xref: /freebsd/sys/vm/vm_mmap.c (revision 9124ddeb4a551977cf6b2218291e7c666ce25f47)
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/capability.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/racct.h>
60 #include <sys/resource.h>
61 #include <sys/resourcevar.h>
62 #include <sys/vnode.h>
63 #include <sys/fcntl.h>
64 #include <sys/file.h>
65 #include <sys/mman.h>
66 #include <sys/mount.h>
67 #include <sys/conf.h>
68 #include <sys/stat.h>
69 #include <sys/sysent.h>
70 #include <sys/vmmeter.h>
71 
72 #include <security/mac/mac_framework.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vm_pageout.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_page.h>
84 #include <vm/vnode_pager.h>
85 
86 #ifdef HWPMC_HOOKS
87 #include <sys/pmckern.h>
88 #endif
89 
90 #ifndef _SYS_SYSPROTO_H_
91 struct sbrk_args {
92 	int incr;
93 };
94 #endif
95 
96 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
97     int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
98 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
99     int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
100 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
101     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
102 
103 /*
104  * MPSAFE
105  */
106 /* ARGSUSED */
107 int
108 sys_sbrk(td, uap)
109 	struct thread *td;
110 	struct sbrk_args *uap;
111 {
112 	/* Not yet implemented */
113 	return (EOPNOTSUPP);
114 }
115 
116 #ifndef _SYS_SYSPROTO_H_
117 struct sstk_args {
118 	int incr;
119 };
120 #endif
121 
122 /*
123  * MPSAFE
124  */
125 /* ARGSUSED */
126 int
127 sys_sstk(td, uap)
128 	struct thread *td;
129 	struct sstk_args *uap;
130 {
131 	/* Not yet implemented */
132 	return (EOPNOTSUPP);
133 }
134 
135 #if defined(COMPAT_43)
136 #ifndef _SYS_SYSPROTO_H_
137 struct getpagesize_args {
138 	int dummy;
139 };
140 #endif
141 
142 int
143 ogetpagesize(td, uap)
144 	struct thread *td;
145 	struct getpagesize_args *uap;
146 {
147 	/* MP SAFE */
148 	td->td_retval[0] = PAGE_SIZE;
149 	return (0);
150 }
151 #endif				/* COMPAT_43 */
152 
153 
154 /*
155  * Memory Map (mmap) system call.  Note that the file offset
156  * and address are allowed to be NOT page aligned, though if
157  * the MAP_FIXED flag it set, both must have the same remainder
158  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
159  * page-aligned, the actual mapping starts at trunc_page(addr)
160  * and the return value is adjusted up by the page offset.
161  *
162  * Generally speaking, only character devices which are themselves
163  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
164  * there would be no cache coherency between a descriptor and a VM mapping
165  * both to the same character device.
166  */
167 #ifndef _SYS_SYSPROTO_H_
168 struct mmap_args {
169 	void *addr;
170 	size_t len;
171 	int prot;
172 	int flags;
173 	int fd;
174 	long pad;
175 	off_t pos;
176 };
177 #endif
178 
179 /*
180  * MPSAFE
181  */
182 int
183 sys_mmap(td, uap)
184 	struct thread *td;
185 	struct mmap_args *uap;
186 {
187 #ifdef HWPMC_HOOKS
188 	struct pmckern_map_in pkm;
189 #endif
190 	struct file *fp;
191 	struct vnode *vp;
192 	vm_offset_t addr;
193 	vm_size_t size, pageoff;
194 	vm_prot_t cap_maxprot, prot, maxprot;
195 	void *handle;
196 	objtype_t handle_type;
197 	int flags, error;
198 	off_t pos;
199 	struct vmspace *vms = td->td_proc->p_vmspace;
200 	cap_rights_t rights;
201 
202 	addr = (vm_offset_t) uap->addr;
203 	size = uap->len;
204 	prot = uap->prot & VM_PROT_ALL;
205 	flags = uap->flags;
206 	pos = uap->pos;
207 
208 	fp = NULL;
209 
210 	/* Make sure mapping fits into numeric range, etc. */
211 	if ((uap->len == 0 && !SV_CURPROC_FLAG(SV_AOUT) &&
212 	     curproc->p_osrel >= P_OSREL_MAP_ANON) ||
213 	    ((flags & MAP_ANON) && (uap->fd != -1 || pos != 0)))
214 		return (EINVAL);
215 
216 	if (flags & MAP_STACK) {
217 		if ((uap->fd != -1) ||
218 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
219 			return (EINVAL);
220 		flags |= MAP_ANON;
221 		pos = 0;
222 	}
223 
224 	/*
225 	 * Align the file position to a page boundary,
226 	 * and save its page offset component.
227 	 */
228 	pageoff = (pos & PAGE_MASK);
229 	pos -= pageoff;
230 
231 	/* Adjust size for rounding (on both ends). */
232 	size += pageoff;			/* low end... */
233 	size = (vm_size_t) round_page(size);	/* hi end */
234 
235 	/*
236 	 * Check for illegal addresses.  Watch out for address wrap... Note
237 	 * that VM_*_ADDRESS are not constants due to casts (argh).
238 	 */
239 	if (flags & MAP_FIXED) {
240 		/*
241 		 * The specified address must have the same remainder
242 		 * as the file offset taken modulo PAGE_SIZE, so it
243 		 * should be aligned after adjustment by pageoff.
244 		 */
245 		addr -= pageoff;
246 		if (addr & PAGE_MASK)
247 			return (EINVAL);
248 
249 		/* Address range must be all in user VM space. */
250 		if (addr < vm_map_min(&vms->vm_map) ||
251 		    addr + size > vm_map_max(&vms->vm_map))
252 			return (EINVAL);
253 		if (addr + size < addr)
254 			return (EINVAL);
255 	} else {
256 		/*
257 		 * XXX for non-fixed mappings where no hint is provided or
258 		 * the hint would fall in the potential heap space,
259 		 * place it after the end of the largest possible heap.
