xref: /freebsd/sys/vm/vm_mmap.c (revision 732a02b4e77866604a120a275c082bb6221bd2ff)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1988 University of Utah.
5  * Copyright (c) 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
37  *
38  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
39  */
40 
41 /*
42  * Mapped file (mmap) interface to VM
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_hwpmc_hooks.h"
49 #include "opt_vm.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/capsicum.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/sysproto.h>
58 #include <sys/elf.h>
59 #include <sys/filedesc.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/procctl.h>
63 #include <sys/racct.h>
64 #include <sys/resource.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
69 #include <sys/fcntl.h>
70 #include <sys/file.h>
71 #include <sys/mman.h>
72 #include <sys/mount.h>
73 #include <sys/conf.h>
74 #include <sys/stat.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/vmmeter.h>
78 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
79 #include <machine/md_var.h>
80 #endif
81 
82 #include <security/audit/audit.h>
83 #include <security/mac/mac_framework.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_pageout.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vm_page.h>
95 #include <vm/vnode_pager.h>
96 
97 #ifdef HWPMC_HOOKS
98 #include <sys/pmckern.h>
99 #endif
100 
101 int old_mlock = 0;
102 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
103     "Do not apply RLIMIT_MEMLOCK on mlockall");
104 static int mincore_mapped = 1;
105 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
106     "mincore reports mappings, not residency");
107 static int imply_prot_max = 0;
108 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0,
109     "Imply maximum page permissions in mmap() when none are specified");
110 
111 #ifdef MAP_32BIT
112 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
113 #endif
114 
115 #ifndef _SYS_SYSPROTO_H_
116 struct sbrk_args {
117 	int incr;
118 };
119 #endif
120 
121 int
122 sys_sbrk(struct thread *td, 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 int
135 sys_sstk(struct thread *td, struct sstk_args *uap)
136 {
137 	/* Not yet implemented */
138 	return (EOPNOTSUPP);
139 }
140 
141 #if defined(COMPAT_43)
142 int
143 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
144 {
145 
146 	td->td_retval[0] = PAGE_SIZE;
147 	return (0);
148 }
149 #endif				/* COMPAT_43 */
150 
151 
152 /*
153  * Memory Map (mmap) system call.  Note that the file offset
154  * and address are allowed to be NOT page aligned, though if
155  * the MAP_FIXED flag it set, both must have the same remainder
156  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
157  * page-aligned, the actual mapping starts at trunc_page(addr)
158  * and the return value is adjusted up by the page offset.
159  *
160  * Generally speaking, only character devices which are themselves
161  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
162  * there would be no cache coherency between a descriptor and a VM mapping
163  * both to the same character device.
164  */
165 #ifndef _SYS_SYSPROTO_H_
166 struct mmap_args {
167 	void *addr;
168 	size_t len;
169 	int prot;
170 	int flags;
171 	int fd;
172 	long pad;
173 	off_t pos;
174 };
175 #endif
176 
177 int
178 sys_mmap(struct thread *td, struct mmap_args *uap)
179 {
180 
181 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
182 	    uap->flags, uap->fd, uap->pos));
183 }
184 
185 int
186 kern_mmap_maxprot(struct proc *p, int prot)
187 {
188 
189 	if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
190 	    (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
191 		return (_PROT_ALL);
192 	if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
193 	    prot != PROT_NONE)
194 		 return (prot);
195 	return (_PROT_ALL);
196 }
197 
198 int
199 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags,
200     int fd, off_t pos)
201 {
202 	struct mmap_req mr = {
203 		.mr_hint = addr0,
204 		.mr_len = len,
205 		.mr_prot = prot,
206 		.mr_flags = flags,
207 		.mr_fd = fd,
208 		.mr_pos = pos
209 	};
210 
211 	return (kern_mmap_req(td, &mr));
212 }
213 
214 int
215 kern_mmap_req(struct thread *td, const struct mmap_req *mrp)
216 {
217 	struct vmspace *vms;
218 	struct file *fp;
219 	struct proc *p;
220 	off_t pos;
221 	vm_offset_t addr;
222 	vm_size_t len, pageoff, size;
223 	vm_prot_t cap_maxprot;
224 	int align, error, fd, flags, max_prot, prot;
225 	cap_rights_t rights;
226 	mmap_check_fp_fn check_fp_fn;
227 
228 	addr  = mrp->mr_hint;
229 	len = mrp->mr_len;
230 	prot = mrp->mr_prot;
231 	flags = mrp->mr_flags;
232 	fd = mrp->mr_fd;
233 	pos = mrp->mr_pos;
234 	check_fp_fn = mrp->mr_check_fp_fn;
235 
236 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
237 		return (EINVAL);
238 	max_prot = PROT_MAX_EXTRACT(prot);
239 	prot = PROT_EXTRACT(prot);
240 	if (max_prot != 0 && (max_prot & prot) != prot)
241 		return (ENOTSUP);
242 
243 	p = td->td_proc;
244 
245 	/*
246 	 * Always honor PROT_MAX if set.  If not, default to all
247 	 * permissions unless we're implying maximum permissions.
248 	 */
249 	if (max_prot == 0)
250 		max_prot = kern_mmap_maxprot(p, prot);
251 
252 	vms = p->p_vmspace;
253 	fp = NULL;
254 	AUDIT_ARG_FD(fd);
255 
256 	/*
257 	 * Ignore old flags that used to be defined but did not do anything.
258 	 */
259 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
260 
261 	/*
262 	 * Enforce the constraints.
263 	 * Mapping of length 0 is only allowed for old binaries.
264 	 * Anonymous mapping shall specify -1 as filedescriptor and
265 	 * zero position for new code. Be nice to ancient a.out
266 	 * binaries and correct pos for anonymous mapping, since old
267 	 * ld.so sometimes issues anonymous map requests with non-zero
268 	 * pos.
