xref: /freebsd/sys/vm/vm_mmap.c (revision 38f0b757fd84d17d0fc24739a7cda160c4516d81)
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_RW | CTLFLAG_TUN, &old_mlock, 0,
96     "Do not apply RLIMIT_MEMLOCK on mlockall");
97 TUNABLE_INT("vm.old_mlock", &old_mlock);
98 
99 #ifdef MAP_32BIT
100 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
101 #endif
102 
103 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
104     int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
105 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
106     int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
107 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
108     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
109 
110 #ifndef _SYS_SYSPROTO_H_
111 struct sbrk_args {
112 	int incr;
113 };
114 #endif
115 
116 /*
117  * MPSAFE
118  */
119 /* ARGSUSED */
120 int
121 sys_sbrk(td, uap)
122 	struct thread *td;
123 	struct sbrk_args *uap;
124 {
125 	/* Not yet implemented */
126 	return (EOPNOTSUPP);
127 }
128 
129 #ifndef _SYS_SYSPROTO_H_
130 struct sstk_args {
131 	int incr;
132 };
133 #endif
134 
135 /*
136  * MPSAFE
137  */
138 /* ARGSUSED */
139 int
140 sys_sstk(td, uap)
141 	struct thread *td;
142 	struct sstk_args *uap;
143 {
144 	/* Not yet implemented */
145 	return (EOPNOTSUPP);
146 }
147 
148 #if defined(COMPAT_43)
149 #ifndef _SYS_SYSPROTO_H_
150 struct getpagesize_args {
151 	int dummy;
152 };
153 #endif
154 
155 int
156 ogetpagesize(td, uap)
157 	struct thread *td;
158 	struct getpagesize_args *uap;
159 {
160 	/* MP SAFE */
161 	td->td_retval[0] = PAGE_SIZE;
162 	return (0);
163 }
164 #endif				/* COMPAT_43 */
165 
166 
167 /*
168  * Memory Map (mmap) system call.  Note that the file offset
169  * and address are allowed to be NOT page aligned, though if
170  * the MAP_FIXED flag it set, both must have the same remainder
171  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
172  * page-aligned, the actual mapping starts at trunc_page(addr)
173  * and the return value is adjusted up by the page offset.
174  *
175  * Generally speaking, only character devices which are themselves
176  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
177  * there would be no cache coherency between a descriptor and a VM mapping
178  * both to the same character device.
179  */
180 #ifndef _SYS_SYSPROTO_H_
181 struct mmap_args {
182 	void *addr;
183 	size_t len;
184 	int prot;
185 	int flags;
186 	int fd;
187 	long pad;
188 	off_t pos;
189 };
190 #endif
191 
192 /*
193  * MPSAFE
194  */
195 int
196 sys_mmap(td, uap)
197 	struct thread *td;
198 	struct mmap_args *uap;
199 {
200 #ifdef HWPMC_HOOKS
201 	struct pmckern_map_in pkm;
202 #endif
203 	struct file *fp;
204 	struct vnode *vp;
205 	vm_offset_t addr;
206 	vm_size_t size, pageoff;
207 	vm_prot_t cap_maxprot, prot, maxprot;
208 	void *handle;
209 	objtype_t handle_type;
210 	int align, error, flags;
211 	off_t pos;
212 	struct vmspace *vms = td->td_proc->p_vmspace;
213 	cap_rights_t rights;
214 
215 	addr = (vm_offset_t) uap->addr;
216 	size = uap->len;
217 	prot = uap->prot & VM_PROT_ALL;
218 	flags = uap->flags;
219 	pos = uap->pos;
220 
221 	fp = NULL;
222 
223 	/*
224 	 * Enforce the constraints.
225 	 * Mapping of length 0 is only allowed for old binaries.
226 	 * Anonymous mapping shall specify -1 as filedescriptor and
227 	 * zero position for new code. Be nice to ancient a.out
228 	 * binaries and correct pos for anonymous mapping, since old
229 	 * ld.so sometimes issues anonymous map requests with non-zero
230 	 * pos.
231 	 */
232 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
233 		if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
234 		    ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
235 			return (EINVAL);
236 	} else {
237 		if ((flags & MAP_ANON) != 0)
238 			pos = 0;
239 	}
240 
241 	if (flags & MAP_STACK) {
242 		if ((uap->fd != -1) ||
243 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
244 			return (EINVAL);
245 		flags |= MAP_ANON;
246 		pos = 0;
247 	}
248 
249 	/*
250 	 * Align the file position to a page boundary,
251 	 * and save its page offset component.
252 	 */
253 	pageoff = (pos & PAGE_MASK);
254 	pos -= pageoff;
255 
256 	/* Adjust size for rounding (on both ends). */
257 	size += pageoff;			/* low end... */
258 	size = (vm_size_t) round_page(size);	/* hi end */
259 
260 	/* Ensure alignment is at least a page and fits in a pointer. */
261 	align = flags & MAP_ALIGNMENT_MASK;
262 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
263 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
264 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
265 		return (EINVAL);
266 
267 	/*
268 	 * Check for illegal addresses.  Watch out for address wrap... Note
269 	 * that VM_*_ADDRESS are not constants due to casts (argh).
270 	 */
271 	if (flags & MAP_FIXED) {
272 		/*
273 		 * The specified address must have the same remainder
274 		 * as the file offset taken modulo PAGE_SIZE, so it
275 		 * should be aligned after adjustment by pageoff.
276 		 */
277 		addr -= pageoff;
278 		if (addr & PAGE_MASK)
279 			return (EINVAL);
280 
281 		/* Address range must be all in user VM space. */
282 		if (addr < vm_map_min(&vms->vm_map) ||
283 		    addr + size > vm_map_max(&vms->vm_map))
284 			return (EINVAL);
285 		if (addr + size < addr)
286 			return (EINVAL);
287 #ifdef MAP_32BIT
288 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
289 			return (EINVAL);
290 	} else if (flags & MAP_32BIT) {
291 		/*
292 		 * For MAP_32BIT, override the hint if it is too high and
293 		 * do not bother moving the mapping past the heap (since
294 		 * the heap is usually above 2GB).
