xref: /freebsd/sys/vm/vm_mmap.c (revision 716fd348e01c5f2ba125f878a634a753436c2994)
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 protections 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 _Static_assert(MAXPAGESIZES <= 4, "MINCORE_SUPER too narrow");
116 
117 #ifndef _SYS_SYSPROTO_H_
118 struct sbrk_args {
119 	int incr;
120 };
121 #endif
122 
123 int
124 sys_sbrk(struct thread *td, struct sbrk_args *uap)
125 {
126 	/* Not yet implemented */
127 	return (EOPNOTSUPP);
128 }
129 
130 #ifndef _SYS_SYSPROTO_H_
131 struct sstk_args {
132 	int incr;
133 };
134 #endif
135 
136 int
137 sys_sstk(struct thread *td, struct sstk_args *uap)
138 {
139 	/* Not yet implemented */
140 	return (EOPNOTSUPP);
141 }
142 
143 #if defined(COMPAT_43)
144 int
145 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
146 {
147 
148 	td->td_retval[0] = PAGE_SIZE;
149 	return (0);
150 }
151 #endif				/* COMPAT_43 */
152 
153 /*
154  * Memory Map (mmap) system call.  Note that the file offset
155  * and address are allowed to be NOT page aligned, though if
156  * the MAP_FIXED flag it set, both must have the same remainder
157  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
158  * page-aligned, the actual mapping starts at trunc_page(addr)
159  * and the return value is adjusted up by the page offset.
160  *
161  * Generally speaking, only character devices which are themselves
162  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
163  * there would be no cache coherency between a descriptor and a VM mapping
164  * both to the same character device.
165  */
166 #ifndef _SYS_SYSPROTO_H_
167 struct mmap_args {
168 	void *addr;
169 	size_t len;
170 	int prot;
171 	int flags;
172 	int fd;
173 	long pad;
174 	off_t pos;
175 };
176 #endif
177 
178 int
179 sys_mmap(struct thread *td, struct mmap_args *uap)
180 {
181 
182 	return (kern_mmap(td, &(struct mmap_req){
183 		.mr_hint = (uintptr_t)uap->addr,
184 		.mr_len = uap->len,
185 		.mr_prot = uap->prot,
186 		.mr_flags = uap->flags,
187 		.mr_fd = uap->fd,
188 		.mr_pos = uap->pos,
189 	    }));
190 }
191 
192 int
193 kern_mmap_maxprot(struct proc *p, int prot)
194 {
195 
196 	if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
197 	    (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
198 		return (_PROT_ALL);
199 	if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
200 	    prot != PROT_NONE)
201 		 return (prot);
202 	return (_PROT_ALL);
203 }
204 
205 int
206 kern_mmap(struct thread *td, const struct mmap_req *mrp)
207 {
208 	struct vmspace *vms;
209 	struct file *fp;
210 	struct proc *p;
211 	off_t pos;
212 	vm_offset_t addr, orig_addr;
213 	vm_size_t len, pageoff, size;
214 	vm_prot_t cap_maxprot;
215 	int align, error, fd, flags, max_prot, prot;
216 	cap_rights_t rights;
217 	mmap_check_fp_fn check_fp_fn;
218 
219 	orig_addr = addr = mrp->mr_hint;
220 	len = mrp->mr_len;
221 	prot = mrp->mr_prot;
222 	flags = mrp->mr_flags;
223 	fd = mrp->mr_fd;
224 	pos = mrp->mr_pos;
225 	check_fp_fn = mrp->mr_check_fp_fn;
226 
227 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
228 		return (EINVAL);
229 	max_prot = PROT_MAX_EXTRACT(prot);
230 	prot = PROT_EXTRACT(prot);
231 	if (max_prot != 0 && (max_prot & prot) != prot)
232 		return (ENOTSUP);
233 
234 	p = td->td_proc;
235 
236 	/*
237 	 * Always honor PROT_MAX if set.  If not, default to all
238 	 * permissions unless we're implying maximum permissions.
239 	 */
240 	if (max_prot == 0)
241 		max_prot = kern_mmap_maxprot(p, prot);
242 
243 	vms = p->p_vmspace;
244 	fp = NULL;
245 	AUDIT_ARG_FD(fd);
246 
247 	/*
248 	 * Ignore old flags that used to be defined but did not do anything.
249 	 */
250 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
251 
252 	/*
253 	 * Enforce the constraints.
254 	 * Mapping of length 0 is only allowed for old binaries.
255 	 * Anonymous mapping shall specify -1 as filedescriptor and
256 	 * zero position for new code. Be nice to ancient a.out
257 	 * binaries and correct pos for anonymous mapping, since old
258 	 * ld.so sometimes issues anonymous map requests with non-zero
259 	 * pos.
260 	 */
261 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
262 		if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
263 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
264 			return (EINVAL);
265 	} else {
266 		if ((flags & MAP_ANON) != 0)
267 			pos = 0;
268 	}
269 
270 	if (flags & MAP_STACK) {
271 		if ((fd != -1) ||
272 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
273 			return (EINVAL);
274 		flags |= MAP_ANON;
275 		pos = 0;
276 	}
277 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
278 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
279 	    MAP_PREFAULT_READ | MAP_GUARD |
280 #ifdef MAP_32BIT
281 	    MAP_32BIT |
282 #endif
283 	    MAP_ALIGNMENT_MASK)) != 0)
284 		return (EINVAL);
285 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
286 		return (EINVAL);
287 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
288 		return (EINVAL);
289 	if (prot != PROT_NONE &&
290 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
291 		return (EINVAL);
292 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
293 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
294 #ifdef MAP_32BIT
295 	    MAP_32BIT |
296 #endif
297 	    MAP_ALIGNMENT_MASK)) != 0))
298 		return (EINVAL);
299 
300 	/*
301 	 * Align the file position to a page boundary,
302 	 * and save its page offset component.
303 	 */
304 	pageoff = (pos & PAGE_MASK);
305 	pos -= pageoff;
306 
307 	/* Compute size from len by rounding (on both ends). */
308 	size = len + pageoff;			/* low end... */
309 	size = round_page(size);		/* hi end */
310 	/* Check for rounding up to zero. */
311 	if (len > size)
312 		return (ENOMEM);
313 
314 	/* Ensure alignment is at least a page and fits in a pointer. */
315 	align = flags & MAP_ALIGNMENT_MASK;
316 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
317 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
318 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
319 		return (EINVAL);
320 
321 	/*
322 	 * Check for illegal addresses.  Watch out for address wrap... Note
323 	 * that VM_*_ADDRESS are not constants due to casts (argh).
