
830 					if (!locked) {
831 						VM_OBJECT_WLOCK(object);
832 						vm_page_lock(m);
833 						goto retry;
834 					}
835 				} else
836 					vm_page_unlock(m);
837 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
838 				    ("mincore: page %p is mapped but invalid",
839 				    m));
840 			} else if (mincoreinfo == 0) {
841 				/*
842 				 * The page is not mapped by this process.  If
843 				 * the object implements managed pages, then
844 				 * determine if the page is resident so that
845 				 * the mappings might be examined.
846 				 */
847 				if (current->object.vm_object != object) {
848 					if (object != NULL)
849 						VM_OBJECT_WUNLOCK(object);
850 					object = current->object.vm_object;
851 					VM_OBJECT_WLOCK(object);
852 				}
853 				if (object->type == OBJT_DEFAULT ||
854 				    object->type == OBJT_SWAP ||
855 				    object->type == OBJT_VNODE) {
856 					pindex = OFF_TO_IDX(current->offset +
857 					    (addr - current->start));
858 					m = vm_page_lookup(object, pindex);
859 					if (m != NULL && m->valid == 0)
860 						m = NULL;
861 					if (m != NULL)
862 						mincoreinfo = MINCORE_INCORE;
863 				}
864 			}
865 			if (m != NULL) {
866 				/* Examine other mappings to the page. */
867 				if (m->dirty == 0 && pmap_is_modified(m))
868 					vm_page_dirty(m);
869 				if (m->dirty != 0)
870 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
871 				/*
872 				 * The first test for PGA_REFERENCED is an
873 				 * optimization.  The second test is
874 				 * required because a concurrent pmap
875 				 * operation could clear the last reference
876 				 * and set PGA_REFERENCED before the call to
877 				 * pmap_is_referenced().
878 				 */
879 				if ((m->aflags & PGA_REFERENCED) != 0 ||
880 				    pmap_is_referenced(m) ||
881 				    (m->aflags & PGA_REFERENCED) != 0)
882 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
883 			}
884 			if (object != NULL)
885 				VM_OBJECT_WUNLOCK(object);
886 
887 			/*
888 			 * subyte may page fault.  In case it needs to modify
889 			 * the map, we release the lock.
890 			 */
891 			vm_map_unlock_read(map);
892 
893 			/*
894 			 * calculate index into user supplied byte vector
895 			 */
896 			vecindex = atop(addr - first_addr);
897 
898 			/*
899 			 * If we have skipped map entries, we need to make sure that
900 			 * the byte vector is zeroed for those skipped entries.
901 			 */
902 			while ((lastvecindex + 1) < vecindex) {
903 				++lastvecindex;
904 				error = subyte(vec + lastvecindex, 0);
905 				if (error) {
906 					error = EFAULT;
907 					goto done2;
908 				}
909 			}
910 
911 			/*
912 			 * Pass the page information to the user
913 			 */
914 			error = subyte(vec + vecindex, mincoreinfo);
915 			if (error) {
916 				error = EFAULT;
917 				goto done2;
918 			}
919 
920 			/*
921 			 * If the map has changed, due to the subyte, the previous
922 			 * output may be invalid.
923 			 */
924 			vm_map_lock_read(map);
925 			if (timestamp != map->timestamp)
926 				goto RestartScan;
927 
928 			lastvecindex = vecindex;
929 			addr += PAGE_SIZE;
930 		}
931 	}
932 
933 	/*
934 	 * subyte may page fault.  In case it needs to modify
935 	 * the map, we release the lock.
936 	 */
937 	vm_map_unlock_read(map);
938 
939 	/*
940 	 * Zero the last entries in the byte vector.
941 	 */
942 	vecindex = atop(end - first_addr);
943 	while ((lastvecindex + 1) < vecindex) {
944 		++lastvecindex;
945 		error = subyte(vec + lastvecindex, 0);
946 		if (error) {
947 			error = EFAULT;
948 			goto done2;
949 		}
950 	}
951 
952 	/*
953 	 * If the map has changed, due to the subyte, the previous
954 	 * output may be invalid.
