xref: /freebsd/sys/vm/vm_mmap.c (revision 2c3632d14fe37fa35c262ee9fb66835be0a52621)
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, (uintptr_t)uap->addr, uap->len, uap->prot,
183 	    uap->flags, uap->fd, uap->pos));
184 }
185 
186 int
187 kern_mmap_maxprot(struct proc *p, int prot)
188 {
189 
190 	if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
191 	    (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
192 		return (_PROT_ALL);
193 	if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
194 	    prot != PROT_NONE)
195 		 return (prot);
196 	return (_PROT_ALL);
197 }
198 
199 int
200 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags,
201     int fd, off_t pos)
202 {
203 	struct mmap_req mr = {
204 		.mr_hint = addr0,
205 		.mr_len = len,
206 		.mr_prot = prot,
207 		.mr_flags = flags,
208 		.mr_fd = fd,
209 		.mr_pos = pos
210 	};
211 
212 	return (kern_mmap_req(td, &mr));
213 }
214 
215 int
216 kern_mmap_req(struct thread *td, const struct mmap_req *mrp)
217 {
218 	struct vmspace *vms;
219 	struct file *fp;
220 	struct proc *p;
221 	off_t pos;
222 	vm_offset_t addr;
223 	vm_size_t len, pageoff, size;
224 	vm_prot_t cap_maxprot;
225 	int align, error, fd, flags, max_prot, prot;
226 	cap_rights_t rights;
227 	mmap_check_fp_fn check_fp_fn;
228 
229 	addr  = mrp->mr_hint;
230 	len = mrp->mr_len;
231 	prot = mrp->mr_prot;
232 	flags = mrp->mr_flags;
233 	fd = mrp->mr_fd;
234 	pos = mrp->mr_pos;
235 	check_fp_fn = mrp->mr_check_fp_fn;
236 
237 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
238 		return (EINVAL);
239 	max_prot = PROT_MAX_EXTRACT(prot);
240 	prot = PROT_EXTRACT(prot);
241 	if (max_prot != 0 && (max_prot & prot) != prot)
242 		return (ENOTSUP);
243 
244 	p = td->td_proc;
245 
246 	/*
247 	 * Always honor PROT_MAX if set.  If not, default to all
248 	 * permissions unless we're implying maximum permissions.
249 	 */
250 	if (max_prot == 0)
251 		max_prot = kern_mmap_maxprot(p, prot);
252 
253 	vms = p->p_vmspace;
254 	fp = NULL;
255 	AUDIT_ARG_FD(fd);
256 
257 	/*
258 	 * Ignore old flags that used to be defined but did not do anything.
259 	 */
260 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
261 
262 	/*
263 	 * Enforce the constraints.
264 	 * Mapping of length 0 is only allowed for old binaries.
265 	 * Anonymous mapping shall specify -1 as filedescriptor and
266 	 * zero position for new code. Be nice to ancient a.out
267 	 * binaries and correct pos for anonymous mapping, since old
268 	 * ld.so sometimes issues anonymous map requests with non-zero
269 	 * pos.
270 	 */
271 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
272 		if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
273 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
274 			return (EINVAL);
275 	} else {
276 		if ((flags & MAP_ANON) != 0)
277 			pos = 0;
278 	}
279 
280 	if (flags & MAP_STACK) {
281 		if ((fd != -1) ||
282 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
283 			return (EINVAL);
284 		flags |= MAP_ANON;
285 		pos = 0;
286 	}
287 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
288 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
289 	    MAP_PREFAULT_READ | MAP_GUARD |
290 #ifdef MAP_32BIT
291 	    MAP_32BIT |
292 #endif
293 	    MAP_ALIGNMENT_MASK)) != 0)
294 		return (EINVAL);
295 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
296 		return (EINVAL);
297 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
298 		return (EINVAL);
299 	if (prot != PROT_NONE &&
300 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
301 		return (EINVAL);
302 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
303 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
304 #ifdef MAP_32BIT
305 	    MAP_32BIT |
306 #endif
307 	    MAP_ALIGNMENT_MASK)) != 0))
308 		return (EINVAL);
309 
310 	/*
311 	 * Align the file position to a page boundary,
312 	 * and save its page offset component.
313 	 */
314 	pageoff = (pos & PAGE_MASK);
315 	pos -= pageoff;
316 
317 	/* Compute size from len by rounding (on both ends). */
318 	size = len + pageoff;			/* low end... */
319 	size = round_page(size);		/* hi end */
320 	/* Check for rounding up to zero. */
321 	if (len > size)
322 		return (ENOMEM);
323 
324 	/* Ensure alignment is at least a page and fits in a pointer. */
325 	align = flags & MAP_ALIGNMENT_MASK;
326 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
327 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
328 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
329 		return (EINVAL);
330 
331 	/*
332 	 * Check for illegal addresses.  Watch out for address wrap... Note
333 	 * that VM_*_ADDRESS are not constants due to casts (argh).
334 	 */
335 	if (flags & MAP_FIXED) {
336 		/*
337 		 * The specified address must have the same remainder
338 		 * as the file offset taken modulo PAGE_SIZE, so it
339 		 * should be aligned after adjustment by pageoff.
340 		 */
341 		addr -= pageoff;
342 		if (addr & PAGE_MASK)
343 			return (EINVAL);
344 
345 		/* Address range must be all in user VM space. */
346 		if (!vm_map_range_valid(&vms->vm_map, addr, addr + size))
347 			return (EINVAL);
348 #ifdef MAP_32BIT
349 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
350 			return (EINVAL);
351 	} else if (flags & MAP_32BIT) {
352 		/*
353 		 * For MAP_32BIT, override the hint if it is too high and
354 		 * do not bother moving the mapping past the heap (since
355 		 * the heap is usually above 2GB).