324 	 */
325 	if (flags & MAP_FIXED) {
326 		/*
327 		 * The specified address must have the same remainder
328 		 * as the file offset taken modulo PAGE_SIZE, so it
329 		 * should be aligned after adjustment by pageoff.
330 		 */
331 		addr -= pageoff;
332 		if (addr & PAGE_MASK)
333 			return (EINVAL);
334 
335 		/* Address range must be all in user VM space. */
336 		if (!vm_map_range_valid(&vms->vm_map, addr, addr + size))
337 			return (EINVAL);
338 #ifdef MAP_32BIT
339 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
340 			return (EINVAL);
341 	} else if (flags & MAP_32BIT) {
342 		/*
343 		 * For MAP_32BIT, override the hint if it is too high and
344 		 * do not bother moving the mapping past the heap (since
345 		 * the heap is usually above 2GB).
346 		 */
347 		if (addr + size > MAP_32BIT_MAX_ADDR)
348 			addr = 0;
349 #endif
350 	} else {
351 		/*
352 		 * XXX for non-fixed mappings where no hint is provided or
353 		 * the hint would fall in the potential heap space,
354 		 * place it after the end of the largest possible heap.
355 		 *
356 		 * There should really be a pmap call to determine a reasonable
357 		 * location.
358 		 */
359 		if (addr == 0 ||
360 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
361 		    addr < round_page((vm_offset_t)vms->vm_daddr +
362 		    lim_max(td, RLIMIT_DATA))))
363 			addr = round_page((vm_offset_t)vms->vm_daddr +
364 			    lim_max(td, RLIMIT_DATA));
365 	}
366 	if (len == 0) {
367 		/*
368 		 * Return success without mapping anything for old
369 		 * binaries that request a page-aligned mapping of
370 		 * length 0.  For modern binaries, this function
371 		 * returns an error earlier.
372 		 */
373 		error = 0;
374 	} else if ((flags & MAP_GUARD) != 0) {
375 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
376 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
377 	} else if ((flags & MAP_ANON) != 0) {
378 		/*
379 		 * Mapping blank space is trivial.
380 		 *
381 		 * This relies on VM_PROT_* matching PROT_*.
382 		 */
383 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
384 		    max_prot, flags, NULL, pos, FALSE, td);
385 	} else {
386 		/*
387 		 * Mapping file, get fp for validation and don't let the
388 		 * descriptor disappear on us if we block. Check capability
389 		 * rights, but also return the maximum rights to be combined
390 		 * with maxprot later.
391 		 */
392 		cap_rights_init_one(&rights, CAP_MMAP);
393 		if (prot & PROT_READ)
394 			cap_rights_set_one(&rights, CAP_MMAP_R);
395 		if ((flags & MAP_SHARED) != 0) {
396 			if (prot & PROT_WRITE)
397 				cap_rights_set_one(&rights, CAP_MMAP_W);
398 		}
399 		if (prot & PROT_EXEC)
400 			cap_rights_set_one(&rights, CAP_MMAP_X);
401 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
402 		if (error != 0)
403 			goto done;
404 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
405 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
406 			error = EINVAL;
407 			goto done;
408 		}
409 		if (check_fp_fn != NULL) {
410 			error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
411 			    flags);
412 			if (error != 0)
413 				goto done;
414 		}
415 		if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data))
416 			addr = orig_addr;
417 		/* This relies on VM_PROT_* matching PROT_*. */
418 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
419 		    max_prot & cap_maxprot, flags, pos, td);
420 	}
421 
422 	if (error == 0)
423 		td->td_retval[0] = addr + pageoff;
424 done:
425 	if (fp)
426 		fdrop(fp, td);
427 
428 	return (error);
429 }
430 
431 #if defined(COMPAT_FREEBSD6)
432 int
433 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
434 {
435 	return (kern_mmap(td, &(struct mmap_req){
436 		.mr_hint = (uintptr_t)uap->addr,
437 		.mr_len = uap->len,
438 		.mr_prot = uap->prot,
439 		.mr_flags = uap->flags,
440 		.mr_fd = uap->fd,
441 		.mr_pos = uap->pos,
442 	    }));
443 }
444 #endif
445 
446 #ifdef COMPAT_43
447 #ifndef _SYS_SYSPROTO_H_
448 struct ommap_args {
449 	caddr_t addr;
450 	int len;
451 	int prot;
452 	int flags;
453 	int fd;
454 	long pos;
455 };
456 #endif
457 int
458 ommap(struct thread *td, struct ommap_args *uap)
459 {
460 	return (kern_ommap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
461 	    uap->flags, uap->fd, uap->pos));
462 }
463 
464 int
465 kern_ommap(struct thread *td, uintptr_t hint, int len, int oprot,
466     int oflags, int fd, long pos)
467 {
468 	static const char cvtbsdprot[8] = {
469 		0,
470 		PROT_EXEC,
471 		PROT_WRITE,
472 		PROT_EXEC | PROT_WRITE,
473 		PROT_READ,
474 		PROT_EXEC | PROT_READ,
475 		PROT_WRITE | PROT_READ,
476 		PROT_EXEC | PROT_WRITE | PROT_READ,
477 	};
478 	int flags, prot;
479 
480 	if (len < 0)
481 		return (EINVAL);
482 
483 #define	OMAP_ANON	0x0002
484 #define	OMAP_COPY	0x0020
485 #define	OMAP_SHARED	0x0010
486 #define	OMAP_FIXED	0x0100
487 
488 	prot = cvtbsdprot[oprot & 0x7];
489 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
490 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
491 	    prot != 0)
492 		prot |= PROT_EXEC;
493 #endif
494 	flags = 0;
495 	if (oflags & OMAP_ANON)
496 		flags |= MAP_ANON;
497 	if (oflags & OMAP_COPY)
498 		flags |= MAP_COPY;
499 	if (oflags & OMAP_SHARED)
500 		flags |= MAP_SHARED;
501 	else
502 		flags |= MAP_PRIVATE;
503 	if (oflags & OMAP_FIXED)
504 		flags |= MAP_FIXED;
505 	return (kern_mmap(td, &(struct mmap_req){
506 		.mr_hint = hint,
507 		.mr_len = len,
508 		.mr_prot = prot,
509 		.mr_flags = flags,
510 		.mr_fd = fd,
511 		.mr_pos = pos,
512 	    }));
513 }
514 #endif				/* COMPAT_43 */
515 
516 #ifndef _SYS_SYSPROTO_H_
517 struct msync_args {
518 	void *addr;
519 	size_t len;
520 	int flags;
521 };
522 #endif
523 int
524 sys_msync(struct thread *td, struct msync_args *uap)
525 {
526 
527 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
528 }
529 
530 int
531 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
532 {
533 	vm_offset_t addr;
534 	vm_size_t pageoff;
535 	vm_map_t map;
536 	int rv;
537 
538 	addr = addr0;
539 	pageoff = (addr & PAGE_MASK);
540 	addr -= pageoff;
541 	size += pageoff;
542 	size = (vm_size_t) round_page(size);
543 	if (addr + size < addr)
544 		return (EINVAL);
545 
546 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
547 		return (EINVAL);
548 
549 	map = &td->td_proc->p_vmspace->vm_map;
550 
551 	/*
552 	 * Clean the pages and interpret the return value.