295 		 */
296 		if (addr + size > MAP_32BIT_MAX_ADDR)
297 			addr = 0;
298 #endif
299 	} else {
300 		/*
301 		 * XXX for non-fixed mappings where no hint is provided or
302 		 * the hint would fall in the potential heap space,
303 		 * place it after the end of the largest possible heap.
304 		 *
305 		 * There should really be a pmap call to determine a reasonable
306 		 * location.
307 		 */
308 		PROC_LOCK(td->td_proc);
309 		if (addr == 0 ||
310 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
311 		    addr < round_page((vm_offset_t)vms->vm_daddr +
312 		    lim_max(td->td_proc, RLIMIT_DATA))))
313 			addr = round_page((vm_offset_t)vms->vm_daddr +
314 			    lim_max(td->td_proc, RLIMIT_DATA));
315 		PROC_UNLOCK(td->td_proc);
316 	}
317 	if (flags & MAP_ANON) {
318 		/*
319 		 * Mapping blank space is trivial.
320 		 */
321 		handle = NULL;
322 		handle_type = OBJT_DEFAULT;
323 		maxprot = VM_PROT_ALL;
324 		cap_maxprot = VM_PROT_ALL;
325 	} else {
326 		/*
327 		 * Mapping file, get fp for validation and don't let the
328 		 * descriptor disappear on us if we block. Check capability
329 		 * rights, but also return the maximum rights to be combined
330 		 * with maxprot later.
331 		 */
332 		cap_rights_init(&rights, CAP_MMAP);
333 		if (prot & PROT_READ)
334 			cap_rights_set(&rights, CAP_MMAP_R);
335 		if ((flags & MAP_SHARED) != 0) {
336 			if (prot & PROT_WRITE)
337 				cap_rights_set(&rights, CAP_MMAP_W);
338 		}
339 		if (prot & PROT_EXEC)
340 			cap_rights_set(&rights, CAP_MMAP_X);
341 		error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
342 		if (error != 0)
343 			goto done;
344 		if (fp->f_type == DTYPE_SHM) {
345 			handle = fp->f_data;
346 			handle_type = OBJT_SWAP;
347 			maxprot = VM_PROT_NONE;
348 
349 			/* FREAD should always be set. */
350 			if (fp->f_flag & FREAD)
351 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
352 			if (fp->f_flag & FWRITE)
353 				maxprot |= VM_PROT_WRITE;
354 			goto map;
355 		}
356 		if (fp->f_type != DTYPE_VNODE) {
357 			error = ENODEV;
358 			goto done;
359 		}
360 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
361     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
362 		/*
363 		 * POSIX shared-memory objects are defined to have
364 		 * kernel persistence, and are not defined to support
365 		 * read(2)/write(2) -- or even open(2).  Thus, we can
366 		 * use MAP_ASYNC to trade on-disk coherence for speed.
367 		 * The shm_open(3) library routine turns on the FPOSIXSHM
368 		 * flag to request this behavior.
369 		 */
370 		if (fp->f_flag & FPOSIXSHM)
371 			flags |= MAP_NOSYNC;
372 #endif
373 		vp = fp->f_vnode;
374 		/*
375 		 * Ensure that file and memory protections are
376 		 * compatible.  Note that we only worry about
377 		 * writability if mapping is shared; in this case,
378 		 * current and max prot are dictated by the open file.
379 		 * XXX use the vnode instead?  Problem is: what
380 		 * credentials do we use for determination? What if
381 		 * proc does a setuid?
382 		 */
383 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
384 			maxprot = VM_PROT_NONE;
385 		else
386 			maxprot = VM_PROT_EXECUTE;
387 		if (fp->f_flag & FREAD) {
388 			maxprot |= VM_PROT_READ;
389 		} else if (prot & PROT_READ) {
390 			error = EACCES;
391 			goto done;
392 		}
393 		/*
394 		 * If we are sharing potential changes (either via
395 		 * MAP_SHARED or via the implicit sharing of character
396 		 * device mappings), and we are trying to get write
397 		 * permission although we opened it without asking
398 		 * for it, bail out.
399 		 */
400 		if ((flags & MAP_SHARED) != 0) {
401 			if ((fp->f_flag & FWRITE) != 0) {
402 				maxprot |= VM_PROT_WRITE;
403 			} else if ((prot & PROT_WRITE) != 0) {
404 				error = EACCES;
405 				goto done;
406 			}
407 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
408 			maxprot |= VM_PROT_WRITE;
409 			cap_maxprot |= VM_PROT_WRITE;
410 		}
411 		handle = (void *)vp;
412 		handle_type = OBJT_VNODE;
413 	}
414 map:
415 	td->td_fpop = fp;
416 	maxprot &= cap_maxprot;
417 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
418 	    flags, handle_type, handle, pos);
419 	td->td_fpop = NULL;
420 #ifdef HWPMC_HOOKS
421 	/* inform hwpmc(4) if an executable is being mapped */
422 	if (error == 0 && handle_type == OBJT_VNODE &&
423 	    (prot & PROT_EXEC)) {
424 		pkm.pm_file = handle;
425 		pkm.pm_address = (uintptr_t) addr;
426 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
427 	}
428 #endif
429 	if (error == 0)
430 		td->td_retval[0] = (register_t) (addr + pageoff);
431 done:
432 	if (fp)
433 		fdrop(fp, td);
434 
435 	return (error);
436 }
437 
438 int
439 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
440 {
441 	struct mmap_args oargs;
442 
443 	oargs.addr = uap->addr;
444 	oargs.len = uap->len;
445 	oargs.prot = uap->prot;
446 	oargs.flags = uap->flags;
447 	oargs.fd = uap->fd;
448 	oargs.pos = uap->pos;
449 	return (sys_mmap(td, &oargs));
450 }
451 
452 #ifdef COMPAT_43
453 #ifndef _SYS_SYSPROTO_H_
454 struct ommap_args {
455 	caddr_t addr;
456 	int len;
457 	int prot;
458 	int flags;
459 	int fd;
460 	long pos;
461 };
462 #endif
463 int
464 ommap(td, uap)
465 	struct thread *td;
466 	struct ommap_args *uap;
467 {
468 	struct mmap_args nargs;
469 	static const char cvtbsdprot[8] = {
470 		0,
471 		PROT_EXEC,
472 		PROT_WRITE,
473 		PROT_EXEC | PROT_WRITE,
474 		PROT_READ,
475 		PROT_EXEC | PROT_READ,
476 		PROT_WRITE | PROT_READ,
477 		PROT_EXEC | PROT_WRITE | PROT_READ,
478 	};
479 
480 #define	OMAP_ANON	0x0002
481 #define	OMAP_COPY	0x0020
482 #define	OMAP_SHARED	0x0010
483 #define	OMAP_FIXED	0x0100
484 
485 	nargs.addr = uap->addr;
486 	nargs.len = uap->len;
487 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
488 #ifdef COMPAT_FREEBSD32