955 	 */
956 	vm_map_lock_read(map);
957 	if (timestamp != map->timestamp)
958 		goto RestartScan;
959 	vm_map_unlock_read(map);
960 done2:
961 	return (error);
962 }
963 
964 #ifndef _SYS_SYSPROTO_H_
965 struct mlock_args {
966 	const void *addr;
967 	size_t len;
968 };
969 #endif
970 int
971 sys_mlock(struct thread *td, struct mlock_args *uap)
972 {
973 
974 	return (kern_mlock(td->td_proc, td->td_ucred,
975 	    __DECONST(uintptr_t, uap->addr), uap->len));
976 }
977 
978 int
979 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
980 {
981 	vm_offset_t addr, end, last, start;
982 	vm_size_t npages, size;
983 	vm_map_t map;
984 	unsigned long nsize;
985 	int error;
986 
987 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
988 	if (error)
989 		return (error);
990 	addr = addr0;
991 	size = len;
992 	last = addr + size;
993 	start = trunc_page(addr);
994 	end = round_page(last);
995 	if (last < addr || end < addr)
996 		return (EINVAL);
997 	npages = atop(end - start);
998 	if (npages > vm_page_max_wired)
999 		return (ENOMEM);
1000 	map = &proc->p_vmspace->vm_map;
1001 	PROC_LOCK(proc);
1002 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1003 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1004 		PROC_UNLOCK(proc);
1005 		return (ENOMEM);
1006 	}
1007 	PROC_UNLOCK(proc);
1008 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1009 		return (EAGAIN);
1010 #ifdef RACCT
1011 	if (racct_enable) {
1012 		PROC_LOCK(proc);
1013 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1014 		PROC_UNLOCK(proc);
1015 		if (error != 0)
1016 			return (ENOMEM);
1017 	}
1018 #endif
1019 	error = vm_map_wire(map, start, end,
1020 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1021 #ifdef RACCT
1022 	if (racct_enable && error != KERN_SUCCESS) {
1023 		PROC_LOCK(proc);
1024 		racct_set(proc, RACCT_MEMLOCK,
1025 		    ptoa(pmap_wired_count(map->pmap)));
1026 		PROC_UNLOCK(proc);
1027 	}
1028 #endif
1029 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1030 }
1031 
1032 #ifndef _SYS_SYSPROTO_H_
1033 struct mlockall_args {
1034 	int	how;
1035 };
1036 #endif
1037 
1038 int
1039 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1040 {
1041 	vm_map_t map;
1042 	int error;
1043 
1044 	map = &td->td_proc->p_vmspace->vm_map;
1045 	error = priv_check(td, PRIV_VM_MLOCK);
1046 	if (error)
1047 		return (error);
1048 
1049 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1050 		return (EINVAL);
1051 
1052 	/*
1053 	 * If wiring all pages in the process would cause it to exceed
1054 	 * a hard resource limit, return ENOMEM.
1055 	 */
1056 	if (!old_mlock && uap->how & MCL_CURRENT) {
1057 		PROC_LOCK(td->td_proc);
1058 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1059 			PROC_UNLOCK(td->td_proc);
1060 			return (ENOMEM);
1061 		}
1062 		PROC_UNLOCK(td->td_proc);
1063 	}
1064 #ifdef RACCT
1065 	if (racct_enable) {
1066 		PROC_LOCK(td->td_proc);
1067 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1068 		PROC_UNLOCK(td->td_proc);
1069 		if (error != 0)
1070 			return (ENOMEM);
1071 	}
1072 #endif
1073 
1074 	if (uap->how & MCL_FUTURE) {
1075 		vm_map_lock(map);
1076 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1077 		vm_map_unlock(map);
1078 		error = 0;
1079 	}
1080 
1081 	if (uap->how & MCL_CURRENT) {
1082 		/*
1083 		 * P1003.1-2001 mandates that all currently mapped pages
1084 		 * will be memory resident and locked (wired) upon return
1085 		 * from mlockall(). vm_map_wire() will wire pages, by
1086 		 * calling vm_fault_wire() for each page in the region.
