xref: /freebsd/sys/vm/vm_mmap.c (revision 6683132d54bd6d589889e43dabdc53d35e38a028)
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/filedesc.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procctl.h>
62 #include <sys/racct.h>
63 #include <sys/resource.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 #include <sys/fcntl.h>
69 #include <sys/file.h>
70 #include <sys/mman.h>
71 #include <sys/mount.h>
72 #include <sys/conf.h>
73 #include <sys/stat.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysent.h>
76 #include <sys/vmmeter.h>
77 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
78 #include <machine/md_var.h>
79 #endif
80 
81 #include <security/audit/audit.h>
82 #include <security/mac/mac_framework.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_pageout.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_page.h>
94 #include <vm/vnode_pager.h>
95 
96 #ifdef HWPMC_HOOKS
97 #include <sys/pmckern.h>
98 #endif
99 
100 int old_mlock = 0;
101 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
102     "Do not apply RLIMIT_MEMLOCK on mlockall");
103 static int mincore_mapped = 1;
104 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
105     "mincore reports mappings, not residency");
106 static int imply_prot_max = 0;
107 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0,
108     "Imply maximum page permissions in mmap() when none are specified");
109 
110 #ifdef MAP_32BIT
111 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
112 #endif
113 
114 #ifndef _SYS_SYSPROTO_H_
115 struct sbrk_args {
116 	int incr;
117 };
118 #endif
119 
120 int
121 sys_sbrk(struct thread *td, struct sbrk_args *uap)
122 {
123 	/* Not yet implemented */
124 	return (EOPNOTSUPP);
125 }
126 
127 #ifndef _SYS_SYSPROTO_H_
128 struct sstk_args {
129 	int incr;
130 };
131 #endif
132 
133 int
134 sys_sstk(struct thread *td, struct sstk_args *uap)
135 {
136 	/* Not yet implemented */
137 	return (EOPNOTSUPP);
138 }
139 
140 #if defined(COMPAT_43)
141 int
142 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
143 {
144 
145 	td->td_retval[0] = PAGE_SIZE;
146 	return (0);
147 }
148 #endif				/* COMPAT_43 */
149 
150 
151 /*
152  * Memory Map (mmap) system call.  Note that the file offset
153  * and address are allowed to be NOT page aligned, though if
154  * the MAP_FIXED flag it set, both must have the same remainder
155  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
156  * page-aligned, the actual mapping starts at trunc_page(addr)
157  * and the return value is adjusted up by the page offset.
158  *
159  * Generally speaking, only character devices which are themselves
160  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
161  * there would be no cache coherency between a descriptor and a VM mapping
162  * both to the same character device.
163  */
164 #ifndef _SYS_SYSPROTO_H_
165 struct mmap_args {
166 	void *addr;
167 	size_t len;
168 	int prot;
169 	int flags;
170 	int fd;
171 	long pad;
172 	off_t pos;
173 };
174 #endif
175 
176 int
177 sys_mmap(struct thread *td, struct mmap_args *uap)
178 {
179 
180 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
181 	    uap->flags, uap->fd, uap->pos));
182 }
183 
184 int
185 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags,
186     int fd, off_t pos)
187 {
188 	struct vmspace *vms;
189 	struct file *fp;
190 	vm_offset_t addr;
191 	vm_size_t pageoff, size;
192 	vm_prot_t cap_maxprot;
193 	int align, error, max_prot;
194 	cap_rights_t rights;
195 
196 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
197 		return (EINVAL);
198 	max_prot = PROT_MAX_EXTRACT(prot);
199 	prot = PROT_EXTRACT(prot);
200 	if (max_prot != 0 && (max_prot & prot) != prot)
201 		return (EINVAL);
202 	/*
203 	 * Always honor PROT_MAX if set.  If not, default to all
204 	 * permissions unless we're implying maximum permissions.
205 	 *
206 	 * XXX: should be tunable per process and ABI.
207 	 */
208 	if (max_prot == 0)
209 		max_prot = (imply_prot_max && prot != PROT_NONE) ?
210 		    prot : _PROT_ALL;
211 
212 	vms = td->td_proc->p_vmspace;
213 	fp = NULL;
214 	AUDIT_ARG_FD(fd);
215 	addr = addr0;
216 
217 	/*
218 	 * Ignore old flags that used to be defined but did not do anything.
219 	 */
220 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
221 
222 	/*
223 	 * Enforce the constraints.
224 	 * Mapping of length 0 is only allowed for old binaries.
225 	 * Anonymous mapping shall specify -1 as filedescriptor and
226 	 * zero position for new code. Be nice to ancient a.out
227 	 * binaries and correct pos for anonymous mapping, since old
228 	 * ld.so sometimes issues anonymous map requests with non-zero
229 	 * pos.
230 	 */
231 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
232 		if ((len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
233 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
234 			return (EINVAL);
235 	} else {
236 		if ((flags & MAP_ANON) != 0)
237 			pos = 0;
238 	}
239 
240 	if (flags & MAP_STACK) {
241 		if ((fd != -1) ||
242 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
243 			return (EINVAL);
244 		flags |= MAP_ANON;
245 		pos = 0;
246 	}
247 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
248 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
249 	    MAP_PREFAULT_READ | MAP_GUARD |
250 #ifdef MAP_32BIT
251 	    MAP_32BIT |
252 #endif
253 	    MAP_ALIGNMENT_MASK)) != 0)
254 		return (EINVAL);
255 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
256 		return (EINVAL);
257 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
258 		return (EINVAL);
259 	if (prot != PROT_NONE &&
260 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
261 		return (EINVAL);
262 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
263 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
264 #ifdef MAP_32BIT
265 	    MAP_32BIT |
266 #endif
267 	    MAP_ALIGNMENT_MASK)) != 0))
268 		return (EINVAL);
269 
270 	/*
271 	 * Align the file position to a page boundary,
272 	 * and save its page offset component.