269 	 */
270 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
271 		if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
272 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
273 			return (EINVAL);
274 	} else {
275 		if ((flags & MAP_ANON) != 0)
276 			pos = 0;
277 	}
278 
279 	if (flags & MAP_STACK) {
280 		if ((fd != -1) ||
281 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
282 			return (EINVAL);
283 		flags |= MAP_ANON;
284 		pos = 0;
285 	}
286 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
287 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
288 	    MAP_PREFAULT_READ | MAP_GUARD |
289 #ifdef MAP_32BIT
290 	    MAP_32BIT |
291 #endif
292 	    MAP_ALIGNMENT_MASK)) != 0)
293 		return (EINVAL);
294 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
295 		return (EINVAL);
296 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
297 		return (EINVAL);
298 	if (prot != PROT_NONE &&
299 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
300 		return (EINVAL);
301 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
302 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
303 #ifdef MAP_32BIT
304 	    MAP_32BIT |
305 #endif
306 	    MAP_ALIGNMENT_MASK)) != 0))
307 		return (EINVAL);
308 
309 	/*
310 	 * Align the file position to a page boundary,
311 	 * and save its page offset component.
312 	 */
313 	pageoff = (pos & PAGE_MASK);
314 	pos -= pageoff;
315 
316 	/* Compute size from len by rounding (on both ends). */
317 	size = len + pageoff;			/* low end... */
318 	size = round_page(size);		/* hi end */
319 	/* Check for rounding up to zero. */
320 	if (len > size)
321 		return (ENOMEM);
322 
323 	/* Ensure alignment is at least a page and fits in a pointer. */
324 	align = flags & MAP_ALIGNMENT_MASK;
325 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
326 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
327 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
328 		return (EINVAL);
329 
330 	/*
331 	 * Check for illegal addresses.  Watch out for address wrap... Note
332 	 * that VM_*_ADDRESS are not constants due to casts (argh).
333 	 */
334 	if (flags & MAP_FIXED) {
335 		/*
336 		 * The specified address must have the same remainder
337 		 * as the file offset taken modulo PAGE_SIZE, so it
338 		 * should be aligned after adjustment by pageoff.
339 		 */
340 		addr -= pageoff;
341 		if (addr & PAGE_MASK)
342 			return (EINVAL);
343 
344 		/* Address range must be all in user VM space. */
345 		if (addr < vm_map_min(&vms->vm_map) ||
346 		    addr + size > vm_map_max(&vms->vm_map))
347 			return (EINVAL);
348 		if (addr + size < addr)
349 			return (EINVAL);
350 #ifdef MAP_32BIT
351 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
352 			return (EINVAL);
353 	} else if (flags & MAP_32BIT) {
354 		/*
355 		 * For MAP_32BIT, override the hint if it is too high and
356 		 * do not bother moving the mapping past the heap (since
357 		 * the heap is usually above 2GB).
358 		 */
359 		if (addr + size > MAP_32BIT_MAX_ADDR)
360 			addr = 0;
361 #endif
362 	} else {
363 		/*
364 		 * XXX for non-fixed mappings where no hint is provided or
365 		 * the hint would fall in the potential heap space,
366 		 * place it after the end of the largest possible heap.
367 		 *
368 		 * There should really be a pmap call to determine a reasonable
369 		 * location.
370 		 */
371 		if (addr == 0 ||
372 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
373 		    addr < round_page((vm_offset_t)vms->vm_daddr +
374 		    lim_max(td, RLIMIT_DATA))))
375 			addr = round_page((vm_offset_t)vms->vm_daddr +
376 			    lim_max(td, RLIMIT_DATA));
377 	}
378 	if (len == 0) {
379 		/*
380 		 * Return success without mapping anything for old
381 		 * binaries that request a page-aligned mapping of
382 		 * length 0.  For modern binaries, this function
383 		 * returns an error earlier.
384 		 */
385 		error = 0;
386 	} else if ((flags & MAP_GUARD) != 0) {
387 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
388 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
389 	} else if ((flags & MAP_ANON) != 0) {
390 		/*
391 		 * Mapping blank space is trivial.
392 		 *
393 		 * This relies on VM_PROT_* matching PROT_*.
394 		 */
395 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
396 		    max_prot, flags, NULL, pos, FALSE, td);
397 	} else {
398 		/*
399 		 * Mapping file, get fp for validation and don't let the
400 		 * descriptor disappear on us if we block. Check capability
401 		 * rights, but also return the maximum rights to be combined
402 		 * with maxprot later.
403 		 */
404 		cap_rights_init_one(&rights, CAP_MMAP);
405 		if (prot & PROT_READ)
406 			cap_rights_set_one(&rights, CAP_MMAP_R);
407 		if ((flags & MAP_SHARED) != 0) {
408 			if (prot & PROT_WRITE)
409 				cap_rights_set_one(&rights, CAP_MMAP_W);
410 		}
411 		if (prot & PROT_EXEC)
412 			cap_rights_set_one(&rights, CAP_MMAP_X);
413 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
414 		if (error != 0)
415 			goto done;
416 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
417 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
418 			error = EINVAL;
419 			goto done;
420 		}
421 		if (check_fp_fn != NULL) {
422 			error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
423 			    flags);
424 			if (error != 0)
425 				goto done;
426 		}
427 		/* This relies on VM_PROT_* matching PROT_*. */
428 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
429 		    max_prot & cap_maxprot, flags, pos, td);
430 	}
431 
432 	if (error == 0)
433 		td->td_retval[0] = (register_t) (addr + pageoff);
434 done:
435 	if (fp)
436 		fdrop(fp, td);
437 
438 	return (error);
439 }
440 
441 #if defined(COMPAT_FREEBSD6)
442 int
443 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
444 {
445 
446 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
447 	    uap->flags, uap->fd, uap->pos));
448 }
449 #endif
450 
451 #ifdef COMPAT_43
452 #ifndef _SYS_SYSPROTO_H_
453 struct ommap_args {
454 	caddr_t addr;
455 	int len;
456 	int prot;
457 	int flags;
458 	int fd;
459 	long pos;
460 };
461 #endif
462 int
463 ommap(struct thread *td, struct ommap_args *uap)
464 {
465 	static const char cvtbsdprot[8] = {
466 		0,
467 		PROT_EXEC,
468 		PROT_WRITE,
469 		PROT_EXEC | PROT_WRITE,
470 		PROT_READ,
471 		PROT_EXEC | PROT_READ,
472 		PROT_WRITE | PROT_READ,
473 		PROT_EXEC | PROT_WRITE | PROT_READ,
474 	};
475 	int flags, prot;
476 
477 #define	OMAP_ANON	0x0002
478 #define	OMAP_COPY	0x0020
479 #define	OMAP_SHARED	0x0010
480 #define	OMAP_FIXED	0x0100
481 
482 	prot = cvtbsdprot[uap->prot & 0x7];
483 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
484 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
485 	    prot != 0)
486 		prot |= PROT_EXEC;
487 #endif
488 	flags = 0;
489 	if (uap->flags & OMAP_ANON)
490 		flags |= MAP_ANON;
491 	if (uap->flags & OMAP_COPY)
492 		flags |= MAP_COPY;
493 	if (uap->flags & OMAP_SHARED)
494 		flags |= MAP_SHARED;
495 	else
496 		flags |= MAP_PRIVATE;
497 	if (uap->flags & OMAP_FIXED)
498 		flags |= MAP_FIXED;
499 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
500 	    uap->fd, uap->pos));
501 }
502 #endif				/* COMPAT_43 */
503 
504 
505 #ifndef _SYS_SYSPROTO_H_
506 struct msync_args {
507 	void *addr;
508 	size_t len;
509 	int flags;
510 };
511 #endif
512 int
513 sys_msync(struct thread *td, struct msync_args *uap)
514 {
515 
516 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
517 }
518 
519 int
520 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
521 {
522 	vm_offset_t addr;
523 	vm_size_t pageoff;
524 	vm_map_t map;
525 	int rv;
526 
527 	addr = addr0;
528 	pageoff = (addr & PAGE_MASK);
529 	addr -= pageoff;
530 	size += pageoff;
531 	size = (vm_size_t) round_page(size);
532 	if (addr + size < addr)
533 		return (EINVAL);
534 
535 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
536 		return (EINVAL);
537 
538 	map = &td->td_proc->p_vmspace->vm_map;
539 
540 	/*
541 	 * Clean the pages and interpret the return value.
542 	 */
543 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
544 	    (flags & MS_INVALIDATE) != 0);
545 	switch (rv) {
546 	case KERN_SUCCESS:
547 		return (0);
548 	case KERN_INVALID_ADDRESS:
549 		return (ENOMEM);
550 	case KERN_INVALID_ARGUMENT:
551 		return (EBUSY);
552 	case KERN_FAILURE:
553 		return (EIO);
554 	default:
555 		return (EINVAL);
556 	}
557 }
558 
559 #ifndef _SYS_SYSPROTO_H_
560 struct munmap_args {
561 	void *addr;
562 	size_t len;
563 };
564 #endif
565 int
566 sys_munmap(struct thread *td, struct munmap_args *uap)
567 {
568 
569 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
570 }
571 
572 int
573 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
574 {
575 #ifdef HWPMC_HOOKS
576 	struct pmckern_map_out pkm;
577 	vm_map_entry_t entry;
578 	bool pmc_handled;
579 #endif
580 	vm_offset_t addr;
581 	vm_size_t pageoff;
582 	vm_map_t map;
583 
584 	if (size == 0)
585 		return (EINVAL);
586 
587 	addr = addr0;
588 	pageoff = (addr & PAGE_MASK);
589 	addr -= pageoff;
590 	size += pageoff;
591 	size = (vm_size_t) round_page(size);
592 	if (addr + size < addr)
593 		return (EINVAL);
594 
595 	/*
596 	 * Check for illegal addresses.  Watch out for address wrap...
597 	 */
598 	map = &td->td_proc->p_vmspace->vm_map;
599 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
600 		return (EINVAL);
601 	vm_map_lock(map);
602 #ifdef HWPMC_HOOKS
603 	pmc_handled = false;
604 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
605 		pmc_handled = true;
606 		/*
607 		 * Inform hwpmc if the address range being unmapped contains
608 		 * an executable region.
609 		 */
610 		pkm.pm_address = (uintptr_t) NULL;
611 		if (vm_map_lookup_entry(map, addr, &entry)) {
612 			for (; entry->start < addr + size;
613 			    entry = vm_map_entry_succ(entry)) {
614 				if (vm_map_check_protection(map, entry->start,
615 					entry->end, VM_PROT_EXECUTE) == TRUE) {
616 					pkm.pm_address = (uintptr_t) addr;
617 					pkm.pm_size = (size_t) size;
618 					break;
619 				}
620 			}
621 		}
622 	}
623 #endif
624 	vm_map_delete(map, addr, addr + size);
625 
626 #ifdef HWPMC_HOOKS
627 	if (__predict_false(pmc_handled)) {
628 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
629 		vm_map_lock_downgrade(map);
630 		if (pkm.pm_address != (uintptr_t) NULL)
631 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
632 		vm_map_unlock_read(map);
633 	} else
634 #endif
635 		vm_map_unlock(map);
636 
637 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
638 	return (0);
639 }
640 
641 #ifndef _SYS_SYSPROTO_H_
642 struct mprotect_args {
643 	const void *addr;
644 	size_t len;
645 	int prot;
646 };
647 #endif
648 int
649 sys_mprotect(struct thread *td, struct mprotect_args *uap)
650 {
651 
652 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
653 }
654 
655 int
656 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
657 {
658 	vm_offset_t addr;
659 	vm_size_t pageoff;
660 	int vm_error, max_prot;
661 
662 	addr = addr0;
663 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
664 		return (EINVAL);
665 	max_prot = PROT_MAX_EXTRACT(prot);
666 	prot = PROT_EXTRACT(prot);
667 	pageoff = (addr & PAGE_MASK);
668 	addr -= pageoff;
669 	size += pageoff;