1087 		 */
1088 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1089 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1090 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1091 	}
1092 #ifdef RACCT
1093 	if (racct_enable && error != KERN_SUCCESS) {
1094 		PROC_LOCK(td->td_proc);
1095 		racct_set(td->td_proc, RACCT_MEMLOCK,
1096 		    ptoa(pmap_wired_count(map->pmap)));
1097 		PROC_UNLOCK(td->td_proc);
1098 	}
1099 #endif
1100 
1101 	return (error);
1102 }
1103 
1104 #ifndef _SYS_SYSPROTO_H_
1105 struct munlockall_args {
1106 	register_t dummy;
1107 };
1108 #endif
1109 
1110 int
1111 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1112 {
1113 	vm_map_t map;
1114 	int error;
1115 
1116 	map = &td->td_proc->p_vmspace->vm_map;
1117 	error = priv_check(td, PRIV_VM_MUNLOCK);
1118 	if (error)
1119 		return (error);
1120 
1121 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1122 	vm_map_lock(map);
1123 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1124 	vm_map_unlock(map);
1125 
1126 	/* Forcibly unwire all pages. */
1127 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1128 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1129 #ifdef RACCT
1130 	if (racct_enable && error == KERN_SUCCESS) {
1131 		PROC_LOCK(td->td_proc);
1132 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1133 		PROC_UNLOCK(td->td_proc);
1134 	}
1135 #endif
1136 
1137 	return (error);
1138 }
1139 
1140 #ifndef _SYS_SYSPROTO_H_
1141 struct munlock_args {
1142 	const void *addr;
1143 	size_t len;
1144 };
1145 #endif
1146 int
1147 sys_munlock(struct thread *td, struct munlock_args *uap)
1148 {
1149 
1150 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1151 }
1152 
1153 int
1154 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1155 {
1156 	vm_offset_t addr, end, last, start;
1157 #ifdef RACCT
1158 	vm_map_t map;
1159 #endif
1160 	int error;
1161 
1162 	error = priv_check(td, PRIV_VM_MUNLOCK);
1163 	if (error)
1164 		return (error);
1165 	addr = addr0;
1166 	last = addr + size;
1167 	start = trunc_page(addr);
1168 	end = round_page(last);
1169 	if (last < addr || end < addr)
1170 		return (EINVAL);
1171 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1172 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1173 #ifdef RACCT
1174 	if (racct_enable && error == KERN_SUCCESS) {
1175 		PROC_LOCK(td->td_proc);
1176 		map = &td->td_proc->p_vmspace->vm_map;
1177 		racct_set(td->td_proc, RACCT_MEMLOCK,
1178 		    ptoa(pmap_wired_count(map->pmap)));
1179 		PROC_UNLOCK(td->td_proc);
1180 	}
1181 #endif
1182 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1183 }
1184 
1185 /*
1186  * vm_mmap_vnode()
1187  *
1188  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1189  * operations on vnodes.
1190  */
1191 int
1192 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1193     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1194     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1195     boolean_t *writecounted)
1196 {
1197 	struct vattr va;
1198 	vm_object_t obj;
1199 	vm_offset_t foff;
1200 	struct ucred *cred;
1201 	int error, flags, locktype;
1202 
1203 	cred = td->td_ucred;
1204 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1205 		locktype = LK_EXCLUSIVE;
1206 	else
1207 		locktype = LK_SHARED;
1208 	if ((error = vget(vp, locktype, td)) != 0)
1209 		return (error);
1210 	AUDIT_ARG_VNODE1(vp);
1211 	foff = *foffp;
1212 	flags = *flagsp;
1213 	obj = vp->v_object;
1214 	if (vp->v_type == VREG) {
1215 		/*
1216 		 * Get the proper underlying object
1217 		 */
1218 		if (obj == NULL) {
1219 			error = EINVAL;
1220 			goto done;
1221 		}
1222 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1223 			vput(vp);
1224 			vp = (struct vnode *)obj->handle;
1225 			/*
1226 			 * Bypass filesystems obey the mpsafety of the
1227 			 * underlying fs.  Tmpfs never bypasses.