273 	 */
274 	pageoff = (pos & PAGE_MASK);
275 	pos -= pageoff;
276 
277 	/* Compute size from len by rounding (on both ends). */
278 	size = len + pageoff;			/* low end... */
279 	size = round_page(size);		/* hi end */
280 	/* Check for rounding up to zero. */
281 	if (len > size)
282 		return (ENOMEM);
283 
284 	/* Ensure alignment is at least a page and fits in a pointer. */
285 	align = flags & MAP_ALIGNMENT_MASK;
286 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
287 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
288 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
289 		return (EINVAL);
290 
291 	/*
292 	 * Check for illegal addresses.  Watch out for address wrap... Note
293 	 * that VM_*_ADDRESS are not constants due to casts (argh).
294 	 */
295 	if (flags & MAP_FIXED) {
296 		/*
297 		 * The specified address must have the same remainder
298 		 * as the file offset taken modulo PAGE_SIZE, so it
299 		 * should be aligned after adjustment by pageoff.
300 		 */
301 		addr -= pageoff;
302 		if (addr & PAGE_MASK)
303 			return (EINVAL);
304 
305 		/* Address range must be all in user VM space. */
306 		if (addr < vm_map_min(&vms->vm_map) ||
307 		    addr + size > vm_map_max(&vms->vm_map))
308 			return (EINVAL);
309 		if (addr + size < addr)
310 			return (EINVAL);
311 #ifdef MAP_32BIT
312 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
313 			return (EINVAL);
314 	} else if (flags & MAP_32BIT) {
315 		/*
316 		 * For MAP_32BIT, override the hint if it is too high and
317 		 * do not bother moving the mapping past the heap (since
318 		 * the heap is usually above 2GB).
319 		 */
320 		if (addr + size > MAP_32BIT_MAX_ADDR)
321 			addr = 0;
322 #endif
323 	} else {
324 		/*
325 		 * XXX for non-fixed mappings where no hint is provided or
326 		 * the hint would fall in the potential heap space,
327 		 * place it after the end of the largest possible heap.
328 		 *
329 		 * There should really be a pmap call to determine a reasonable
330 		 * location.
331 		 */
332 		if (addr == 0 ||
333 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
334 		    addr < round_page((vm_offset_t)vms->vm_daddr +
335 		    lim_max(td, RLIMIT_DATA))))
336 			addr = round_page((vm_offset_t)vms->vm_daddr +
337 			    lim_max(td, RLIMIT_DATA));
338 	}
339 	if (len == 0) {
340 		/*
341 		 * Return success without mapping anything for old
342 		 * binaries that request a page-aligned mapping of
343 		 * length 0.  For modern binaries, this function
344 		 * returns an error earlier.
345 		 */
346 		error = 0;
347 	} else if ((flags & MAP_GUARD) != 0) {
348 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
349 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
350 	} else if ((flags & MAP_ANON) != 0) {
351 		/*
352 		 * Mapping blank space is trivial.
353 		 *
354 		 * This relies on VM_PROT_* matching PROT_*.
355 		 */
356 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
357 		    max_prot, flags, NULL, pos, FALSE, td);
358 	} else {
359 		/*
360 		 * Mapping file, get fp for validation and don't let the
361 		 * descriptor disappear on us if we block. Check capability
362 		 * rights, but also return the maximum rights to be combined
363 		 * with maxprot later.
364 		 */
365 		cap_rights_init(&rights, CAP_MMAP);
366 		if (prot & PROT_READ)
367 			cap_rights_set(&rights, CAP_MMAP_R);
368 		if ((flags & MAP_SHARED) != 0) {
369 			if (prot & PROT_WRITE)
370 				cap_rights_set(&rights, CAP_MMAP_W);
371 		}
372 		if (prot & PROT_EXEC)
373 			cap_rights_set(&rights, CAP_MMAP_X);
374 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
375 		if (error != 0)
376 			goto done;
377 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
378 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
379 			error = EINVAL;
380 			goto done;
381 		}
382 
383 		/* This relies on VM_PROT_* matching PROT_*. */
384 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
385 		    max_prot & cap_maxprot, flags, pos, td);
386 	}
387 
388 	if (error == 0)
389 		td->td_retval[0] = (register_t) (addr + pageoff);
390 done:
391 	if (fp)
392 		fdrop(fp, td);
393 
394 	return (error);
395 }
396 
397 #if defined(COMPAT_FREEBSD6)
398 int
399 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
400 {
401 
402 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
403 	    uap->flags, uap->fd, uap->pos));
404 }
405 #endif
406 
407 #ifdef COMPAT_43
408 #ifndef _SYS_SYSPROTO_H_
409 struct ommap_args {
410 	caddr_t addr;
411 	int len;
412 	int prot;
413 	int flags;
414 	int fd;
415 	long pos;
416 };
417 #endif
418 int
419 ommap(struct thread *td, struct ommap_args *uap)
420 {
421 	static const char cvtbsdprot[8] = {
422 		0,
423 		PROT_EXEC,
424 		PROT_WRITE,
425 		PROT_EXEC | PROT_WRITE,
426 		PROT_READ,
427 		PROT_EXEC | PROT_READ,
428 		PROT_WRITE | PROT_READ,
429 		PROT_EXEC | PROT_WRITE | PROT_READ,
430 	};
431 	int flags, prot;
432 
433 #define	OMAP_ANON	0x0002
434 #define	OMAP_COPY	0x0020
435 #define	OMAP_SHARED	0x0010
436 #define	OMAP_FIXED	0x0100
437 
438 	prot = cvtbsdprot[uap->prot & 0x7];
439 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
440 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
441 	    prot != 0)
442 		prot |= PROT_EXEC;
443 #endif
444 	flags = 0;
445 	if (uap->flags & OMAP_ANON)
446 		flags |= MAP_ANON;
447 	if (uap->flags & OMAP_COPY)
448 		flags |= MAP_COPY;
449 	if (uap->flags & OMAP_SHARED)
450 		flags |= MAP_SHARED;
451 	else
452 		flags |= MAP_PRIVATE;
453 	if (uap->flags & OMAP_FIXED)
454 		flags |= MAP_FIXED;
455 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
456 	    uap->fd, uap->pos));
457 }
458 #endif				/* COMPAT_43 */
459 
460 
461 #ifndef _SYS_SYSPROTO_H_
462 struct msync_args {
463 	void *addr;
464 	size_t len;
465 	int flags;
466 };
467 #endif
468 int
469 sys_msync(struct thread *td, struct msync_args *uap)
470 {
471 
472 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
473 }
474 
475 int
476 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
477 {
478 	vm_offset_t addr;
479 	vm_size_t pageoff;
480 	vm_map_t map;
481 	int rv;
482 
483 	addr = addr0;
484 	pageoff = (addr & PAGE_MASK);
485 	addr -= pageoff;
486 	size += pageoff;
487 	size = (vm_size_t) round_page(size);
488 	if (addr + size < addr)
489 		return (EINVAL);
490 
491 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
492 		return (EINVAL);
493 
494 	map = &td->td_proc->p_vmspace->vm_map;
495 
496 	/*
497 	 * Clean the pages and interpret the return value.
