xref: /freebsd/sys/vm/vm_mmap.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
1 /*-
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35  *
36  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
37  */
38 
39 /*
40  * Mapped file (mmap) interface to VM
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capsicum.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/procctl.h>
60 #include <sys/racct.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/rwlock.h>
64 #include <sys/sysctl.h>
65 #include <sys/vnode.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/mman.h>
69 #include <sys/mount.h>
70 #include <sys/conf.h>
71 #include <sys/stat.h>
72 #include <sys/syscallsubr.h>
73 #include <sys/sysent.h>
74 #include <sys/vmmeter.h>
75 
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pager.h>
85 #include <vm/vm_pageout.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_page.h>
88 #include <vm/vnode_pager.h>
89 
90 #ifdef HWPMC_HOOKS
91 #include <sys/pmckern.h>
92 #endif
93 
94 int old_mlock = 0;
95 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
96     "Do not apply RLIMIT_MEMLOCK on mlockall");
97 
98 #ifdef MAP_32BIT
99 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
100 #endif
101 
102 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
103     int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
104 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
105     int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
106 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
107     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
108 
109 #ifndef _SYS_SYSPROTO_H_
110 struct sbrk_args {
111 	int incr;
112 };
113 #endif
114 
115 /*
116  * MPSAFE
117  */
118 /* ARGSUSED */
119 int
120 sys_sbrk(td, uap)
121 	struct thread *td;
122 	struct sbrk_args *uap;
123 {
124 	/* Not yet implemented */
125 	return (EOPNOTSUPP);
126 }
127 
128 #ifndef _SYS_SYSPROTO_H_
129 struct sstk_args {
130 	int incr;
131 };
132 #endif
133 
134 /*
135  * MPSAFE
136  */
137 /* ARGSUSED */
138 int
139 sys_sstk(td, uap)
140 	struct thread *td;
141 	struct sstk_args *uap;
142 {
143 	/* Not yet implemented */
144 	return (EOPNOTSUPP);
145 }
146 
147 #if defined(COMPAT_43)
148 #ifndef _SYS_SYSPROTO_H_
149 struct getpagesize_args {
150 	int dummy;
151 };
152 #endif
153 
154 int
155 ogetpagesize(td, uap)
156 	struct thread *td;
157 	struct getpagesize_args *uap;
158 {
159 	/* MP SAFE */
160 	td->td_retval[0] = PAGE_SIZE;
161 	return (0);
162 }
163 #endif				/* COMPAT_43 */
164 
165 
166 /*
167  * Memory Map (mmap) system call.  Note that the file offset
168  * and address are allowed to be NOT page aligned, though if
169  * the MAP_FIXED flag it set, both must have the same remainder
170  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
171  * page-aligned, the actual mapping starts at trunc_page(addr)
172  * and the return value is adjusted up by the page offset.
173  *
174  * Generally speaking, only character devices which are themselves
175  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
176  * there would be no cache coherency between a descriptor and a VM mapping
177  * both to the same character device.
178  */
179 #ifndef _SYS_SYSPROTO_H_
180 struct mmap_args {
181 	void *addr;
182 	size_t len;
183 	int prot;
184 	int flags;
185 	int fd;
186 	long pad;
187 	off_t pos;
188 };
189 #endif
190 
191 /*
192  * MPSAFE
193  */
194 int
195 sys_mmap(td, uap)
196 	struct thread *td;
197 	struct mmap_args *uap;
198 {
199 #ifdef HWPMC_HOOKS
200 	struct pmckern_map_in pkm;
201 #endif
202 	struct file *fp;
203 	struct vnode *vp;
204 	vm_offset_t addr;
205 	vm_size_t size, pageoff;
206 	vm_prot_t cap_maxprot, maxprot;
207 	void *handle;
208 	objtype_t handle_type;
209 	int align, error, flags, prot;
210 	off_t pos;
211 	struct vmspace *vms = td->td_proc->p_vmspace;
212 	cap_rights_t rights;
213 
214 	addr = (vm_offset_t) uap->addr;
215 	size = uap->len;
216 	prot = uap->prot;
217 	flags = uap->flags;
218 	pos = uap->pos;
219 
220 	fp = NULL;
221 
222 	/*
223 	 * Ignore old flags that used to be defined but did not do anything.
224 	 */
225 	if (td->td_proc->p_osrel < P_OSREL_MAP_RENAME)
226 		flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
227 
228 	/*
229 	 * Enforce the constraints.
230 	 * Mapping of length 0 is only allowed for old binaries.
231 	 * Anonymous mapping shall specify -1 as filedescriptor and
232 	 * zero position for new code. Be nice to ancient a.out
233 	 * binaries and correct pos for anonymous mapping, since old
234 	 * ld.so sometimes issues anonymous map requests with non-zero
235 	 * pos.
236 	 */
237 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
238 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
239 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
240 			return (EINVAL);
241 	} else {
242 		if ((flags & MAP_ANON) != 0)
243 			pos = 0;
244 	}
245 
246 	if (flags & MAP_STACK) {
247 		if ((uap->fd != -1) ||
248 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
249 			return (EINVAL);
250 		flags |= MAP_ANON;
251 		pos = 0;
252 	}
253 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
254 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
255 	    MAP_PREFAULT_READ |
256 #ifdef MAP_32BIT
257 	    MAP_32BIT |
258 #endif
259 	    MAP_ALIGNMENT_MASK)) != 0)
260 		return (EINVAL);
261 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
262 		return (EINVAL);
263 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
264 		return (EINVAL);
265 	if (prot != PROT_NONE &&
266 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
267 		return (EINVAL);
268 
269 	/*
270 	 * Align the file position to a page boundary,
271 	 * and save its page offset component.
272 	 */
273 	pageoff = (pos & PAGE_MASK);
274 	pos -= pageoff;
275 
276 	/* Adjust size for rounding (on both ends). */
277 	size += pageoff;			/* low end... */
278 	size = (vm_size_t) round_page(size);	/* hi end */
279 
280 	/* Ensure alignment is at least a page and fits in a pointer. */
281 	align = flags & MAP_ALIGNMENT_MASK;
282 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
283 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
284 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
285 		return (EINVAL);
286 
287 	/*
288 	 * Check for illegal addresses.  Watch out for address wrap... Note
289 	 * that VM_*_ADDRESS are not constants due to casts (argh).