489 #if defined(__amd64__) || defined(__ia64__)
490 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
491 	    nargs.prot != 0)
492 		nargs.prot |= PROT_EXEC;
493 #endif
494 #endif
495 	nargs.flags = 0;
496 	if (uap->flags & OMAP_ANON)
497 		nargs.flags |= MAP_ANON;
498 	if (uap->flags & OMAP_COPY)
499 		nargs.flags |= MAP_COPY;
500 	if (uap->flags & OMAP_SHARED)
501 		nargs.flags |= MAP_SHARED;
502 	else
503 		nargs.flags |= MAP_PRIVATE;
504 	if (uap->flags & OMAP_FIXED)
505 		nargs.flags |= MAP_FIXED;
506 	nargs.fd = uap->fd;
507 	nargs.pos = uap->pos;
508 	return (sys_mmap(td, &nargs));
509 }
510 #endif				/* COMPAT_43 */
511 
512 
513 #ifndef _SYS_SYSPROTO_H_
514 struct msync_args {
515 	void *addr;
516 	size_t len;
517 	int flags;
518 };
519 #endif
520 /*
521  * MPSAFE
522  */
523 int
524 sys_msync(td, uap)
525 	struct thread *td;
526 	struct msync_args *uap;
527 {
528 	vm_offset_t addr;
529 	vm_size_t size, pageoff;
530 	int flags;
531 	vm_map_t map;
532 	int rv;
533 
534 	addr = (vm_offset_t) uap->addr;
535 	size = uap->len;
536 	flags = uap->flags;
537 
538 	pageoff = (addr & PAGE_MASK);
539 	addr -= pageoff;
540 	size += pageoff;
541 	size = (vm_size_t) round_page(size);
542 	if (addr + size < addr)
543 		return (EINVAL);
544 
545 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
546 		return (EINVAL);
547 
548 	map = &td->td_proc->p_vmspace->vm_map;
549 
550 	/*
551 	 * Clean the pages and interpret the return value.
552 	 */
553 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
554 	    (flags & MS_INVALIDATE) != 0);
555 	switch (rv) {
556 	case KERN_SUCCESS:
557 		return (0);
558 	case KERN_INVALID_ADDRESS:
559 		return (EINVAL);	/* Sun returns ENOMEM? */
560 	case KERN_INVALID_ARGUMENT:
561 		return (EBUSY);
562 	case KERN_FAILURE:
563 		return (EIO);
564 	default:
565 		return (EINVAL);
566 	}
567 }
568 
569 #ifndef _SYS_SYSPROTO_H_
570 struct munmap_args {
571 	void *addr;
572 	size_t len;
573 };
574 #endif
575 /*
576  * MPSAFE
577  */
578 int
579 sys_munmap(td, uap)
580 	struct thread *td;
581 	struct munmap_args *uap;
582 {
583 #ifdef HWPMC_HOOKS
584 	struct pmckern_map_out pkm;
585 	vm_map_entry_t entry;
586 #endif
587 	vm_offset_t addr;
588 	vm_size_t size, pageoff;
589 	vm_map_t map;
590 
591 	addr = (vm_offset_t) uap->addr;
592 	size = uap->len;
593 	if (size == 0)
594 		return (EINVAL);
595 
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 	/*
604 	 * Check for illegal addresses.  Watch out for address wrap...
605 	 */
606 	map = &td->td_proc->p_vmspace->vm_map;
607 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
608 		return (EINVAL);
609 	vm_map_lock(map);
610 #ifdef HWPMC_HOOKS
611 	/*
612 	 * Inform hwpmc if the address range being unmapped contains
613 	 * an executable region.
614 	 */
615 	pkm.pm_address = (uintptr_t) NULL;
616 	if (vm_map_lookup_entry(map, addr, &entry)) {
617 		for (;
618 		     entry != &map->header && entry->start < addr + size;
619 		     entry = entry->next) {
620 			if (vm_map_check_protection(map, entry->start,
621 				entry->end, VM_PROT_EXECUTE) == TRUE) {
622 				pkm.pm_address = (uintptr_t) addr;
623 				pkm.pm_size = (size_t) size;
624 				break;
625 			}
626 		}
627 	}
628 #endif
629 	vm_map_delete(map, addr, addr + size);
630 
631 #ifdef HWPMC_HOOKS
632 	/* downgrade the lock to prevent a LOR with the pmc-sx lock */
633 	vm_map_lock_downgrade(map);
634 	if (pkm.pm_address != (uintptr_t) NULL)
635 		PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
636 	vm_map_unlock_read(map);
637 #else
638 	vm_map_unlock(map);
639 #endif
640 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
641 	return (0);
642 }
643 
644 #ifndef _SYS_SYSPROTO_H_
645 struct mprotect_args {
646 	const void *addr;
647 	size_t len;
648 	int prot;
649 };
650 #endif
651 /*
652  * MPSAFE
653  */
654 int
655 sys_mprotect(td, uap)
656 	struct thread *td;
657 	struct mprotect_args *uap;
658 {
659 	vm_offset_t addr;
660 	vm_size_t size, pageoff;
661 	vm_prot_t prot;
662 
663 	addr = (vm_offset_t) uap->addr;
664 	size = uap->len;
665 	prot = uap->prot & VM_PROT_ALL;
666 
667 	pageoff = (addr & PAGE_MASK);
668 	addr -= pageoff;
669 	size += pageoff;
670 	size = (vm_size_t) round_page(size);
671 	if (addr + size < addr)
672 		return (EINVAL);
673 
674 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
675 	    addr + size, prot, FALSE)) {
676 	case KERN_SUCCESS:
677 		return (0);
678 	case KERN_PROTECTION_FAILURE:
679 		return (EACCES);
680 	case KERN_RESOURCE_SHORTAGE:
681 		return (ENOMEM);
682 	}
683 	return (EINVAL);
684 }
685 
686 #ifndef _SYS_SYSPROTO_H_
687 struct minherit_args {
688 	void *addr;
689 	size_t len;
690 	int inherit;
691 };
692 #endif
693 /*
694  * MPSAFE
695  */
696 int
697 sys_minherit(td, uap)
698 	struct thread *td;
699 	struct minherit_args *uap;
700 {
701 	vm_offset_t addr;
702 	vm_size_t size, pageoff;
703 	vm_inherit_t inherit;
704 
705 	addr = (vm_offset_t)uap->addr;
706 	size = uap->len;
707 	inherit = uap->inherit;
708 
709 	pageoff = (addr & PAGE_MASK);
710 	addr -= pageoff;
711 	size += pageoff;
712 	size = (vm_size_t) round_page(size);
713 	if (addr + size < addr)
714 		return (EINVAL);
715 
716 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
717 	    addr + size, inherit)) {
718 	case KERN_SUCCESS:
719 		return (0);
720 	case KERN_PROTECTION_FAILURE:
721 		return (EACCES);
722 	}
723 	return (EINVAL);
724 }
725 
726 #ifndef _SYS_SYSPROTO_H_
727 struct madvise_args {
728 	void *addr;
729 	size_t len;
730 	int behav;
731 };
732 #endif
733 
734 /*
735  * MPSAFE
736  */
737 int
738 sys_madvise(td, uap)
739 	struct thread *td;
740 	struct madvise_args *uap;
741 {
742 	vm_offset_t start, end;
743 	vm_map_t map;
744 	int flags;
745 
746 	/*
747 	 * Check for our special case, advising the swap pager we are
748 	 * "immortal."