670 	size = (vm_size_t) round_page(size);
671 #ifdef COMPAT_FREEBSD32
672 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
673 		if (((addr + size) & 0xffffffff) < addr)
674 			return (EINVAL);
675 	} else
676 #endif
677 	if (addr + size < addr)
678 		return (EINVAL);
679 
680 	vm_error = KERN_SUCCESS;
681 	if (max_prot != 0) {
682 		if ((max_prot & prot) != prot)
683 			return (ENOTSUP);
684 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
685 		    addr, addr + size, max_prot, TRUE);
686 	}
687 	if (vm_error == KERN_SUCCESS)
688 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
689 		    addr, addr + size, prot, FALSE);
690 
691 	switch (vm_error) {
692 	case KERN_SUCCESS:
693 		return (0);
694 	case KERN_PROTECTION_FAILURE:
695 		return (EACCES);
696 	case KERN_RESOURCE_SHORTAGE:
697 		return (ENOMEM);
698 	}
699 	return (EINVAL);
700 }
701 
702 #ifndef _SYS_SYSPROTO_H_
703 struct minherit_args {
704 	void *addr;
705 	size_t len;
706 	int inherit;
707 };
708 #endif
709 int
710 sys_minherit(struct thread *td, struct minherit_args *uap)
711 {
712 	vm_offset_t addr;
713 	vm_size_t size, pageoff;
714 	vm_inherit_t inherit;
715 
716 	addr = (vm_offset_t)uap->addr;
717 	size = uap->len;
718 	inherit = uap->inherit;
719 
720 	pageoff = (addr & PAGE_MASK);
721 	addr -= pageoff;
722 	size += pageoff;
723 	size = (vm_size_t) round_page(size);
724 	if (addr + size < addr)
725 		return (EINVAL);
726 
727 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
728 	    addr + size, inherit)) {
729 	case KERN_SUCCESS:
730 		return (0);
731 	case KERN_PROTECTION_FAILURE:
732 		return (EACCES);
733 	}
734 	return (EINVAL);
735 }
736 
737 #ifndef _SYS_SYSPROTO_H_
738 struct madvise_args {
739 	void *addr;
740 	size_t len;
741 	int behav;
742 };
743 #endif
744 
745 int
746 sys_madvise(struct thread *td, struct madvise_args *uap)
747 {
748 
749 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
750 }
751 
752 int
753 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
754 {
755 	vm_map_t map;
756 	vm_offset_t addr, end, start;
757 	int flags;
758 
759 	/*
760 	 * Check for our special case, advising the swap pager we are
761 	 * "immortal."
762 	 */
763 	if (behav == MADV_PROTECT) {
764 		flags = PPROT_SET;
765 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
766 		    PROC_SPROTECT, &flags));
767 	}
768 
769 	/*
770 	 * Check for illegal addresses.  Watch out for address wrap... Note
771 	 * that VM_*_ADDRESS are not constants due to casts (argh).
772 	 */
773 	map = &td->td_proc->p_vmspace->vm_map;
774 	addr = addr0;
775 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
776 		return (EINVAL);
777 	if ((addr + len) < addr)
778 		return (EINVAL);
779 
780 	/*
781 	 * Since this routine is only advisory, we default to conservative
782 	 * behavior.
783 	 */
784 	start = trunc_page(addr);
785 	end = round_page(addr + len);
786 
787 	/*
788 	 * vm_map_madvise() checks for illegal values of behav.
789 	 */
790 	return (vm_map_madvise(map, start, end, behav));
791 }
792 
793 #ifndef _SYS_SYSPROTO_H_
794 struct mincore_args {
795 	const void *addr;
796 	size_t len;
797 	char *vec;
798 };
799 #endif
800 
801 int
802 sys_mincore(struct thread *td, struct mincore_args *uap)
803 {
804 
805 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
806 }
807 
808 int
809 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
810 {
811 	pmap_t pmap;
812 	vm_map_t map;
813 	vm_map_entry_t current, entry;
814 	vm_object_t object;
815 	vm_offset_t addr, cend, end, first_addr;
816 	vm_paddr_t pa;
817 	vm_page_t m;
818 	vm_pindex_t pindex;
819 	int error, lastvecindex, mincoreinfo, vecindex;
820 	unsigned int timestamp;
821 
822 	/*
823 	 * Make sure that the addresses presented are valid for user
824 	 * mode.
825 	 */
826 	first_addr = addr = trunc_page(addr0);
827 	end = round_page(addr0 + len);
828 	map = &td->td_proc->p_vmspace->vm_map;
829 	if (end > vm_map_max(map) || end < addr)
830 		return (ENOMEM);
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 	while (entry->start < end) {
850 
851 		/*
852 		 * check for contiguity
853 		 */
854 		current = entry;
855 		entry = vm_map_entry_succ(current);
856 		if (current->end < end &&
857 		    entry->start > current->end) {
858 			vm_map_unlock_read(map);
859 			return (ENOMEM);
860 		}
861 
862 		/*
863 		 * ignore submaps (for now) or null objects
864 		 */
865 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
866 		    current->object.vm_object == NULL)
867 			continue;
868 
869 		/*
870 		 * limit this scan to the current map entry and the
871 		 * limits for the mincore call
872 		 */
873 		if (addr < current->start)
874 			addr = current->start;
875 		cend = current->end;
876 		if (cend > end)
877 			cend = end;
878 
879 		for (; addr < cend; addr += PAGE_SIZE) {
880 			/*
881 			 * Check pmap first, it is likely faster, also
882 			 * it can provide info as to whether we are the
883 			 * one referencing or modifying the page.
884 			 */
885 			m = NULL;
886 			object = NULL;
887 retry:
888 			pa = 0;
889 			mincoreinfo = pmap_mincore(pmap, addr, &pa);
890 			if (mincore_mapped) {
891 				/*
892 				 * We only care about this pmap's
893 				 * mapping of the page, if any.
894 				 */
895 				;
896 			} else if (pa != 0) {
897 				/*
898 				 * The page is mapped by this process but not
899 				 * both accessed and modified.  It is also
900 				 * managed.  Acquire the object lock so that
901 				 * other mappings might be examined.  The page's
902 				 * identity may change at any point before its
903 				 * object lock is acquired, so re-validate if
904 				 * necessary.