356 		 */
357 		if (addr + size > MAP_32BIT_MAX_ADDR)
358 			addr = 0;
359 #endif
360 	} else {
361 		/*
362 		 * XXX for non-fixed mappings where no hint is provided or
363 		 * the hint would fall in the potential heap space,
364 		 * place it after the end of the largest possible heap.
365 		 *
366 		 * There should really be a pmap call to determine a reasonable
367 		 * location.
368 		 */
369 		if (addr == 0 ||
370 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
371 		    addr < round_page((vm_offset_t)vms->vm_daddr +
372 		    lim_max(td, RLIMIT_DATA))))
373 			addr = round_page((vm_offset_t)vms->vm_daddr +
374 			    lim_max(td, RLIMIT_DATA));
375 	}
376 	if (len == 0) {
377 		/*
378 		 * Return success without mapping anything for old
379 		 * binaries that request a page-aligned mapping of
380 		 * length 0.  For modern binaries, this function
381 		 * returns an error earlier.
382 		 */
383 		error = 0;
384 	} else if ((flags & MAP_GUARD) != 0) {
385 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
386 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
387 	} else if ((flags & MAP_ANON) != 0) {
388 		/*
389 		 * Mapping blank space is trivial.
390 		 *
391 		 * This relies on VM_PROT_* matching PROT_*.
392 		 */
393 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
394 		    max_prot, flags, NULL, pos, FALSE, td);
395 	} else {
396 		/*
397 		 * Mapping file, get fp for validation and don't let the
398 		 * descriptor disappear on us if we block. Check capability
399 		 * rights, but also return the maximum rights to be combined
400 		 * with maxprot later.
401 		 */
402 		cap_rights_init_one(&rights, CAP_MMAP);
403 		if (prot & PROT_READ)
404 			cap_rights_set_one(&rights, CAP_MMAP_R);
405 		if ((flags & MAP_SHARED) != 0) {
406 			if (prot & PROT_WRITE)
407 				cap_rights_set_one(&rights, CAP_MMAP_W);
408 		}
409 		if (prot & PROT_EXEC)
410 			cap_rights_set_one(&rights, CAP_MMAP_X);
411 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
412 		if (error != 0)
413 			goto done;
414 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
415 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
416 			error = EINVAL;
417 			goto done;
418 		}
419 		if (check_fp_fn != NULL) {
420 			error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
421 			    flags);
422 			if (error != 0)
423 				goto done;
424 		}
425 		/* This relies on VM_PROT_* matching PROT_*. */
426 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
427 		    max_prot & cap_maxprot, flags, pos, td);
428 	}
429 
430 	if (error == 0)
431 		td->td_retval[0] = (register_t) (addr + pageoff);
432 done:
433 	if (fp)
434 		fdrop(fp, td);
435 
436 	return (error);
437 }
438 
439 #if defined(COMPAT_FREEBSD6)
440 int
441 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
442 {
443 
444 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
445 	    uap->flags, uap->fd, uap->pos));
446 }
447 #endif
448 
449 #ifdef COMPAT_43
450 #ifndef _SYS_SYSPROTO_H_
451 struct ommap_args {
452 	caddr_t addr;
453 	int len;
454 	int prot;
455 	int flags;
456 	int fd;
457 	long pos;
458 };
459 #endif
460 int
461 ommap(struct thread *td, struct ommap_args *uap)
462 {
463 	static const char cvtbsdprot[8] = {
464 		0,
465 		PROT_EXEC,
466 		PROT_WRITE,
467 		PROT_EXEC | PROT_WRITE,
468 		PROT_READ,
469 		PROT_EXEC | PROT_READ,
470 		PROT_WRITE | PROT_READ,
471 		PROT_EXEC | PROT_WRITE | PROT_READ,
472 	};
473 	int flags, prot;
474 
475 #define	OMAP_ANON	0x0002
476 #define	OMAP_COPY	0x0020
477 #define	OMAP_SHARED	0x0010
478 #define	OMAP_FIXED	0x0100
479 
480 	prot = cvtbsdprot[uap->prot & 0x7];
481 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
482 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
483 	    prot != 0)
484 		prot |= PROT_EXEC;
485 #endif
486 	flags = 0;
487 	if (uap->flags & OMAP_ANON)
488 		flags |= MAP_ANON;
489 	if (uap->flags & OMAP_COPY)
490 		flags |= MAP_COPY;
491 	if (uap->flags & OMAP_SHARED)
492 		flags |= MAP_SHARED;
493 	else
494 		flags |= MAP_PRIVATE;
495 	if (uap->flags & OMAP_FIXED)
496 		flags |= MAP_FIXED;
497 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
498 	    uap->fd, uap->pos));
499 }
500 #endif				/* COMPAT_43 */
501 
502 #ifndef _SYS_SYSPROTO_H_
503 struct msync_args {
504 	void *addr;
505 	size_t len;
506 	int flags;
507 };
508 #endif
509 int
510 sys_msync(struct thread *td, struct msync_args *uap)
511 {
512 
513 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
514 }
515 
516 int
517 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
518 {
519 	vm_offset_t addr;
520 	vm_size_t pageoff;
521 	vm_map_t map;
522 	int rv;
523 
524 	addr = addr0;
525 	pageoff = (addr & PAGE_MASK);
526 	addr -= pageoff;
527 	size += pageoff;
528 	size = (vm_size_t) round_page(size);
529 	if (addr + size < addr)
530 		return (EINVAL);
531 
532 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
533 		return (EINVAL);
534 
535 	map = &td->td_proc->p_vmspace->vm_map;
536 
537 	/*
538 	 * Clean the pages and interpret the return value.