749 	 */
750 	if (uap->behav == MADV_PROTECT) {
751 		flags = PPROT_SET;
752 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
753 		    PROC_SPROTECT, &flags));
754 	}
755 
756 	/*
757 	 * Check for illegal behavior
758 	 */
759 	if (uap->behav < 0 || uap->behav > MADV_CORE)
760 		return (EINVAL);
761 	/*
762 	 * Check for illegal addresses.  Watch out for address wrap... Note
763 	 * that VM_*_ADDRESS are not constants due to casts (argh).
764 	 */
765 	map = &td->td_proc->p_vmspace->vm_map;
766 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
767 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
768 		return (EINVAL);
769 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
770 		return (EINVAL);
771 
772 	/*
773 	 * Since this routine is only advisory, we default to conservative
774 	 * behavior.
775 	 */
776 	start = trunc_page((vm_offset_t) uap->addr);
777 	end = round_page((vm_offset_t) uap->addr + uap->len);
778 
779 	if (vm_map_madvise(map, start, end, uap->behav))
780 		return (EINVAL);
781 	return (0);
782 }
783 
784 #ifndef _SYS_SYSPROTO_H_
785 struct mincore_args {
786 	const void *addr;
787 	size_t len;
788 	char *vec;
789 };
790 #endif
791 
792 /*
793  * MPSAFE
794  */
795 int
796 sys_mincore(td, uap)
797 	struct thread *td;
798 	struct mincore_args *uap;
799 {
800 	vm_offset_t addr, first_addr;
801 	vm_offset_t end, cend;
802 	pmap_t pmap;
803 	vm_map_t map;
804 	char *vec;
805 	int error = 0;
806 	int vecindex, lastvecindex;
807 	vm_map_entry_t current;
808 	vm_map_entry_t entry;
809 	vm_object_t object;
810 	vm_paddr_t locked_pa;
811 	vm_page_t m;
812 	vm_pindex_t pindex;
813 	int mincoreinfo;
814 	unsigned int timestamp;
815 	boolean_t locked;
816 
817 	/*
818 	 * Make sure that the addresses presented are valid for user
819 	 * mode.
820 	 */
821 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
822 	end = addr + (vm_size_t)round_page(uap->len);
823 	map = &td->td_proc->p_vmspace->vm_map;
824 	if (end > vm_map_max(map) || end < addr)
825 		return (ENOMEM);
826 
827 	/*
828 	 * Address of byte vector
829 	 */
830 	vec = uap->vec;
831 
832 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
833 
834 	vm_map_lock_read(map);
835 RestartScan:
836 	timestamp = map->timestamp;
837 
838 	if (!vm_map_lookup_entry(map, addr, &entry)) {
839 		vm_map_unlock_read(map);
840 		return (ENOMEM);
841 	}
842 
843 	/*
844 	 * Do this on a map entry basis so that if the pages are not
845 	 * in the current processes address space, we can easily look
846 	 * up the pages elsewhere.
847 	 */
848 	lastvecindex = -1;
849 	for (current = entry;
850 	    (current != &map->header) && (current->start < end);
851 	    current = current->next) {
852 
853 		/*
854 		 * check for contiguity
855 		 */
856 		if (current->end < end &&
857 		    (entry->next == &map->header ||
858 		     current->next->start > current->end)) {
859 			vm_map_unlock_read(map);
860 			return (ENOMEM);
861 		}
862 
863 		/*
864 		 * ignore submaps (for now) or null objects
865 		 */
866 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
867 			current->object.vm_object == NULL)
868 			continue;
869 
870 		/*
871 		 * limit this scan to the current map entry and the
872 		 * limits for the mincore call
873 		 */
874 		if (addr < current->start)
875 			addr = current->start;
876 		cend = current->end;
877 		if (cend > end)
878 			cend = end;
879 
880 		/*
881 		 * scan this entry one page at a time
882 		 */
883 		while (addr < cend) {
884 			/*
885 			 * Check pmap first, it is likely faster, also
886 			 * it can provide info as to whether we are the
887 			 * one referencing or modifying the page.
888 			 */
889 			object = NULL;
890 			locked_pa = 0;
891 		retry:
892 			m = NULL;
893 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
894 			if (locked_pa != 0) {
895 				/*
896 				 * The page is mapped by this process but not
897 				 * both accessed and modified.  It is also
898 				 * managed.  Acquire the object lock so that
899 				 * other mappings might be examined.