1228 			 */
1229 			error = vget(vp, locktype, td);
1230 			if (error != 0)
1231 				return (error);
1232 		}
1233 		if (locktype == LK_EXCLUSIVE) {
1234 			*writecounted = TRUE;
1235 			vnode_pager_update_writecount(obj, 0, objsize);
1236 		}
1237 	} else {
1238 		error = EINVAL;
1239 		goto done;
1240 	}
1241 	if ((error = VOP_GETATTR(vp, &va, cred)))
1242 		goto done;
1243 #ifdef MAC
1244 	/* This relies on VM_PROT_* matching PROT_*. */
1245 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1246 	if (error != 0)
1247 		goto done;
1248 #endif
1249 	if ((flags & MAP_SHARED) != 0) {
1250 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1251 			if (prot & VM_PROT_WRITE) {
1252 				error = EPERM;
1253 				goto done;
1254 			}
1255 			*maxprotp &= ~VM_PROT_WRITE;
1256 		}
1257 	}
1258 	/*
1259 	 * If it is a regular file without any references
1260 	 * we do not need to sync it.
1261 	 * Adjust object size to be the size of actual file.
1262 	 */
1263 	objsize = round_page(va.va_size);
1264 	if (va.va_nlink == 0)
1265 		flags |= MAP_NOSYNC;
1266 	if (obj->type == OBJT_VNODE) {
1267 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1268 		    cred);
1269 		if (obj == NULL) {
1270 			error = ENOMEM;
1271 			goto done;
1272 		}
1273 	} else {
1274 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1275 		    ("wrong object type"));
1276 		VM_OBJECT_WLOCK(obj);
1277 		vm_object_reference_locked(obj);
1278 #if VM_NRESERVLEVEL > 0
1279 		vm_object_color(obj, 0);
1280 #endif
1281 		VM_OBJECT_WUNLOCK(obj);
1282 	}
1283 	*objp = obj;
1284 	*flagsp = flags;
1285 
1286 	vfs_mark_atime(vp, cred);
1287 
1288 done:
1289 	if (error != 0 && *writecounted) {
1290 		*writecounted = FALSE;
1291 		vnode_pager_update_writecount(obj, objsize, 0);
1292 	}
1293 	vput(vp);
1294 	return (error);
1295 }
1296 
1297 /*
1298  * vm_mmap_cdev()
1299  *
1300  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1301  * operations on cdevs.
1302  */
1303 int
1304 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1305     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1306     vm_ooffset_t *foff, vm_object_t *objp)
1307 {
1308 	vm_object_t obj;
1309 	int error, flags;
1310 
1311 	flags = *flagsp;
1312 
1313 	if (dsw->d_flags & D_MMAP_ANON) {
1314 		*objp = NULL;
1315 		*foff = 0;
1316 		*maxprotp = VM_PROT_ALL;
1317 		*flagsp |= MAP_ANON;
1318 		return (0);
1319 	}
1320 	/*
1321 	 * cdevs do not provide private mappings of any kind.
1322 	 */
1323 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1324 	    (prot & VM_PROT_WRITE) != 0)
1325 		return (EACCES);
1326 	if (flags & (MAP_PRIVATE|MAP_COPY))
1327 		return (EINVAL);
1328 	/*
1329 	 * Force device mappings to be shared.
1330 	 */
1331 	flags |= MAP_SHARED;
1332 #ifdef MAC_XXX
1333 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1334 	if (error != 0)
1335 		return (error);
1336 #endif
1337 	/*
1338 	 * First, try d_mmap_single().  If that is not implemented
1339 	 * (returns ENODEV), fall back to using the device pager.
1340 	 * Note that d_mmap_single() must return a reference to the
1341 	 * object (it needs to bump the reference count of the object
1342 	 * it returns somehow).