260 		 *
261 		 * There should really be a pmap call to determine a reasonable
262 		 * location.
263 		 */
264 		PROC_LOCK(td->td_proc);
265 		if (addr == 0 ||
266 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
267 		    addr < round_page((vm_offset_t)vms->vm_daddr +
268 		    lim_max(td->td_proc, RLIMIT_DATA))))
269 			addr = round_page((vm_offset_t)vms->vm_daddr +
270 			    lim_max(td->td_proc, RLIMIT_DATA));
271 		PROC_UNLOCK(td->td_proc);
272 	}
273 	if (flags & MAP_ANON) {
274 		/*
275 		 * Mapping blank space is trivial.
276 		 */
277 		handle = NULL;
278 		handle_type = OBJT_DEFAULT;
279 		maxprot = VM_PROT_ALL;
280 		cap_maxprot = VM_PROT_ALL;
281 	} else {
282 		/*
283 		 * Mapping file, get fp for validation and don't let the
284 		 * descriptor disappear on us if we block. Check capability
285 		 * rights, but also return the maximum rights to be combined
286 		 * with maxprot later.
287 		 */
288 		rights = CAP_MMAP;
289 		if (prot & PROT_READ)
290 			rights |= CAP_READ;
291 		if ((flags & MAP_SHARED) != 0) {
292 			if (prot & PROT_WRITE)
293 				rights |= CAP_WRITE;
294 		}
295 		if (prot & PROT_EXEC)
296 			rights |= CAP_MAPEXEC;
297 		if ((error = fget_mmap(td, uap->fd, rights, &cap_maxprot,
298 		    &fp)) != 0)
299 			goto done;
300 		if (fp->f_type == DTYPE_SHM) {
301 			handle = fp->f_data;
302 			handle_type = OBJT_SWAP;
303 			maxprot = VM_PROT_NONE;
304 
305 			/* FREAD should always be set. */
306 			if (fp->f_flag & FREAD)
307 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
308 			if (fp->f_flag & FWRITE)
309 				maxprot |= VM_PROT_WRITE;
310 			goto map;
311 		}
312 		if (fp->f_type != DTYPE_VNODE) {
313 			error = ENODEV;
314 			goto done;
315 		}
316 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
317     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
318 		/*
319 		 * POSIX shared-memory objects are defined to have
320 		 * kernel persistence, and are not defined to support
321 		 * read(2)/write(2) -- or even open(2).  Thus, we can
322 		 * use MAP_ASYNC to trade on-disk coherence for speed.
323 		 * The shm_open(3) library routine turns on the FPOSIXSHM
324 		 * flag to request this behavior.
325 		 */
326 		if (fp->f_flag & FPOSIXSHM)
327 			flags |= MAP_NOSYNC;
328 #endif
329 		vp = fp->f_vnode;
330 		/*
331 		 * Ensure that file and memory protections are
332 		 * compatible.  Note that we only worry about
333 		 * writability if mapping is shared; in this case,
334 		 * current and max prot are dictated by the open file.
335 		 * XXX use the vnode instead?  Problem is: what
336 		 * credentials do we use for determination? What if
337 		 * proc does a setuid?
338 		 */
339 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
340 			maxprot = VM_PROT_NONE;
341 		else
342 			maxprot = VM_PROT_EXECUTE;
343 		if (fp->f_flag & FREAD) {
344 			maxprot |= VM_PROT_READ;
345 		} else if (prot & PROT_READ) {
346 			error = EACCES;
347 			goto done;
348 		}
349 		/*
350 		 * If we are sharing potential changes (either via
351 		 * MAP_SHARED or via the implicit sharing of character
352 		 * device mappings), and we are trying to get write
353 		 * permission although we opened it without asking
354 		 * for it, bail out.
355 		 */
356 		if ((flags & MAP_SHARED) != 0) {
357 			if ((fp->f_flag & FWRITE) != 0) {
358 				maxprot |= VM_PROT_WRITE;
359 			} else if ((prot & PROT_WRITE) != 0) {
360 				error = EACCES;
361 				goto done;
362 			}
363 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
364 			maxprot |= VM_PROT_WRITE;
365 			cap_maxprot |= VM_PROT_WRITE;
366 		}
367 		handle = (void *)vp;
368 		handle_type = OBJT_VNODE;
369 	}
370 map:
371 	td->td_fpop = fp;
372 	maxprot &= cap_maxprot;
373 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
374 	    flags, handle_type, handle, pos);
375 	td->td_fpop = NULL;
376 #ifdef HWPMC_HOOKS
377 	/* inform hwpmc(4) if an executable is being mapped */
378 	if (error == 0 && handle_type == OBJT_VNODE &&
379 	    (prot & PROT_EXEC)) {
380 		pkm.pm_file = handle;
381 		pkm.pm_address = (uintptr_t) addr;
382 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
383 	}
384 #endif
385 	if (error == 0)
386 		td->td_retval[0] = (register_t) (addr + pageoff);
387 done:
388 	if (fp)
389 		fdrop(fp, td);
390 
391 	return (error);
392 }
393 
394 int
395 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
396 {
397 	struct mmap_args oargs;
398 
399 	oargs.addr = uap->addr;
400 	oargs.len = uap->len;
401 	oargs.prot = uap->prot;
402 	oargs.flags = uap->flags;
403 	oargs.fd = uap->fd;
404 	oargs.pos = uap->pos;
405 	return (sys_mmap(td, &oargs));
406 }
407 
408 #ifdef COMPAT_43
409 #ifndef _SYS_SYSPROTO_H_
410 struct ommap_args {
411 	caddr_t addr;
412 	int len;
413 	int prot;
414 	int flags;
415 	int fd;
416 	long pos;
417 };
418 #endif
419 int
420 ommap(td, uap)
421 	struct thread *td;
422 	struct ommap_args *uap;
423 {
424 	struct mmap_args nargs;
425 	static const char cvtbsdprot[8] = {