900 				 */
901 				m = PHYS_TO_VM_PAGE(locked_pa);
902 				if (m->object != object) {
903 					if (object != NULL)
904 						VM_OBJECT_WUNLOCK(object);
905 					object = m->object;
906 					locked = VM_OBJECT_TRYWLOCK(object);
907 					vm_page_unlock(m);
908 					if (!locked) {
909 						VM_OBJECT_WLOCK(object);
910 						vm_page_lock(m);
911 						goto retry;
912 					}
913 				} else
914 					vm_page_unlock(m);
915 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
916 				    ("mincore: page %p is mapped but invalid",
917 				    m));
918 			} else if (mincoreinfo == 0) {
919 				/*
920 				 * The page is not mapped by this process.  If
921 				 * the object implements managed pages, then
922 				 * determine if the page is resident so that
923 				 * the mappings might be examined.
924 				 */
925 				if (current->object.vm_object != object) {
926 					if (object != NULL)
927 						VM_OBJECT_WUNLOCK(object);
928 					object = current->object.vm_object;
929 					VM_OBJECT_WLOCK(object);
930 				}
931 				if (object->type == OBJT_DEFAULT ||
932 				    object->type == OBJT_SWAP ||
933 				    object->type == OBJT_VNODE) {
934 					pindex = OFF_TO_IDX(current->offset +
935 					    (addr - current->start));
936 					m = vm_page_lookup(object, pindex);
937 					if (m == NULL &&
938 					    vm_page_is_cached(object, pindex))
939 						mincoreinfo = MINCORE_INCORE;
940 					if (m != NULL && m->valid == 0)
941 						m = NULL;
942 					if (m != NULL)
943 						mincoreinfo = MINCORE_INCORE;
944 				}
945 			}
946 			if (m != NULL) {
947 				/* Examine other mappings to the page. */
948 				if (m->dirty == 0 && pmap_is_modified(m))
949 					vm_page_dirty(m);
950 				if (m->dirty != 0)
951 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
952 				/*
953 				 * The first test for PGA_REFERENCED is an
954 				 * optimization.  The second test is
955 				 * required because a concurrent pmap
956 				 * operation could clear the last reference
957 				 * and set PGA_REFERENCED before the call to
958 				 * pmap_is_referenced().
959 				 */
960 				if ((m->aflags & PGA_REFERENCED) != 0 ||
961 				    pmap_is_referenced(m) ||
962 				    (m->aflags & PGA_REFERENCED) != 0)
963 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
964 			}
965 			if (object != NULL)
966 				VM_OBJECT_WUNLOCK(object);
967 
968 			/*
969 			 * subyte may page fault.  In case it needs to modify
970 			 * the map, we release the lock.
971 			 */
972 			vm_map_unlock_read(map);
973 
974 			/*
975 			 * calculate index into user supplied byte vector
976 			 */
977 			vecindex = OFF_TO_IDX(addr - first_addr);
978 
979 			/*
980 			 * If we have skipped map entries, we need to make sure that
981 			 * the byte vector is zeroed for those skipped entries.
982 			 */
983 			while ((lastvecindex + 1) < vecindex) {
984 				++lastvecindex;
985 				error = subyte(vec + lastvecindex, 0);
986 				if (error) {
987 					error = EFAULT;
988 					goto done2;
989 				}
990 			}
991 
992 			/*
993 			 * Pass the page information to the user
994 			 */
995 			error = subyte(vec + vecindex, mincoreinfo);
996 			if (error) {
997 				error = EFAULT;
998 				goto done2;
999 			}
1000 
1001 			/*
1002 			 * If the map has changed, due to the subyte, the previous
1003 			 * output may be invalid.
1004 			 */
1005 			vm_map_lock_read(map);
1006 			if (timestamp != map->timestamp)
1007 				goto RestartScan;
1008 
1009 			lastvecindex = vecindex;
1010 			addr += PAGE_SIZE;
1011 		}
1012 	}
1013 
1014 	/*
1015 	 * subyte may page fault.  In case it needs to modify
1016 	 * the map, we release the lock.
1017 	 */
1018 	vm_map_unlock_read(map);
1019 
1020 	/*
1021 	 * Zero the last entries in the byte vector.
1022 	 */
1023 	vecindex = OFF_TO_IDX(end - first_addr);
1024 	while ((lastvecindex + 1) < vecindex) {
1025 		++lastvecindex;
1026 		error = subyte(vec + lastvecindex, 0);
1027 		if (error) {
1028 			error = EFAULT;
1029 			goto done2;
1030 		}
1031 	}
1032 
1033 	/*
1034 	 * If the map has changed, due to the subyte, the previous
1035 	 * output may be invalid.
1036 	 */
1037 	vm_map_lock_read(map);
1038 	if (timestamp != map->timestamp)
1039 		goto RestartScan;
1040 	vm_map_unlock_read(map);
1041 done2:
1042 	return (error);
1043 }
1044 
1045 #ifndef _SYS_SYSPROTO_H_
1046 struct mlock_args {
1047 	const void *addr;
1048 	size_t len;
1049 };
1050 #endif
1051 /*
1052  * MPSAFE
1053  */
1054 int
1055 sys_mlock(td, uap)
1056 	struct thread *td;
1057 	struct mlock_args *uap;
1058 {
1059 
1060 	return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1061 }
1062 
1063 int
1064 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1065 {
1066 	vm_offset_t addr, end, last, start;
1067 	vm_size_t npages, size;
1068 	vm_map_t map;
1069 	unsigned long nsize;
1070 	int error;
1071 
1072 	error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1073 	if (error)
1074 		return (error);
1075 	addr = (vm_offset_t)addr0;
1076 	size = len;
1077 	last = addr + size;
1078 	start = trunc_page(addr);
1079 	end = round_page(last);
1080 	if (last < addr || end < addr)
1081 		return (EINVAL);
1082 	npages = atop(end - start);
1083 	if (npages > vm_page_max_wired)
1084 		return (ENOMEM);
1085 	map = &proc->p_vmspace->vm_map;
1086 	PROC_LOCK(proc);
1087 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1088 	if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1089 		PROC_UNLOCK(proc);
1090 		return (ENOMEM);
1091 	}
1092 	PROC_UNLOCK(proc);
1093 	if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1094 		return (EAGAIN);
1095 #ifdef RACCT
1096 	PROC_LOCK(proc);
1097 	error = racct_set(proc, RACCT_MEMLOCK, nsize);
1098 	PROC_UNLOCK(proc);
1099 	if (error != 0)
1100 		return (ENOMEM);
1101 #endif
1102 	error = vm_map_wire(map, start, end,
1103 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1104 #ifdef RACCT
1105 	if (error != KERN_SUCCESS) {
1106 		PROC_LOCK(proc);
1107 		racct_set(proc, RACCT_MEMLOCK,
1108 		    ptoa(pmap_wired_count(map->pmap)));
1109 		PROC_UNLOCK(proc);
1110 	}
1111 #endif
1112 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1113 }
1114 
1115 #ifndef _SYS_SYSPROTO_H_
1116 struct mlockall_args {
1117 	int	how;
1118 };
1119 #endif
1120 
1121 /*
1122  * MPSAFE
1123  */
1124 int
1125 sys_mlockall(td, uap)
1126 	struct thread *td;
1127 	struct mlockall_args *uap;
1128 {
1129 	vm_map_t map;
1130 	int error;
1131 
1132 	map = &td->td_proc->p_vmspace->vm_map;
1133 	error = priv_check(td, PRIV_VM_MLOCK);
1134 	if (error)
1135 		return (error);
1136 
1137 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1138 		return (EINVAL);
1139 
1140 	/*
1141 	 * If wiring all pages in the process would cause it to exceed
1142 	 * a hard resource limit, return ENOMEM.