553 	 */
554 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
555 	    (flags & MS_INVALIDATE) != 0);
556 	switch (rv) {
557 	case KERN_SUCCESS:
558 		return (0);
559 	case KERN_INVALID_ADDRESS:
560 		return (ENOMEM);
561 	case KERN_INVALID_ARGUMENT:
562 		return (EBUSY);
563 	case KERN_FAILURE:
564 		return (EIO);
565 	default:
566 		return (EINVAL);
567 	}
568 }
569 
570 #ifndef _SYS_SYSPROTO_H_
571 struct munmap_args {
572 	void *addr;
573 	size_t len;
574 };
575 #endif
576 int
577 sys_munmap(struct thread *td, struct munmap_args *uap)
578 {
579 
580 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
581 }
582 
583 int
584 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
585 {
586 #ifdef HWPMC_HOOKS
587 	struct pmckern_map_out pkm;
588 	vm_map_entry_t entry;
589 	bool pmc_handled;
590 #endif
591 	vm_offset_t addr, end;
592 	vm_size_t pageoff;
593 	vm_map_t map;
594 	int rv;
595 
596 	if (size == 0)
597 		return (EINVAL);
598 
599 	addr = addr0;
600 	pageoff = (addr & PAGE_MASK);
601 	addr -= pageoff;
602 	size += pageoff;
603 	size = (vm_size_t) round_page(size);
604 	end = addr + size;
605 	map = &td->td_proc->p_vmspace->vm_map;
606 	if (!vm_map_range_valid(map, addr, end))
607 		return (EINVAL);
608 
609 	vm_map_lock(map);
610 #ifdef HWPMC_HOOKS
611 	pmc_handled = false;
612 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
613 		pmc_handled = true;
614 		/*
615 		 * Inform hwpmc if the address range being unmapped contains
616 		 * an executable region.
617 		 */
618 		pkm.pm_address = (uintptr_t) NULL;
619 		if (vm_map_lookup_entry(map, addr, &entry)) {
620 			for (; entry->start < end;
621 			    entry = vm_map_entry_succ(entry)) {
622 				if (vm_map_check_protection(map, entry->start,
623 					entry->end, VM_PROT_EXECUTE) == TRUE) {
624 					pkm.pm_address = (uintptr_t) addr;
625 					pkm.pm_size = (size_t) size;
626 					break;
627 				}
628 			}
629 		}
630 	}
631 #endif
632 	rv = vm_map_delete(map, addr, end);
633 
634 #ifdef HWPMC_HOOKS
635 	if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) {
636 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
637 		vm_map_lock_downgrade(map);
638 		if (pkm.pm_address != (uintptr_t) NULL)
639 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
640 		vm_map_unlock_read(map);
641 	} else
642 #endif
643 		vm_map_unlock(map);
644 
645 	return (vm_mmap_to_errno(rv));
646 }
647 
648 #ifndef _SYS_SYSPROTO_H_
649 struct mprotect_args {
650 	const void *addr;
651 	size_t len;
652 	int prot;
653 };
654 #endif
655 int
656 sys_mprotect(struct thread *td, struct mprotect_args *uap)
657 {
658 
659 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
660 }
661 
662 int
663 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
664 {
665 	vm_offset_t addr;
666 	vm_size_t pageoff;
667 	int vm_error, max_prot;
668 	int flags;
669 
670 	addr = addr0;
671 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
672 		return (EINVAL);
673 	max_prot = PROT_MAX_EXTRACT(prot);
674 	prot = PROT_EXTRACT(prot);
675 	pageoff = (addr & PAGE_MASK);
676 	addr -= pageoff;
677 	size += pageoff;
678 	size = (vm_size_t) round_page(size);
679 #ifdef COMPAT_FREEBSD32
680 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
681 		if (((addr + size) & 0xffffffff) < addr)
682 			return (EINVAL);
683 	} else
684 #endif
685 	if (addr + size < addr)
686 		return (EINVAL);
687 
688 	flags = VM_MAP_PROTECT_SET_PROT;
689 	if (max_prot != 0)
690 		flags |= VM_MAP_PROTECT_SET_MAXPROT;
691 	vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
692 	    addr, addr + size, prot, max_prot, flags);
693 
694 	switch (vm_error) {
695 	case KERN_SUCCESS:
696 		return (0);
697 	case KERN_PROTECTION_FAILURE:
698 		return (EACCES);
699 	case KERN_RESOURCE_SHORTAGE:
700 		return (ENOMEM);
701 	case KERN_OUT_OF_BOUNDS:
702 		return (ENOTSUP);
703 	}
704 	return (EINVAL);
705 }
706 
707 #ifndef _SYS_SYSPROTO_H_
708 struct minherit_args {
709 	void *addr;
710 	size_t len;
711 	int inherit;
712 };
713 #endif
714 int
715 sys_minherit(struct thread *td, struct minherit_args *uap)
716 {
717 
718 	return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
719 	    uap->inherit));
720 }
721 
722 int
723 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
724 {
725 	vm_offset_t addr;
726 	vm_size_t size, pageoff;
727 	vm_inherit_t inherit;
728 
729 	addr = (vm_offset_t)addr0;
730 	size = len;
731 	inherit = inherit0;
732 
733 	pageoff = (addr & PAGE_MASK);
734 	addr -= pageoff;
735 	size += pageoff;
736 	size = (vm_size_t) round_page(size);
737 	if (addr + size < addr)
738 		return (EINVAL);
739 
740 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
741 	    addr + size, inherit)) {
742 	case KERN_SUCCESS:
743 		return (0);
744 	case KERN_PROTECTION_FAILURE:
745 		return (EACCES);
746 	}
747 	return (EINVAL);
748 }
749 
750 #ifndef _SYS_SYSPROTO_H_
751 struct madvise_args {
752 	void *addr;
753 	size_t len;
754 	int behav;
755 };
756 #endif
757 
758 int
759 sys_madvise(struct thread *td, struct madvise_args *uap)
760 {
761 
762 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
763 }
764 
765 int
766 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
767 {
768 	vm_map_t map;
769 	vm_offset_t addr, end, start;
770 	int flags;
771 
772 	/*
773 	 * Check for our special case, advising the swap pager we are
774 	 * "immortal."