426 		0,
427 		PROT_EXEC,
428 		PROT_WRITE,
429 		PROT_EXEC | PROT_WRITE,
430 		PROT_READ,
431 		PROT_EXEC | PROT_READ,
432 		PROT_WRITE | PROT_READ,
433 		PROT_EXEC | PROT_WRITE | PROT_READ,
434 	};
435 
436 #define	OMAP_ANON	0x0002
437 #define	OMAP_COPY	0x0020
438 #define	OMAP_SHARED	0x0010
439 #define	OMAP_FIXED	0x0100
440 
441 	nargs.addr = uap->addr;
442 	nargs.len = uap->len;
443 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
444 	nargs.flags = 0;
445 	if (uap->flags & OMAP_ANON)
446 		nargs.flags |= MAP_ANON;
447 	if (uap->flags & OMAP_COPY)
448 		nargs.flags |= MAP_COPY;
449 	if (uap->flags & OMAP_SHARED)
450 		nargs.flags |= MAP_SHARED;
451 	else
452 		nargs.flags |= MAP_PRIVATE;
453 	if (uap->flags & OMAP_FIXED)
454 		nargs.flags |= MAP_FIXED;
455 	nargs.fd = uap->fd;
456 	nargs.pos = uap->pos;
457 	return (sys_mmap(td, &nargs));
458 }
459 #endif				/* COMPAT_43 */
460 
461 
462 #ifndef _SYS_SYSPROTO_H_
463 struct msync_args {
464 	void *addr;
465 	size_t len;
466 	int flags;
467 };
468 #endif
469 /*
470  * MPSAFE
471  */
472 int
473 sys_msync(td, uap)
474 	struct thread *td;
475 	struct msync_args *uap;
476 {
477 	vm_offset_t addr;
478 	vm_size_t size, pageoff;
479 	int flags;
480 	vm_map_t map;
481 	int rv;
482 
483 	addr = (vm_offset_t) uap->addr;
484 	size = uap->len;
485 	flags = uap->flags;
486 
487 	pageoff = (addr & PAGE_MASK);
488 	addr -= pageoff;
489 	size += pageoff;
490 	size = (vm_size_t) round_page(size);
491 	if (addr + size < addr)
492 		return (EINVAL);
493 
494 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
495 		return (EINVAL);
496 
497 	map = &td->td_proc->p_vmspace->vm_map;
498 
499 	/*
500 	 * Clean the pages and interpret the return value.
501 	 */
502 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
503 	    (flags & MS_INVALIDATE) != 0);
504 	switch (rv) {
505 	case KERN_SUCCESS:
506 		return (0);
507 	case KERN_INVALID_ADDRESS:
508 		return (EINVAL);	/* Sun returns ENOMEM? */
509 	case KERN_INVALID_ARGUMENT:
510 		return (EBUSY);
511 	default:
512 		return (EINVAL);
513 	}
514 }
515 
516 #ifndef _SYS_SYSPROTO_H_
517 struct munmap_args {
518 	void *addr;
519 	size_t len;
520 };
521 #endif
522 /*
523  * MPSAFE
524  */
525 int
526 sys_munmap(td, uap)
527 	struct thread *td;
528 	struct munmap_args *uap;
529 {
530 #ifdef HWPMC_HOOKS
531 	struct pmckern_map_out pkm;
532 	vm_map_entry_t entry;
533 #endif
534 	vm_offset_t addr;
535 	vm_size_t size, pageoff;
536 	vm_map_t map;
537 
538 	addr = (vm_offset_t) uap->addr;
539 	size = uap->len;
540 	if (size == 0)
541 		return (EINVAL);
542 
543 	pageoff = (addr & PAGE_MASK);
544 	addr -= pageoff;
545 	size += pageoff;
546 	size = (vm_size_t) round_page(size);
547 	if (addr + size < addr)
548 		return (EINVAL);
549 
550 	/*
551 	 * Check for illegal addresses.  Watch out for address wrap...
552 	 */
553 	map = &td->td_proc->p_vmspace->vm_map;
554 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
555 		return (EINVAL);
556 	vm_map_lock(map);
557 #ifdef HWPMC_HOOKS
558 	/*
559 	 * Inform hwpmc if the address range being unmapped contains
560 	 * an executable region.
561 	 */
562 	pkm.pm_address = (uintptr_t) NULL;
563 	if (vm_map_lookup_entry(map, addr, &entry)) {
564 		for (;
565 		     entry != &map->header && entry->start < addr + size;
566 		     entry = entry->next) {
567 			if (vm_map_check_protection(map, entry->start,
568 				entry->end, VM_PROT_EXECUTE) == TRUE) {
569 				pkm.pm_address = (uintptr_t) addr;
570 				pkm.pm_size = (size_t) size;
571 				break;
572 			}
573 		}
574 	}
575 #endif
576 	vm_map_delete(map, addr, addr + size);
577 
578 #ifdef HWPMC_HOOKS
579 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
580 	vm_map_lock_downgrade(map);
581 	if (pkm.pm_address != (uintptr_t) NULL)
582 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
583 	vm_map_unlock_read(map);
584 #else
585 	vm_map_unlock(map);
586 #endif
587 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
588 	return (0);
589 }
590 
591 #ifndef _SYS_SYSPROTO_H_
592 struct mprotect_args {
593 	const void *addr;
594 	size_t len;
595 	int prot;
596 };
597 #endif
598 /*
599  * MPSAFE
600  */
601 int
602 sys_mprotect(td, uap)
603 	struct thread *td;
604 	struct mprotect_args *uap;
605 {
606 	vm_offset_t addr;
607 	vm_size_t size, pageoff;
608 	vm_prot_t prot;
609 
610 	addr = (vm_offset_t) uap->addr;
611 	size = uap->len;
612 	prot = uap->prot & VM_PROT_ALL;
613 
614 	pageoff = (addr & PAGE_MASK);
615 	addr -= pageoff;
616 	size += pageoff;
617 	size = (vm_size_t) round_page(size);
618 	if (addr + size < addr)
619 		return (EINVAL);
620 
621 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
622 	    addr + size, prot, FALSE)) {
623 	case KERN_SUCCESS:
624 		return (0);
625 	case KERN_PROTECTION_FAILURE:
626 		return (EACCES);
627 	case KERN_RESOURCE_SHORTAGE:
628 		return (ENOMEM);
629 	}
630 	return (EINVAL);
631 }
632 
633 #ifndef _SYS_SYSPROTO_H_
634 struct minherit_args {