290 	 */
291 	if (flags & MAP_FIXED) {
292 		/*
293 		 * The specified address must have the same remainder
294 		 * as the file offset taken modulo PAGE_SIZE, so it
295 		 * should be aligned after adjustment by pageoff.
296 		 */
297 		addr -= pageoff;
298 		if (addr & PAGE_MASK)
299 			return (EINVAL);
300 
301 		/* Address range must be all in user VM space. */
302 		if (addr < vm_map_min(&vms->vm_map) ||
303 		    addr + size > vm_map_max(&vms->vm_map))
304 			return (EINVAL);
305 		if (addr + size < addr)
306 			return (EINVAL);
307 #ifdef MAP_32BIT
308 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
309 			return (EINVAL);
310 	} else if (flags & MAP_32BIT) {
311 		/*
312 		 * For MAP_32BIT, override the hint if it is too high and
313 		 * do not bother moving the mapping past the heap (since
314 		 * the heap is usually above 2GB).
315 		 */
316 		if (addr + size > MAP_32BIT_MAX_ADDR)
317 			addr = 0;
318 #endif
319 	} else {
320 		/*
321 		 * XXX for non-fixed mappings where no hint is provided or
322 		 * the hint would fall in the potential heap space,
323 		 * place it after the end of the largest possible heap.
324 		 *
325 		 * There should really be a pmap call to determine a reasonable
326 		 * location.
327 		 */
328 		PROC_LOCK(td->td_proc);
329 		if (addr == 0 ||
330 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
331 		    addr < round_page((vm_offset_t)vms->vm_daddr +
332 		    lim_max(td->td_proc, RLIMIT_DATA))))
333 			addr = round_page((vm_offset_t)vms->vm_daddr +
334 			    lim_max(td->td_proc, RLIMIT_DATA));
335 		PROC_UNLOCK(td->td_proc);
336 	}
337 	if (flags & MAP_ANON) {
338 		/*
339 		 * Mapping blank space is trivial.
340 		 */
341 		handle = NULL;
342 		handle_type = OBJT_DEFAULT;
343 		maxprot = VM_PROT_ALL;
344 		cap_maxprot = VM_PROT_ALL;
345 	} else {
346 		/*
347 		 * Mapping file, get fp for validation and don't let the
348 		 * descriptor disappear on us if we block. Check capability
349 		 * rights, but also return the maximum rights to be combined
350 		 * with maxprot later.
351 		 */
352 		cap_rights_init(&rights, CAP_MMAP);
353 		if (prot & PROT_READ)
354 			cap_rights_set(&rights, CAP_MMAP_R);
355 		if ((flags & MAP_SHARED) != 0) {
356 			if (prot & PROT_WRITE)
357 				cap_rights_set(&rights, CAP_MMAP_W);
358 		}
359 		if (prot & PROT_EXEC)
360 			cap_rights_set(&rights, CAP_MMAP_X);
361 		error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
362 		if (error != 0)
363 			goto done;
364 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
365 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
366 			error = EINVAL;
367 			goto done;
368 		}
369 		if (fp->f_type == DTYPE_SHM) {
370 			handle = fp->f_data;
371 			handle_type = OBJT_SWAP;
372 			maxprot = VM_PROT_NONE;
373 
374 			/* FREAD should always be set. */
375 			if (fp->f_flag & FREAD)
376 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
377 			if (fp->f_flag & FWRITE)
378 				maxprot |= VM_PROT_WRITE;
379 			goto map;
380 		}
381 		if (fp->f_type != DTYPE_VNODE) {
382 			error = ENODEV;
383 			goto done;
384 		}
385 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
386     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
387 		/*
388 		 * POSIX shared-memory objects are defined to have
389 		 * kernel persistence, and are not defined to support
390 		 * read(2)/write(2) -- or even open(2).  Thus, we can
391 		 * use MAP_ASYNC to trade on-disk coherence for speed.
392 		 * The shm_open(3) library routine turns on the FPOSIXSHM
393 		 * flag to request this behavior.
394 		 */
395 		if (fp->f_flag & FPOSIXSHM)
396 			flags |= MAP_NOSYNC;
397 #endif
398 		vp = fp->f_vnode;
399 		/*
400 		 * Ensure that file and memory protections are
401 		 * compatible.  Note that we only worry about
402 		 * writability if mapping is shared; in this case,
403 		 * current and max prot are dictated by the open file.
404 		 * XXX use the vnode instead?  Problem is: what
405 		 * credentials do we use for determination? What if
406 		 * proc does a setuid?
407 		 */
408 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
409 			maxprot = VM_PROT_NONE;
410 		else
411 			maxprot = VM_PROT_EXECUTE;
412 		if (fp->f_flag & FREAD) {
413 			maxprot |= VM_PROT_READ;
414 		} else if (prot & PROT_READ) {
415 			error = EACCES;
416 			goto done;
417 		}
418 		/*
419 		 * If we are sharing potential changes (either via
420 		 * MAP_SHARED or via the implicit sharing of character
421 		 * device mappings), and we are trying to get write
422 		 * permission although we opened it without asking
423 		 * for it, bail out.