1143 	 */
1144 	if (!old_mlock && uap->how & MCL_CURRENT) {
1145 		PROC_LOCK(td->td_proc);
1146 		if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1147 			PROC_UNLOCK(td->td_proc);
1148 			return (ENOMEM);
1149 		}
1150 		PROC_UNLOCK(td->td_proc);
1151 	}
1152 #ifdef RACCT
1153 	PROC_LOCK(td->td_proc);
1154 	error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1155 	PROC_UNLOCK(td->td_proc);
1156 	if (error != 0)
1157 		return (ENOMEM);
1158 #endif
1159 
1160 	if (uap->how & MCL_FUTURE) {
1161 		vm_map_lock(map);
1162 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1163 		vm_map_unlock(map);
1164 		error = 0;
1165 	}
1166 
1167 	if (uap->how & MCL_CURRENT) {
1168 		/*
1169 		 * P1003.1-2001 mandates that all currently mapped pages
1170 		 * will be memory resident and locked (wired) upon return
1171 		 * from mlockall(). vm_map_wire() will wire pages, by
1172 		 * calling vm_fault_wire() for each page in the region.
1173 		 */
1174 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1175 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1176 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1177 	}
1178 #ifdef RACCT
1179 	if (error != KERN_SUCCESS) {
1180 		PROC_LOCK(td->td_proc);
1181 		racct_set(td->td_proc, RACCT_MEMLOCK,
1182 		    ptoa(pmap_wired_count(map->pmap)));
1183 		PROC_UNLOCK(td->td_proc);
1184 	}
1185 #endif
1186 
1187 	return (error);
1188 }
1189 
1190 #ifndef _SYS_SYSPROTO_H_
1191 struct munlockall_args {
1192 	register_t dummy;
1193 };
1194 #endif
1195 
1196 /*
1197  * MPSAFE
1198  */
1199 int
1200 sys_munlockall(td, uap)
1201 	struct thread *td;
1202 	struct munlockall_args *uap;
1203 {
1204 	vm_map_t map;
1205 	int error;
1206 
1207 	map = &td->td_proc->p_vmspace->vm_map;
1208 	error = priv_check(td, PRIV_VM_MUNLOCK);
1209 	if (error)
1210 		return (error);
1211 
1212 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1213 	vm_map_lock(map);
1214 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1215 	vm_map_unlock(map);
1216 
1217 	/* Forcibly unwire all pages. */
1218 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1219 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1220 #ifdef RACCT
1221 	if (error == KERN_SUCCESS) {
1222 		PROC_LOCK(td->td_proc);
1223 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1224 		PROC_UNLOCK(td->td_proc);
1225 	}
1226 #endif
1227 
1228 	return (error);
1229 }
1230 
1231 #ifndef _SYS_SYSPROTO_H_
1232 struct munlock_args {
1233 	const void *addr;
1234 	size_t len;
1235 };
1236 #endif
1237 /*
1238  * MPSAFE
1239  */
1240 int
1241 sys_munlock(td, uap)
1242 	struct thread *td;
1243 	struct munlock_args *uap;
1244 {
1245 	vm_offset_t addr, end, last, start;
1246 	vm_size_t size;
1247 #ifdef RACCT
1248 	vm_map_t map;
1249 #endif
1250 	int error;
1251 
1252 	error = priv_check(td, PRIV_VM_MUNLOCK);
1253 	if (error)
1254 		return (error);
1255 	addr = (vm_offset_t)uap->addr;
1256 	size = uap->len;
1257 	last = addr + size;
1258 	start = trunc_page(addr);
1259 	end = round_page(last);
1260 	if (last < addr || end < addr)
1261 		return (EINVAL);
1262 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1263 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1264 #ifdef RACCT
1265 	if (error == KERN_SUCCESS) {
1266 		PROC_LOCK(td->td_proc);
1267 		map = &td->td_proc->p_vmspace->vm_map;
1268 		racct_set(td->td_proc, RACCT_MEMLOCK,
1269 		    ptoa(pmap_wired_count(map->pmap)));
1270 		PROC_UNLOCK(td->td_proc);
1271 	}
1272 #endif
1273 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1274 }
1275 
1276 /*
1277  * vm_mmap_vnode()
1278  *
1279  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1280  * operations on vnodes.
1281  *
1282  * For VCHR vnodes, the vnode lock is held over the call to
1283  * vm_mmap_cdev() to keep vp->v_rdev valid.