635 	void *addr;
636 	size_t len;
637 	int inherit;
638 };
639 #endif
640 /*
641  * MPSAFE
642  */
643 int
644 sys_minherit(td, uap)
645 	struct thread *td;
646 	struct minherit_args *uap;
647 {
648 	vm_offset_t addr;
649 	vm_size_t size, pageoff;
650 	vm_inherit_t inherit;
651 
652 	addr = (vm_offset_t)uap->addr;
653 	size = uap->len;
654 	inherit = uap->inherit;
655 
656 	pageoff = (addr & PAGE_MASK);
657 	addr -= pageoff;
658 	size += pageoff;
659 	size = (vm_size_t) round_page(size);
660 	if (addr + size < addr)
661 		return (EINVAL);
662 
663 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
664 	    addr + size, inherit)) {
665 	case KERN_SUCCESS:
666 		return (0);
667 	case KERN_PROTECTION_FAILURE:
668 		return (EACCES);
669 	}
670 	return (EINVAL);
671 }
672 
673 #ifndef _SYS_SYSPROTO_H_
674 struct madvise_args {
675 	void *addr;
676 	size_t len;
677 	int behav;
678 };
679 #endif
680 
681 /*
682  * MPSAFE
683  */
684 int
685 sys_madvise(td, uap)
686 	struct thread *td;
687 	struct madvise_args *uap;
688 {
689 	vm_offset_t start, end;
690 	vm_map_t map;
691 	struct proc *p;
692 	int error;
693 
694 	/*
695 	 * Check for our special case, advising the swap pager we are
696 	 * "immortal."
697 	 */
698 	if (uap->behav == MADV_PROTECT) {
699 		error = priv_check(td, PRIV_VM_MADV_PROTECT);
700 		if (error == 0) {
701 			p = td->td_proc;
702 			PROC_LOCK(p);
703 			p->p_flag |= P_PROTECTED;
704 			PROC_UNLOCK(p);
705 		}
706 		return (error);
707 	}
708 	/*
709 	 * Check for illegal behavior
710 	 */
711 	if (uap->behav < 0 || uap->behav > MADV_CORE)
712 		return (EINVAL);
713 	/*
714 	 * Check for illegal addresses.  Watch out for address wrap... Note
715 	 * that VM_*_ADDRESS are not constants due to casts (argh).
716 	 */
717 	map = &td->td_proc->p_vmspace->vm_map;
718 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
719 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
720 		return (EINVAL);
721 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
722 		return (EINVAL);
723 
724 	/*
725 	 * Since this routine is only advisory, we default to conservative
726 	 * behavior.
727 	 */
728 	start = trunc_page((vm_offset_t) uap->addr);
729 	end = round_page((vm_offset_t) uap->addr + uap->len);
730 
731 	if (vm_map_madvise(map, start, end, uap->behav))
732 		return (EINVAL);
733 	return (0);
734 }
735 
736 #ifndef _SYS_SYSPROTO_H_
737 struct mincore_args {
738 	const void *addr;
739 	size_t len;
740 	char *vec;
741 };
742 #endif
743 
744 /*
745  * MPSAFE
746  */
747 int
748 sys_mincore(td, uap)
749 	struct thread *td;
750 	struct mincore_args *uap;
751 {
752 	vm_offset_t addr, first_addr;
753 	vm_offset_t end, cend;
754 	pmap_t pmap;
755 	vm_map_t map;
756 	char *vec;
757 	int error = 0;
758 	int vecindex, lastvecindex;
759 	vm_map_entry_t current;
760 	vm_map_entry_t entry;
761 	vm_object_t object;
762 	vm_paddr_t locked_pa;
763 	vm_page_t m;
764 	vm_pindex_t pindex;
765 	int mincoreinfo;
766 	unsigned int timestamp;
767 	boolean_t locked;
768 
769 	/*
770 	 * Make sure that the addresses presented are valid for user
771 	 * mode.
772 	 */
773 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
774 	end = addr + (vm_size_t)round_page(uap->len);
775 	map = &td->td_proc->p_vmspace->vm_map;
776 	if (end > vm_map_max(map) || end < addr)
777 		return (ENOMEM);
778 
779 	/*
780 	 * Address of byte vector
781 	 */
782 	vec = uap->vec;
783 
784 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
785 
786 	vm_map_lock_read(map);
787 RestartScan:
788 	timestamp = map->timestamp;
789 
790 	if (!vm_map_lookup_entry(map, addr, &entry)) {
791 		vm_map_unlock_read(map);
792 		return (ENOMEM);
793 	}
794 
795 	/*
796 	 * Do this on a map entry basis so that if the pages are not
797 	 * in the current processes address space, we can easily look
798 	 * up the pages elsewhere.
799 	 */
800 	lastvecindex = -1;
801 	for (current = entry;
802 	    (current != &map->header) && (current->start < end);
803 	    current = current->next) {
804 
805 		/*
806 		 * check for contiguity
807 		 */
808 		if (current->end < end &&
809 		    (entry->next == &map->header ||
810 		     current->next->start > current->end)) {
811 			vm_map_unlock_read(map);
812 			return (ENOMEM);
813 		}
814 
815 		/*
816 		 * ignore submaps (for now) or null objects
817 		 */
818 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
819 			current->object.vm_object == NULL)
820 			continue;
821 
822 		/*
823 		 * limit this scan to the current map entry and the
824 		 * limits for the mincore call
825 		 */
826 		if (addr < current->start)
827 			addr = current->start;
828 		cend = current->end;
829 		if (cend > end)
830 			cend = end;
831 
832 		/*
833 		 * scan this entry one page at a time
834 		 */
835 		while (addr < cend) {
836 			/*
837 			 * Check pmap first, it is likely faster, also
838 			 * it can provide info as to whether we are the
839 			 * one referencing or modifying the page.