539 	 */
540 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
541 	    (flags & MS_INVALIDATE) != 0);
542 	switch (rv) {
543 	case KERN_SUCCESS:
544 		return (0);
545 	case KERN_INVALID_ADDRESS:
546 		return (ENOMEM);
547 	case KERN_INVALID_ARGUMENT:
548 		return (EBUSY);
549 	case KERN_FAILURE:
550 		return (EIO);
551 	default:
552 		return (EINVAL);
553 	}
554 }
555 
556 #ifndef _SYS_SYSPROTO_H_
557 struct munmap_args {
558 	void *addr;
559 	size_t len;
560 };
561 #endif
562 int
563 sys_munmap(struct thread *td, struct munmap_args *uap)
564 {
565 
566 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
567 }
568 
569 int
570 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
571 {
572 #ifdef HWPMC_HOOKS
573 	struct pmckern_map_out pkm;
574 	vm_map_entry_t entry;
575 	bool pmc_handled;
576 #endif
577 	vm_offset_t addr, end;
578 	vm_size_t pageoff;
579 	vm_map_t map;
580 
581 	if (size == 0)
582 		return (EINVAL);
583 
584 	addr = addr0;
585 	pageoff = (addr & PAGE_MASK);
586 	addr -= pageoff;
587 	size += pageoff;
588 	size = (vm_size_t) round_page(size);
589 	end = addr + size;
590 	map = &td->td_proc->p_vmspace->vm_map;
591 	if (!vm_map_range_valid(map, addr, end))
592 		return (EINVAL);
593 
594 	vm_map_lock(map);
595 #ifdef HWPMC_HOOKS
596 	pmc_handled = false;
597 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
598 		pmc_handled = true;
599 		/*
600 		 * Inform hwpmc if the address range being unmapped contains
601 		 * an executable region.
602 		 */
603 		pkm.pm_address = (uintptr_t) NULL;
604 		if (vm_map_lookup_entry(map, addr, &entry)) {
605 			for (; entry->start < end;
606 			    entry = vm_map_entry_succ(entry)) {
607 				if (vm_map_check_protection(map, entry->start,
608 					entry->end, VM_PROT_EXECUTE) == TRUE) {
609 					pkm.pm_address = (uintptr_t) addr;
610 					pkm.pm_size = (size_t) size;
611 					break;
612 				}
613 			}
614 		}
615 	}
616 #endif
617 	vm_map_delete(map, addr, end);
618 
619 #ifdef HWPMC_HOOKS
620 	if (__predict_false(pmc_handled)) {
621 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
622 		vm_map_lock_downgrade(map);
623 		if (pkm.pm_address != (uintptr_t) NULL)
624 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
625 		vm_map_unlock_read(map);
626 	} else
627 #endif
628 		vm_map_unlock(map);
629 
630 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
631 	return (0);
632 }
633 
634 #ifndef _SYS_SYSPROTO_H_
635 struct mprotect_args {
636 	const void *addr;
637 	size_t len;
638 	int prot;
639 };
640 #endif
641 int
642 sys_mprotect(struct thread *td, struct mprotect_args *uap)
643 {
644 
645 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
646 }
647 
648 int
649 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
650 {
651 	vm_offset_t addr;
652 	vm_size_t pageoff;
653 	int vm_error, max_prot;
654 
655 	addr = addr0;
656 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
657 		return (EINVAL);
658 	max_prot = PROT_MAX_EXTRACT(prot);
659 	prot = PROT_EXTRACT(prot);
660 	pageoff = (addr & PAGE_MASK);
661 	addr -= pageoff;
662 	size += pageoff;
663 	size = (vm_size_t) round_page(size);
664 #ifdef COMPAT_FREEBSD32
665 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
666 		if (((addr + size) & 0xffffffff) < addr)
667 			return (EINVAL);
668 	} else
669 #endif
670 	if (addr + size < addr)
671 		return (EINVAL);
672 
673 	vm_error = KERN_SUCCESS;
674 	if (max_prot != 0) {
675 		if ((max_prot & prot) != prot)
676 			return (ENOTSUP);
677 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
678 		    addr, addr + size, max_prot, TRUE);
679 	}
680 	if (vm_error == KERN_SUCCESS)
681 		vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
682 		    addr, addr + size, prot, FALSE);
683 
684 	switch (vm_error) {
685 	case KERN_SUCCESS:
686 		return (0);
687 	case KERN_PROTECTION_FAILURE:
688 		return (EACCES);
689 	case KERN_RESOURCE_SHORTAGE:
690 		return (ENOMEM);
691 	}
692 	return (EINVAL);
693 }
694 
695 #ifndef _SYS_SYSPROTO_H_
696 struct minherit_args {
697 	void *addr;
698 	size_t len;
699 	int inherit;
700 };
701 #endif
702 int
703 sys_minherit(struct thread *td, struct minherit_args *uap)
704 {
705 
706 	return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
707 	    uap->inherit));
708 }
709 
710 int
711 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
712 {
713 	vm_offset_t addr;
714 	vm_size_t size, pageoff;
715 	vm_inherit_t inherit;
716 
717 	addr = (vm_offset_t)addr0;
718 	size = len;
719 	inherit = inherit0;
720 
721 	pageoff = (addr & PAGE_MASK);
722 	addr -= pageoff;
723 	size += pageoff;
724 	size = (vm_size_t) round_page(size);
725 	if (addr + size < addr)
726 		return (EINVAL);
727 
728 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
729 	    addr + size, inherit)) {
730 	case KERN_SUCCESS:
731 		return (0);
732 	case KERN_PROTECTION_FAILURE:
733 		return (EACCES);
734 	}
735 	return (EINVAL);
736 }
737 
738 #ifndef _SYS_SYSPROTO_H_
739 struct madvise_args {
740 	void *addr;
741 	size_t len;
742 	int behav;
743 };
744 #endif
745 
746 int
747 sys_madvise(struct thread *td, struct madvise_args *uap)
748 {
749 
750 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
751 }
752 
753 int
754 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
755 {
756 	vm_map_t map;
757 	vm_offset_t addr, end, start;
758 	int flags;
759 
760 	/*
761 	 * Check for our special case, advising the swap pager we are
762 	 * "immortal."
