xref: /freebsd/sys/vm/vm_mmap.c (revision 7899f917b1c0ea178f1d2be0cfb452086d079d23)
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 
39 /*
40  * Mapped file (mmap) interface to VM
41  */
42 
43 #include <sys/cdefs.h>
44 #include "opt_hwpmc_hooks.h"
45 #include "opt_vm.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/capsicum.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/sysproto.h>
54 #include <sys/elf.h>
55 #include <sys/filedesc.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/procctl.h>
59 #include <sys/racct.h>
60 #include <sys/resource.h>
61 #include <sys/resourcevar.h>
62 #include <sys/rwlock.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 #include <sys/fcntl.h>
66 #include <sys/file.h>
67 #include <sys/mman.h>
68 #include <sys/mount.h>
69 #include <sys/conf.h>
70 #include <sys/stat.h>
71 #include <sys/syscallsubr.h>
72 #include <sys/sysent.h>
73 #include <sys/vmmeter.h>
74 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
75 #include <machine/md_var.h>
76 #endif
77 
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_extern.h>
90 #include <vm/vm_page.h>
91 #include <vm/vnode_pager.h>
92 
93 #ifdef HWPMC_HOOKS
94 #include <sys/pmckern.h>
95 #endif
96 
97 int old_mlock = 0;
98 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
99     "Do not apply RLIMIT_MEMLOCK on mlockall");
100 static int mincore_mapped = 1;
101 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
102     "mincore reports mappings, not residency");
103 static int imply_prot_max = 0;
104 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0,
105     "Imply maximum page protections in mmap() when none are specified");
106 
107 _Static_assert(MAXPAGESIZES <= 4, "MINCORE_SUPER too narrow");
108 
109 #if defined(COMPAT_43)
110 int
111 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
112 {
113 
114 	td->td_retval[0] = PAGE_SIZE;
115 	return (0);
116 }
117 #endif				/* COMPAT_43 */
118 
119 /*
120  * Memory Map (mmap) system call.  Note that the file offset
121  * and address are allowed to be NOT page aligned, though if
122  * the MAP_FIXED flag it set, both must have the same remainder
123  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
124  * page-aligned, the actual mapping starts at trunc_page(addr)
125  * and the return value is adjusted up by the page offset.
126  *
127  * Generally speaking, only character devices which are themselves
128  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
129  * there would be no cache coherency between a descriptor and a VM mapping
130  * both to the same character device.
131  */
132 #ifndef _SYS_SYSPROTO_H_
133 struct mmap_args {
134 	void *addr;
135 	size_t len;
136 	int prot;
137 	int flags;
138 	int fd;
139 	long pad;
140 	off_t pos;
141 };
142 #endif
143 
144 int
145 sys_mmap(struct thread *td, struct mmap_args *uap)
146 {
147 
148 	return (kern_mmap(td, &(struct mmap_req){
149 		.mr_hint = (uintptr_t)uap->addr,
150 		.mr_len = uap->len,
151 		.mr_prot = uap->prot,
152 		.mr_flags = uap->flags,
153 		.mr_fd = uap->fd,
154 		.mr_pos = uap->pos,
155 	    }));
156 }
157 
158 int
159 kern_mmap_maxprot(struct proc *p, int prot)
160 {
161 
162 	if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
163 	    (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
164 		return (_PROT_ALL);
165 	if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
166 	    prot != PROT_NONE)
167 		 return (prot);
168 	return (_PROT_ALL);
169 }
170 
171 int
172 kern_mmap(struct thread *td, const struct mmap_req *mrp)
173 {
174 	struct vmspace *vms;
175 	struct file *fp;
176 	struct proc *p;
177 	off_t pos;
178 	vm_offset_t addr, orig_addr;
179 	vm_size_t len, pageoff, size;
180 	vm_prot_t cap_maxprot;
181 	int align, error, fd, flags, max_prot, prot;
182 	cap_rights_t rights;
183 	mmap_check_fp_fn check_fp_fn;
184 
185 	orig_addr = addr = mrp->mr_hint;
186 	len = mrp->mr_len;
187 	prot = mrp->mr_prot;
188 	flags = mrp->mr_flags;
189 	fd = mrp->mr_fd;
190 	pos = mrp->mr_pos;
191 	check_fp_fn = mrp->mr_check_fp_fn;
192 
193 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
194 		return (EINVAL);
195 	max_prot = PROT_MAX_EXTRACT(prot);
196 	prot = PROT_EXTRACT(prot);
197 	if (max_prot != 0 && (max_prot & prot) != prot)
198 		return (ENOTSUP);
199 
200 	p = td->td_proc;
201 
202 	/*
203 	 * Always honor PROT_MAX if set.  If not, default to all
204 	 * permissions unless we're implying maximum permissions.
205 	 */
206 	if (max_prot == 0)
207 		max_prot = kern_mmap_maxprot(p, prot);
208 
209 	vms = p->p_vmspace;
210 	fp = NULL;
211 	AUDIT_ARG_FD(fd);
212 
213 	/*
214 	 * Ignore old flags that used to be defined but did not do anything.
215 	 */
216 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
217 
218 	/*
219 	 * Enforce the constraints.
220 	 * Mapping of length 0 is only allowed for old binaries.
221 	 * Anonymous mapping shall specify -1 as filedescriptor and
222 	 * zero position for new code. Be nice to ancient a.out
223 	 * binaries and correct pos for anonymous mapping, since old
224 	 * ld.so sometimes issues anonymous map requests with non-zero
225 	 * pos.