840 			 */
841 			object = NULL;
842 			locked_pa = 0;
843 		retry:
844 			m = NULL;
845 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
846 			if (locked_pa != 0) {
847 				/*
848 				 * The page is mapped by this process but not
849 				 * both accessed and modified.  It is also
850 				 * managed.  Acquire the object lock so that
851 				 * other mappings might be examined.
852 				 */
853 				m = PHYS_TO_VM_PAGE(locked_pa);
854 				if (m->object != object) {
855 					if (object != NULL)
856 						VM_OBJECT_UNLOCK(object);
857 					object = m->object;
858 					locked = VM_OBJECT_TRYLOCK(object);
859 					vm_page_unlock(m);
860 					if (!locked) {
861 						VM_OBJECT_LOCK(object);
862 						vm_page_lock(m);
863 						goto retry;
864 					}
865 				} else
866 					vm_page_unlock(m);
867 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
868 				    ("mincore: page %p is mapped but invalid",
869 				    m));
870 			} else if (mincoreinfo == 0) {
871 				/*
872 				 * The page is not mapped by this process.  If
873 				 * the object implements managed pages, then
874 				 * determine if the page is resident so that
875 				 * the mappings might be examined.
876 				 */
877 				if (current->object.vm_object != object) {
878 					if (object != NULL)
879 						VM_OBJECT_UNLOCK(object);
880 					object = current->object.vm_object;
881 					VM_OBJECT_LOCK(object);
882 				}
883 				if (object->type == OBJT_DEFAULT ||
884 				    object->type == OBJT_SWAP ||
885 				    object->type == OBJT_VNODE) {
886 					pindex = OFF_TO_IDX(current->offset +
887 					    (addr - current->start));
888 					m = vm_page_lookup(object, pindex);
889 					if (m != NULL && m->valid == 0)
890 						m = NULL;
891 					if (m != NULL)
892 						mincoreinfo = MINCORE_INCORE;
893 				}
894 			}
895 			if (m != NULL) {
896 				/* Examine other mappings to the page. */
897 				if (m->dirty == 0 && pmap_is_modified(m))
898 					vm_page_dirty(m);
899 				if (m->dirty != 0)
900 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
901 				/*
902 				 * The first test for PGA_REFERENCED is an
903 				 * optimization.  The second test is
904 				 * required because a concurrent pmap
905 				 * operation could clear the last reference
906 				 * and set PGA_REFERENCED before the call to
907 				 * pmap_is_referenced().
908 				 */
909 				if ((m->aflags & PGA_REFERENCED) != 0 ||
910 				    pmap_is_referenced(m) ||
911 				    (m->aflags & PGA_REFERENCED) != 0)
912 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
913 			}
914 			if (object != NULL)
915 				VM_OBJECT_UNLOCK(object);
916 
917 			/*
918 			 * subyte may page fault.  In case it needs to modify
919 			 * the map, we release the lock.
920 			 */
921 			vm_map_unlock_read(map);
922 
923 			/*
924 			 * calculate index into user supplied byte vector
925 			 */
926 			vecindex = OFF_TO_IDX(addr - first_addr);
927 
928 			/*
929 			 * If we have skipped map entries, we need to make sure that
930 			 * the byte vector is zeroed for those skipped entries.
931 			 */
932 			while ((lastvecindex + 1) < vecindex) {
933 				error = subyte(vec + lastvecindex, 0);
934 				if (error) {
935 					error = EFAULT;
936 					goto done2;
937 				}
938 				++lastvecindex;
939 			}
940 
941 			/*
942 			 * Pass the page information to the user
943 			 */
944 			error = subyte(vec + vecindex, mincoreinfo);
945 			if (error) {
946 				error = EFAULT;
947 				goto done2;
948 			}
949 
950 			/*
951 			 * If the map has changed, due to the subyte, the previous
952 			 * output may be invalid.
953 			 */
954 			vm_map_lock_read(map);
955 			if (timestamp != map->timestamp)
956 				goto RestartScan;
957 
958 			lastvecindex = vecindex;
959 			addr += PAGE_SIZE;
960 		}
961 	}
962 
963 	/*
964 	 * subyte may page fault.  In case it needs to modify
965 	 * the map, we release the lock.
966 	 */
967 	vm_map_unlock_read(map);
968 
969 	/*
970 	 * Zero the last entries in the byte vector.
971 	 */
972 	vecindex = OFF_TO_IDX(end - first_addr);
973 	while ((lastvecindex + 1) < vecindex) {
974 		error = subyte(vec + lastvecindex, 0);
975 		if (error) {
976 			error = EFAULT;
977 			goto done2;
978 		}
979 		++lastvecindex;
980 	}
981 
982 	/*
983 	 * If the map has changed, due to the subyte, the previous
984 	 * output may be invalid.