763 	 */
764 	if (behav == MADV_PROTECT) {
765 		flags = PPROT_SET;
766 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
767 		    PROC_SPROTECT, &flags));
768 	}
769 
770 	/*
771 	 * Check for illegal addresses.  Watch out for address wrap... Note
772 	 * that VM_*_ADDRESS are not constants due to casts (argh).
773 	 */
774 	map = &td->td_proc->p_vmspace->vm_map;
775 	addr = addr0;
776 	if (!vm_map_range_valid(map, addr, addr + len))
777 		return (EINVAL);
778 
779 	/*
780 	 * Since this routine is only advisory, we default to conservative
781 	 * behavior.
782 	 */
783 	start = trunc_page(addr);
784 	end = round_page(addr + len);
785 
786 	/*
787 	 * vm_map_madvise() checks for illegal values of behav.
788 	 */
789 	return (vm_map_madvise(map, start, end, behav));
790 }
791 
792 #ifndef _SYS_SYSPROTO_H_
793 struct mincore_args {
794 	const void *addr;
795 	size_t len;
796 	char *vec;
797 };
798 #endif
799 
800 int
801 sys_mincore(struct thread *td, struct mincore_args *uap)
802 {
803 
804 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
805 }
806 
807 int
808 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
809 {
810 	pmap_t pmap;
811 	vm_map_t map;
812 	vm_map_entry_t current, entry;
813 	vm_object_t object;
814 	vm_offset_t addr, cend, end, first_addr;
815 	vm_paddr_t pa;
816 	vm_page_t m;
817 	vm_pindex_t pindex;
818 	int error, lastvecindex, mincoreinfo, vecindex;
819 	unsigned int timestamp;
820 
821 	/*
822 	 * Make sure that the addresses presented are valid for user
823 	 * mode.
824 	 */
825 	first_addr = addr = trunc_page(addr0);
826 	end = round_page(addr0 + len);
827 	map = &td->td_proc->p_vmspace->vm_map;
828 	if (end > vm_map_max(map) || end < addr)
829 		return (ENOMEM);
830 
831 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
832 
833 	vm_map_lock_read(map);
834 RestartScan:
835 	timestamp = map->timestamp;
836 
837 	if (!vm_map_lookup_entry(map, addr, &entry)) {
838 		vm_map_unlock_read(map);
839 		return (ENOMEM);
840 	}
841 
842 	/*
843 	 * Do this on a map entry basis so that if the pages are not
844 	 * in the current processes address space, we can easily look
845 	 * up the pages elsewhere.
846 	 */
847 	lastvecindex = -1;
848 	while (entry->start < end) {
849 		/*
850 		 * check for contiguity
851 		 */
852 		current = entry;
853 		entry = vm_map_entry_succ(current);
854 		if (current->end < end &&
855 		    entry->start > current->end) {
856 			vm_map_unlock_read(map);
857 			return (ENOMEM);
858 		}
859 
860 		/*
861 		 * ignore submaps (for now) or null objects
862 		 */
863 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
864 		    current->object.vm_object == NULL)
865 			continue;
866 
867 		/*
868 		 * limit this scan to the current map entry and the
869 		 * limits for the mincore call
870 		 */
871 		if (addr < current->start)
872 			addr = current->start;
873 		cend = current->end;
874 		if (cend > end)
875 			cend = end;
876 
877 		for (; addr < cend; addr += PAGE_SIZE) {
878 			/*
879 			 * Check pmap first, it is likely faster, also
880 			 * it can provide info as to whether we are the
881 			 * one referencing or modifying the page.
882 			 */
883 			m = NULL;
884 			object = NULL;
885 retry:
886 			pa = 0;
887 			mincoreinfo = pmap_mincore(pmap, addr, &pa);
888 			if (mincore_mapped) {
889 				/*
890 				 * We only care about this pmap's
891 				 * mapping of the page, if any.
892 				 */
893 				;
894 			} else if (pa != 0) {
895 				/*
896 				 * The page is mapped by this process but not
897 				 * both accessed and modified.  It is also
898 				 * managed.  Acquire the object lock so that
899 				 * other mappings might be examined.  The page's
900 				 * identity may change at any point before its
901 				 * object lock is acquired, so re-validate if
902 				 * necessary.
903 				 */
904 				m = PHYS_TO_VM_PAGE(pa);
905 				while (object == NULL || m->object != object) {
906 					if (object != NULL)
907 						VM_OBJECT_WUNLOCK(object);
908 					object = atomic_load_ptr(&m->object);
909 					if (object == NULL)
910 						goto retry;
911 					VM_OBJECT_WLOCK(object);
912 				}
913 				if (pa != pmap_extract(pmap, addr))
914 					goto retry;
915 				KASSERT(vm_page_all_valid(m),
916 				    ("mincore: page %p is mapped but invalid",
917 				    m));
918 			} else if (mincoreinfo == 0) {
919 				/*
920 				 * The page is not mapped by this process.  If
921 				 * the object implements managed pages, then
922 				 * determine if the page is resident so that
923 				 * the mappings might be examined.
924 				 */
925 				if (current->object.vm_object != object) {
926 					if (object != NULL)
927 						VM_OBJECT_WUNLOCK(object);
928 					object = current->object.vm_object;
929 					VM_OBJECT_WLOCK(object);
930 				}
931 				if (object->type == OBJT_DEFAULT ||
932 				    object->type == OBJT_SWAP ||
933 				    object->type == OBJT_VNODE) {
934 					pindex = OFF_TO_IDX(current->offset +
935 					    (addr - current->start));
936 					m = vm_page_lookup(object, pindex);
937 					if (m != NULL && vm_page_none_valid(m))
938 						m = NULL;
939 					if (m != NULL)
940 						mincoreinfo = MINCORE_INCORE;
941 				}
942 			}
943 			if (m != NULL) {
944 				VM_OBJECT_ASSERT_WLOCKED(m->object);
945 
946 				/* Examine other mappings of the page. */
947 				if (m->dirty == 0 && pmap_is_modified(m))
948 					vm_page_dirty(m);
949 				if (m->dirty != 0)
950 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
951 
952 				/*
953 				 * The first test for PGA_REFERENCED is an
954 				 * optimization.  The second test is
955 				 * required because a concurrent pmap
956 				 * operation could clear the last reference
957 				 * and set PGA_REFERENCED before the call to
958 				 * pmap_is_referenced().