1284  */
1285 int
1286 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1287     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1288     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1289     boolean_t *writecounted)
1290 {
1291 	struct vattr va;
1292 	vm_object_t obj;
1293 	vm_offset_t foff;
1294 	struct mount *mp;
1295 	struct ucred *cred;
1296 	int error, flags, locktype;
1297 
1298 	mp = vp->v_mount;
1299 	cred = td->td_ucred;
1300 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1301 		locktype = LK_EXCLUSIVE;
1302 	else
1303 		locktype = LK_SHARED;
1304 	if ((error = vget(vp, locktype, td)) != 0)
1305 		return (error);
1306 	foff = *foffp;
1307 	flags = *flagsp;
1308 	obj = vp->v_object;
1309 	if (vp->v_type == VREG) {
1310 		/*
1311 		 * Get the proper underlying object
1312 		 */
1313 		if (obj == NULL) {
1314 			error = EINVAL;
1315 			goto done;
1316 		}
1317 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1318 			vput(vp);
1319 			vp = (struct vnode *)obj->handle;
1320 			/*
1321 			 * Bypass filesystems obey the mpsafety of the
1322 			 * underlying fs.  Tmpfs never bypasses.
1323 			 */
1324 			error = vget(vp, locktype, td);
1325 			if (error != 0)
1326 				return (error);
1327 		}
1328 		if (locktype == LK_EXCLUSIVE) {
1329 			*writecounted = TRUE;
1330 			vnode_pager_update_writecount(obj, 0, objsize);
1331 		}
1332 	} else if (vp->v_type == VCHR) {
1333 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1334 		    vp->v_rdev, foffp, objp);
1335 		if (error == 0)
1336 			goto mark_atime;
1337 		goto done;
1338 	} else {
1339 		error = EINVAL;
1340 		goto done;
1341 	}
1342 	if ((error = VOP_GETATTR(vp, &va, cred)))
1343 		goto done;
1344 #ifdef MAC
1345 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1346 	if (error != 0)
1347 		goto done;
1348 #endif
1349 	if ((flags & MAP_SHARED) != 0) {
1350 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1351 			if (prot & PROT_WRITE) {
1352 				error = EPERM;
1353 				goto done;
1354 			}
1355 			*maxprotp &= ~VM_PROT_WRITE;
1356 		}
1357 	}
1358 	/*
1359 	 * If it is a regular file without any references
1360 	 * we do not need to sync it.
1361 	 * Adjust object size to be the size of actual file.
1362 	 */
1363 	objsize = round_page(va.va_size);
1364 	if (va.va_nlink == 0)
1365 		flags |= MAP_NOSYNC;
1366 	if (obj->type == OBJT_VNODE)
1367 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1368 		    cred);
1369 	else {
1370 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1371 		    ("wrong object type"));
1372 		vm_object_reference(obj);
1373 	}
1374 	if (obj == NULL) {
1375 		error = ENOMEM;
1376 		goto done;
1377 	}
1378 	*objp = obj;
1379 	*flagsp = flags;
1380 
1381 mark_atime:
1382 	vfs_mark_atime(vp, cred);
1383 
1384 done:
1385 	if (error != 0 && *writecounted) {
1386 		*writecounted = FALSE;
1387 		vnode_pager_update_writecount(obj, objsize, 0);
1388 	}
1389 	vput(vp);
1390 	return (error);
1391 }
1392 
1393 /*
1394  * vm_mmap_cdev()
1395  *
1396  * MPSAFE
1397  *
1398  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1399  * operations on cdevs.
1400  */
1401 int
1402 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1403     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1404     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1405 {
1406 	vm_object_t obj;
1407 	struct cdevsw *dsw;
1408 	int error, flags, ref;
1409 
1410 	flags = *flagsp;
1411 
1412 	dsw = dev_refthread(cdev, &ref);
1413 	if (dsw == NULL)
1414 		return (ENXIO);
1415 	if (dsw->d_flags & D_MMAP_ANON) {
1416 		dev_relthread(cdev, ref);
1417 		*maxprotp = VM_PROT_ALL;
1418 		*flagsp |= MAP_ANON;
1419 		return (0);
1420 	}
1421 	/*
1422 	 * cdevs do not provide private mappings of any kind.
1423 	 */
1424 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1425 	    (prot & PROT_WRITE) != 0) {
1426 		dev_relthread(cdev, ref);
1427 		return (EACCES);
1428 	}
1429 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1430 		dev_relthread(cdev, ref);
1431 		return (EINVAL);
1432 	}
1433 	/*
1434 	 * Force device mappings to be shared.
1435 	 */
1436 	flags |= MAP_SHARED;
1437 #ifdef MAC_XXX
1438 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1439 	if (error != 0) {
1440 		dev_relthread(cdev, ref);
1441 		return (error);
1442 	}
1443 #endif
1444 	/*
1445 	 * First, try d_mmap_single().  If that is not implemented
1446 	 * (returns ENODEV), fall back to using the device pager.
1447 	 * Note that d_mmap_single() must return a reference to the
1448 	 * object (it needs to bump the reference count of the object
1449 	 * it returns somehow).
1450 	 *
1451 	 * XXX assumes VM_PROT_* == PROT_*
1452 	 */
1453 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1454 	dev_relthread(cdev, ref);
1455 	if (error != ENODEV)
1456 		return (error);
1457 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1458 	    td->td_ucred);
1459 	if (obj == NULL)
1460 		return (EINVAL);
1461 	*objp = obj;
1462 	*flagsp = flags;
1463 	return (0);
1464 }
1465 
1466 /*
1467  * vm_mmap_shm()
1468  *
1469  * MPSAFE
1470  *
1471  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1472  * operations on shm file descriptors.