985 	 */
986 	vm_map_lock_read(map);
987 	if (timestamp != map->timestamp)
988 		goto RestartScan;
989 	vm_map_unlock_read(map);
990 done2:
991 	return (error);
992 }
993 
994 #ifndef _SYS_SYSPROTO_H_
995 struct mlock_args {
996 	const void *addr;
997 	size_t len;
998 };
999 #endif
1000 /*
1001  * MPSAFE
1002  */
1003 int
1004 sys_mlock(td, uap)
1005 	struct thread *td;
1006 	struct mlock_args *uap;
1007 {
1008 	struct proc *proc;
1009 	vm_offset_t addr, end, last, start;
1010 	vm_size_t npages, size;
1011 	unsigned long nsize;
1012 	int error;
1013 
1014 	error = priv_check(td, PRIV_VM_MLOCK);
1015 	if (error)
1016 		return (error);
1017 	addr = (vm_offset_t)uap->addr;
1018 	size = uap->len;
1019 	last = addr + size;
1020 	start = trunc_page(addr);
1021 	end = round_page(last);
1022 	if (last < addr || end < addr)
1023 		return (EINVAL);
1024 	npages = atop(end - start);
1025 	if (npages > vm_page_max_wired)
1026 		return (ENOMEM);
1027 	proc = td->td_proc;
1028 	PROC_LOCK(proc);
1029 	nsize = ptoa(npages +
1030 	    pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map)));
1031 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1032 		PROC_UNLOCK(proc);
1033 		return (ENOMEM);
1034 	}
1035 	PROC_UNLOCK(proc);
1036 	if (npages + cnt.v_wire_count > vm_page_max_wired)
1037 		return (EAGAIN);
1038 #ifdef RACCT
1039 	PROC_LOCK(proc);
1040 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1041 	PROC_UNLOCK(proc);
1042 	if (error != 0)
1043 		return (ENOMEM);
1044 #endif
1045 	error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1046 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1047 #ifdef RACCT
1048 	if (error != KERN_SUCCESS) {
1049 		PROC_LOCK(proc);
1050 		racct_set(proc, RACCT_MEMLOCK,
1051 		    ptoa(pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))));
1052 		PROC_UNLOCK(proc);
1053 	}
1054 #endif
1055 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1056 }
1057 
1058 #ifndef _SYS_SYSPROTO_H_
1059 struct mlockall_args {
1060 	int	how;
1061 };
1062 #endif
1063 
1064 /*
1065  * MPSAFE
1066  */
1067 int
1068 sys_mlockall(td, uap)
1069 	struct thread *td;
1070 	struct mlockall_args *uap;
1071 {
1072 	vm_map_t map;
1073 	int error;
1074 
1075 	map = &td->td_proc->p_vmspace->vm_map;
1076 	error = 0;
1077 
1078 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1079 		return (EINVAL);
1080 
1081 #if 0
1082 	/*
1083 	 * If wiring all pages in the process would cause it to exceed
1084 	 * a hard resource limit, return ENOMEM.
1085 	 */
1086 	PROC_LOCK(td->td_proc);
1087 	if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1088 		PROC_UNLOCK(td->td_proc);
1089 		return (ENOMEM);
1090 	}
1091 	PROC_UNLOCK(td->td_proc);
1092 #else
1093 	error = priv_check(td, PRIV_VM_MLOCK);
1094 	if (error)
1095 		return (error);
1096 #endif
1097 #ifdef RACCT
1098 	PROC_LOCK(td->td_proc);
1099 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1100 	PROC_UNLOCK(td->td_proc);
1101 	if (error != 0)
1102 		return (ENOMEM);
1103 #endif
1104 
1105 	if (uap->how & MCL_FUTURE) {
1106 		vm_map_lock(map);
1107 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1108 		vm_map_unlock(map);
1109 		error = 0;
1110 	}
1111 
1112 	if (uap->how & MCL_CURRENT) {
1113 		/*
1114 		 * P1003.1-2001 mandates that all currently mapped pages
1115 		 * will be memory resident and locked (wired) upon return
1116 		 * from mlockall(). vm_map_wire() will wire pages, by
1117 		 * calling vm_fault_wire() for each page in the region.
1118 		 */
1119 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1120 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1121 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1122 	}
1123 #ifdef RACCT
1124 	if (error != KERN_SUCCESS) {
1125 		PROC_LOCK(td->td_proc);
1126 		racct_set(td->td_proc, RACCT_MEMLOCK,
1127 		    ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))));
1128 		PROC_UNLOCK(td->td_proc);
1129 	}
1130 #endif
1131 
1132 	return (error);
1133 }
1134 
1135 #ifndef _SYS_SYSPROTO_H_
1136 struct munlockall_args {
1137 	register_t dummy;
1138 };
1139 #endif
1140 
1141 /*
1142  * MPSAFE
1143  */
1144 int
1145 sys_munlockall(td, uap)
1146 	struct thread *td;
1147 	struct munlockall_args *uap;
1148 {
1149 	vm_map_t map;
1150 	int error;
1151 
1152 	map = &td->td_proc->p_vmspace->vm_map;
1153 	error = priv_check(td, PRIV_VM_MUNLOCK);
1154 	if (error)
1155 		return (error);
1156 
1157 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1158 	vm_map_lock(map);
1159 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1160 	vm_map_unlock(map);
1161 
1162 	/* Forcibly unwire all pages. */
1163 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1164 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1165 #ifdef RACCT
1166 	if (error == KERN_SUCCESS) {
1167 		PROC_LOCK(td->td_proc);
1168 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1169 		PROC_UNLOCK(td->td_proc);
1170 	}
1171 #endif
1172 
1173 	return (error);
1174 }
1175 
1176 #ifndef _SYS_SYSPROTO_H_
1177 struct munlock_args {
1178 	const void *addr;
1179 	size_t len;
1180 };
1181 #endif
1182 /*
1183  * MPSAFE
1184  */
1185 int
1186 sys_munlock(td, uap)
1187 	struct thread *td;
1188 	struct munlock_args *uap;
1189 {
1190 	vm_offset_t addr, end, last, start;
1191 	vm_size_t size;
1192 	int error;
1193 
1194 	error = priv_check(td, PRIV_VM_MUNLOCK);
1195 	if (error)
1196 		return (error);
1197 	addr = (vm_offset_t)uap->addr;
1198 	size = uap->len;
1199 	last = addr + size;
1200 	start = trunc_page(addr);
1201 	end = round_page(last);
1202 	if (last < addr || end < addr)
1203 		return (EINVAL);
1204 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1205 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1206 #ifdef RACCT
1207 	if (error == KERN_SUCCESS) {
1208 		PROC_LOCK(td->td_proc);
1209 		racct_sub(td->td_proc, RACCT_MEMLOCK, ptoa(end - start));
1210 		PROC_UNLOCK(td->td_proc);
1211 	}
1212 #endif
1213 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1214 }
1215 
1216 /*
1217  * vm_mmap_vnode()
1218  *
1219  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1220  * operations on vnodes.