959 				 */
960 				if ((m->a.flags & PGA_REFERENCED) != 0 ||
961 				    pmap_is_referenced(m) ||
962 				    (m->a.flags & PGA_REFERENCED) != 0)
963 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
964 			}
965 			if (object != NULL)
966 				VM_OBJECT_WUNLOCK(object);
967 
968 			/*
969 			 * subyte may page fault.  In case it needs to modify
970 			 * the map, we release the lock.
971 			 */
972 			vm_map_unlock_read(map);
973 
974 			/*
975 			 * calculate index into user supplied byte vector
976 			 */
977 			vecindex = atop(addr - first_addr);
978 
979 			/*
980 			 * If we have skipped map entries, we need to make sure that
981 			 * the byte vector is zeroed for those skipped entries.
982 			 */
983 			while ((lastvecindex + 1) < vecindex) {
984 				++lastvecindex;
985 				error = subyte(vec + lastvecindex, 0);
986 				if (error) {
987 					error = EFAULT;
988 					goto done2;
989 				}
990 			}
991 
992 			/*
993 			 * Pass the page information to the user
994 			 */
995 			error = subyte(vec + vecindex, mincoreinfo);
996 			if (error) {
997 				error = EFAULT;
998 				goto done2;
999 			}
1000 
1001 			/*
1002 			 * If the map has changed, due to the subyte, the previous
1003 			 * output may be invalid.
1004 			 */
1005 			vm_map_lock_read(map);
1006 			if (timestamp != map->timestamp)
1007 				goto RestartScan;
1008 
1009 			lastvecindex = vecindex;
1010 		}
1011 	}
1012 
1013 	/*
1014 	 * subyte may page fault.  In case it needs to modify
1015 	 * the map, we release the lock.
1016 	 */
1017 	vm_map_unlock_read(map);
1018 
1019 	/*
1020 	 * Zero the last entries in the byte vector.
1021 	 */
1022 	vecindex = atop(end - first_addr);
1023 	while ((lastvecindex + 1) < vecindex) {
1024 		++lastvecindex;
1025 		error = subyte(vec + lastvecindex, 0);
1026 		if (error) {
1027 			error = EFAULT;
1028 			goto done2;
1029 		}
1030 	}
1031 
1032 	/*
1033 	 * If the map has changed, due to the subyte, the previous
1034 	 * output may be invalid.
1035 	 */
1036 	vm_map_lock_read(map);
1037 	if (timestamp != map->timestamp)
1038 		goto RestartScan;
1039 	vm_map_unlock_read(map);
1040 done2:
1041 	return (error);
1042 }
1043 
1044 #ifndef _SYS_SYSPROTO_H_
1045 struct mlock_args {
1046 	const void *addr;
1047 	size_t len;
1048 };
1049 #endif
1050 int
1051 sys_mlock(struct thread *td, struct mlock_args *uap)
1052 {
1053 
1054 	return (kern_mlock(td->td_proc, td->td_ucred,
1055 	    __DECONST(uintptr_t, uap->addr), uap->len));
1056 }
1057 
1058 int
1059 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1060 {
1061 	vm_offset_t addr, end, last, start;
1062 	vm_size_t npages, size;
1063 	vm_map_t map;
1064 	unsigned long nsize;
1065 	int error;
1066 
1067 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1068 	if (error)
1069 		return (error);
1070 	addr = addr0;
1071 	size = len;
1072 	last = addr + size;
1073 	start = trunc_page(addr);
1074 	end = round_page(last);
1075 	if (last < addr || end < addr)
1076 		return (EINVAL);
1077 	npages = atop(end - start);
1078 	if (npages > vm_page_max_user_wired)
1079 		return (ENOMEM);
1080 	map = &proc->p_vmspace->vm_map;
1081 	PROC_LOCK(proc);
1082 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1083 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1084 		PROC_UNLOCK(proc);
1085 		return (ENOMEM);
1086 	}
1087 	PROC_UNLOCK(proc);
1088 #ifdef RACCT
1089 	if (racct_enable) {
1090 		PROC_LOCK(proc);
1091 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1092 		PROC_UNLOCK(proc);
1093 		if (error != 0)
1094 			return (ENOMEM);
1095 	}
1096 #endif
1097 	error = vm_map_wire(map, start, end,
1098 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1099 #ifdef RACCT
1100 	if (racct_enable && error != KERN_SUCCESS) {
1101 		PROC_LOCK(proc);
1102 		racct_set(proc, RACCT_MEMLOCK,
1103 		    ptoa(pmap_wired_count(map->pmap)));
1104 		PROC_UNLOCK(proc);
1105 	}
1106 #endif
1107 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1108 }
1109 
1110 #ifndef _SYS_SYSPROTO_H_
1111 struct mlockall_args {
1112 	int	how;
1113 };
1114 #endif
1115 
1116 int
1117 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1118 {
1119 	vm_map_t map;
1120 	int error;
1121 
1122 	map = &td->td_proc->p_vmspace->vm_map;
1123 	error = priv_check(td, PRIV_VM_MLOCK);
1124 	if (error)
1125 		return (error);
1126 
1127 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1128 		return (EINVAL);
1129 
1130 	/*
1131 	 * If wiring all pages in the process would cause it to exceed
1132 	 * a hard resource limit, return ENOMEM.