424 		 */
425 		if ((flags & MAP_SHARED) != 0) {
426 			if ((fp->f_flag & FWRITE) != 0) {
427 				maxprot |= VM_PROT_WRITE;
428 			} else if ((prot & PROT_WRITE) != 0) {
429 				error = EACCES;
430 				goto done;
431 			}
432 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
433 			maxprot |= VM_PROT_WRITE;
434 			cap_maxprot |= VM_PROT_WRITE;
435 		}
436 		handle = (void *)vp;
437 		handle_type = OBJT_VNODE;
438 	}
439 map:
440 	td->td_fpop = fp;
441 	maxprot &= cap_maxprot;
442 
443 	/* This relies on VM_PROT_* matching PROT_*. */
444 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
445 	    flags, handle_type, handle, pos);
446 	td->td_fpop = NULL;
447 #ifdef HWPMC_HOOKS
448 	/* inform hwpmc(4) if an executable is being mapped */
449 	if (error == 0 && handle_type == OBJT_VNODE &&
450 	    (prot & PROT_EXEC)) {
451 		pkm.pm_file = handle;
452 		pkm.pm_address = (uintptr_t) addr;
453 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
454 	}
455 #endif
456 	if (error == 0)
457 		td->td_retval[0] = (register_t) (addr + pageoff);
458 done:
459 	if (fp)
460 		fdrop(fp, td);
461 
462 	return (error);
463 }
464 
465 int
466 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
467 {
468 	struct mmap_args oargs;
469 
470 	oargs.addr = uap->addr;
471 	oargs.len = uap->len;
472 	oargs.prot = uap->prot;
473 	oargs.flags = uap->flags;
474 	oargs.fd = uap->fd;
475 	oargs.pos = uap->pos;
476 	return (sys_mmap(td, &oargs));
477 }
478 
479 #ifdef COMPAT_43
480 #ifndef _SYS_SYSPROTO_H_
481 struct ommap_args {
482 	caddr_t addr;
483 	int len;
484 	int prot;
485 	int flags;
486 	int fd;
487 	long pos;
488 };
489 #endif
490 int
491 ommap(td, uap)
492 	struct thread *td;
493 	struct ommap_args *uap;
494 {
495 	struct mmap_args nargs;
496 	static const char cvtbsdprot[8] = {
497 		0,
498 		PROT_EXEC,
499 		PROT_WRITE,
500 		PROT_EXEC | PROT_WRITE,
501 		PROT_READ,
502 		PROT_EXEC | PROT_READ,
503 		PROT_WRITE | PROT_READ,
504 		PROT_EXEC | PROT_WRITE | PROT_READ,
505 	};
506 
507 #define	OMAP_ANON	0x0002
508 #define	OMAP_COPY	0x0020
509 #define	OMAP_SHARED	0x0010
510 #define	OMAP_FIXED	0x0100
511 
512 	nargs.addr = uap->addr;
513 	nargs.len = uap->len;
514 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
515 #ifdef COMPAT_FREEBSD32
516 #if defined(__amd64__)
517 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
518 	    nargs.prot != 0)
519 		nargs.prot |= PROT_EXEC;
520 #endif
521 #endif
522 	nargs.flags = 0;
523 	if (uap->flags & OMAP_ANON)
524 		nargs.flags |= MAP_ANON;
525 	if (uap->flags & OMAP_COPY)
526 		nargs.flags |= MAP_COPY;
527 	if (uap->flags & OMAP_SHARED)
528 		nargs.flags |= MAP_SHARED;
529 	else
530 		nargs.flags |= MAP_PRIVATE;
531 	if (uap->flags & OMAP_FIXED)
532 		nargs.flags |= MAP_FIXED;
533 	nargs.fd = uap->fd;
534 	nargs.pos = uap->pos;
535 	return (sys_mmap(td, &nargs));
536 }
537 #endif				/* COMPAT_43 */
538 
539 
540 #ifndef _SYS_SYSPROTO_H_
541 struct msync_args {
542 	void *addr;
543 	size_t len;
544 	int flags;
545 };
546 #endif
547 /*
548  * MPSAFE
549  */
550 int
551 sys_msync(td, uap)
552 	struct thread *td;
553 	struct msync_args *uap;
554 {
555 	vm_offset_t addr;
556 	vm_size_t size, pageoff;
557 	int flags;
558 	vm_map_t map;
559 	int rv;
560 
561 	addr = (vm_offset_t) uap->addr;
562 	size = uap->len;
563 	flags = uap->flags;
564 
565 	pageoff = (addr & PAGE_MASK);
566 	addr -= pageoff;
567 	size += pageoff;
568 	size = (vm_size_t) round_page(size);
569 	if (addr + size < addr)
570 		return (EINVAL);
571 
572 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
573 		return (EINVAL);
574 
575 	map = &td->td_proc->p_vmspace->vm_map;
576 
577 	/*
578 	 * Clean the pages and interpret the return value.
579 	 */
580 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
581 	    (flags & MS_INVALIDATE) != 0);
582 	switch (rv) {
583 	case KERN_SUCCESS:
584 		return (0);
585 	case KERN_INVALID_ADDRESS:
586 		return (ENOMEM);
587 	case KERN_INVALID_ARGUMENT:
588 		return (EBUSY);
589 	case KERN_FAILURE:
590 		return (EIO);
591 	default:
592 		return (EINVAL);
593 	}
594 }
595 
596 #ifndef _SYS_SYSPROTO_H_
597 struct munmap_args {
598 	void *addr;
599 	size_t len;
600 };
601 #endif
602 /*
603  * MPSAFE
604  */
605 int
606 sys_munmap(td, uap)
607 	struct thread *td;
608 	struct munmap_args *uap;
609 {
610 #ifdef HWPMC_HOOKS
611 	struct pmckern_map_out pkm;
612 	vm_map_entry_t entry;
613 #endif
614 	vm_offset_t addr;
615 	vm_size_t size, pageoff;
616 	vm_map_t map;
617 
618 	addr = (vm_offset_t) uap->addr;
619 	size = uap->len;
620 	if (size == 0)
621 		return (EINVAL);
622 
623 	pageoff = (addr & PAGE_MASK);
624 	addr -= pageoff;
625 	size += pageoff;
626 	size = (vm_size_t) round_page(size);
627 	if (addr + size < addr)
628 		return (EINVAL);
629 
630 	/*
631 	 * Check for illegal addresses.  Watch out for address wrap...
632 	 */
633 	map = &td->td_proc->p_vmspace->vm_map;
634 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
635 		return (EINVAL);
636 	vm_map_lock(map);
637 #ifdef HWPMC_HOOKS
638 	/*
639 	 * Inform hwpmc if the address range being unmapped contains
640 	 * an executable region.