775 	 */
776 	if (behav == MADV_PROTECT) {
777 		flags = PPROT_SET;
778 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
779 		    PROC_SPROTECT, &flags));
780 	}
781 
782 	/*
783 	 * Check for illegal addresses.  Watch out for address wrap... Note
784 	 * that VM_*_ADDRESS are not constants due to casts (argh).
785 	 */
786 	map = &td->td_proc->p_vmspace->vm_map;
787 	addr = addr0;
788 	if (!vm_map_range_valid(map, addr, addr + len))
789 		return (EINVAL);
790 
791 	/*
792 	 * Since this routine is only advisory, we default to conservative
793 	 * behavior.
794 	 */
795 	start = trunc_page(addr);
796 	end = round_page(addr + len);
797 
798 	/*
799 	 * vm_map_madvise() checks for illegal values of behav.
800 	 */
801 	return (vm_map_madvise(map, start, end, behav));
802 }
803 
804 #ifndef _SYS_SYSPROTO_H_
805 struct mincore_args {
806 	const void *addr;
807 	size_t len;
808 	char *vec;
809 };
810 #endif
811 
812 int
813 sys_mincore(struct thread *td, struct mincore_args *uap)
814 {
815 
816 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
817 }
818 
819 int
820 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
821 {
822 	pmap_t pmap;
823 	vm_map_t map;
824 	vm_map_entry_t current, entry;
825 	vm_object_t object;
826 	vm_offset_t addr, cend, end, first_addr;
827 	vm_paddr_t pa;
828 	vm_page_t m;
829 	vm_pindex_t pindex;
830 	int error, lastvecindex, mincoreinfo, vecindex;
831 	unsigned int timestamp;
832 
833 	/*
834 	 * Make sure that the addresses presented are valid for user
835 	 * mode.
836 	 */
837 	first_addr = addr = trunc_page(addr0);
838 	end = round_page(addr0 + len);
839 	map = &td->td_proc->p_vmspace->vm_map;
840 	if (end > vm_map_max(map) || end < addr)
841 		return (ENOMEM);
842 
843 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
844 
845 	vm_map_lock_read(map);
846 RestartScan:
847 	timestamp = map->timestamp;
848 
849 	if (!vm_map_lookup_entry(map, addr, &entry)) {
850 		vm_map_unlock_read(map);
851 		return (ENOMEM);
852 	}
853 
854 	/*
855 	 * Do this on a map entry basis so that if the pages are not
856 	 * in the current processes address space, we can easily look
857 	 * up the pages elsewhere.
858 	 */
859 	lastvecindex = -1;
860 	while (entry->start < end) {
861 		/*
862 		 * check for contiguity
863 		 */
864 		current = entry;
865 		entry = vm_map_entry_succ(current);
866 		if (current->end < end &&
867 		    entry->start > current->end) {
868 			vm_map_unlock_read(map);
869 			return (ENOMEM);
870 		}
871 
872 		/*
873 		 * ignore submaps (for now) or null objects
874 		 */
875 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
876 		    current->object.vm_object == NULL)
877 			continue;
878 
879 		/*
880 		 * limit this scan to the current map entry and the
881 		 * limits for the mincore call
882 		 */
883 		if (addr < current->start)
884 			addr = current->start;
885 		cend = current->end;
886 		if (cend > end)
887 			cend = end;
888 
889 		for (; addr < cend; addr += PAGE_SIZE) {
890 			/*
891 			 * Check pmap first, it is likely faster, also
892 			 * it can provide info as to whether we are the
893 			 * one referencing or modifying the page.
894 			 */
895 			m = NULL;
896 			object = NULL;
897 retry:
898 			pa = 0;
899 			mincoreinfo = pmap_mincore(pmap, addr, &pa);
900 			if (mincore_mapped) {
901 				/*
902 				 * We only care about this pmap's
903 				 * mapping of the page, if any.
904 				 */
905 				;
906 			} else if (pa != 0) {
907 				/*
908 				 * The page is mapped by this process but not
909 				 * both accessed and modified.  It is also
910 				 * managed.  Acquire the object lock so that
911 				 * other mappings might be examined.  The page's
912 				 * identity may change at any point before its
913 				 * object lock is acquired, so re-validate if
914 				 * necessary.
915 				 */
916 				m = PHYS_TO_VM_PAGE(pa);
917 				while (object == NULL || m->object != object) {
918 					if (object != NULL)
919 						VM_OBJECT_WUNLOCK(object);
920 					object = atomic_load_ptr(&m->object);
921 					if (object == NULL)
922 						goto retry;
923 					VM_OBJECT_WLOCK(object);
924 				}
925 				if (pa != pmap_extract(pmap, addr))
926 					goto retry;
927 				KASSERT(vm_page_all_valid(m),
928 				    ("mincore: page %p is mapped but invalid",
929 				    m));
930 			} else if (mincoreinfo == 0) {
931 				/*
932 				 * The page is not mapped by this process.  If
933 				 * the object implements managed pages, then
934 				 * determine if the page is resident so that
935 				 * the mappings might be examined.