1221  *
1222  * For VCHR vnodes, the vnode lock is held over the call to
1223  * vm_mmap_cdev() to keep vp->v_rdev valid.
1224  */
1225 int
1226 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1227     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1228     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1229     boolean_t *writecounted)
1230 {
1231 	struct vattr va;
1232 	vm_object_t obj;
1233 	vm_offset_t foff;
1234 	struct mount *mp;
1235 	struct ucred *cred;
1236 	int error, flags, locktype, vfslocked;
1237 
1238 	mp = vp->v_mount;
1239 	cred = td->td_ucred;
1240 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1241 		locktype = LK_EXCLUSIVE;
1242 	else
1243 		locktype = LK_SHARED;
1244 	vfslocked = VFS_LOCK_GIANT(mp);
1245 	if ((error = vget(vp, locktype, td)) != 0) {
1246 		VFS_UNLOCK_GIANT(vfslocked);
1247 		return (error);
1248 	}
1249 	foff = *foffp;
1250 	flags = *flagsp;
1251 	obj = vp->v_object;
1252 	if (vp->v_type == VREG) {
1253 		/*
1254 		 * Get the proper underlying object
1255 		 */
1256 		if (obj == NULL) {
1257 			error = EINVAL;
1258 			goto done;
1259 		}
1260 		if (obj->handle != vp) {
1261 			vput(vp);
1262 			vp = (struct vnode *)obj->handle;
1263 			/*
1264 			 * Bypass filesystems obey the mpsafety of the
1265 			 * underlying fs.
1266 			 */
1267 			error = vget(vp, locktype, td);
1268 			if (error != 0) {
1269 				VFS_UNLOCK_GIANT(vfslocked);
1270 				return (error);
1271 			}
1272 		}
1273 		if (locktype == LK_EXCLUSIVE) {
1274 			*writecounted = TRUE;
1275 			vnode_pager_update_writecount(obj, 0, objsize);
1276 		}
1277 	} else if (vp->v_type == VCHR) {
1278 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1279 		    vp->v_rdev, foffp, objp);
1280 		if (error == 0)
1281 			goto mark_atime;
1282 		goto done;
1283 	} else {
1284 		error = EINVAL;
1285 		goto done;
1286 	}
1287 	if ((error = VOP_GETATTR(vp, &va, cred)))
1288 		goto done;
1289 #ifdef MAC
1290 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1291 	if (error != 0)
1292 		goto done;
1293 #endif
1294 	if ((flags & MAP_SHARED) != 0) {
1295 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1296 			if (prot & PROT_WRITE) {
1297 				error = EPERM;
1298 				goto done;
1299 			}
1300 			*maxprotp &= ~VM_PROT_WRITE;
1301 		}
1302 	}
1303 	/*
1304 	 * If it is a regular file without any references
1305 	 * we do not need to sync it.
1306 	 * Adjust object size to be the size of actual file.
1307 	 */
1308 	objsize = round_page(va.va_size);
1309 	if (va.va_nlink == 0)
1310 		flags |= MAP_NOSYNC;
1311 	obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, cred);
1312 	if (obj == NULL) {
1313 		error = ENOMEM;
1314 		goto done;
1315 	}
1316 	*objp = obj;
1317 	*flagsp = flags;
1318 
1319 mark_atime:
1320 	vfs_mark_atime(vp, cred);
1321 
1322 done:
1323 	vput(vp);
1324 	VFS_UNLOCK_GIANT(vfslocked);
1325 	return (error);
1326 }
1327 
1328 /*
1329  * vm_mmap_cdev()
1330  *
1331  * MPSAFE
1332  *
1333  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1334  * operations on cdevs.
1335  */
1336 int
1337 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1338     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1339     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1340 {
1341 	vm_object_t obj;
1342 	struct cdevsw *dsw;
1343 	int error, flags, ref;
1344 
1345 	flags = *flagsp;
1346 
1347 	dsw = dev_refthread(cdev, &ref);
1348 	if (dsw == NULL)
1349 		return (ENXIO);
1350 	if (dsw->d_flags & D_MMAP_ANON) {
1351 		dev_relthread(cdev, ref);
1352 		*maxprotp = VM_PROT_ALL;
1353 		*flagsp |= MAP_ANON;
1354 		return (0);
1355 	}
1356 	/*
1357 	 * cdevs do not provide private mappings of any kind.
1358 	 */
1359 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1360 	    (prot & PROT_WRITE) != 0) {
1361 		dev_relthread(cdev, ref);
1362 		return (EACCES);
1363 	}
1364 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1365 		dev_relthread(cdev, ref);
1366 		return (EINVAL);
1367 	}
1368 	/*
1369 	 * Force device mappings to be shared.
1370 	 */
1371 	flags |= MAP_SHARED;
1372 #ifdef MAC_XXX
1373 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1374 	if (error != 0) {
1375 		dev_relthread(cdev, ref);
1376 		return (error);
1377 	}
1378 #endif
1379 	/*
1380 	 * First, try d_mmap_single().  If that is not implemented
1381 	 * (returns ENODEV), fall back to using the device pager.
1382 	 * Note that d_mmap_single() must return a reference to the
1383 	 * object (it needs to bump the reference count of the object
1384 	 * it returns somehow).
1385 	 *
1386 	 * XXX assumes VM_PROT_* == PROT_*
1387 	 */
1388 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1389 	dev_relthread(cdev, ref);
1390 	if (error != ENODEV)
1391 		return (error);
1392 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1393 	    td->td_ucred);
1394 	if (obj == NULL)
1395 		return (EINVAL);
1396 	*objp = obj;
1397 	*flagsp = flags;
1398 	return (0);
1399 }
1400 
1401 /*
1402  * vm_mmap_shm()
1403  *
1404  * MPSAFE
1405  *
1406  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1407  * operations on shm file descriptors.