905 				 */
906 				m = PHYS_TO_VM_PAGE(pa);
907 				while (object == NULL || m->object != object) {
908 					if (object != NULL)
909 						VM_OBJECT_WUNLOCK(object);
910 					object = atomic_load_ptr(&m->object);
911 					if (object == NULL)
912 						goto retry;
913 					VM_OBJECT_WLOCK(object);
914 				}
915 				if (pa != pmap_extract(pmap, addr))
916 					goto retry;
917 				KASSERT(vm_page_all_valid(m),
918 				    ("mincore: page %p is mapped but invalid",
919 				    m));
920 			} else if (mincoreinfo == 0) {
921 				/*
922 				 * The page is not mapped by this process.  If
923 				 * the object implements managed pages, then
924 				 * determine if the page is resident so that
925 				 * the mappings might be examined.
926 				 */
927 				if (current->object.vm_object != object) {
928 					if (object != NULL)
929 						VM_OBJECT_WUNLOCK(object);
930 					object = current->object.vm_object;
931 					VM_OBJECT_WLOCK(object);
932 				}
933 				if (object->type == OBJT_DEFAULT ||
934 				    object->type == OBJT_SWAP ||
935 				    object->type == OBJT_VNODE) {
936 					pindex = OFF_TO_IDX(current->offset +
937 					    (addr - current->start));
938 					m = vm_page_lookup(object, pindex);
939 					if (m != NULL && vm_page_none_valid(m))
940 						m = NULL;
941 					if (m != NULL)
942 						mincoreinfo = MINCORE_INCORE;
943 				}
944 			}
945 			if (m != NULL) {
946 				VM_OBJECT_ASSERT_WLOCKED(m->object);
947 
948 				/* Examine other mappings of the page. */
949 				if (m->dirty == 0 && pmap_is_modified(m))
950 					vm_page_dirty(m);
951 				if (m->dirty != 0)
952 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
953 
954 				/*
955 				 * The first test for PGA_REFERENCED is an
956 				 * optimization.  The second test is
957 				 * required because a concurrent pmap
958 				 * operation could clear the last reference
959 				 * and set PGA_REFERENCED before the call to
960 				 * pmap_is_referenced().
961 				 */
962 				if ((m->a.flags & PGA_REFERENCED) != 0 ||
963 				    pmap_is_referenced(m) ||
964 				    (m->a.flags & PGA_REFERENCED) != 0)
965 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
966 			}
967 			if (object != NULL)
968 				VM_OBJECT_WUNLOCK(object);
969 
970 			/*
971 			 * subyte may page fault.  In case it needs to modify
972 			 * the map, we release the lock.
973 			 */
974 			vm_map_unlock_read(map);
975 
976 			/*
977 			 * calculate index into user supplied byte vector
978 			 */
979 			vecindex = atop(addr - first_addr);
980 
981 			/*
982 			 * If we have skipped map entries, we need to make sure that
983 			 * the byte vector is zeroed for those skipped entries.
984 			 */
985 			while ((lastvecindex + 1) < vecindex) {
986 				++lastvecindex;
987 				error = subyte(vec + lastvecindex, 0);
988 				if (error) {
989 					error = EFAULT;
990 					goto done2;
991 				}
992 			}
993 
994 			/*
995 			 * Pass the page information to the user
996 			 */
997 			error = subyte(vec + vecindex, mincoreinfo);
998 			if (error) {
999 				error = EFAULT;
1000 				goto done2;
1001 			}
1002 
1003 			/*
1004 			 * If the map has changed, due to the subyte, the previous
1005 			 * output may be invalid.
1006 			 */
1007 			vm_map_lock_read(map);
1008 			if (timestamp != map->timestamp)
1009 				goto RestartScan;
1010 
1011 			lastvecindex = vecindex;
1012 		}
1013 	}
1014 
1015 	/*
1016 	 * subyte may page fault.  In case it needs to modify
1017 	 * the map, we release the lock.
1018 	 */
1019 	vm_map_unlock_read(map);
1020 
1021 	/*
1022 	 * Zero the last entries in the byte vector.
1023 	 */
1024 	vecindex = atop(end - first_addr);
1025 	while ((lastvecindex + 1) < vecindex) {
1026 		++lastvecindex;
1027 		error = subyte(vec + lastvecindex, 0);
1028 		if (error) {
1029 			error = EFAULT;
1030 			goto done2;
1031 		}
1032 	}
1033 
1034 	/*
1035 	 * If the map has changed, due to the subyte, the previous
1036 	 * output may be invalid.
1037 	 */
1038 	vm_map_lock_read(map);
1039 	if (timestamp != map->timestamp)
1040 		goto RestartScan;
1041 	vm_map_unlock_read(map);
1042 done2:
1043 	return (error);
1044 }
1045 
1046 #ifndef _SYS_SYSPROTO_H_
1047 struct mlock_args {
1048 	const void *addr;
1049 	size_t len;
1050 };
1051 #endif
1052 int
1053 sys_mlock(struct thread *td, struct mlock_args *uap)
1054 {
1055 
1056 	return (kern_mlock(td->td_proc, td->td_ucred,
1057 	    __DECONST(uintptr_t, uap->addr), uap->len));
1058 }
1059 
1060 int
1061 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1062 {
1063 	vm_offset_t addr, end, last, start;
1064 	vm_size_t npages, size;
1065 	vm_map_t map;
1066 	unsigned long nsize;
1067 	int error;
1068 
1069 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1070 	if (error)
1071 		return (error);
1072 	addr = addr0;
1073 	size = len;
1074 	last = addr + size;
1075 	start = trunc_page(addr);
1076 	end = round_page(last);
1077 	if (last < addr || end < addr)
1078 		return (EINVAL);
1079 	npages = atop(end - start);
1080 	if (npages > vm_page_max_user_wired)
1081 		return (ENOMEM);
1082 	map = &proc->p_vmspace->vm_map;
1083 	PROC_LOCK(proc);
1084 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1085 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1086 		PROC_UNLOCK(proc);
1087 		return (ENOMEM);
1088 	}
1089 	PROC_UNLOCK(proc);
1090 #ifdef RACCT
1091 	if (racct_enable) {
1092 		PROC_LOCK(proc);
1093 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1094 		PROC_UNLOCK(proc);
1095 		if (error != 0)
1096 			return (ENOMEM);
1097 	}
1098 #endif
1099 	error = vm_map_wire(map, start, end,
1100 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1101 #ifdef RACCT
1102 	if (racct_enable && error != KERN_SUCCESS) {
1103 		PROC_LOCK(proc);
1104 		racct_set(proc, RACCT_MEMLOCK,
1105 		    ptoa(pmap_wired_count(map->pmap)));
1106 		PROC_UNLOCK(proc);
1107 	}
1108 #endif
1109 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1110 }
1111 
1112 #ifndef _SYS_SYSPROTO_H_
1113 struct mlockall_args {
1114 	int	how;
1115 };
1116 #endif
1117 
1118 int
1119 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1120 {
1121 	vm_map_t map;
1122 	int error;
1123 
1124 	map = &td->td_proc->p_vmspace->vm_map;
1125 	error = priv_check(td, PRIV_VM_MLOCK);
1126 	if (error)
1127 		return (error);
1128 
1129 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1130 		return (EINVAL);
1131 
1132 	/*
1133 	 * If wiring all pages in the process would cause it to exceed
1134 	 * a hard resource limit, return ENOMEM.