1473  */
1474 int
1475 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1476     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1477     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1478 {
1479 	int error;
1480 
1481 	if ((*flagsp & MAP_SHARED) != 0 &&
1482 	    (*maxprotp & VM_PROT_WRITE) == 0 &&
1483 	    (prot & PROT_WRITE) != 0)
1484 		return (EACCES);
1485 #ifdef MAC
1486 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1487 	if (error != 0)
1488 		return (error);
1489 #endif
1490 	error = shm_mmap(shmfd, objsize, foff, objp);
1491 	if (error)
1492 		return (error);
1493 	return (0);
1494 }
1495 
1496 /*
1497  * vm_mmap()
1498  *
1499  * MPSAFE
1500  *
1501  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1502  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1503  */
1504 int
1505 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1506 	vm_prot_t maxprot, int flags,
1507 	objtype_t handle_type, void *handle,
1508 	vm_ooffset_t foff)
1509 {
1510 	boolean_t fitit;
1511 	vm_object_t object = NULL;
1512 	struct thread *td = curthread;
1513 	int docow, error, findspace, rv;
1514 	boolean_t writecounted;
1515 
1516 	if (size == 0)
1517 		return (0);
1518 
1519 	size = round_page(size);
1520 
1521 	if (map == &td->td_proc->p_vmspace->vm_map) {
1522 		PROC_LOCK(td->td_proc);
1523 		if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1524 			PROC_UNLOCK(td->td_proc);
1525 			return (ENOMEM);
1526 		}
1527 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1528 			PROC_UNLOCK(td->td_proc);
1529 			return (ENOMEM);
1530 		}
1531 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1532 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1533 			    lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1534 				racct_set_force(td->td_proc, RACCT_VMEM,
1535 				    map->size);
1536 				PROC_UNLOCK(td->td_proc);
1537 				return (ENOMEM);
1538 			}
1539 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1540 			    ptoa(pmap_wired_count(map->pmap)) + size);
1541 			if (error != 0) {
1542 				racct_set_force(td->td_proc, RACCT_VMEM,
1543 				    map->size);
1544 				PROC_UNLOCK(td->td_proc);
1545 				return (error);
1546 			}
1547 		}
1548 		PROC_UNLOCK(td->td_proc);
1549 	}
1550 
1551 	/*
1552 	 * We currently can only deal with page aligned file offsets.
1553 	 * The check is here rather than in the syscall because the
1554 	 * kernel calls this function internally for other mmaping
1555 	 * operations (such as in exec) and non-aligned offsets will
1556 	 * cause pmap inconsistencies...so we want to be sure to
1557 	 * disallow this in all cases.
1558 	 */
1559 	if (foff & PAGE_MASK)
1560 		return (EINVAL);
1561 
1562 	if ((flags & MAP_FIXED) == 0) {
1563 		fitit = TRUE;
1564 		*addr = round_page(*addr);
1565 	} else {
1566 		if (*addr != trunc_page(*addr))
1567 			return (EINVAL);
1568 		fitit = FALSE;
1569 	}
1570 	writecounted = FALSE;
1571 
1572 	/*
1573 	 * Lookup/allocate object.
1574 	 */
1575 	switch (handle_type) {
1576 	case OBJT_DEVICE:
1577 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1578 		    handle, &foff, &object);
1579 		break;
1580 	case OBJT_VNODE:
1581 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1582 		    handle, &foff, &object, &writecounted);
1583 		break;
1584 	case OBJT_SWAP:
1585 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1586 		    handle, foff, &object);
1587 		break;
1588 	case OBJT_DEFAULT:
1589 		if (handle == NULL) {
1590 			error = 0;
1591 			break;
1592 		}
1593 		/* FALLTHROUGH */
1594 	default:
1595 		error = EINVAL;
1596 		break;
1597 	}
1598 	if (error)
1599 		return (error);
1600 	if (flags & MAP_ANON) {
1601 		object = NULL;
1602 		docow = 0;
1603 		/*
1604 		 * Unnamed anonymous regions always start at 0.
1605 		 */
1606 		if (handle == 0)
1607 			foff = 0;
1608 	} else if (flags & MAP_PREFAULT_READ)
1609 		docow = MAP_PREFAULT;
1610 	else
1611 		docow = MAP_PREFAULT_PARTIAL;
1612 
1613 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1614 		docow |= MAP_COPY_ON_WRITE;
1615 	if (flags & MAP_NOSYNC)
1616 		docow |= MAP_DISABLE_SYNCER;
1617 	if (flags & MAP_NOCORE)
1618 		docow |= MAP_DISABLE_COREDUMP;
1619 	/* Shared memory is also shared with children. */
1620 	if (flags & MAP_SHARED)
1621 		docow |= MAP_INHERIT_SHARE;
1622 	if (writecounted)
1623 		docow |= MAP_VN_WRITECOUNT;
1624 
1625 	if (flags & MAP_STACK)
1626 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1627 		    docow | MAP_STACK_GROWS_DOWN);
1628 	else if (fitit) {
1629 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1630 			findspace = VMFS_SUPER_SPACE;
1631 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1632 			findspace = VMFS_ALIGNED_SPACE(flags >>
1633 			    MAP_ALIGNMENT_SHIFT);
1634 		else
1635 			findspace = VMFS_OPTIMAL_SPACE;
1636 		rv = vm_map_find(map, object, foff, addr, size,
1637 #ifdef MAP_32BIT
1638 		    flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1639 #endif
1640 		    0, findspace, prot, maxprot, docow);
1641 	} else
1642 		rv = vm_map_fixed(map, object, foff, *addr, size,
1643 				 prot, maxprot, docow);
1644 
1645 	if (rv == KERN_SUCCESS) {
1646 		/*
1647 		 * If the process has requested that all future mappings
1648 		 * be wired, then heed this.
1649 		 */
1650 		if (map->flags & MAP_WIREFUTURE) {
1651 			vm_map_wire(map, *addr, *addr + size,
1652 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1653 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1654 		}
1655 	} else {
1656 		/*
1657 		 * If this mapping was accounted for in the vnode's
1658 		 * writecount, then undo that now.
1659 		 */
1660 		if (writecounted)
1661 			vnode_pager_release_writecount(object, 0, size);
1662 		/*
1663 		 * Lose the object reference.  Will destroy the
1664 		 * object if it's an unnamed anonymous mapping
1665 		 * or named anonymous without other references.
1666 		 */
1667 		vm_object_deallocate(object);
1668 	}
1669 	return (vm_mmap_to_errno(rv));
1670 }
1671 
1672 /*
1673  * Translate a Mach VM return code to zero on success or the appropriate errno
1674  * on failure.
1675  */
1676 int
1677 vm_mmap_to_errno(int rv)
1678 {
1679 
1680 	switch (rv) {
1681 	case KERN_SUCCESS:
1682 		return (0);
1683 	case KERN_INVALID_ADDRESS:
1684 	case KERN_NO_SPACE:
1685 		return (ENOMEM);
1686 	case KERN_PROTECTION_FAILURE:
1687 		return (EACCES);
1688 	default:
1689 		return (EINVAL);
1690 	}
1691 }
1692