936 				 */
937 				if (current->object.vm_object != object) {
938 					if (object != NULL)
939 						VM_OBJECT_WUNLOCK(object);
940 					object = current->object.vm_object;
941 					VM_OBJECT_WLOCK(object);
942 				}
943 				if (object->type == OBJT_DEFAULT ||
944 				    (object->flags & OBJ_SWAP) != 0 ||
945 				    object->type == OBJT_VNODE) {
946 					pindex = OFF_TO_IDX(current->offset +
947 					    (addr - current->start));
948 					m = vm_page_lookup(object, pindex);
949 					if (m != NULL && vm_page_none_valid(m))
950 						m = NULL;
951 					if (m != NULL)
952 						mincoreinfo = MINCORE_INCORE;
953 				}
954 			}
955 			if (m != NULL) {
956 				VM_OBJECT_ASSERT_WLOCKED(m->object);
957 
958 				/* Examine other mappings of the page. */
959 				if (m->dirty == 0 && pmap_is_modified(m))
960 					vm_page_dirty(m);
961 				if (m->dirty != 0)
962 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
963 
964 				/*
965 				 * The first test for PGA_REFERENCED is an
966 				 * optimization.  The second test is
967 				 * required because a concurrent pmap
968 				 * operation could clear the last reference
969 				 * and set PGA_REFERENCED before the call to
970 				 * pmap_is_referenced().
971 				 */
972 				if ((m->a.flags & PGA_REFERENCED) != 0 ||
973 				    pmap_is_referenced(m) ||
974 				    (m->a.flags & PGA_REFERENCED) != 0)
975 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
976 			}
977 			if (object != NULL)
978 				VM_OBJECT_WUNLOCK(object);
979 
980 			/*
981 			 * subyte may page fault.  In case it needs to modify
982 			 * the map, we release the lock.
983 			 */
984 			vm_map_unlock_read(map);
985 
986 			/*
987 			 * calculate index into user supplied byte vector
988 			 */
989 			vecindex = atop(addr - first_addr);
990 
991 			/*
992 			 * If we have skipped map entries, we need to make sure that
993 			 * the byte vector is zeroed for those skipped entries.
994 			 */
995 			while ((lastvecindex + 1) < vecindex) {
996 				++lastvecindex;
997 				error = subyte(vec + lastvecindex, 0);
998 				if (error) {
999 					error = EFAULT;
1000 					goto done2;
1001 				}
1002 			}
1003 
1004 			/*
1005 			 * Pass the page information to the user
1006 			 */
1007 			error = subyte(vec + vecindex, mincoreinfo);
1008 			if (error) {
1009 				error = EFAULT;
1010 				goto done2;
1011 			}
1012 
1013 			/*
1014 			 * If the map has changed, due to the subyte, the previous
1015 			 * output may be invalid.
1016 			 */
1017 			vm_map_lock_read(map);
1018 			if (timestamp != map->timestamp)
1019 				goto RestartScan;
1020 
1021 			lastvecindex = vecindex;
1022 		}
1023 	}
1024 
1025 	/*
1026 	 * subyte may page fault.  In case it needs to modify
1027 	 * the map, we release the lock.
1028 	 */
1029 	vm_map_unlock_read(map);
1030 
1031 	/*
1032 	 * Zero the last entries in the byte vector.
1033 	 */
1034 	vecindex = atop(end - first_addr);
1035 	while ((lastvecindex + 1) < vecindex) {
1036 		++lastvecindex;
1037 		error = subyte(vec + lastvecindex, 0);
1038 		if (error) {
1039 			error = EFAULT;
1040 			goto done2;
1041 		}
1042 	}
1043 
1044 	/*
1045 	 * If the map has changed, due to the subyte, the previous
1046 	 * output may be invalid.
1047 	 */
1048 	vm_map_lock_read(map);
1049 	if (timestamp != map->timestamp)
1050 		goto RestartScan;
1051 	vm_map_unlock_read(map);
1052 done2:
1053 	return (error);
1054 }
1055 
1056 #ifndef _SYS_SYSPROTO_H_
1057 struct mlock_args {
1058 	const void *addr;
1059 	size_t len;
1060 };
1061 #endif
1062 int
1063 sys_mlock(struct thread *td, struct mlock_args *uap)
1064 {
1065 
1066 	return (kern_mlock(td->td_proc, td->td_ucred,
1067 	    __DECONST(uintptr_t, uap->addr), uap->len));
1068 }
1069 
1070 int
1071 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1072 {
1073 	vm_offset_t addr, end, last, start;
1074 	vm_size_t npages, size;
1075 	vm_map_t map;
1076 	unsigned long nsize;
1077 	int error;
1078 
1079 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1080 	if (error)
1081 		return (error);
1082 	addr = addr0;
1083 	size = len;
1084 	last = addr + size;
1085 	start = trunc_page(addr);
1086 	end = round_page(last);
1087 	if (last < addr || end < addr)
1088 		return (EINVAL);
1089 	npages = atop(end - start);
1090 	if (npages > vm_page_max_user_wired)
1091 		return (ENOMEM);
1092 	map = &proc->p_vmspace->vm_map;
1093 	PROC_LOCK(proc);
1094 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1095 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1096 		PROC_UNLOCK(proc);
1097 		return (ENOMEM);
1098 	}
1099 	PROC_UNLOCK(proc);
1100 #ifdef RACCT
1101 	if (racct_enable) {
1102 		PROC_LOCK(proc);
1103 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1104 		PROC_UNLOCK(proc);
1105 		if (error != 0)
1106 			return (ENOMEM);
1107 	}
1108 #endif
1109 	error = vm_map_wire(map, start, end,
1110 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1111 #ifdef RACCT
1112 	if (racct_enable && error != KERN_SUCCESS) {
1113 		PROC_LOCK(proc);
1114 		racct_set(proc, RACCT_MEMLOCK,
1115 		    ptoa(pmap_wired_count(map->pmap)));
1116 		PROC_UNLOCK(proc);
1117 	}
1118 #endif
1119 	switch (error) {
1120 	case KERN_SUCCESS:
1121 		return (0);
1122 	case KERN_INVALID_ARGUMENT:
1123 		return (EINVAL);
1124 	default:
1125 		return (ENOMEM);
1126 	}
1127 }
1128 
1129 #ifndef _SYS_SYSPROTO_H_
1130 struct mlockall_args {
1131 	int	how;
1132 };
1133 #endif
1134 
1135 int
1136 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1137 {
1138 	vm_map_t map;
1139 	int error;
1140 
1141 	map = &td->td_proc->p_vmspace->vm_map;
1142 	error = priv_check(td, PRIV_VM_MLOCK);
1143 	if (error)
1144 		return (error);
1145 
1146 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1147 		return (EINVAL);
1148 
1149 	/*
1150 	 * If wiring all pages in the process would cause it to exceed
1151 	 * a hard resource limit, return ENOMEM.