498 	 */
499 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
500 	    (flags & MS_INVALIDATE) != 0);
501 	switch (rv) {
502 	case KERN_SUCCESS:
503 		return (0);
504 	case KERN_INVALID_ADDRESS:
505 		return (ENOMEM);
506 	case KERN_INVALID_ARGUMENT:
507 		return (EBUSY);
508 	case KERN_FAILURE:
509 		return (EIO);
510 	default:
511 		return (EINVAL);
512 	}
513 }
514 
515 #ifndef _SYS_SYSPROTO_H_
516 struct munmap_args {
517 	void *addr;
518 	size_t len;
519 };
520 #endif
521 int
522 sys_munmap(struct thread *td, struct munmap_args *uap)
523 {
524 
525 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
526 }
527 
528 int
529 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
530 {
531 #ifdef HWPMC_HOOKS
532 	struct pmckern_map_out pkm;
533 	vm_map_entry_t entry;
534 	bool pmc_handled;
535 #endif
536 	vm_offset_t addr;
537 	vm_size_t pageoff;
538 	vm_map_t map;
539 
540 	if (size == 0)
541 		return (EINVAL);
542 
543 	addr = addr0;
544 	pageoff = (addr & PAGE_MASK);
545 	addr -= pageoff;
546 	size += pageoff;
547 	size = (vm_size_t) round_page(size);
548 	if (addr + size < addr)
549 		return (EINVAL);
550 
551 	/*
552 	 * Check for illegal addresses.  Watch out for address wrap...
553 	 */
554 	map = &td->td_proc->p_vmspace->vm_map;
555 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
556 		return (EINVAL);
557 	vm_map_lock(map);
558 #ifdef HWPMC_HOOKS
559 	pmc_handled = false;
560 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
561 		pmc_handled = true;
562 		/*
563 		 * Inform hwpmc if the address range being unmapped contains
564 		 * an executable region.
565 		 */
566 		pkm.pm_address = (uintptr_t) NULL;
567 		if (vm_map_lookup_entry(map, addr, &entry)) {
568 			for (; entry->start < addr + size;
569 			    entry = entry->next) {
570 				if (vm_map_check_protection(map, entry->start,
571 					entry->end, VM_PROT_EXECUTE) == TRUE) {
572 					pkm.pm_address = (uintptr_t) addr;
573 					pkm.pm_size = (size_t) size;
574 					break;
575 				}
576 			}
577 		}
578 	}
579 #endif
580 	vm_map_delete(map, addr, addr + size);
581 
582 #ifdef HWPMC_HOOKS
583 	if (__predict_false(pmc_handled)) {
584 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
585 		vm_map_lock_downgrade(map);
586 		if (pkm.pm_address != (uintptr_t) NULL)
587 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
588 		vm_map_unlock_read(map);
589 	} else
590 #endif
591 		vm_map_unlock(map);
592 
593 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
594 	return (0);
595 }
596 
597 #ifndef _SYS_SYSPROTO_H_
598 struct mprotect_args {
599 	const void *addr;
600 	size_t len;
601 	int prot;
602 };
603 #endif
604 int
605 sys_mprotect(struct thread *td, struct mprotect_args *uap)
606 {
607 
608 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
609 }
610 
611 int
612 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
613 {
614 	vm_offset_t addr;
615 	vm_size_t pageoff;
616 	int vm_error, max_prot;
617 
618 	addr = addr0;
619 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
620 		return (EINVAL);
621 	max_prot = PROT_MAX_EXTRACT(prot);
622 	prot = PROT_EXTRACT(prot);
623 	pageoff = (addr & PAGE_MASK);
624 	addr -= pageoff;
625 	size += pageoff;
626 	size = (vm_size_t) round_page(size);
627 #ifdef COMPAT_FREEBSD32
628 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
629 		if (((addr + size) & 0xffffffff) < addr)
630 			return (EINVAL);
631 	} else
632 #endif
633 	if (addr + size < addr)
634 		return (EINVAL);
635 
636 	vm_error = KERN_SUCCESS;
637 	if (max_prot != 0) {
638 		if ((max_prot & prot) != prot)
639 			return (EINVAL);
640 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
641 		    addr, addr + size, max_prot, TRUE);
642 	}
643 	if (vm_error == KERN_SUCCESS)
644 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
645 		    addr, addr + size, prot, FALSE);
646 
647 	switch (vm_error) {
648 	case KERN_SUCCESS:
649 		return (0);
650 	case KERN_PROTECTION_FAILURE:
651 		return (EACCES);
652 	case KERN_RESOURCE_SHORTAGE:
653 		return (ENOMEM);
654 	}
655 	return (EINVAL);
656 }
657 
658 #ifndef _SYS_SYSPROTO_H_
659 struct minherit_args {
660 	void *addr;
661 	size_t len;
662 	int inherit;
663 };
664 #endif
665 int
666 sys_minherit(struct thread *td, struct minherit_args *uap)
667 {
668 	vm_offset_t addr;
669 	vm_size_t size, pageoff;
670 	vm_inherit_t inherit;
671 
672 	addr = (vm_offset_t)uap->addr;
673 	size = uap->len;
674 	inherit = uap->inherit;
675 
676 	pageoff = (addr & PAGE_MASK);
677 	addr -= pageoff;
678 	size += pageoff;
679 	size = (vm_size_t) round_page(size);
680 	if (addr + size < addr)
681 		return (EINVAL);
682 
683 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
684 	    addr + size, inherit)) {
685 	case KERN_SUCCESS:
686 		return (0);
687 	case KERN_PROTECTION_FAILURE:
688 		return (EACCES);
689 	}
690 	return (EINVAL);
691 }
692 
693 #ifndef _SYS_SYSPROTO_H_
694 struct madvise_args {
695 	void *addr;
696 	size_t len;
697 	int behav;
698 };
699 #endif
700 
701 int
702 sys_madvise(struct thread *td, struct madvise_args *uap)
703 {
704 
705 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
706 }
707 
708 int
709 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
710 {
711 	vm_map_t map;
712 	vm_offset_t addr, end, start;
713 	int flags;
714 
715 	/*
716 	 * Check for our special case, advising the swap pager we are
717 	 * "immortal."