1343 	 *
1344 	 * XXX assumes VM_PROT_* == PROT_*
1345 	 */
1346 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1347 	if (error != ENODEV)
1348 		return (error);
1349 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1350 	    td->td_ucred);
1351 	if (obj == NULL)
1352 		return (EINVAL);
1353 	*objp = obj;
1354 	*flagsp = flags;
1355 	return (0);
1356 }
1357 
1358 /*
1359  * vm_mmap()
1360  *
1361  * Internal version of mmap used by exec, sys5 shared memory, and
1362  * various device drivers.  Handle is either a vnode pointer, a
1363  * character device, or NULL for MAP_ANON.
1364  */
1365 int
1366 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1367 	vm_prot_t maxprot, int flags,
1368 	objtype_t handle_type, void *handle,
1369 	vm_ooffset_t foff)
1370 {
1371 	vm_object_t object;
1372 	struct thread *td = curthread;
1373 	int error;
1374 	boolean_t writecounted;
1375 
1376 	if (size == 0)
1377 		return (EINVAL);
1378 
1379 	size = round_page(size);
1380 	object = NULL;
1381 	writecounted = FALSE;
1382 
1383 	/*
1384 	 * Lookup/allocate object.
1385 	 */
1386 	switch (handle_type) {
1387 	case OBJT_DEVICE: {
1388 		struct cdevsw *dsw;
1389 		struct cdev *cdev;
1390 		int ref;
1391 
1392 		cdev = handle;
1393 		dsw = dev_refthread(cdev, &ref);
1394 		if (dsw == NULL)
1395 			return (ENXIO);
1396 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1397 		    dsw, &foff, &object);
1398 		dev_relthread(cdev, ref);
1399 		break;
1400 	}
1401 	case OBJT_VNODE:
1402 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1403 		    handle, &foff, &object, &writecounted);
1404 		break;
1405 	case OBJT_DEFAULT:
1406 		if (handle == NULL) {
1407 			error = 0;
1408 			break;
1409 		}
1410 		/* FALLTHROUGH */
1411 	default:
1412 		error = EINVAL;
1413 		break;
1414 	}
1415 	if (error)
1416 		return (error);
1417 
1418 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1419 	    foff, writecounted, td);
1420 	if (error != 0 && object != NULL) {
1421 		/*
1422 		 * If this mapping was accounted for in the vnode's
1423 		 * writecount, then undo that now.
1424 		 */
1425 		if (writecounted)
1426 			vnode_pager_release_writecount(object, 0, size);
1427 		vm_object_deallocate(object);
1428 	}
1429 	return (error);
1430 }
1431 
1432 /*
1433  * Internal version of mmap that maps a specific VM object into an
1434  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1435  */
1436 int
1437 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1438     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1439     boolean_t writecounted, struct thread *td)
1440 {
1441 	boolean_t curmap, fitit;
1442 	vm_offset_t max_addr;
1443 	int docow, error, findspace, rv;
1444 
1445 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1446 	if (curmap) {
1447 		PROC_LOCK(td->td_proc);
1448 		if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1449 			PROC_UNLOCK(td->td_proc);
1450 			return (ENOMEM);
1451 		}
1452 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1453 			PROC_UNLOCK(td->td_proc);
1454 			return (ENOMEM);
1455 		}
1456 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1457 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1458 			    lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1459 				racct_set_force(td->td_proc, RACCT_VMEM,
1460 				    map->size);
1461 				PROC_UNLOCK(td->td_proc);
1462 				return (ENOMEM);
1463 			}
1464 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1465 			    ptoa(pmap_wired_count(map->pmap)) + size);
1466 			if (error != 0) {
1467 				racct_set_force(td->td_proc, RACCT_VMEM,
1468 				    map->size);
1469 				PROC_UNLOCK(td->td_proc);
1470 				return (error);
1471 			}
1472 		}
1473 		PROC_UNLOCK(td->td_proc);
1474 	}
1475 
1476 	/*
1477 	 * We currently can only deal with page aligned file offsets.