1152 	 */
1153 	if (!old_mlock && uap->how & MCL_CURRENT) {
1154 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1155 			return (ENOMEM);
1156 	}
1157 #ifdef RACCT
1158 	if (racct_enable) {
1159 		PROC_LOCK(td->td_proc);
1160 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1161 		PROC_UNLOCK(td->td_proc);
1162 		if (error != 0)
1163 			return (ENOMEM);
1164 	}
1165 #endif
1166 
1167 	if (uap->how & MCL_FUTURE) {
1168 		vm_map_lock(map);
1169 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1170 		vm_map_unlock(map);
1171 		error = 0;
1172 	}
1173 
1174 	if (uap->how & MCL_CURRENT) {
1175 		/*
1176 		 * P1003.1-2001 mandates that all currently mapped pages
1177 		 * will be memory resident and locked (wired) upon return
1178 		 * from mlockall(). vm_map_wire() will wire pages, by
1179 		 * calling vm_fault_wire() for each page in the region.
1180 		 */
1181 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1182 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1183 		if (error == KERN_SUCCESS)
1184 			error = 0;
1185 		else if (error == KERN_RESOURCE_SHORTAGE)
1186 			error = ENOMEM;
1187 		else
1188 			error = EAGAIN;
1189 	}
1190 #ifdef RACCT
1191 	if (racct_enable && error != KERN_SUCCESS) {
1192 		PROC_LOCK(td->td_proc);
1193 		racct_set(td->td_proc, RACCT_MEMLOCK,
1194 		    ptoa(pmap_wired_count(map->pmap)));
1195 		PROC_UNLOCK(td->td_proc);
1196 	}
1197 #endif
1198 
1199 	return (error);
1200 }
1201 
1202 #ifndef _SYS_SYSPROTO_H_
1203 struct munlockall_args {
1204 	register_t dummy;
1205 };
1206 #endif
1207 
1208 int
1209 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1210 {
1211 	vm_map_t map;
1212 	int error;
1213 
1214 	map = &td->td_proc->p_vmspace->vm_map;
1215 	error = priv_check(td, PRIV_VM_MUNLOCK);
1216 	if (error)
1217 		return (error);
1218 
1219 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1220 	vm_map_lock(map);
1221 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1222 	vm_map_unlock(map);
1223 
1224 	/* Forcibly unwire all pages. */
1225 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1226 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1227 #ifdef RACCT
1228 	if (racct_enable && error == KERN_SUCCESS) {
1229 		PROC_LOCK(td->td_proc);
1230 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1231 		PROC_UNLOCK(td->td_proc);
1232 	}
1233 #endif
1234 
1235 	return (error);
1236 }
1237 
1238 #ifndef _SYS_SYSPROTO_H_
1239 struct munlock_args {
1240 	const void *addr;
1241 	size_t len;
1242 };
1243 #endif
1244 int
1245 sys_munlock(struct thread *td, struct munlock_args *uap)
1246 {
1247 
1248 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1249 }
1250 
1251 int
1252 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1253 {
1254 	vm_offset_t addr, end, last, start;
1255 #ifdef RACCT
1256 	vm_map_t map;
1257 #endif
1258 	int error;
1259 
1260 	error = priv_check(td, PRIV_VM_MUNLOCK);
1261 	if (error)
1262 		return (error);
1263 	addr = addr0;
1264 	last = addr + size;
1265 	start = trunc_page(addr);
1266 	end = round_page(last);
1267 	if (last < addr || end < addr)
1268 		return (EINVAL);
1269 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1270 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1271 #ifdef RACCT
1272 	if (racct_enable && error == KERN_SUCCESS) {
1273 		PROC_LOCK(td->td_proc);
1274 		map = &td->td_proc->p_vmspace->vm_map;
1275 		racct_set(td->td_proc, RACCT_MEMLOCK,
1276 		    ptoa(pmap_wired_count(map->pmap)));
1277 		PROC_UNLOCK(td->td_proc);
1278 	}
1279 #endif
1280 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1281 }
1282 
1283 /*
1284  * vm_mmap_vnode()
1285  *
1286  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1287  * operations on vnodes.
1288  */
1289 int
1290 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1291     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1292     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1293     boolean_t *writecounted)
1294 {
1295 	struct vattr va;
1296 	vm_object_t obj;
1297 	vm_ooffset_t foff;
1298 	struct ucred *cred;
1299 	int error, flags;
1300 	bool writex;
1301 
1302 	cred = td->td_ucred;
1303 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1304 	    (*flagsp & MAP_SHARED) != 0;
1305 	if ((error = vget(vp, LK_SHARED)) != 0)
1306 		return (error);
1307 	AUDIT_ARG_VNODE1(vp);
1308 	foff = *foffp;
1309 	flags = *flagsp;
1310 	obj = vp->v_object;
1311 	if (vp->v_type == VREG) {
1312 		/*
1313 		 * Get the proper underlying object
1314 		 */
1315 		if (obj == NULL) {
1316 			error = EINVAL;
1317 			goto done;
1318 		}
1319 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1320 			vput(vp);
1321 			vp = (struct vnode *)obj->handle;
1322 			/*
1323 			 * Bypass filesystems obey the mpsafety of the
1324 			 * underlying fs.  Tmpfs never bypasses.