718 	 */
719 	if (behav == MADV_PROTECT) {
720 		flags = PPROT_SET;
721 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
722 		    PROC_SPROTECT, &flags));
723 	}
724 
725 	/*
726 	 * Check for illegal addresses.  Watch out for address wrap... Note
727 	 * that VM_*_ADDRESS are not constants due to casts (argh).
728 	 */
729 	map = &td->td_proc->p_vmspace->vm_map;
730 	addr = addr0;
731 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
732 		return (EINVAL);
733 	if ((addr + len) < addr)
734 		return (EINVAL);
735 
736 	/*
737 	 * Since this routine is only advisory, we default to conservative
738 	 * behavior.
739 	 */
740 	start = trunc_page(addr);
741 	end = round_page(addr + len);
742 
743 	/*
744 	 * vm_map_madvise() checks for illegal values of behav.
745 	 */
746 	return (vm_map_madvise(map, start, end, behav));
747 }
748 
749 #ifndef _SYS_SYSPROTO_H_
750 struct mincore_args {
751 	const void *addr;
752 	size_t len;
753 	char *vec;
754 };
755 #endif
756 
757 int
758 sys_mincore(struct thread *td, struct mincore_args *uap)
759 {
760 
761 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
762 }
763 
764 int
765 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
766 {
767 	vm_offset_t addr, first_addr;
768 	vm_offset_t end, cend;
769 	pmap_t pmap;
770 	vm_map_t map;
771 	int error = 0;
772 	int vecindex, lastvecindex;
773 	vm_map_entry_t current;
774 	vm_map_entry_t entry;
775 	vm_object_t object;
776 	vm_paddr_t locked_pa;
777 	vm_page_t m;
778 	vm_pindex_t pindex;
779 	int mincoreinfo;
780 	unsigned int timestamp;
781 	boolean_t locked;
782 
783 	/*
784 	 * Make sure that the addresses presented are valid for user
785 	 * mode.
786 	 */
787 	first_addr = addr = trunc_page(addr0);
788 	end = addr + (vm_size_t)round_page(len);
789 	map = &td->td_proc->p_vmspace->vm_map;
790 	if (end > vm_map_max(map) || end < addr)
791 		return (ENOMEM);
792 
793 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
794 
795 	vm_map_lock_read(map);
796 RestartScan:
797 	timestamp = map->timestamp;
798 
799 	if (!vm_map_lookup_entry(map, addr, &entry)) {
800 		vm_map_unlock_read(map);
801 		return (ENOMEM);
802 	}
803 
804 	/*
805 	 * Do this on a map entry basis so that if the pages are not
806 	 * in the current processes address space, we can easily look
807 	 * up the pages elsewhere.
808 	 */
809 	lastvecindex = -1;
810 	for (current = entry; current->start < end; current = current->next) {
811 
812 		/*
813 		 * check for contiguity
814 		 */
815 		if (current->end < end && current->next->start > current->end) {
816 			vm_map_unlock_read(map);
817 			return (ENOMEM);
818 		}
819 
820 		/*
821 		 * ignore submaps (for now) or null objects
822 		 */
823 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
824 			current->object.vm_object == NULL)
825 			continue;
826 
827 		/*
828 		 * limit this scan to the current map entry and the
829 		 * limits for the mincore call
830 		 */
831 		if (addr < current->start)
832 			addr = current->start;
833 		cend = current->end;
834 		if (cend > end)
835 			cend = end;
836 
837 		/*
838 		 * scan this entry one page at a time
839 		 */
840 		while (addr < cend) {
841 			/*
842 			 * Check pmap first, it is likely faster, also
843 			 * it can provide info as to whether we are the
844 			 * one referencing or modifying the page.
845 			 */
846 			object = NULL;
847 			locked_pa = 0;
848 		retry:
849 			m = NULL;
850 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
851 			if (mincore_mapped) {
852 				/*
853 				 * We only care about this pmap's
854 				 * mapping of the page, if any.
855 				 */
856 				if (locked_pa != 0) {
857 					vm_page_unlock(PHYS_TO_VM_PAGE(
858 					    locked_pa));
859 				}
860 			} else if (locked_pa != 0) {
861 				/*
862 				 * The page is mapped by this process but not
863 				 * both accessed and modified.  It is also
864 				 * managed.  Acquire the object lock so that
865 				 * other mappings might be examined.
866 				 */
867 				m = PHYS_TO_VM_PAGE(locked_pa);
868 				if (m->object != object) {
869 					if (object != NULL)
870 						VM_OBJECT_WUNLOCK(object);
871 					object = m->object;
872 					locked = VM_OBJECT_TRYWLOCK(object);
873 					vm_page_unlock(m);
874 					if (!locked) {
875 						VM_OBJECT_WLOCK(object);
876 						vm_page_lock(m);
877 						goto retry;
878 					}
879 				} else
880 					vm_page_unlock(m);
881 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
882 				    ("mincore: page %p is mapped but invalid",
883 				    m));
884 			} else if (mincoreinfo == 0) {
885 				/*
886 				 * The page is not mapped by this process.  If
887 				 * the object implements managed pages, then
888 				 * determine if the page is resident so that
889 				 * the mappings might be examined.