1133 	 */
1134 	if (!old_mlock && uap->how & MCL_CURRENT) {
1135 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1136 			return (ENOMEM);
1137 	}
1138 #ifdef RACCT
1139 	if (racct_enable) {
1140 		PROC_LOCK(td->td_proc);
1141 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1142 		PROC_UNLOCK(td->td_proc);
1143 		if (error != 0)
1144 			return (ENOMEM);
1145 	}
1146 #endif
1147 
1148 	if (uap->how & MCL_FUTURE) {
1149 		vm_map_lock(map);
1150 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1151 		vm_map_unlock(map);
1152 		error = 0;
1153 	}
1154 
1155 	if (uap->how & MCL_CURRENT) {
1156 		/*
1157 		 * P1003.1-2001 mandates that all currently mapped pages
1158 		 * will be memory resident and locked (wired) upon return
1159 		 * from mlockall(). vm_map_wire() will wire pages, by
1160 		 * calling vm_fault_wire() for each page in the region.
1161 		 */
1162 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1163 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1164 		if (error == KERN_SUCCESS)
1165 			error = 0;
1166 		else if (error == KERN_RESOURCE_SHORTAGE)
1167 			error = ENOMEM;
1168 		else
1169 			error = EAGAIN;
1170 	}
1171 #ifdef RACCT
1172 	if (racct_enable && error != KERN_SUCCESS) {
1173 		PROC_LOCK(td->td_proc);
1174 		racct_set(td->td_proc, RACCT_MEMLOCK,
1175 		    ptoa(pmap_wired_count(map->pmap)));
1176 		PROC_UNLOCK(td->td_proc);
1177 	}
1178 #endif
1179 
1180 	return (error);
1181 }
1182 
1183 #ifndef _SYS_SYSPROTO_H_
1184 struct munlockall_args {
1185 	register_t dummy;
1186 };
1187 #endif
1188 
1189 int
1190 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1191 {
1192 	vm_map_t map;
1193 	int error;
1194 
1195 	map = &td->td_proc->p_vmspace->vm_map;
1196 	error = priv_check(td, PRIV_VM_MUNLOCK);
1197 	if (error)
1198 		return (error);
1199 
1200 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1201 	vm_map_lock(map);
1202 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1203 	vm_map_unlock(map);
1204 
1205 	/* Forcibly unwire all pages. */
1206 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1207 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1208 #ifdef RACCT
1209 	if (racct_enable && error == KERN_SUCCESS) {
1210 		PROC_LOCK(td->td_proc);
1211 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1212 		PROC_UNLOCK(td->td_proc);
1213 	}
1214 #endif
1215 
1216 	return (error);
1217 }
1218 
1219 #ifndef _SYS_SYSPROTO_H_
1220 struct munlock_args {
1221 	const void *addr;
1222 	size_t len;
1223 };
1224 #endif
1225 int
1226 sys_munlock(struct thread *td, struct munlock_args *uap)
1227 {
1228 
1229 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1230 }
1231 
1232 int
1233 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1234 {
1235 	vm_offset_t addr, end, last, start;
1236 #ifdef RACCT
1237 	vm_map_t map;
1238 #endif
1239 	int error;
1240 
1241 	error = priv_check(td, PRIV_VM_MUNLOCK);
1242 	if (error)
1243 		return (error);
1244 	addr = addr0;
1245 	last = addr + size;
1246 	start = trunc_page(addr);
1247 	end = round_page(last);
1248 	if (last < addr || end < addr)
1249 		return (EINVAL);
1250 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1251 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1252 #ifdef RACCT
1253 	if (racct_enable && error == KERN_SUCCESS) {
1254 		PROC_LOCK(td->td_proc);
1255 		map = &td->td_proc->p_vmspace->vm_map;
1256 		racct_set(td->td_proc, RACCT_MEMLOCK,
1257 		    ptoa(pmap_wired_count(map->pmap)));
1258 		PROC_UNLOCK(td->td_proc);
1259 	}
1260 #endif
1261 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1262 }
1263 
1264 /*
1265  * vm_mmap_vnode()
1266  *
1267  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1268  * operations on vnodes.
1269  */
1270 int
1271 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1272     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1273     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1274     boolean_t *writecounted)
1275 {
1276 	struct vattr va;
1277 	vm_object_t obj;
1278 	vm_ooffset_t foff;
1279 	struct ucred *cred;
1280 	int error, flags;
1281 	bool writex;
1282 
1283 	cred = td->td_ucred;
1284 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1285 	    (*flagsp & MAP_SHARED) != 0;
1286 	if ((error = vget(vp, LK_SHARED)) != 0)
1287 		return (error);
1288 	AUDIT_ARG_VNODE1(vp);
1289 	foff = *foffp;
1290 	flags = *flagsp;
1291 	obj = vp->v_object;
1292 	if (vp->v_type == VREG) {
1293 		/*
1294 		 * Get the proper underlying object
1295 		 */
1296 		if (obj == NULL) {
1297 			error = EINVAL;
1298 			goto done;
1299 		}
1300 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1301 			vput(vp);
1302 			vp = (struct vnode *)obj->handle;
1303 			/*
1304 			 * Bypass filesystems obey the mpsafety of the
1305 			 * underlying fs.  Tmpfs never bypasses.
1306 			 */
1307 			error = vget(vp, LK_SHARED);
1308 			if (error != 0)
1309 				return (error);
1310 		}
1311 		if (writex) {
1312 			*writecounted = TRUE;
1313 			vm_pager_update_writecount(obj, 0, objsize);
1314 		}
1315 	} else {
1316 		error = EINVAL;
1317 		goto done;
1318 	}
1319 	if ((error = VOP_GETATTR(vp, &va, cred)))
1320 		goto done;
1321 #ifdef MAC
1322 	/* This relies on VM_PROT_* matching PROT_*. */
1323 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1324 	if (error != 0)
1325 		goto done;
1326 #endif
1327 	if ((flags & MAP_SHARED) != 0) {
1328 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1329 			if (prot & VM_PROT_WRITE) {
1330 				error = EPERM;
1331 				goto done;
1332 			}
1333 			*maxprotp &= ~VM_PROT_WRITE;
1334 		}
1335 	}
1336 	/*
1337 	 * If it is a regular file without any references
1338 	 * we do not need to sync it.