1325 			 */
1326 			error = vget(vp, LK_SHARED);
1327 			if (error != 0)
1328 				return (error);
1329 		}
1330 		if (writex) {
1331 			*writecounted = TRUE;
1332 			vm_pager_update_writecount(obj, 0, objsize);
1333 		}
1334 	} else {
1335 		error = EINVAL;
1336 		goto done;
1337 	}
1338 	if ((error = VOP_GETATTR(vp, &va, cred)))
1339 		goto done;
1340 #ifdef MAC
1341 	/* This relies on VM_PROT_* matching PROT_*. */
1342 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1343 	if (error != 0)
1344 		goto done;
1345 #endif
1346 	if ((flags & MAP_SHARED) != 0) {
1347 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1348 			if (prot & VM_PROT_WRITE) {
1349 				error = EPERM;
1350 				goto done;
1351 			}
1352 			*maxprotp &= ~VM_PROT_WRITE;
1353 		}
1354 	}
1355 	/*
1356 	 * If it is a regular file without any references
1357 	 * we do not need to sync it.
1358 	 * Adjust object size to be the size of actual file.
1359 	 */
1360 	objsize = round_page(va.va_size);
1361 	if (va.va_nlink == 0)
1362 		flags |= MAP_NOSYNC;
1363 	if (obj->type == OBJT_VNODE) {
1364 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1365 		    cred);
1366 		if (obj == NULL) {
1367 			error = ENOMEM;
1368 			goto done;
1369 		}
1370 	} else {
1371 		KASSERT(obj->type == OBJT_DEFAULT ||
1372 		    (obj->flags & OBJ_SWAP) != 0,
1373 		    ("wrong object type"));
1374 		vm_object_reference(obj);
1375 #if VM_NRESERVLEVEL > 0
1376 		if ((obj->flags & OBJ_COLORED) == 0) {
1377 			VM_OBJECT_WLOCK(obj);
1378 			vm_object_color(obj, 0);
1379 			VM_OBJECT_WUNLOCK(obj);
1380 		}
1381 #endif
1382 	}
1383 	*objp = obj;
1384 	*flagsp = flags;
1385 
1386 	VOP_MMAPPED(vp);
1387 
1388 done:
1389 	if (error != 0 && *writecounted) {
1390 		*writecounted = FALSE;
1391 		vm_pager_update_writecount(obj, objsize, 0);
1392 	}
1393 	vput(vp);
1394 	return (error);
1395 }
1396 
1397 /*
1398  * vm_mmap_cdev()
1399  *
1400  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1401  * operations on cdevs.
1402  */
1403 int
1404 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1405     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1406     vm_ooffset_t *foff, vm_object_t *objp)
1407 {
1408 	vm_object_t obj;
1409 	int error, flags;
1410 
1411 	flags = *flagsp;
1412 
1413 	if (dsw->d_flags & D_MMAP_ANON) {
1414 		*objp = NULL;
1415 		*foff = 0;
1416 		*maxprotp = VM_PROT_ALL;
1417 		*flagsp |= MAP_ANON;
1418 		return (0);
1419 	}
1420 	/*
1421 	 * cdevs do not provide private mappings of any kind.
1422 	 */
1423 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1424 	    (prot & VM_PROT_WRITE) != 0)
1425 		return (EACCES);
1426 	if (flags & (MAP_PRIVATE|MAP_COPY))
1427 		return (EINVAL);
1428 	/*
1429 	 * Force device mappings to be shared.
1430 	 */
1431 	flags |= MAP_SHARED;
1432 #ifdef MAC_XXX
1433 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1434 	if (error != 0)
1435 		return (error);
1436 #endif
1437 	/*
1438 	 * First, try d_mmap_single().  If that is not implemented
1439 	 * (returns ENODEV), fall back to using the device pager.
1440 	 * Note that d_mmap_single() must return a reference to the
1441 	 * object (it needs to bump the reference count of the object
1442 	 * it returns somehow).
1443 	 *
1444 	 * XXX assumes VM_PROT_* == PROT_*
1445 	 */
1446 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1447 	if (error != ENODEV)
1448 		return (error);
1449 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1450 	    td->td_ucred);
1451 	if (obj == NULL)
1452 		return (EINVAL);
1453 	*objp = obj;
1454 	*flagsp = flags;
1455 	return (0);
1456 }
1457 
1458 /*
1459  * vm_mmap()
1460  *
1461  * Internal version of mmap used by exec, sys5 shared memory, and
1462  * various device drivers.  Handle is either a vnode pointer, a
1463  * character device, or NULL for MAP_ANON.
1464  */
1465 int
1466 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1467 	vm_prot_t maxprot, int flags,
1468 	objtype_t handle_type, void *handle,
1469 	vm_ooffset_t foff)
1470 {
1471 	vm_object_t object;
1472 	struct thread *td = curthread;
1473 	int error;
1474 	boolean_t writecounted;
1475 
1476 	if (size == 0)
1477 		return (EINVAL);
1478 
1479 	size = round_page(size);
1480 	object = NULL;
1481 	writecounted = FALSE;
1482 
1483 	/*
1484 	 * Lookup/allocate object.
1485 	 */
1486 	switch (handle_type) {
1487 	case OBJT_DEVICE: {
1488 		struct cdevsw *dsw;
1489 		struct cdev *cdev;
1490 		int ref;
1491 
1492 		cdev = handle;
1493 		dsw = dev_refthread(cdev, &ref);
1494 		if (dsw == NULL)
1495 			return (ENXIO);
1496 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1497 		    dsw, &foff, &object);
1498 		dev_relthread(cdev, ref);
1499 		break;
1500 	}
1501 	case OBJT_VNODE:
1502 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1503 		    handle, &foff, &object, &writecounted);
1504 		break;
1505 	case OBJT_DEFAULT:
1506 		if (handle == NULL) {
1507 			error = 0;
1508 			break;
1509 		}
1510 		/* FALLTHROUGH */
1511 	default:
1512 		error = EINVAL;
1513 		break;
1514 	}
1515 	if (error)
1516 		return (error);
1517 
1518 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1519 	    foff, writecounted, td);
1520 	if (error != 0 && object != NULL) {
1521 		/*
1522 		 * If this mapping was accounted for in the vnode's
1523 		 * writecount, then undo that now.