890 				 */
891 				if (current->object.vm_object != object) {
892 					if (object != NULL)
893 						VM_OBJECT_WUNLOCK(object);
894 					object = current->object.vm_object;
895 					VM_OBJECT_WLOCK(object);
896 				}
897 				if (object->type == OBJT_DEFAULT ||
898 				    object->type == OBJT_SWAP ||
899 				    object->type == OBJT_VNODE) {
900 					pindex = OFF_TO_IDX(current->offset +
901 					    (addr - current->start));
902 					m = vm_page_lookup(object, pindex);
903 					if (m != NULL && m->valid == 0)
904 						m = NULL;
905 					if (m != NULL)
906 						mincoreinfo = MINCORE_INCORE;
907 				}
908 			}
909 			if (m != NULL) {
910 				/* Examine other mappings to the page. */
911 				if (m->dirty == 0 && pmap_is_modified(m))
912 					vm_page_dirty(m);
913 				if (m->dirty != 0)
914 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
915 				/*
916 				 * The first test for PGA_REFERENCED is an
917 				 * optimization.  The second test is
918 				 * required because a concurrent pmap
919 				 * operation could clear the last reference
920 				 * and set PGA_REFERENCED before the call to
921 				 * pmap_is_referenced().
922 				 */
923 				if ((m->aflags & PGA_REFERENCED) != 0 ||
924 				    pmap_is_referenced(m) ||
925 				    (m->aflags & PGA_REFERENCED) != 0)
926 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
927 			}
928 			if (object != NULL)
929 				VM_OBJECT_WUNLOCK(object);
930 
931 			/*
932 			 * subyte may page fault.  In case it needs to modify
933 			 * the map, we release the lock.
934 			 */
935 			vm_map_unlock_read(map);
936 
937 			/*
938 			 * calculate index into user supplied byte vector
939 			 */
940 			vecindex = atop(addr - first_addr);
941 
942 			/*
943 			 * If we have skipped map entries, we need to make sure that
944 			 * the byte vector is zeroed for those skipped entries.
945 			 */
946 			while ((lastvecindex + 1) < vecindex) {
947 				++lastvecindex;
948 				error = subyte(vec + lastvecindex, 0);
949 				if (error) {
950 					error = EFAULT;
951 					goto done2;
952 				}
953 			}
954 
955 			/*
956 			 * Pass the page information to the user
957 			 */
958 			error = subyte(vec + vecindex, mincoreinfo);
959 			if (error) {
960 				error = EFAULT;
961 				goto done2;
962 			}
963 
964 			/*
965 			 * If the map has changed, due to the subyte, the previous
966 			 * output may be invalid.
967 			 */
968 			vm_map_lock_read(map);
969 			if (timestamp != map->timestamp)
970 				goto RestartScan;
971 
972 			lastvecindex = vecindex;
973 			addr += PAGE_SIZE;
974 		}
975 	}
976 
977 	/*
978 	 * subyte may page fault.  In case it needs to modify
979 	 * the map, we release the lock.
980 	 */
981 	vm_map_unlock_read(map);
982 
983 	/*
984 	 * Zero the last entries in the byte vector.
985 	 */
986 	vecindex = atop(end - first_addr);
987 	while ((lastvecindex + 1) < vecindex) {
988 		++lastvecindex;
989 		error = subyte(vec + lastvecindex, 0);
990 		if (error) {
991 			error = EFAULT;
992 			goto done2;
993 		}
994 	}
995 
996 	/*
997 	 * If the map has changed, due to the subyte, the previous
998 	 * output may be invalid.
999 	 */
1000 	vm_map_lock_read(map);
1001 	if (timestamp != map->timestamp)
1002 		goto RestartScan;
1003 	vm_map_unlock_read(map);
1004 done2:
1005 	return (error);
1006 }
1007 
1008 #ifndef _SYS_SYSPROTO_H_
1009 struct mlock_args {
1010 	const void *addr;
1011 	size_t len;
1012 };
1013 #endif
1014 int
1015 sys_mlock(struct thread *td, struct mlock_args *uap)
1016 {
1017 
1018 	return (kern_mlock(td->td_proc, td->td_ucred,
1019 	    __DECONST(uintptr_t, uap->addr), uap->len));
1020 }
1021 
1022 int
1023 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1024 {
1025 	vm_offset_t addr, end, last, start;
1026 	vm_size_t npages, size;
1027 	vm_map_t map;
1028 	unsigned long nsize;
1029 	int error;
1030 
1031 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1032 	if (error)
1033 		return (error);
1034 	addr = addr0;
1035 	size = len;
1036 	last = addr + size;
1037 	start = trunc_page(addr);
1038 	end = round_page(last);
1039 	if (last < addr || end < addr)
1040 		return (EINVAL);
1041 	npages = atop(end - start);
1042 	if (npages > vm_page_max_user_wired)
1043 		return (ENOMEM);
1044 	map = &proc->p_vmspace->vm_map;
1045 	PROC_LOCK(proc);
1046 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1047 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1048 		PROC_UNLOCK(proc);
1049 		return (ENOMEM);
1050 	}
1051 	PROC_UNLOCK(proc);
1052 #ifdef RACCT
1053 	if (racct_enable) {
1054 		PROC_LOCK(proc);
1055 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1056 		PROC_UNLOCK(proc);
1057 		if (error != 0)
1058 			return (ENOMEM);
1059 	}
1060 #endif
1061 	error = vm_map_wire(map, start, end,
1062 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1063 #ifdef RACCT
1064 	if (racct_enable && error != KERN_SUCCESS) {
1065 		PROC_LOCK(proc);
1066 		racct_set(proc, RACCT_MEMLOCK,
1067 		    ptoa(pmap_wired_count(map->pmap)));
1068 		PROC_UNLOCK(proc);
1069 	}
1070 #endif
1071 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1072 }
1073 
1074 #ifndef _SYS_SYSPROTO_H_
1075 struct mlockall_args {
1076 	int	how;
1077 };
1078 #endif
1079 
1080 int
1081 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1082 {
1083 	vm_map_t map;
1084 	int error;
1085 
1086 	map = &td->td_proc->p_vmspace->vm_map;
1087 	error = priv_check(td, PRIV_VM_MLOCK);
1088 	if (error)
1089 		return (error);
1090 
1091 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1092 		return (EINVAL);
1093 
1094 	/*
1095 	 * If wiring all pages in the process would cause it to exceed
1096 	 * a hard resource limit, return ENOMEM.