1135 	 */
1136 	if (!old_mlock && uap->how & MCL_CURRENT) {
1137 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1138 			return (ENOMEM);
1139 	}
1140 #ifdef RACCT
1141 	if (racct_enable) {
1142 		PROC_LOCK(td->td_proc);
1143 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1144 		PROC_UNLOCK(td->td_proc);
1145 		if (error != 0)
1146 			return (ENOMEM);
1147 	}
1148 #endif
1149 
1150 	if (uap->how & MCL_FUTURE) {
1151 		vm_map_lock(map);
1152 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1153 		vm_map_unlock(map);
1154 		error = 0;
1155 	}
1156 
1157 	if (uap->how & MCL_CURRENT) {
1158 		/*
1159 		 * P1003.1-2001 mandates that all currently mapped pages
1160 		 * will be memory resident and locked (wired) upon return
1161 		 * from mlockall(). vm_map_wire() will wire pages, by
1162 		 * calling vm_fault_wire() for each page in the region.
1163 		 */
1164 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1165 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1166 		if (error == KERN_SUCCESS)
1167 			error = 0;
1168 		else if (error == KERN_RESOURCE_SHORTAGE)
1169 			error = ENOMEM;
1170 		else
1171 			error = EAGAIN;
1172 	}
1173 #ifdef RACCT
1174 	if (racct_enable && error != KERN_SUCCESS) {
1175 		PROC_LOCK(td->td_proc);
1176 		racct_set(td->td_proc, RACCT_MEMLOCK,
1177 		    ptoa(pmap_wired_count(map->pmap)));
1178 		PROC_UNLOCK(td->td_proc);
1179 	}
1180 #endif
1181 
1182 	return (error);
1183 }
1184 
1185 #ifndef _SYS_SYSPROTO_H_
1186 struct munlockall_args {
1187 	register_t dummy;
1188 };
1189 #endif
1190 
1191 int
1192 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1193 {
1194 	vm_map_t map;
1195 	int error;
1196 
1197 	map = &td->td_proc->p_vmspace->vm_map;
1198 	error = priv_check(td, PRIV_VM_MUNLOCK);
1199 	if (error)
1200 		return (error);
1201 
1202 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1203 	vm_map_lock(map);
1204 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1205 	vm_map_unlock(map);
1206 
1207 	/* Forcibly unwire all pages. */
1208 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1209 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1210 #ifdef RACCT
1211 	if (racct_enable && error == KERN_SUCCESS) {
1212 		PROC_LOCK(td->td_proc);
1213 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1214 		PROC_UNLOCK(td->td_proc);
1215 	}
1216 #endif
1217 
1218 	return (error);
1219 }
1220 
1221 #ifndef _SYS_SYSPROTO_H_
1222 struct munlock_args {
1223 	const void *addr;
1224 	size_t len;
1225 };
1226 #endif
1227 int
1228 sys_munlock(struct thread *td, struct munlock_args *uap)
1229 {
1230 
1231 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1232 }
1233 
1234 int
1235 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1236 {
1237 	vm_offset_t addr, end, last, start;
1238 #ifdef RACCT
1239 	vm_map_t map;
1240 #endif
1241 	int error;
1242 
1243 	error = priv_check(td, PRIV_VM_MUNLOCK);
1244 	if (error)
1245 		return (error);
1246 	addr = addr0;
1247 	last = addr + size;
1248 	start = trunc_page(addr);
1249 	end = round_page(last);
1250 	if (last < addr || end < addr)
1251 		return (EINVAL);
1252 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1253 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1254 #ifdef RACCT
1255 	if (racct_enable && error == KERN_SUCCESS) {
1256 		PROC_LOCK(td->td_proc);
1257 		map = &td->td_proc->p_vmspace->vm_map;
1258 		racct_set(td->td_proc, RACCT_MEMLOCK,
1259 		    ptoa(pmap_wired_count(map->pmap)));
1260 		PROC_UNLOCK(td->td_proc);
1261 	}
1262 #endif
1263 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1264 }
1265 
1266 /*
1267  * vm_mmap_vnode()
1268  *
1269  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1270  * operations on vnodes.
1271  */
1272 int
1273 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1274     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1275     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1276     boolean_t *writecounted)
1277 {
1278 	struct vattr va;
1279 	vm_object_t obj;
1280 	vm_ooffset_t foff;
1281 	struct ucred *cred;
1282 	int error, flags;
1283 	bool writex;
1284 
1285 	cred = td->td_ucred;
1286 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1287 	    (*flagsp & MAP_SHARED) != 0;
1288 	if ((error = vget(vp, LK_SHARED, td)) != 0)
1289 		return (error);
1290 	AUDIT_ARG_VNODE1(vp);
1291 	foff = *foffp;
1292 	flags = *flagsp;
1293 	obj = vp->v_object;
1294 	if (vp->v_type == VREG) {
1295 		/*
1296 		 * Get the proper underlying object
1297 		 */
1298 		if (obj == NULL) {
1299 			error = EINVAL;
1300 			goto done;
1301 		}
1302 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1303 			vput(vp);
1304 			vp = (struct vnode *)obj->handle;
1305 			/*
1306 			 * Bypass filesystems obey the mpsafety of the
1307 			 * underlying fs.  Tmpfs never bypasses.