1339 	 * Adjust object size to be the size of actual file.
1340 	 */
1341 	objsize = round_page(va.va_size);
1342 	if (va.va_nlink == 0)
1343 		flags |= MAP_NOSYNC;
1344 	if (obj->type == OBJT_VNODE) {
1345 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1346 		    cred);
1347 		if (obj == NULL) {
1348 			error = ENOMEM;
1349 			goto done;
1350 		}
1351 	} else {
1352 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1353 		    ("wrong object type"));
1354 		vm_object_reference(obj);
1355 #if VM_NRESERVLEVEL > 0
1356 		if ((obj->flags & OBJ_COLORED) == 0) {
1357 			VM_OBJECT_WLOCK(obj);
1358 			vm_object_color(obj, 0);
1359 			VM_OBJECT_WUNLOCK(obj);
1360 		}
1361 #endif
1362 	}
1363 	*objp = obj;
1364 	*flagsp = flags;
1365 
1366 	VOP_MMAPPED(vp);
1367 
1368 done:
1369 	if (error != 0 && *writecounted) {
1370 		*writecounted = FALSE;
1371 		vm_pager_update_writecount(obj, objsize, 0);
1372 	}
1373 	vput(vp);
1374 	return (error);
1375 }
1376 
1377 /*
1378  * vm_mmap_cdev()
1379  *
1380  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1381  * operations on cdevs.
1382  */
1383 int
1384 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1385     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1386     vm_ooffset_t *foff, vm_object_t *objp)
1387 {
1388 	vm_object_t obj;
1389 	int error, flags;
1390 
1391 	flags = *flagsp;
1392 
1393 	if (dsw->d_flags & D_MMAP_ANON) {
1394 		*objp = NULL;
1395 		*foff = 0;
1396 		*maxprotp = VM_PROT_ALL;
1397 		*flagsp |= MAP_ANON;
1398 		return (0);
1399 	}
1400 	/*
1401 	 * cdevs do not provide private mappings of any kind.
1402 	 */
1403 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1404 	    (prot & VM_PROT_WRITE) != 0)
1405 		return (EACCES);
1406 	if (flags & (MAP_PRIVATE|MAP_COPY))
1407 		return (EINVAL);
1408 	/*
1409 	 * Force device mappings to be shared.
1410 	 */
1411 	flags |= MAP_SHARED;
1412 #ifdef MAC_XXX
1413 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1414 	if (error != 0)
1415 		return (error);
1416 #endif
1417 	/*
1418 	 * First, try d_mmap_single().  If that is not implemented
1419 	 * (returns ENODEV), fall back to using the device pager.
1420 	 * Note that d_mmap_single() must return a reference to the
1421 	 * object (it needs to bump the reference count of the object
1422 	 * it returns somehow).
1423 	 *
1424 	 * XXX assumes VM_PROT_* == PROT_*
1425 	 */
1426 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1427 	if (error != ENODEV)
1428 		return (error);
1429 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1430 	    td->td_ucred);
1431 	if (obj == NULL)
1432 		return (EINVAL);
1433 	*objp = obj;
1434 	*flagsp = flags;
1435 	return (0);
1436 }
1437 
1438 /*
1439  * vm_mmap()
1440  *
1441  * Internal version of mmap used by exec, sys5 shared memory, and
1442  * various device drivers.  Handle is either a vnode pointer, a
1443  * character device, or NULL for MAP_ANON.
1444  */
1445 int
1446 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1447 	vm_prot_t maxprot, int flags,
1448 	objtype_t handle_type, void *handle,
1449 	vm_ooffset_t foff)
1450 {
1451 	vm_object_t object;
1452 	struct thread *td = curthread;
1453 	int error;
1454 	boolean_t writecounted;
1455 
1456 	if (size == 0)
1457 		return (EINVAL);
1458 
1459 	size = round_page(size);
1460 	object = NULL;
1461 	writecounted = FALSE;
1462 
1463 	/*
1464 	 * Lookup/allocate object.
1465 	 */
1466 	switch (handle_type) {
1467 	case OBJT_DEVICE: {
1468 		struct cdevsw *dsw;
1469 		struct cdev *cdev;
1470 		int ref;
1471 
1472 		cdev = handle;
1473 		dsw = dev_refthread(cdev, &ref);
1474 		if (dsw == NULL)
1475 			return (ENXIO);
1476 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1477 		    dsw, &foff, &object);
1478 		dev_relthread(cdev, ref);
1479 		break;
1480 	}
1481 	case OBJT_VNODE:
1482 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1483 		    handle, &foff, &object, &writecounted);
1484 		break;
1485 	case OBJT_DEFAULT:
1486 		if (handle == NULL) {
1487 			error = 0;
1488 			break;
1489 		}
1490 		/* FALLTHROUGH */
1491 	default:
1492 		error = EINVAL;
1493 		break;
1494 	}
1495 	if (error)
1496 		return (error);
1497 
1498 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1499 	    foff, writecounted, td);
1500 	if (error != 0 && object != NULL) {
1501 		/*
1502 		 * If this mapping was accounted for in the vnode's
1503 		 * writecount, then undo that now.