1097 	 */
1098 	if (!old_mlock && uap->how & MCL_CURRENT) {
1099 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1100 			return (ENOMEM);
1101 	}
1102 #ifdef RACCT
1103 	if (racct_enable) {
1104 		PROC_LOCK(td->td_proc);
1105 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1106 		PROC_UNLOCK(td->td_proc);
1107 		if (error != 0)
1108 			return (ENOMEM);
1109 	}
1110 #endif
1111 
1112 	if (uap->how & MCL_FUTURE) {
1113 		vm_map_lock(map);
1114 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1115 		vm_map_unlock(map);
1116 		error = 0;
1117 	}
1118 
1119 	if (uap->how & MCL_CURRENT) {
1120 		/*
1121 		 * P1003.1-2001 mandates that all currently mapped pages
1122 		 * will be memory resident and locked (wired) upon return
1123 		 * from mlockall(). vm_map_wire() will wire pages, by
1124 		 * calling vm_fault_wire() for each page in the region.
1125 		 */
1126 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1127 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1128 		if (error == KERN_SUCCESS)
1129 			error = 0;
1130 		else if (error == KERN_RESOURCE_SHORTAGE)
1131 			error = ENOMEM;
1132 		else
1133 			error = EAGAIN;
1134 	}
1135 #ifdef RACCT
1136 	if (racct_enable && error != KERN_SUCCESS) {
1137 		PROC_LOCK(td->td_proc);
1138 		racct_set(td->td_proc, RACCT_MEMLOCK,
1139 		    ptoa(pmap_wired_count(map->pmap)));
1140 		PROC_UNLOCK(td->td_proc);
1141 	}
1142 #endif
1143 
1144 	return (error);
1145 }
1146 
1147 #ifndef _SYS_SYSPROTO_H_
1148 struct munlockall_args {
1149 	register_t dummy;
1150 };
1151 #endif
1152 
1153 int
1154 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1155 {
1156 	vm_map_t map;
1157 	int error;
1158 
1159 	map = &td->td_proc->p_vmspace->vm_map;
1160 	error = priv_check(td, PRIV_VM_MUNLOCK);
1161 	if (error)
1162 		return (error);
1163 
1164 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1165 	vm_map_lock(map);
1166 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1167 	vm_map_unlock(map);
1168 
1169 	/* Forcibly unwire all pages. */
1170 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1171 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1172 #ifdef RACCT
1173 	if (racct_enable && error == KERN_SUCCESS) {
1174 		PROC_LOCK(td->td_proc);
1175 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1176 		PROC_UNLOCK(td->td_proc);
1177 	}
1178 #endif
1179 
1180 	return (error);
1181 }
1182 
1183 #ifndef _SYS_SYSPROTO_H_
1184 struct munlock_args {
1185 	const void *addr;
1186 	size_t len;
1187 };
1188 #endif
1189 int
1190 sys_munlock(struct thread *td, struct munlock_args *uap)
1191 {
1192 
1193 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1194 }
1195 
1196 int
1197 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1198 {
1199 	vm_offset_t addr, end, last, start;
1200 #ifdef RACCT
1201 	vm_map_t map;
1202 #endif
1203 	int error;
1204 
1205 	error = priv_check(td, PRIV_VM_MUNLOCK);
1206 	if (error)
1207 		return (error);
1208 	addr = addr0;
1209 	last = addr + size;
1210 	start = trunc_page(addr);
1211 	end = round_page(last);
1212 	if (last < addr || end < addr)
1213 		return (EINVAL);
1214 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1215 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1216 #ifdef RACCT
1217 	if (racct_enable && error == KERN_SUCCESS) {
1218 		PROC_LOCK(td->td_proc);
1219 		map = &td->td_proc->p_vmspace->vm_map;
1220 		racct_set(td->td_proc, RACCT_MEMLOCK,
1221 		    ptoa(pmap_wired_count(map->pmap)));
1222 		PROC_UNLOCK(td->td_proc);
1223 	}
1224 #endif
1225 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1226 }
1227 
1228 /*
1229  * vm_mmap_vnode()
1230  *
1231  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1232  * operations on vnodes.
1233  */
1234 int
1235 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1236     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1237     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1238     boolean_t *writecounted)
1239 {
1240 	struct vattr va;
1241 	vm_object_t obj;
1242 	vm_ooffset_t foff;
1243 	struct ucred *cred;
1244 	int error, flags;
1245 	bool writex;
1246 
1247 	cred = td->td_ucred;
1248 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1249 	    (*flagsp & MAP_SHARED) != 0;
1250 	if ((error = vget(vp, LK_SHARED, td)) != 0)
1251 		return (error);
1252 	AUDIT_ARG_VNODE1(vp);
1253 	foff = *foffp;
1254 	flags = *flagsp;
1255 	obj = vp->v_object;
1256 	if (vp->v_type == VREG) {
1257 		/*
1258 		 * Get the proper underlying object
1259 		 */
1260 		if (obj == NULL) {
1261 			error = EINVAL;
1262 			goto done;
1263 		}
1264 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1265 			vput(vp);
1266 			vp = (struct vnode *)obj->handle;
1267 			/*
1268 			 * Bypass filesystems obey the mpsafety of the
1269 			 * underlying fs.  Tmpfs never bypasses.