641 	 */
642 	pkm.pm_address = (uintptr_t) NULL;
643 	if (vm_map_lookup_entry(map, addr, &entry)) {
644 		for (;
645 		     entry != &map->header && entry->start < addr + size;
646 		     entry = entry->next) {
647 			if (vm_map_check_protection(map, entry->start,
648 				entry->end, VM_PROT_EXECUTE) == TRUE) {
649 				pkm.pm_address = (uintptr_t) addr;
650 				pkm.pm_size = (size_t) size;
651 				break;
652 			}
653 		}
654 	}
655 #endif
656 	vm_map_delete(map, addr, addr + size);
657 
658 #ifdef HWPMC_HOOKS
659 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
660 	vm_map_lock_downgrade(map);
661 	if (pkm.pm_address != (uintptr_t) NULL)
662 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
663 	vm_map_unlock_read(map);
664 #else
665 	vm_map_unlock(map);
666 #endif
667 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
668 	return (0);
669 }
670 
671 #ifndef _SYS_SYSPROTO_H_
672 struct mprotect_args {
673 	const void *addr;
674 	size_t len;
675 	int prot;
676 };
677 #endif
678 /*
679  * MPSAFE
680  */
681 int
682 sys_mprotect(td, uap)
683 	struct thread *td;
684 	struct mprotect_args *uap;
685 {
686 	vm_offset_t addr;
687 	vm_size_t size, pageoff;
688 	vm_prot_t prot;
689 
690 	addr = (vm_offset_t) uap->addr;
691 	size = uap->len;
692 	prot = uap->prot & VM_PROT_ALL;
693 
694 	pageoff = (addr & PAGE_MASK);
695 	addr -= pageoff;
696 	size += pageoff;
697 	size = (vm_size_t) round_page(size);
698 	if (addr + size < addr)
699 		return (EINVAL);
700 
701 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
702 	    addr + size, prot, FALSE)) {
703 	case KERN_SUCCESS:
704 		return (0);
705 	case KERN_PROTECTION_FAILURE:
706 		return (EACCES);
707 	case KERN_RESOURCE_SHORTAGE:
708 		return (ENOMEM);
709 	}
710 	return (EINVAL);
711 }
712 
713 #ifndef _SYS_SYSPROTO_H_
714 struct minherit_args {
715 	void *addr;
716 	size_t len;
717 	int inherit;
718 };
719 #endif
720 /*
721  * MPSAFE
722  */
723 int
724 sys_minherit(td, uap)
725 	struct thread *td;
726 	struct minherit_args *uap;
727 {
728 	vm_offset_t addr;
729 	vm_size_t size, pageoff;
730 	vm_inherit_t inherit;
731 
732 	addr = (vm_offset_t)uap->addr;
733 	size = uap->len;
734 	inherit = uap->inherit;
735 
736 	pageoff = (addr & PAGE_MASK);
737 	addr -= pageoff;
738 	size += pageoff;
739 	size = (vm_size_t) round_page(size);
740 	if (addr + size < addr)
741 		return (EINVAL);
742 
743 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
744 	    addr + size, inherit)) {
745 	case KERN_SUCCESS:
746 		return (0);
747 	case KERN_PROTECTION_FAILURE:
748 		return (EACCES);
749 	}
750 	return (EINVAL);
751 }
752 
753 #ifndef _SYS_SYSPROTO_H_
754 struct madvise_args {
755 	void *addr;
756 	size_t len;
757 	int behav;
758 };
759 #endif
760 
761 /*
762  * MPSAFE
763  */
764 int
765 sys_madvise(td, uap)
766 	struct thread *td;
767 	struct madvise_args *uap;
768 {
769 	vm_offset_t start, end;
770 	vm_map_t map;
771 	int flags;
772 
773 	/*
774 	 * Check for our special case, advising the swap pager we are
775 	 * "immortal."
776 	 */
777 	if (uap->behav == MADV_PROTECT) {
778 		flags = PPROT_SET;
779 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
780 		    PROC_SPROTECT, &flags));
781 	}
782 
783 	/*
784 	 * Check for illegal behavior
785 	 */
786 	if (uap->behav < 0 || uap->behav > MADV_CORE)
787 		return (EINVAL);
788 	/*
789 	 * Check for illegal addresses.  Watch out for address wrap... Note
790 	 * that VM_*_ADDRESS are not constants due to casts (argh).
791 	 */
792 	map = &td->td_proc->p_vmspace->vm_map;
793 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
794 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
795 		return (EINVAL);
796 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
797 		return (EINVAL);
798 
799 	/*
800 	 * Since this routine is only advisory, we default to conservative
801 	 * behavior.
802 	 */
803 	start = trunc_page((vm_offset_t) uap->addr);
804 	end = round_page((vm_offset_t) uap->addr + uap->len);
805 
806 	if (vm_map_madvise(map, start, end, uap->behav))
807 		return (EINVAL);
808 	return (0);
809 }
810 
811 #ifndef _SYS_SYSPROTO_H_
812 struct mincore_args {
813 	const void *addr;
814 	size_t len;
815 	char *vec;
816 };
817 #endif
818 
819 /*
820  * MPSAFE
821  */
822 int
823 sys_mincore(td, uap)
824 	struct thread *td;
825 	struct mincore_args *uap;
826 {
827 	vm_offset_t addr, first_addr;
828 	vm_offset_t end, cend;
829 	pmap_t pmap;
830 	vm_map_t map;
831 	char *vec;
832 	int error = 0;
833 	int vecindex, lastvecindex;
834 	vm_map_entry_t current;
835 	vm_map_entry_t entry;
836 	vm_object_t object;
837 	vm_paddr_t locked_pa;
838 	vm_page_t m;
839 	vm_pindex_t pindex;
840 	int mincoreinfo;
841 	unsigned int timestamp;
842 	boolean_t locked;
843 
844 	/*
845 	 * Make sure that the addresses presented are valid for user
846 	 * mode.
847 	 */
848 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
849 	end = addr + (vm_size_t)round_page(uap->len);
850 	map = &td->td_proc->p_vmspace->vm_map;
851 	if (end > vm_map_max(map) || end < addr)
852 		return (ENOMEM);
853 
854 	/*
855 	 * Address of byte vector
856 	 */
857 	vec = uap->vec;
858 
859 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
860 
861 	vm_map_lock_read(map);
862 RestartScan:
863 	timestamp = map->timestamp;
864 
865 	if (!vm_map_lookup_entry(map, addr, &entry)) {
866 		vm_map_unlock_read(map);
867 		return (ENOMEM);
868 	}
869 
870 	/*
871 	 * Do this on a map entry basis so that if the pages are not
872 	 * in the current processes address space, we can easily look
873 	 * up the pages elsewhere.
874 	 */
875 	lastvecindex = -1;
876 	for (current = entry;
877 	    (current != &map->header) && (current->start < end);
878 	    current = current->next) {
879 
880 		/*
881 		 * check for contiguity
882 		 */
883 		if (current->end < end &&
884 		    (entry->next == &map->header ||
885 		     current->next->start > current->end)) {
886 			vm_map_unlock_read(map);
887 			return (ENOMEM);
888 		}
889 
890 		/*
891 		 * ignore submaps (for now) or null objects
892 		 */
893 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
894 			current->object.vm_object == NULL)
895 			continue;
896 
897 		/*
898 		 * limit this scan to the current map entry and the
899 		 * limits for the mincore call
900 		 */
901 		if (addr < current->start)
902 			addr = current->start;
903 		cend = current->end;
904 		if (cend > end)
905 			cend = end;
906 
907 		/*
908 		 * scan this entry one page at a time
909 		 */
910 		while (addr < cend) {
911 			/*
912 			 * Check pmap first, it is likely faster, also
913 			 * it can provide info as to whether we are the
914 			 * one referencing or modifying the page.