1308 			 */
1309 			error = vget(vp, LK_SHARED, td);
1310 			if (error != 0)
1311 				return (error);
1312 		}
1313 		if (writex) {
1314 			*writecounted = TRUE;
1315 			vm_pager_update_writecount(obj, 0, objsize);
1316 		}
1317 	} else {
1318 		error = EINVAL;
1319 		goto done;
1320 	}
1321 	if ((error = VOP_GETATTR(vp, &va, cred)))
1322 		goto done;
1323 #ifdef MAC
1324 	/* This relies on VM_PROT_* matching PROT_*. */
1325 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1326 	if (error != 0)
1327 		goto done;
1328 #endif
1329 	if ((flags & MAP_SHARED) != 0) {
1330 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1331 			if (prot & VM_PROT_WRITE) {
1332 				error = EPERM;
1333 				goto done;
1334 			}
1335 			*maxprotp &= ~VM_PROT_WRITE;
1336 		}
1337 	}
1338 	/*
1339 	 * If it is a regular file without any references
1340 	 * we do not need to sync it.
1341 	 * Adjust object size to be the size of actual file.
1342 	 */
1343 	objsize = round_page(va.va_size);
1344 	if (va.va_nlink == 0)
1345 		flags |= MAP_NOSYNC;
1346 	if (obj->type == OBJT_VNODE) {
1347 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1348 		    cred);
1349 		if (obj == NULL) {
1350 			error = ENOMEM;
1351 			goto done;
1352 		}
1353 	} else {
1354 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1355 		    ("wrong object type"));
1356 		vm_object_reference(obj);
1357 #if VM_NRESERVLEVEL > 0
1358 		if ((obj->flags & OBJ_COLORED) == 0) {
1359 			VM_OBJECT_WLOCK(obj);
1360 			vm_object_color(obj, 0);
1361 			VM_OBJECT_WUNLOCK(obj);
1362 		}
1363 #endif
1364 	}
1365 	*objp = obj;
1366 	*flagsp = flags;
1367 
1368 	VOP_MMAPPED(vp);
1369 
1370 done:
1371 	if (error != 0 && *writecounted) {
1372 		*writecounted = FALSE;
1373 		vm_pager_update_writecount(obj, objsize, 0);
1374 	}
1375 	vput(vp);
1376 	return (error);
1377 }
1378 
1379 /*
1380  * vm_mmap_cdev()
1381  *
1382  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1383  * operations on cdevs.
1384  */
1385 int
1386 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1387     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1388     vm_ooffset_t *foff, vm_object_t *objp)
1389 {
1390 	vm_object_t obj;
1391 	int error, flags;
1392 
1393 	flags = *flagsp;
1394 
1395 	if (dsw->d_flags & D_MMAP_ANON) {
1396 		*objp = NULL;
1397 		*foff = 0;
1398 		*maxprotp = VM_PROT_ALL;
1399 		*flagsp |= MAP_ANON;
1400 		return (0);
1401 	}
1402 	/*
1403 	 * cdevs do not provide private mappings of any kind.
1404 	 */
1405 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1406 	    (prot & VM_PROT_WRITE) != 0)
1407 		return (EACCES);
1408 	if (flags & (MAP_PRIVATE|MAP_COPY))
1409 		return (EINVAL);
1410 	/*
1411 	 * Force device mappings to be shared.
1412 	 */
1413 	flags |= MAP_SHARED;
1414 #ifdef MAC_XXX
1415 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1416 	if (error != 0)
1417 		return (error);
1418 #endif
1419 	/*
1420 	 * First, try d_mmap_single().  If that is not implemented
1421 	 * (returns ENODEV), fall back to using the device pager.
1422 	 * Note that d_mmap_single() must return a reference to the
1423 	 * object (it needs to bump the reference count of the object
1424 	 * it returns somehow).
1425 	 *
1426 	 * XXX assumes VM_PROT_* == PROT_*
1427 	 */
1428 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1429 	if (error != ENODEV)
1430 		return (error);
1431 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1432 	    td->td_ucred);
1433 	if (obj == NULL)
1434 		return (EINVAL);
1435 	*objp = obj;
1436 	*flagsp = flags;
1437 	return (0);
1438 }
1439 
1440 /*
1441  * vm_mmap()
1442  *
1443  * Internal version of mmap used by exec, sys5 shared memory, and
1444  * various device drivers.  Handle is either a vnode pointer, a
1445  * character device, or NULL for MAP_ANON.
1446  */
1447 int
1448 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1449 	vm_prot_t maxprot, int flags,
1450 	objtype_t handle_type, void *handle,
1451 	vm_ooffset_t foff)
1452 {
1453 	vm_object_t object;
1454 	struct thread *td = curthread;
1455 	int error;
1456 	boolean_t writecounted;
1457 
1458 	if (size == 0)
1459 		return (EINVAL);
1460 
1461 	size = round_page(size);
1462 	object = NULL;
1463 	writecounted = FALSE;
1464 
1465 	/*
1466 	 * Lookup/allocate object.
1467 	 */
1468 	switch (handle_type) {
1469 	case OBJT_DEVICE: {
1470 		struct cdevsw *dsw;
1471 		struct cdev *cdev;
1472 		int ref;
1473 
1474 		cdev = handle;
1475 		dsw = dev_refthread(cdev, &ref);
1476 		if (dsw == NULL)
1477 			return (ENXIO);
1478 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1479 		    dsw, &foff, &object);
1480 		dev_relthread(cdev, ref);
1481 		break;
1482 	}
1483 	case OBJT_VNODE:
1484 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1485 		    handle, &foff, &object, &writecounted);
1486 		break;
1487 	case OBJT_DEFAULT:
1488 		if (handle == NULL) {
1489 			error = 0;
1490 			break;
1491 		}
1492 		/* FALLTHROUGH */
1493 	default:
1494 		error = EINVAL;
1495 		break;
1496 	}
1497 	if (error)
1498 		return (error);
1499 
1500 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1501 	    foff, writecounted, td);
1502 	if (error != 0 && object != NULL) {
1503 		/*
1504 		 * If this mapping was accounted for in the vnode's
1505 		 * writecount, then undo that now.