1270 			 */
1271 			error = vget(vp, LK_SHARED, td);
1272 			if (error != 0)
1273 				return (error);
1274 		}
1275 		if (writex) {
1276 			*writecounted = TRUE;
1277 			vnode_pager_update_writecount(obj, 0, objsize);
1278 		}
1279 	} else {
1280 		error = EINVAL;
1281 		goto done;
1282 	}
1283 	if ((error = VOP_GETATTR(vp, &va, cred)))
1284 		goto done;
1285 #ifdef MAC
1286 	/* This relies on VM_PROT_* matching PROT_*. */
1287 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1288 	if (error != 0)
1289 		goto done;
1290 #endif
1291 	if ((flags & MAP_SHARED) != 0) {
1292 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1293 			if (prot & VM_PROT_WRITE) {
1294 				error = EPERM;
1295 				goto done;
1296 			}
1297 			*maxprotp &= ~VM_PROT_WRITE;
1298 		}
1299 	}
1300 	/*
1301 	 * If it is a regular file without any references
1302 	 * we do not need to sync it.
1303 	 * Adjust object size to be the size of actual file.
1304 	 */
1305 	objsize = round_page(va.va_size);
1306 	if (va.va_nlink == 0)
1307 		flags |= MAP_NOSYNC;
1308 	if (obj->type == OBJT_VNODE) {
1309 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1310 		    cred);
1311 		if (obj == NULL) {
1312 			error = ENOMEM;
1313 			goto done;
1314 		}
1315 	} else {
1316 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1317 		    ("wrong object type"));
1318 		VM_OBJECT_WLOCK(obj);
1319 		vm_object_reference_locked(obj);
1320 #if VM_NRESERVLEVEL > 0
1321 		vm_object_color(obj, 0);
1322 #endif
1323 		VM_OBJECT_WUNLOCK(obj);
1324 	}
1325 	*objp = obj;
1326 	*flagsp = flags;
1327 
1328 	vfs_mark_atime(vp, cred);
1329 
1330 done:
1331 	if (error != 0 && *writecounted) {
1332 		*writecounted = FALSE;
1333 		vnode_pager_update_writecount(obj, objsize, 0);
1334 	}
1335 	vput(vp);
1336 	return (error);
1337 }
1338 
1339 /*
1340  * vm_mmap_cdev()
1341  *
1342  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1343  * operations on cdevs.
1344  */
1345 int
1346 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1347     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1348     vm_ooffset_t *foff, vm_object_t *objp)
1349 {
1350 	vm_object_t obj;
1351 	int error, flags;
1352 
1353 	flags = *flagsp;
1354 
1355 	if (dsw->d_flags & D_MMAP_ANON) {
1356 		*objp = NULL;
1357 		*foff = 0;
1358 		*maxprotp = VM_PROT_ALL;
1359 		*flagsp |= MAP_ANON;
1360 		return (0);
1361 	}
1362 	/*
1363 	 * cdevs do not provide private mappings of any kind.
1364 	 */
1365 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1366 	    (prot & VM_PROT_WRITE) != 0)
1367 		return (EACCES);
1368 	if (flags & (MAP_PRIVATE|MAP_COPY))
1369 		return (EINVAL);
1370 	/*
1371 	 * Force device mappings to be shared.
1372 	 */
1373 	flags |= MAP_SHARED;
1374 #ifdef MAC_XXX
1375 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1376 	if (error != 0)
1377 		return (error);
1378 #endif
1379 	/*
1380 	 * First, try d_mmap_single().  If that is not implemented
1381 	 * (returns ENODEV), fall back to using the device pager.
1382 	 * Note that d_mmap_single() must return a reference to the
1383 	 * object (it needs to bump the reference count of the object
1384 	 * it returns somehow).
1385 	 *
1386 	 * XXX assumes VM_PROT_* == PROT_*
1387 	 */
1388 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1389 	if (error != ENODEV)
1390 		return (error);
1391 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1392 	    td->td_ucred);
1393 	if (obj == NULL)
1394 		return (EINVAL);
1395 	*objp = obj;
1396 	*flagsp = flags;
1397 	return (0);
1398 }
1399 
1400 /*
1401  * vm_mmap()
1402  *
1403  * Internal version of mmap used by exec, sys5 shared memory, and
1404  * various device drivers.  Handle is either a vnode pointer, a
1405  * character device, or NULL for MAP_ANON.
1406  */
1407 int
1408 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1409 	vm_prot_t maxprot, int flags,
1410 	objtype_t handle_type, void *handle,
1411 	vm_ooffset_t foff)
1412 {
1413 	vm_object_t object;
1414 	struct thread *td = curthread;
1415 	int error;
1416 	boolean_t writecounted;
1417 
1418 	if (size == 0)
1419 		return (EINVAL);
1420 
1421 	size = round_page(size);
1422 	object = NULL;
1423 	writecounted = FALSE;
1424 
1425 	/*
1426 	 * Lookup/allocate object.
1427 	 */
1428 	switch (handle_type) {
1429 	case OBJT_DEVICE: {
1430 		struct cdevsw *dsw;
1431 		struct cdev *cdev;
1432 		int ref;
1433 
1434 		cdev = handle;
1435 		dsw = dev_refthread(cdev, &ref);
1436 		if (dsw == NULL)
1437 			return (ENXIO);
1438 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1439 		    dsw, &foff, &object);
1440 		dev_relthread(cdev, ref);
1441 		break;
1442 	}
1443 	case OBJT_VNODE:
1444 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1445 		    handle, &foff, &object, &writecounted);
1446 		break;
1447 	case OBJT_DEFAULT:
1448 		if (handle == NULL) {
1449 			error = 0;
1450 			break;
1451 		}
1452 		/* FALLTHROUGH */
1453 	default:
1454 		error = EINVAL;
1455 		break;
1456 	}
1457 	if (error)
1458 		return (error);
1459 
1460 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1461 	    foff, writecounted, td);
1462 	if (error != 0 && object != NULL) {
1463 		/*
1464 		 * If this mapping was accounted for in the vnode's
1465 		 * writecount, then undo that now.