915 			 */
916 			object = NULL;
917 			locked_pa = 0;
918 		retry:
919 			m = NULL;
920 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
921 			if (locked_pa != 0) {
922 				/*
923 				 * The page is mapped by this process but not
924 				 * both accessed and modified.  It is also
925 				 * managed.  Acquire the object lock so that
926 				 * other mappings might be examined.
927 				 */
928 				m = PHYS_TO_VM_PAGE(locked_pa);
929 				if (m->object != object) {
930 					if (object != NULL)
931 						VM_OBJECT_WUNLOCK(object);
932 					object = m->object;
933 					locked = VM_OBJECT_TRYWLOCK(object);
934 					vm_page_unlock(m);
935 					if (!locked) {
936 						VM_OBJECT_WLOCK(object);
937 						vm_page_lock(m);
938 						goto retry;
939 					}
940 				} else
941 					vm_page_unlock(m);
942 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
943 				    ("mincore: page %p is mapped but invalid",
944 				    m));
945 			} else if (mincoreinfo == 0) {
946 				/*
947 				 * The page is not mapped by this process.  If
948 				 * the object implements managed pages, then
949 				 * determine if the page is resident so that
950 				 * the mappings might be examined.
951 				 */
952 				if (current->object.vm_object != object) {
953 					if (object != NULL)
954 						VM_OBJECT_WUNLOCK(object);
955 					object = current->object.vm_object;
956 					VM_OBJECT_WLOCK(object);
957 				}
958 				if (object->type == OBJT_DEFAULT ||
959 				    object->type == OBJT_SWAP ||
960 				    object->type == OBJT_VNODE) {
961 					pindex = OFF_TO_IDX(current->offset +
962 					    (addr - current->start));
963 					m = vm_page_lookup(object, pindex);
964 					if (m == NULL &&
965 					    vm_page_is_cached(object, pindex))
966 						mincoreinfo = MINCORE_INCORE;
967 					if (m != NULL && m->valid == 0)
968 						m = NULL;
969 					if (m != NULL)
970 						mincoreinfo = MINCORE_INCORE;
971 				}
972 			}
973 			if (m != NULL) {
974 				/* Examine other mappings to the page. */
975 				if (m->dirty == 0 && pmap_is_modified(m))
976 					vm_page_dirty(m);
977 				if (m->dirty != 0)
978 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
979 				/*
980 				 * The first test for PGA_REFERENCED is an
981 				 * optimization.  The second test is
982 				 * required because a concurrent pmap
983 				 * operation could clear the last reference
984 				 * and set PGA_REFERENCED before the call to
985 				 * pmap_is_referenced().
986 				 */
987 				if ((m->aflags & PGA_REFERENCED) != 0 ||
988 				    pmap_is_referenced(m) ||
989 				    (m->aflags & PGA_REFERENCED) != 0)
990 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
991 			}
992 			if (object != NULL)
993 				VM_OBJECT_WUNLOCK(object);
994 
995 			/*
996 			 * subyte may page fault.  In case it needs to modify
997 			 * the map, we release the lock.
998 			 */
999 			vm_map_unlock_read(map);
1000 
1001 			/*
1002 			 * calculate index into user supplied byte vector
1003 			 */
1004 			vecindex = OFF_TO_IDX(addr - first_addr);
1005 
1006 			/*
1007 			 * If we have skipped map entries, we need to make sure that
1008 			 * the byte vector is zeroed for those skipped entries.
1009 			 */
1010 			while ((lastvecindex + 1) < vecindex) {
1011 				++lastvecindex;
1012 				error = subyte(vec + lastvecindex, 0);
1013 				if (error) {
1014 					error = EFAULT;
1015 					goto done2;
1016 				}
1017 			}
1018 
1019 			/*
1020 			 * Pass the page information to the user
1021 			 */
1022 			error = subyte(vec + vecindex, mincoreinfo);
1023 			if (error) {
1024 				error = EFAULT;
1025 				goto done2;
1026 			}
1027 
1028 			/*
1029 			 * If the map has changed, due to the subyte, the previous
1030 			 * output may be invalid.
1031 			 */
1032 			vm_map_lock_read(map);
1033 			if (timestamp != map->timestamp)
1034 				goto RestartScan;
1035 
1036 			lastvecindex = vecindex;
1037 			addr += PAGE_SIZE;
1038 		}
1039 	}
1040 
1041 	/*
1042 	 * subyte may page fault.  In case it needs to modify
1043 	 * the map, we release the lock.
1044 	 */
1045 	vm_map_unlock_read(map);
1046 
1047 	/*
1048 	 * Zero the last entries in the byte vector.
1049 	 */
1050 	vecindex = OFF_TO_IDX(end - first_addr);
1051 	while ((lastvecindex + 1) < vecindex) {
1052 		++lastvecindex;
1053 		error = subyte(vec + lastvecindex, 0);
1054 		if (error) {
1055 			error = EFAULT;
1056 			goto done2;
1057 		}
1058 	}
1059 
1060 	/*
1061 	 * If the map has changed, due to the subyte, the previous
1062 	 * output may be invalid.