1478 	 * The mmap() system call already enforces this by subtracting
1479 	 * the page offset from the file offset, but checking here
1480 	 * catches errors in device drivers (e.g. d_single_mmap()
1481 	 * callbacks) and other internal mapping requests (such as in
1482 	 * exec).
1483 	 */
1484 	if (foff & PAGE_MASK)
1485 		return (EINVAL);
1486 
1487 	if ((flags & MAP_FIXED) == 0) {
1488 		fitit = TRUE;
1489 		*addr = round_page(*addr);
1490 	} else {
1491 		if (*addr != trunc_page(*addr))
1492 			return (EINVAL);
1493 		fitit = FALSE;
1494 	}
1495 
1496 	if (flags & MAP_ANON) {
1497 		if (object != NULL || foff != 0)
1498 			return (EINVAL);
1499 		docow = 0;
1500 	} else if (flags & MAP_PREFAULT_READ)
1501 		docow = MAP_PREFAULT;
1502 	else
1503 		docow = MAP_PREFAULT_PARTIAL;
1504 
1505 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1506 		docow |= MAP_COPY_ON_WRITE;
1507 	if (flags & MAP_NOSYNC)
1508 		docow |= MAP_DISABLE_SYNCER;
1509 	if (flags & MAP_NOCORE)
1510 		docow |= MAP_DISABLE_COREDUMP;
1511 	/* Shared memory is also shared with children. */
1512 	if (flags & MAP_SHARED)
1513 		docow |= MAP_INHERIT_SHARE;
1514 	if (writecounted)
1515 		docow |= MAP_VN_WRITECOUNT;
1516 	if (flags & MAP_STACK) {
1517 		if (object != NULL)
1518 			return (EINVAL);
1519 		docow |= MAP_STACK_GROWS_DOWN;
1520 	}
1521 	if ((flags & MAP_EXCL) != 0)
1522 		docow |= MAP_CHECK_EXCL;
1523 	if ((flags & MAP_GUARD) != 0)
1524 		docow |= MAP_CREATE_GUARD;
1525 
1526 	if (fitit) {
1527 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1528 			findspace = VMFS_SUPER_SPACE;
1529 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1530 			findspace = VMFS_ALIGNED_SPACE(flags >>
1531 			    MAP_ALIGNMENT_SHIFT);
1532 		else
1533 			findspace = VMFS_OPTIMAL_SPACE;
1534 		max_addr = 0;
1535 #ifdef MAP_32BIT
1536 		if ((flags & MAP_32BIT) != 0)
1537 			max_addr = MAP_32BIT_MAX_ADDR;
1538 #endif
1539 		if (curmap) {
1540 			rv = vm_map_find_min(map, object, foff, addr, size,
1541 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1542 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1543 			    findspace, prot, maxprot, docow);
1544 		} else {
1545 			rv = vm_map_find(map, object, foff, addr, size,
1546 			    max_addr, findspace, prot, maxprot, docow);
1547 		}
1548 	} else {
1549 		rv = vm_map_fixed(map, object, foff, *addr, size,
1550 		    prot, maxprot, docow);
1551 	}
1552 
1553 	if (rv == KERN_SUCCESS) {
1554 		/*
1555 		 * If the process has requested that all future mappings
1556 		 * be wired, then heed this.
1557 		 */
1558 		if (map->flags & MAP_WIREFUTURE) {
1559 			vm_map_wire(map, *addr, *addr + size,
1560 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1561 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1562 		}
1563 	}
1564 	return (vm_mmap_to_errno(rv));
1565 }
1566 
1567 /*
1568  * Translate a Mach VM return code to zero on success or the appropriate errno
1569  * on failure.
1570  */
1571 int
1572 vm_mmap_to_errno(int rv)
1573 {
1574 
1575 	switch (rv) {
1576 	case KERN_SUCCESS:
1577 		return (0);
1578 	case KERN_INVALID_ADDRESS:
1579 	case KERN_NO_SPACE:
1580 		return (ENOMEM);
1581 	case KERN_PROTECTION_FAILURE:
1582 		return (EACCES);
1583 	default:
1584 		return (EINVAL);
1585 	}
1586 }
1587