1506 		 */
1507 		if (writecounted)
1508 			vm_pager_release_writecount(object, 0, size);
1509 		vm_object_deallocate(object);
1510 	}
1511 	return (error);
1512 }
1513 
1514 /*
1515  * Internal version of mmap that maps a specific VM object into an
1516  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1517  */
1518 int
1519 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1520     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1521     boolean_t writecounted, struct thread *td)
1522 {
1523 	boolean_t curmap, fitit;
1524 	vm_offset_t max_addr;
1525 	int docow, error, findspace, rv;
1526 
1527 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1528 	if (curmap) {
1529 		RACCT_PROC_LOCK(td->td_proc);
1530 		if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1531 			RACCT_PROC_UNLOCK(td->td_proc);
1532 			return (ENOMEM);
1533 		}
1534 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1535 			RACCT_PROC_UNLOCK(td->td_proc);
1536 			return (ENOMEM);
1537 		}
1538 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1539 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1540 			    lim_cur(td, RLIMIT_MEMLOCK)) {
1541 				racct_set_force(td->td_proc, RACCT_VMEM,
1542 				    map->size);
1543 				RACCT_PROC_UNLOCK(td->td_proc);
1544 				return (ENOMEM);
1545 			}
1546 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1547 			    ptoa(pmap_wired_count(map->pmap)) + size);
1548 			if (error != 0) {
1549 				racct_set_force(td->td_proc, RACCT_VMEM,
1550 				    map->size);
1551 				RACCT_PROC_UNLOCK(td->td_proc);
1552 				return (error);
1553 			}
1554 		}
1555 		RACCT_PROC_UNLOCK(td->td_proc);
1556 	}
1557 
1558 	/*
1559 	 * We currently can only deal with page aligned file offsets.
1560 	 * The mmap() system call already enforces this by subtracting
1561 	 * the page offset from the file offset, but checking here
1562 	 * catches errors in device drivers (e.g. d_single_mmap()
1563 	 * callbacks) and other internal mapping requests (such as in
1564 	 * exec).
1565 	 */
1566 	if (foff & PAGE_MASK)
1567 		return (EINVAL);
1568 
1569 	if ((flags & MAP_FIXED) == 0) {
1570 		fitit = TRUE;
1571 		*addr = round_page(*addr);
1572 	} else {
1573 		if (*addr != trunc_page(*addr))
1574 			return (EINVAL);
1575 		fitit = FALSE;
1576 	}
1577 
1578 	if (flags & MAP_ANON) {
1579 		if (object != NULL || foff != 0)
1580 			return (EINVAL);
1581 		docow = 0;
1582 	} else if (flags & MAP_PREFAULT_READ)
1583 		docow = MAP_PREFAULT;
1584 	else
1585 		docow = MAP_PREFAULT_PARTIAL;
1586 
1587 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1588 		docow |= MAP_COPY_ON_WRITE;
1589 	if (flags & MAP_NOSYNC)
1590 		docow |= MAP_DISABLE_SYNCER;
1591 	if (flags & MAP_NOCORE)
1592 		docow |= MAP_DISABLE_COREDUMP;
1593 	/* Shared memory is also shared with children. */
1594 	if (flags & MAP_SHARED)
1595 		docow |= MAP_INHERIT_SHARE;
1596 	if (writecounted)
1597 		docow |= MAP_WRITECOUNT;
1598 	if (flags & MAP_STACK) {
1599 		if (object != NULL)
1600 			return (EINVAL);
1601 		docow |= MAP_STACK_GROWS_DOWN;
1602 	}
1603 	if ((flags & MAP_EXCL) != 0)
1604 		docow |= MAP_CHECK_EXCL;
1605 	if ((flags & MAP_GUARD) != 0)
1606 		docow |= MAP_CREATE_GUARD;
1607 
1608 	if (fitit) {
1609 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1610 			findspace = VMFS_SUPER_SPACE;
1611 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1612 			findspace = VMFS_ALIGNED_SPACE(flags >>
1613 			    MAP_ALIGNMENT_SHIFT);
1614 		else
1615 			findspace = VMFS_OPTIMAL_SPACE;
1616 		max_addr = 0;
1617 #ifdef MAP_32BIT
1618 		if ((flags & MAP_32BIT) != 0)
1619 			max_addr = MAP_32BIT_MAX_ADDR;
1620 #endif
1621 		if (curmap) {
1622 			rv = vm_map_find_min(map, object, foff, addr, size,
1623 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1624 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1625 			    findspace, prot, maxprot, docow);
1626 		} else {
1627 			rv = vm_map_find(map, object, foff, addr, size,
1628 			    max_addr, findspace, prot, maxprot, docow);
1629 		}
1630 	} else {
1631 		rv = vm_map_fixed(map, object, foff, *addr, size,
1632 		    prot, maxprot, docow);
1633 	}
1634 
1635 	if (rv == KERN_SUCCESS) {
1636 		/*
1637 		 * If the process has requested that all future mappings
1638 		 * be wired, then heed this.
1639 		 */
1640 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1641 			vm_map_lock(map);
1642 			if ((map->flags & MAP_WIREFUTURE) != 0)
1643 				(void)vm_map_wire_locked(map, *addr,
1644 				    *addr + size, VM_MAP_WIRE_USER |
1645 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1646 				    VM_MAP_WIRE_NOHOLES));
1647 			vm_map_unlock(map);
1648 		}
1649 	}
1650 	return (vm_mmap_to_errno(rv));
1651 }
1652 
1653 /*
1654  * Translate a Mach VM return code to zero on success or the appropriate errno
1655  * on failure.
1656  */
1657 int
1658 vm_mmap_to_errno(int rv)
1659 {
1660 
1661 	switch (rv) {
1662 	case KERN_SUCCESS:
1663 		return (0);
1664 	case KERN_INVALID_ADDRESS:
1665 	case KERN_NO_SPACE:
1666 		return (ENOMEM);
1667 	case KERN_PROTECTION_FAILURE:
1668 		return (EACCES);
1669 	default:
1670 		return (EINVAL);
1671 	}
1672 }
1673