1504 		 */
1505 		if (writecounted)
1506 			vm_pager_release_writecount(object, 0, size);
1507 		vm_object_deallocate(object);
1508 	}
1509 	return (error);
1510 }
1511 
1512 /*
1513  * Internal version of mmap that maps a specific VM object into an
1514  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1515  */
1516 int
1517 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1518     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1519     boolean_t writecounted, struct thread *td)
1520 {
1521 	boolean_t curmap, fitit;
1522 	vm_offset_t max_addr;
1523 	int docow, error, findspace, rv;
1524 
1525 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1526 	if (curmap) {
1527 		RACCT_PROC_LOCK(td->td_proc);
1528 		if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1529 			RACCT_PROC_UNLOCK(td->td_proc);
1530 			return (ENOMEM);
1531 		}
1532 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1533 			RACCT_PROC_UNLOCK(td->td_proc);
1534 			return (ENOMEM);
1535 		}
1536 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1537 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1538 			    lim_cur(td, RLIMIT_MEMLOCK)) {
1539 				racct_set_force(td->td_proc, RACCT_VMEM,
1540 				    map->size);
1541 				RACCT_PROC_UNLOCK(td->td_proc);
1542 				return (ENOMEM);
1543 			}
1544 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1545 			    ptoa(pmap_wired_count(map->pmap)) + size);
1546 			if (error != 0) {
1547 				racct_set_force(td->td_proc, RACCT_VMEM,
1548 				    map->size);
1549 				RACCT_PROC_UNLOCK(td->td_proc);
1550 				return (error);
1551 			}
1552 		}
1553 		RACCT_PROC_UNLOCK(td->td_proc);
1554 	}
1555 
1556 	/*
1557 	 * We currently can only deal with page aligned file offsets.
1558 	 * The mmap() system call already enforces this by subtracting
1559 	 * the page offset from the file offset, but checking here
1560 	 * catches errors in device drivers (e.g. d_single_mmap()
1561 	 * callbacks) and other internal mapping requests (such as in
1562 	 * exec).
1563 	 */
1564 	if (foff & PAGE_MASK)
1565 		return (EINVAL);
1566 
1567 	if ((flags & MAP_FIXED) == 0) {
1568 		fitit = TRUE;
1569 		*addr = round_page(*addr);
1570 	} else {
1571 		if (*addr != trunc_page(*addr))
1572 			return (EINVAL);
1573 		fitit = FALSE;
1574 	}
1575 
1576 	if (flags & MAP_ANON) {
1577 		if (object != NULL || foff != 0)
1578 			return (EINVAL);
1579 		docow = 0;
1580 	} else if (flags & MAP_PREFAULT_READ)
1581 		docow = MAP_PREFAULT;
1582 	else
1583 		docow = MAP_PREFAULT_PARTIAL;
1584 
1585 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1586 		docow |= MAP_COPY_ON_WRITE;
1587 	if (flags & MAP_NOSYNC)
1588 		docow |= MAP_DISABLE_SYNCER;
1589 	if (flags & MAP_NOCORE)
1590 		docow |= MAP_DISABLE_COREDUMP;
1591 	/* Shared memory is also shared with children. */
1592 	if (flags & MAP_SHARED)
1593 		docow |= MAP_INHERIT_SHARE;
1594 	if (writecounted)
1595 		docow |= MAP_WRITECOUNT;
1596 	if (flags & MAP_STACK) {
1597 		if (object != NULL)
1598 			return (EINVAL);
1599 		docow |= MAP_STACK_GROWS_DOWN;
1600 	}
1601 	if ((flags & MAP_EXCL) != 0)
1602 		docow |= MAP_CHECK_EXCL;
1603 	if ((flags & MAP_GUARD) != 0)
1604 		docow |= MAP_CREATE_GUARD;
1605 
1606 	if (fitit) {
1607 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1608 			findspace = VMFS_SUPER_SPACE;
1609 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1610 			findspace = VMFS_ALIGNED_SPACE(flags >>
1611 			    MAP_ALIGNMENT_SHIFT);
1612 		else
1613 			findspace = VMFS_OPTIMAL_SPACE;
1614 		max_addr = 0;
1615 #ifdef MAP_32BIT
1616 		if ((flags & MAP_32BIT) != 0)
1617 			max_addr = MAP_32BIT_MAX_ADDR;
1618 #endif
1619 		if (curmap) {
1620 			rv = vm_map_find_min(map, object, foff, addr, size,
1621 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1622 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1623 			    findspace, prot, maxprot, docow);
1624 		} else {
1625 			rv = vm_map_find(map, object, foff, addr, size,
1626 			    max_addr, findspace, prot, maxprot, docow);
1627 		}
1628 	} else {
1629 		rv = vm_map_fixed(map, object, foff, *addr, size,
1630 		    prot, maxprot, docow);
1631 	}
1632 
1633 	if (rv == KERN_SUCCESS) {
1634 		/*
1635 		 * If the process has requested that all future mappings
1636 		 * be wired, then heed this.
1637 		 */
1638 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1639 			vm_map_lock(map);
1640 			if ((map->flags & MAP_WIREFUTURE) != 0)
1641 				(void)vm_map_wire_locked(map, *addr,
1642 				    *addr + size, VM_MAP_WIRE_USER |
1643 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1644 				    VM_MAP_WIRE_NOHOLES));
1645 			vm_map_unlock(map);
1646 		}
1647 	}
1648 	return (vm_mmap_to_errno(rv));
1649 }
1650 
1651 /*
1652  * Translate a Mach VM return code to zero on success or the appropriate errno
1653  * on failure.
1654  */
1655 int
1656 vm_mmap_to_errno(int rv)
1657 {
1658 
1659 	switch (rv) {
1660 	case KERN_SUCCESS:
1661 		return (0);
1662 	case KERN_INVALID_ADDRESS:
1663 	case KERN_NO_SPACE:
1664 		return (ENOMEM);
1665 	case KERN_PROTECTION_FAILURE:
1666 		return (EACCES);
1667 	default:
1668 		return (EINVAL);
1669 	}
1670 }
1671