1063 	 */
1064 	vm_map_lock_read(map);
1065 	if (timestamp != map->timestamp)
1066 		goto RestartScan;
1067 	vm_map_unlock_read(map);
1068 done2:
1069 	return (error);
1070 }
1071 
1072 #ifndef _SYS_SYSPROTO_H_
1073 struct mlock_args {
1074 	const void *addr;
1075 	size_t len;
1076 };
1077 #endif
1078 /*
1079  * MPSAFE
1080  */
1081 int
1082 sys_mlock(td, uap)
1083 	struct thread *td;
1084 	struct mlock_args *uap;
1085 {
1086 
1087 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1088 }
1089 
1090 int
1091 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1092 {
1093 	vm_offset_t addr, end, last, start;
1094 	vm_size_t npages, size;
1095 	vm_map_t map;
1096 	unsigned long nsize;
1097 	int error;
1098 
1099 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1100 	if (error)
1101 		return (error);
1102 	addr = (vm_offset_t)addr0;
1103 	size = len;
1104 	last = addr + size;
1105 	start = trunc_page(addr);
1106 	end = round_page(last);
1107 	if (last < addr || end < addr)
1108 		return (EINVAL);
1109 	npages = atop(end - start);
1110 	if (npages > vm_page_max_wired)
1111 		return (ENOMEM);
1112 	map = &proc->p_vmspace->vm_map;
1113 	PROC_LOCK(proc);
1114 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1115 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1116 		PROC_UNLOCK(proc);
1117 		return (ENOMEM);
1118 	}
1119 	PROC_UNLOCK(proc);
1120 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1121 		return (EAGAIN);
1122 #ifdef RACCT
1123 	PROC_LOCK(proc);
1124 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1125 	PROC_UNLOCK(proc);
1126 	if (error != 0)
1127 		return (ENOMEM);
1128 #endif
1129 	error = vm_map_wire(map, start, end,
1130 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1131 #ifdef RACCT
1132 	if (error != KERN_SUCCESS) {
1133 		PROC_LOCK(proc);
1134 		racct_set(proc, RACCT_MEMLOCK,
1135 		    ptoa(pmap_wired_count(map->pmap)));
1136 		PROC_UNLOCK(proc);
1137 	}
1138 #endif
1139 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1140 }
1141 
1142 #ifndef _SYS_SYSPROTO_H_
1143 struct mlockall_args {
1144 	int	how;
1145 };
1146 #endif
1147 
1148 /*
1149  * MPSAFE
1150  */
1151 int
1152 sys_mlockall(td, uap)
1153 	struct thread *td;
1154 	struct mlockall_args *uap;
1155 {
1156 	vm_map_t map;
1157 	int error;
1158 
1159 	map = &td->td_proc->p_vmspace->vm_map;
1160 	error = priv_check(td, PRIV_VM_MLOCK);
1161 	if (error)
1162 		return (error);
1163 
1164 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1165 		return (EINVAL);
1166 
1167 	/*
1168 	 * If wiring all pages in the process would cause it to exceed
1169 	 * a hard resource limit, return ENOMEM.
1170 	 */
1171 	if (!old_mlock && uap->how & MCL_CURRENT) {
1172 		PROC_LOCK(td->td_proc);
1173 		if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1174 			PROC_UNLOCK(td->td_proc);
1175 			return (ENOMEM);
1176 		}
1177 		PROC_UNLOCK(td->td_proc);
1178 	}
1179 #ifdef RACCT
1180 	PROC_LOCK(td->td_proc);
1181 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1182 	PROC_UNLOCK(td->td_proc);
1183 	if (error != 0)
1184 		return (ENOMEM);
1185 #endif
1186 
1187 	if (uap->how & MCL_FUTURE) {
1188 		vm_map_lock(map);
1189 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1190 		vm_map_unlock(map);
1191 		error = 0;
1192 	}
1193 
1194 	if (uap->how & MCL_CURRENT) {
1195 		/*
1196 		 * P1003.1-2001 mandates that all currently mapped pages
1197 		 * will be memory resident and locked (wired) upon return
1198 		 * from mlockall(). vm_map_wire() will wire pages, by
1199 		 * calling vm_fault_wire() for each page in the region.
1200 		 */
1201 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1202 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1203 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1204 	}
1205 #ifdef RACCT
1206 	if (error != KERN_SUCCESS) {
1207 		PROC_LOCK(td->td_proc);
1208 		racct_set(td->td_proc, RACCT_MEMLOCK,
1209 		    ptoa(pmap_wired_count(map->pmap)));
1210 		PROC_UNLOCK(td->td_proc);
1211 	}
1212 #endif
1213 
1214 	return (error);
1215 }
1216 
1217 #ifndef _SYS_SYSPROTO_H_
1218 struct munlockall_args {
1219 	register_t dummy;
1220 };
1221 #endif
1222 
1223 /*
1224  * MPSAFE
1225  */
1226 int
1227 sys_munlockall(td, uap)
1228 	struct thread *td;
1229 	struct munlockall_args *uap;
1230 {
1231 	vm_map_t map;
1232 	int error;
1233 
1234 	map = &td->td_proc->p_vmspace->vm_map;
1235 	error = priv_check(td, PRIV_VM_MUNLOCK);
1236 	if (error)
1237 		return (error);
1238 
1239 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1240 	vm_map_lock(map);
1241 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1242 	vm_map_unlock(map);
1243 
1244 	/* Forcibly unwire all pages. */
1245 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1246 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1247 #ifdef RACCT
1248 	if (error == KERN_SUCCESS) {
1249 		PROC_LOCK(td->td_proc);
1250 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1251 		PROC_UNLOCK(td->td_proc);
1252 	}
1253 #endif
1254 
1255 	return (error);
1256 }
1257 
1258 #ifndef _SYS_SYSPROTO_H_
1259 struct munlock_args {
1260 	const void *addr;
1261 	size_t len;
1262 };
1263 #endif
1264 /*
1265  * MPSAFE
1266  */
1267 int
1268 sys_munlock(td, uap)
1269 	struct thread *td;
1270 	struct munlock_args *uap;
1271 {
1272 	vm_offset_t addr, end, last, start;
1273 	vm_size_t size;
1274 #ifdef RACCT
1275 	vm_map_t map;
1276 #endif
1277 	int error;
1278 
1279 	error = priv_check(td, PRIV_VM_MUNLOCK);
1280 	if (error)
1281 		return (error);
1282 	addr = (vm_offset_t)uap->addr;
1283 	size = uap->len;
1284 	last = addr + size;
1285 	start = trunc_page(addr);
1286 	end = round_page(last);
1287 	if (last < addr || end < addr)
1288 		return (EINVAL);
1289 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1290 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1291 #ifdef RACCT
1292 	if (error == KERN_SUCCESS) {
1293 		PROC_LOCK(td->td_proc);
1294 		map = &td->td_proc->p_vmspace->vm_map;
1295 		racct_set(td->td_proc, RACCT_MEMLOCK,
1296 		    ptoa(pmap_wired_count(map->pmap)));
1297 		PROC_UNLOCK(td->td_proc);
1298 	}
1299 #endif
1300 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1301 }
1302 
1303 /*
1304  * vm_mmap_vnode()
1305  *
1306  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1307  * operations on vnodes.
1308  *
1309  * For VCHR vnodes, the vnode lock is held over the call to
1310  * vm_mmap_cdev() to keep vp->v_rdev valid.