1524 		 */
1525 		if (writecounted)
1526 			vm_pager_release_writecount(object, 0, size);
1527 		vm_object_deallocate(object);
1528 	}
1529 	return (error);
1530 }
1531 
1532 int
1533 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size)
1534 {
1535 	int error;
1536 
1537 	RACCT_PROC_LOCK(td->td_proc);
1538 	if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1539 		RACCT_PROC_UNLOCK(td->td_proc);
1540 		return (ENOMEM);
1541 	}
1542 	if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1543 		RACCT_PROC_UNLOCK(td->td_proc);
1544 		return (ENOMEM);
1545 	}
1546 	if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1547 		if (ptoa(pmap_wired_count(map->pmap)) + size >
1548 		    lim_cur(td, RLIMIT_MEMLOCK)) {
1549 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1550 			RACCT_PROC_UNLOCK(td->td_proc);
1551 			return (ENOMEM);
1552 		}
1553 		error = racct_set(td->td_proc, RACCT_MEMLOCK,
1554 		    ptoa(pmap_wired_count(map->pmap)) + size);
1555 		if (error != 0) {
1556 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1557 			RACCT_PROC_UNLOCK(td->td_proc);
1558 			return (error);
1559 		}
1560 	}
1561 	RACCT_PROC_UNLOCK(td->td_proc);
1562 	return (0);
1563 }
1564 
1565 /*
1566  * Internal version of mmap that maps a specific VM object into an
1567  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1568  */
1569 int
1570 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1571     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1572     boolean_t writecounted, struct thread *td)
1573 {
1574 	vm_offset_t max_addr;
1575 	int docow, error, findspace, rv;
1576 	bool curmap, fitit;
1577 
1578 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1579 	if (curmap) {
1580 		error = kern_mmap_racct_check(td, map, size);
1581 		if (error != 0)
1582 			return (error);
1583 	}
1584 
1585 	/*
1586 	 * We currently can only deal with page aligned file offsets.
1587 	 * The mmap() system call already enforces this by subtracting
1588 	 * the page offset from the file offset, but checking here
1589 	 * catches errors in device drivers (e.g. d_single_mmap()
1590 	 * callbacks) and other internal mapping requests (such as in
1591 	 * exec).
1592 	 */
1593 	if (foff & PAGE_MASK)
1594 		return (EINVAL);
1595 
1596 	if ((flags & MAP_FIXED) == 0) {
1597 		fitit = TRUE;
1598 		*addr = round_page(*addr);
1599 	} else {
1600 		if (*addr != trunc_page(*addr))
1601 			return (EINVAL);
1602 		fitit = FALSE;
1603 	}
1604 
1605 	if (flags & MAP_ANON) {
1606 		if (object != NULL || foff != 0)
1607 			return (EINVAL);
1608 		docow = 0;
1609 	} else if (flags & MAP_PREFAULT_READ)
1610 		docow = MAP_PREFAULT;
1611 	else
1612 		docow = MAP_PREFAULT_PARTIAL;
1613 
1614 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1615 		docow |= MAP_COPY_ON_WRITE;
1616 	if (flags & MAP_NOSYNC)
1617 		docow |= MAP_DISABLE_SYNCER;
1618 	if (flags & MAP_NOCORE)
1619 		docow |= MAP_DISABLE_COREDUMP;
1620 	/* Shared memory is also shared with children. */
1621 	if (flags & MAP_SHARED)
1622 		docow |= MAP_INHERIT_SHARE;
1623 	if (writecounted)
1624 		docow |= MAP_WRITECOUNT;
1625 	if (flags & MAP_STACK) {
1626 		if (object != NULL)
1627 			return (EINVAL);
1628 		docow |= MAP_STACK_GROWS_DOWN;
1629 	}
1630 	if ((flags & MAP_EXCL) != 0)
1631 		docow |= MAP_CHECK_EXCL;
1632 	if ((flags & MAP_GUARD) != 0)
1633 		docow |= MAP_CREATE_GUARD;
1634 
1635 	if (fitit) {
1636 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1637 			findspace = VMFS_SUPER_SPACE;
1638 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1639 			findspace = VMFS_ALIGNED_SPACE(flags >>
1640 			    MAP_ALIGNMENT_SHIFT);
1641 		else
1642 			findspace = VMFS_OPTIMAL_SPACE;
1643 		max_addr = 0;
1644 #ifdef MAP_32BIT
1645 		if ((flags & MAP_32BIT) != 0)
1646 			max_addr = MAP_32BIT_MAX_ADDR;
1647 #endif
1648 		if (curmap) {
1649 			rv = vm_map_find_min(map, object, foff, addr, size,
1650 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1651 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1652 			    findspace, prot, maxprot, docow);
1653 		} else {
1654 			rv = vm_map_find(map, object, foff, addr, size,
1655 			    max_addr, findspace, prot, maxprot, docow);
1656 		}
1657 	} else {
1658 		rv = vm_map_fixed(map, object, foff, *addr, size,
1659 		    prot, maxprot, docow);
1660 	}
1661 
1662 	if (rv == KERN_SUCCESS) {
1663 		/*
1664 		 * If the process has requested that all future mappings
1665 		 * be wired, then heed this.
1666 		 */
1667 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1668 			vm_map_lock(map);
1669 			if ((map->flags & MAP_WIREFUTURE) != 0)
1670 				(void)vm_map_wire_locked(map, *addr,
1671 				    *addr + size, VM_MAP_WIRE_USER |
1672 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1673 				    VM_MAP_WIRE_NOHOLES));
1674 			vm_map_unlock(map);
1675 		}
1676 	}
1677 	return (vm_mmap_to_errno(rv));
1678 }
1679 
1680 /*
1681  * Translate a Mach VM return code to zero on success or the appropriate errno
1682  * on failure.
1683  */
1684 int
1685 vm_mmap_to_errno(int rv)
1686 {
1687 
1688 	switch (rv) {
1689 	case KERN_SUCCESS:
1690 		return (0);
1691 	case KERN_INVALID_ADDRESS:
1692 	case KERN_NO_SPACE:
1693 		return (ENOMEM);
1694 	case KERN_PROTECTION_FAILURE:
1695 		return (EACCES);
1696 	default:
1697 		return (EINVAL);
1698 	}
1699 }
1700