1466 		 */
1467 		if (writecounted)
1468 			vnode_pager_release_writecount(object, 0, size);
1469 		vm_object_deallocate(object);
1470 	}
1471 	return (error);
1472 }
1473 
1474 /*
1475  * Internal version of mmap that maps a specific VM object into an
1476  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1477  */
1478 int
1479 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1480     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1481     boolean_t writecounted, struct thread *td)
1482 {
1483 	boolean_t curmap, fitit;
1484 	vm_offset_t max_addr;
1485 	int docow, error, findspace, rv;
1486 
1487 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1488 	if (curmap) {
1489 		RACCT_PROC_LOCK(td->td_proc);
1490 		if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1491 			RACCT_PROC_UNLOCK(td->td_proc);
1492 			return (ENOMEM);
1493 		}
1494 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1495 			RACCT_PROC_UNLOCK(td->td_proc);
1496 			return (ENOMEM);
1497 		}
1498 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1499 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1500 			    lim_cur(td, RLIMIT_MEMLOCK)) {
1501 				racct_set_force(td->td_proc, RACCT_VMEM,
1502 				    map->size);
1503 				RACCT_PROC_UNLOCK(td->td_proc);
1504 				return (ENOMEM);
1505 			}
1506 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1507 			    ptoa(pmap_wired_count(map->pmap)) + size);
1508 			if (error != 0) {
1509 				racct_set_force(td->td_proc, RACCT_VMEM,
1510 				    map->size);
1511 				RACCT_PROC_UNLOCK(td->td_proc);
1512 				return (error);
1513 			}
1514 		}
1515 		RACCT_PROC_UNLOCK(td->td_proc);
1516 	}
1517 
1518 	/*
1519 	 * We currently can only deal with page aligned file offsets.
1520 	 * The mmap() system call already enforces this by subtracting
1521 	 * the page offset from the file offset, but checking here
1522 	 * catches errors in device drivers (e.g. d_single_mmap()
1523 	 * callbacks) and other internal mapping requests (such as in
1524 	 * exec).
1525 	 */
1526 	if (foff & PAGE_MASK)
1527 		return (EINVAL);
1528 
1529 	if ((flags & MAP_FIXED) == 0) {
1530 		fitit = TRUE;
1531 		*addr = round_page(*addr);
1532 	} else {
1533 		if (*addr != trunc_page(*addr))
1534 			return (EINVAL);
1535 		fitit = FALSE;
1536 	}
1537 
1538 	if (flags & MAP_ANON) {
1539 		if (object != NULL || foff != 0)
1540 			return (EINVAL);
1541 		docow = 0;
1542 	} else if (flags & MAP_PREFAULT_READ)
1543 		docow = MAP_PREFAULT;
1544 	else
1545 		docow = MAP_PREFAULT_PARTIAL;
1546 
1547 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1548 		docow |= MAP_COPY_ON_WRITE;
1549 	if (flags & MAP_NOSYNC)
1550 		docow |= MAP_DISABLE_SYNCER;
1551 	if (flags & MAP_NOCORE)
1552 		docow |= MAP_DISABLE_COREDUMP;
1553 	/* Shared memory is also shared with children. */
1554 	if (flags & MAP_SHARED)
1555 		docow |= MAP_INHERIT_SHARE;
1556 	if (writecounted)
1557 		docow |= MAP_VN_WRITECOUNT;
1558 	if (flags & MAP_STACK) {
1559 		if (object != NULL)
1560 			return (EINVAL);
1561 		docow |= MAP_STACK_GROWS_DOWN;
1562 	}
1563 	if ((flags & MAP_EXCL) != 0)
1564 		docow |= MAP_CHECK_EXCL;
1565 	if ((flags & MAP_GUARD) != 0)
1566 		docow |= MAP_CREATE_GUARD;
1567 
1568 	if (fitit) {
1569 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1570 			findspace = VMFS_SUPER_SPACE;
1571 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1572 			findspace = VMFS_ALIGNED_SPACE(flags >>
1573 			    MAP_ALIGNMENT_SHIFT);
1574 		else
1575 			findspace = VMFS_OPTIMAL_SPACE;
1576 		max_addr = 0;
1577 #ifdef MAP_32BIT
1578 		if ((flags & MAP_32BIT) != 0)
1579 			max_addr = MAP_32BIT_MAX_ADDR;
1580 #endif
1581 		if (curmap) {
1582 			rv = vm_map_find_min(map, object, foff, addr, size,
1583 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1584 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1585 			    findspace, prot, maxprot, docow);
1586 		} else {
1587 			rv = vm_map_find(map, object, foff, addr, size,
1588 			    max_addr, findspace, prot, maxprot, docow);
1589 		}
1590 	} else {
1591 		rv = vm_map_fixed(map, object, foff, *addr, size,
1592 		    prot, maxprot, docow);
1593 	}
1594 
1595 	if (rv == KERN_SUCCESS) {
1596 		/*
1597 		 * If the process has requested that all future mappings
1598 		 * be wired, then heed this.
1599 		 */
1600 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1601 			vm_map_lock(map);
1602 			if ((map->flags & MAP_WIREFUTURE) != 0)
1603 				(void)vm_map_wire_locked(map, *addr,
1604 				    *addr + size, VM_MAP_WIRE_USER |
1605 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1606 				    VM_MAP_WIRE_NOHOLES));
1607 			vm_map_unlock(map);
1608 		}
1609 	}
1610 	return (vm_mmap_to_errno(rv));
1611 }
1612 
1613 /*
1614  * Translate a Mach VM return code to zero on success or the appropriate errno
1615  * on failure.
1616  */
1617 int
1618 vm_mmap_to_errno(int rv)
1619 {
1620 
1621 	switch (rv) {
1622 	case KERN_SUCCESS:
1623 		return (0);
1624 	case KERN_INVALID_ADDRESS:
1625 	case KERN_NO_SPACE:
1626 		return (ENOMEM);
1627 	case KERN_PROTECTION_FAILURE:
1628 		return (EACCES);
1629 	default:
1630 		return (EINVAL);
1631 	}
1632 }
1633