1311  */
1312 int
1313 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1314     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1315     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1316     boolean_t *writecounted)
1317 {
1318 	struct vattr va;
1319 	vm_object_t obj;
1320 	vm_offset_t foff;
1321 	struct mount *mp;
1322 	struct ucred *cred;
1323 	int error, flags, locktype;
1324 
1325 	mp = vp->v_mount;
1326 	cred = td->td_ucred;
1327 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1328 		locktype = LK_EXCLUSIVE;
1329 	else
1330 		locktype = LK_SHARED;
1331 	if ((error = vget(vp, locktype, td)) != 0)
1332 		return (error);
1333 	foff = *foffp;
1334 	flags = *flagsp;
1335 	obj = vp->v_object;
1336 	if (vp->v_type == VREG) {
1337 		/*
1338 		 * Get the proper underlying object
1339 		 */
1340 		if (obj == NULL) {
1341 			error = EINVAL;
1342 			goto done;
1343 		}
1344 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1345 			vput(vp);
1346 			vp = (struct vnode *)obj->handle;
1347 			/*
1348 			 * Bypass filesystems obey the mpsafety of the
1349 			 * underlying fs.  Tmpfs never bypasses.
1350 			 */
1351 			error = vget(vp, locktype, td);
1352 			if (error != 0)
1353 				return (error);
1354 		}
1355 		if (locktype == LK_EXCLUSIVE) {
1356 			*writecounted = TRUE;
1357 			vnode_pager_update_writecount(obj, 0, objsize);
1358 		}
1359 	} else if (vp->v_type == VCHR) {
1360 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1361 		    vp->v_rdev, foffp, objp);
1362 		if (error == 0)
1363 			goto mark_atime;
1364 		goto done;
1365 	} else {
1366 		error = EINVAL;
1367 		goto done;
1368 	}
1369 	if ((error = VOP_GETATTR(vp, &va, cred)))
1370 		goto done;
1371 #ifdef MAC
1372 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1373 	if (error != 0)
1374 		goto done;
1375 #endif
1376 	if ((flags & MAP_SHARED) != 0) {
1377 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1378 			if (prot & PROT_WRITE) {
1379 				error = EPERM;
1380 				goto done;
1381 			}
1382 			*maxprotp &= ~VM_PROT_WRITE;
1383 		}
1384 	}
1385 	/*
1386 	 * If it is a regular file without any references
1387 	 * we do not need to sync it.
1388 	 * Adjust object size to be the size of actual file.
1389 	 */
1390 	objsize = round_page(va.va_size);
1391 	if (va.va_nlink == 0)
1392 		flags |= MAP_NOSYNC;
1393 	if (obj->type == OBJT_VNODE)
1394 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1395 		    cred);
1396 	else {
1397 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1398 		    ("wrong object type"));
1399 		vm_object_reference(obj);
1400 	}
1401 	if (obj == NULL) {
1402 		error = ENOMEM;
1403 		goto done;
1404 	}
1405 	*objp = obj;
1406 	*flagsp = flags;
1407 
1408 mark_atime:
1409 	vfs_mark_atime(vp, cred);
1410 
1411 done:
1412 	if (error != 0 && *writecounted) {
1413 		*writecounted = FALSE;
1414 		vnode_pager_update_writecount(obj, objsize, 0);
1415 	}
1416 	vput(vp);
1417 	return (error);
1418 }
1419 
1420 /*
1421  * vm_mmap_cdev()
1422  *
1423  * MPSAFE
1424  *
1425  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1426  * operations on cdevs.
1427  */
1428 int
1429 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1430     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1431     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1432 {
1433 	vm_object_t obj;
1434 	struct cdevsw *dsw;
1435 	int error, flags, ref;
1436 
1437 	flags = *flagsp;
1438 
1439 	dsw = dev_refthread(cdev, &ref);
1440 	if (dsw == NULL)
1441 		return (ENXIO);
1442 	if (dsw->d_flags & D_MMAP_ANON) {
1443 		dev_relthread(cdev, ref);
1444 		*maxprotp = VM_PROT_ALL;
1445 		*flagsp |= MAP_ANON;
1446 		return (0);
1447 	}
1448 	/*
1449 	 * cdevs do not provide private mappings of any kind.
1450 	 */
1451 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1452 	    (prot & PROT_WRITE) != 0) {
1453 		dev_relthread(cdev, ref);
1454 		return (EACCES);
1455 	}
1456 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1457 		dev_relthread(cdev, ref);
1458 		return (EINVAL);
1459 	}
1460 	/*
1461 	 * Force device mappings to be shared.
1462 	 */
1463 	flags |= MAP_SHARED;
1464 #ifdef MAC_XXX
1465 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1466 	if (error != 0) {
1467 		dev_relthread(cdev, ref);
1468 		return (error);
1469 	}
1470 #endif
1471 	/*
1472 	 * First, try d_mmap_single().  If that is not implemented
1473 	 * (returns ENODEV), fall back to using the device pager.
1474 	 * Note that d_mmap_single() must return a reference to the
1475 	 * object (it needs to bump the reference count of the object
1476 	 * it returns somehow).
1477 	 *
1478 	 * XXX assumes VM_PROT_* == PROT_*
1479 	 */
1480 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1481 	dev_relthread(cdev, ref);
1482 	if (error != ENODEV)
1483 		return (error);
1484 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1485 	    td->td_ucred);
1486 	if (obj == NULL)
1487 		return (EINVAL);
1488 	*objp = obj;
1489 	*flagsp = flags;
1490 	return (0);
1491 }
1492 
1493 /*
1494  * vm_mmap_shm()
1495  *
1496  * MPSAFE
1497  *
1498  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1499  * operations on shm file descriptors.