1408  */
1409 int
1410 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1411     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1412     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1413 {
1414 	int error;
1415 
1416 	if ((*flagsp & MAP_SHARED) != 0 &&
1417 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1418 	    (prot & PROT_WRITE) != 0)
1419 		return (EACCES);
1420 #ifdef MAC
1421 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1422 	if (error != 0)
1423 		return (error);
1424 #endif
1425 	error = shm_mmap(shmfd, objsize, foff, objp);
1426 	if (error)
1427 		return (error);
1428 	return (0);
1429 }
1430 
1431 /*
1432  * vm_mmap()
1433  *
1434  * MPSAFE
1435  *
1436  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1437  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1438  */
1439 int
1440 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1441 	vm_prot_t maxprot, int flags,
1442 	objtype_t handle_type, void *handle,
1443 	vm_ooffset_t foff)
1444 {
1445 	boolean_t fitit;
1446 	vm_object_t object = NULL;
1447 	struct thread *td = curthread;
1448 	int docow, error, rv;
1449 	boolean_t writecounted;
1450 
1451 	if (size == 0)
1452 		return (0);
1453 
1454 	size = round_page(size);
1455 
1456 	if (map == &td->td_proc->p_vmspace->vm_map) {
1457 		PROC_LOCK(td->td_proc);
1458 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1459 			PROC_UNLOCK(td->td_proc);
1460 			return (ENOMEM);
1461 		}
1462 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1463 			PROC_UNLOCK(td->td_proc);
1464 			return (ENOMEM);
1465 		}
1466 		PROC_UNLOCK(td->td_proc);
1467 	}
1468 
1469 	/*
1470 	 * We currently can only deal with page aligned file offsets.
1471 	 * The check is here rather than in the syscall because the
1472 	 * kernel calls this function internally for other mmaping
1473 	 * operations (such as in exec) and non-aligned offsets will
1474 	 * cause pmap inconsistencies...so we want to be sure to
1475 	 * disallow this in all cases.
1476 	 */
1477 	if (foff & PAGE_MASK)
1478 		return (EINVAL);
1479 
1480 	if ((flags & MAP_FIXED) == 0) {
1481 		fitit = TRUE;
1482 		*addr = round_page(*addr);
1483 	} else {
1484 		if (*addr != trunc_page(*addr))
1485 			return (EINVAL);
1486 		fitit = FALSE;
1487 	}
1488 	writecounted = FALSE;
1489 
1490 	/*
1491 	 * Lookup/allocate object.
1492 	 */
1493 	switch (handle_type) {
1494 	case OBJT_DEVICE:
1495 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1496 		    handle, &foff, &object);
1497 		break;
1498 	case OBJT_VNODE:
1499 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1500 		    handle, &foff, &object, &writecounted);
1501 		break;
1502 	case OBJT_SWAP:
1503 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1504 		    handle, foff, &object);
1505 		break;
1506 	case OBJT_DEFAULT:
1507 		if (handle == NULL) {
1508 			error = 0;
1509 			break;
1510 		}
1511 		/* FALLTHROUGH */
1512 	default:
1513 		error = EINVAL;
1514 		break;
1515 	}
1516 	if (error)
1517 		return (error);
1518 	if (flags & MAP_ANON) {
1519 		object = NULL;
1520 		docow = 0;
1521 		/*
1522 		 * Unnamed anonymous regions always start at 0.
1523 		 */
1524 		if (handle == 0)
1525 			foff = 0;
1526 	} else if (flags & MAP_PREFAULT_READ)
1527 		docow = MAP_PREFAULT;
1528 	else
1529 		docow = MAP_PREFAULT_PARTIAL;
1530 
1531 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1532 		docow |= MAP_COPY_ON_WRITE;
1533 	if (flags & MAP_NOSYNC)
1534 		docow |= MAP_DISABLE_SYNCER;
1535 	if (flags & MAP_NOCORE)
1536 		docow |= MAP_DISABLE_COREDUMP;
1537 	/* Shared memory is also shared with children. */
1538 	if (flags & MAP_SHARED)
1539 		docow |= MAP_INHERIT_SHARE;
1540 	if (writecounted)
1541 		docow |= MAP_VN_WRITECOUNT;
1542 
1543 	if (flags & MAP_STACK)
1544 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1545 		    docow | MAP_STACK_GROWS_DOWN);
1546 	else if (fitit)
1547 		rv = vm_map_find(map, object, foff, addr, size,
1548 		    object != NULL && object->type == OBJT_DEVICE ?
1549 		    VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1550 	else
1551 		rv = vm_map_fixed(map, object, foff, *addr, size,
1552 				 prot, maxprot, docow);
1553 
1554 	if (rv == KERN_SUCCESS) {
1555 		/*
1556 		 * If the process has requested that all future mappings
1557 		 * be wired, then heed this.
1558 		 */
1559 		if (map->flags & MAP_WIREFUTURE)
1560 			vm_map_wire(map, *addr, *addr + size,
1561 			    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1562 	} else {
1563 		/*
1564 		 * If this mapping was accounted for in the vnode's
1565 		 * writecount, then undo that now.
1566 		 */
1567 		if (writecounted)
1568 			vnode_pager_release_writecount(object, 0, size);
1569 		/*
1570 		 * Lose the object reference.  Will destroy the
1571 		 * object if it's an unnamed anonymous mapping
1572 		 * or named anonymous without other references.
1573 		 */
1574 		vm_object_deallocate(object);
1575 	}
1576 	return (vm_mmap_to_errno(rv));
1577 }
1578 
1579 /*
1580  * Translate a Mach VM return code to zero on success or the appropriate errno
1581  * on failure.
1582  */
1583 int
1584 vm_mmap_to_errno(int rv)
1585 {
1586 
1587 	switch (rv) {
1588 	case KERN_SUCCESS:
1589 		return (0);
1590 	case KERN_INVALID_ADDRESS:
1591 	case KERN_NO_SPACE:
1592 		return (ENOMEM);
1593 	case KERN_PROTECTION_FAILURE:
1594 		return (EACCES);
1595 	default:
1596 		return (EINVAL);
1597 	}
1598 }
1599