1500  */
1501 int
1502 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1503     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1504     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1505 {
1506 	int error;
1507 
1508 	if ((*flagsp & MAP_SHARED) != 0 &&
1509 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1510 	    (prot & PROT_WRITE) != 0)
1511 		return (EACCES);
1512 #ifdef MAC
1513 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1514 	if (error != 0)
1515 		return (error);
1516 #endif
1517 	error = shm_mmap(shmfd, objsize, foff, objp);
1518 	if (error)
1519 		return (error);
1520 	return (0);
1521 }
1522 
1523 /*
1524  * vm_mmap()
1525  *
1526  * MPSAFE
1527  *
1528  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1529  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1530  */
1531 int
1532 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1533 	vm_prot_t maxprot, int flags,
1534 	objtype_t handle_type, void *handle,
1535 	vm_ooffset_t foff)
1536 {
1537 	boolean_t fitit;
1538 	vm_object_t object = NULL;
1539 	struct thread *td = curthread;
1540 	int docow, error, findspace, rv;
1541 	boolean_t writecounted;
1542 
1543 	if (size == 0)
1544 		return (0);
1545 
1546 	size = round_page(size);
1547 
1548 	if (map == &td->td_proc->p_vmspace->vm_map) {
1549 		PROC_LOCK(td->td_proc);
1550 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1551 			PROC_UNLOCK(td->td_proc);
1552 			return (ENOMEM);
1553 		}
1554 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1555 			PROC_UNLOCK(td->td_proc);
1556 			return (ENOMEM);
1557 		}
1558 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1559 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1560 			    lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1561 				racct_set_force(td->td_proc, RACCT_VMEM,
1562 				    map->size);
1563 				PROC_UNLOCK(td->td_proc);
1564 				return (ENOMEM);
1565 			}
1566 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1567 			    ptoa(pmap_wired_count(map->pmap)) + size);
1568 			if (error != 0) {
1569 				racct_set_force(td->td_proc, RACCT_VMEM,
1570 				    map->size);
1571 				PROC_UNLOCK(td->td_proc);
1572 				return (error);
1573 			}
1574 		}
1575 		PROC_UNLOCK(td->td_proc);
1576 	}
1577 
1578 	/*
1579 	 * We currently can only deal with page aligned file offsets.
1580 	 * The check is here rather than in the syscall because the
1581 	 * kernel calls this function internally for other mmaping
1582 	 * operations (such as in exec) and non-aligned offsets will
1583 	 * cause pmap inconsistencies...so we want to be sure to
1584 	 * disallow this in all cases.
1585 	 */
1586 	if (foff & PAGE_MASK)
1587 		return (EINVAL);
1588 
1589 	if ((flags & MAP_FIXED) == 0) {
1590 		fitit = TRUE;
1591 		*addr = round_page(*addr);
1592 	} else {
1593 		if (*addr != trunc_page(*addr))
1594 			return (EINVAL);
1595 		fitit = FALSE;
1596 	}
1597 	writecounted = FALSE;
1598 
1599 	/*
1600 	 * Lookup/allocate object.
1601 	 */
1602 	switch (handle_type) {
1603 	case OBJT_DEVICE:
1604 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1605 		    handle, &foff, &object);
1606 		break;
1607 	case OBJT_VNODE:
1608 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1609 		    handle, &foff, &object, &writecounted);
1610 		break;
1611 	case OBJT_SWAP:
1612 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1613 		    handle, foff, &object);
1614 		break;
1615 	case OBJT_DEFAULT:
1616 		if (handle == NULL) {
1617 			error = 0;
1618 			break;
1619 		}
1620 		/* FALLTHROUGH */
1621 	default:
1622 		error = EINVAL;
1623 		break;
1624 	}
1625 	if (error)
1626 		return (error);
1627 	if (flags & MAP_ANON) {
1628 		object = NULL;
1629 		docow = 0;
1630 		/*
1631 		 * Unnamed anonymous regions always start at 0.
1632 		 */
1633 		if (handle == 0)
1634 			foff = 0;
1635 	} else if (flags & MAP_PREFAULT_READ)
1636 		docow = MAP_PREFAULT;
1637 	else
1638 		docow = MAP_PREFAULT_PARTIAL;
1639 
1640 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1641 		docow |= MAP_COPY_ON_WRITE;
1642 	if (flags & MAP_NOSYNC)
1643 		docow |= MAP_DISABLE_SYNCER;
1644 	if (flags & MAP_NOCORE)
1645 		docow |= MAP_DISABLE_COREDUMP;
1646 	/* Shared memory is also shared with children. */
1647 	if (flags & MAP_SHARED)
1648 		docow |= MAP_INHERIT_SHARE;
1649 	if (writecounted)
1650 		docow |= MAP_VN_WRITECOUNT;
1651 	if (flags & MAP_STACK) {
1652 		if (object != NULL)
1653 			return (EINVAL);
1654 		docow |= MAP_STACK_GROWS_DOWN;
1655 	}
1656 	if ((flags & MAP_EXCL) != 0)
1657 		docow |= MAP_CHECK_EXCL;
1658 
1659 	if (fitit) {
1660 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1661 			findspace = VMFS_SUPER_SPACE;
1662 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1663 			findspace = VMFS_ALIGNED_SPACE(flags >>
1664 			    MAP_ALIGNMENT_SHIFT);
1665 		else
1666 			findspace = VMFS_OPTIMAL_SPACE;
1667 		rv = vm_map_find(map, object, foff, addr, size,
1668 #ifdef MAP_32BIT
1669 		    flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1670 #endif
1671 		    0, findspace, prot, maxprot, docow);
1672 	} else {
1673 		rv = vm_map_fixed(map, object, foff, *addr, size,
1674 		    prot, maxprot, docow);
1675 	}
1676 
1677 	if (rv == KERN_SUCCESS) {
1678 		/*
1679 		 * If the process has requested that all future mappings
1680 		 * be wired, then heed this.
1681 		 */
1682 		if (map->flags & MAP_WIREFUTURE) {
1683 			vm_map_wire(map, *addr, *addr + size,
1684 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1685 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1686 		}
1687 	} else {
1688 		/*
1689 		 * If this mapping was accounted for in the vnode's
1690 		 * writecount, then undo that now.
1691 		 */
1692 		if (writecounted)
1693 			vnode_pager_release_writecount(object, 0, size);
1694 		/*
1695 		 * Lose the object reference.  Will destroy the
1696 		 * object if it's an unnamed anonymous mapping
1697 		 * or named anonymous without other references.
1698 		 */
1699 		vm_object_deallocate(object);
1700 	}
1701 	return (vm_mmap_to_errno(rv));
1702 }
1703 
1704 /*
1705  * Translate a Mach VM return code to zero on success or the appropriate errno
1706  * on failure.
1707  */
1708 int
1709 vm_mmap_to_errno(int rv)
1710 {
1711 
1712 	switch (rv) {
1713 	case KERN_SUCCESS:
1714 		return (0);
1715 	case KERN_INVALID_ADDRESS:
1716 	case KERN_NO_SPACE:
1717 		return (ENOMEM);
1718 	case KERN_PROTECTION_FAILURE:
1719 		return (EACCES);
1720 	default:
1721 		return (EINVAL);
1722 	}
1723 }
1724