226 	 */
227 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
228 		if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
229 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
230 			return (EINVAL);
231 	} else {
232 		if ((flags & MAP_ANON) != 0)
233 			pos = 0;
234 	}
235 
236 	if (flags & MAP_STACK) {
237 		if ((fd != -1) ||
238 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
239 			return (EINVAL);
240 		flags |= MAP_ANON;
241 		pos = 0;
242 	}
243 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
244 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
245 	    MAP_PREFAULT_READ | MAP_GUARD | MAP_32BIT | MAP_ALIGNMENT_MASK)) != 0)
246 		return (EINVAL);
247 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
248 		return (EINVAL);
249 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
250 		return (EINVAL);
251 	if (prot != PROT_NONE &&
252 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
253 		return (EINVAL);
254 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
255 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
256 	    MAP_32BIT | MAP_ALIGNMENT_MASK)) != 0))
257 		return (EINVAL);
258 
259 	/*
260 	 * Align the file position to a page boundary,
261 	 * and save its page offset component.
262 	 */
263 	pageoff = (pos & PAGE_MASK);
264 	pos -= pageoff;
265 
266 	/* Compute size from len by rounding (on both ends). */
267 	size = len + pageoff;			/* low end... */
268 	size = round_page(size);		/* hi end */
269 	/* Check for rounding up to zero. */
270 	if (len > size)
271 		return (ENOMEM);
272 
273 	/* Ensure alignment is at least a page and fits in a pointer. */
274 	align = flags & MAP_ALIGNMENT_MASK;
275 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
276 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
277 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
278 		return (EINVAL);
279 
280 	/*
281 	 * Check for illegal addresses.  Watch out for address wrap... Note
282 	 * that VM_*_ADDRESS are not constants due to casts (argh).
283 	 */
284 	if (flags & MAP_FIXED) {
285 		/*
286 		 * The specified address must have the same remainder
287 		 * as the file offset taken modulo PAGE_SIZE, so it
288 		 * should be aligned after adjustment by pageoff.
289 		 */
290 		addr -= pageoff;
291 		if (addr & PAGE_MASK)
292 			return (EINVAL);
293 
294 		/* Address range must be all in user VM space. */
295 		if (!vm_map_range_valid(&vms->vm_map, addr, addr + size))
296 			return (EINVAL);
297 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
298 			return (EINVAL);
299 	} else if (flags & MAP_32BIT) {
300 		/*
301 		 * For MAP_32BIT, override the hint if it is too high and
302 		 * do not bother moving the mapping past the heap (since
303 		 * the heap is usually above 2GB).
304 		 */
305 		if (addr + size > MAP_32BIT_MAX_ADDR)
306 			addr = 0;
307 	} else {
308 		/*
309 		 * XXX for non-fixed mappings where no hint is provided or
310 		 * the hint would fall in the potential heap space,
311 		 * place it after the end of the largest possible heap.
312 		 *
313 		 * For anonymous mappings within the address space of the
314 		 * calling process, the absence of a hint is handled at a
315 		 * lower level in order to implement different clustering
316 		 * strategies for ASLR.
317 		 */
318 		if (((flags & MAP_ANON) == 0 && addr == 0) ||
319 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
320 		    addr < round_page((vm_offset_t)vms->vm_daddr +
321 		    lim_max(td, RLIMIT_DATA))))
322 			addr = round_page((vm_offset_t)vms->vm_daddr +
323 			    lim_max(td, RLIMIT_DATA));
324 	}
325 	if (len == 0) {
326 		/*
327 		 * Return success without mapping anything for old
328 		 * binaries that request a page-aligned mapping of
329 		 * length 0.  For modern binaries, this function
330 		 * returns an error earlier.
331 		 */
332 		error = 0;
333 	} else if ((flags & MAP_GUARD) != 0) {
334 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
335 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
336 	} else if ((flags & MAP_ANON) != 0) {
337 		/*
338 		 * Mapping blank space is trivial.
339 		 *
340 		 * This relies on VM_PROT_* matching PROT_*.
341 		 */
342 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
343 		    max_prot, flags, NULL, pos, FALSE, td);
344 	} else {
345 		/*
346 		 * Mapping file, get fp for validation and don't let the
347 		 * descriptor disappear on us if we block. Check capability
348 		 * rights, but also return the maximum rights to be combined
349 		 * with maxprot later.
350 		 */
351 		cap_rights_init_one(&rights, CAP_MMAP);
352 		if (prot & PROT_READ)
353 			cap_rights_set_one(&rights, CAP_MMAP_R);
354 		if ((flags & MAP_SHARED) != 0) {
355 			if (prot & PROT_WRITE)
356 				cap_rights_set_one(&rights, CAP_MMAP_W);
357 		}
358 		if (prot & PROT_EXEC)
359 			cap_rights_set_one(&rights, CAP_MMAP_X);
360 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
361 		if (error != 0)
362 			goto done;
363 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
364 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
365 			error = EINVAL;
366 			goto done;
367 		}
368 		if (check_fp_fn != NULL) {
369 			error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
370 			    flags);
371 			if (error != 0)
372 				goto done;
373 		}
374 		if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data))
375 			addr = orig_addr;
376 		/* This relies on VM_PROT_* matching PROT_*. */
377 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
378 		    max_prot & cap_maxprot, flags, pos, td);
379 	}
380 
381 	if (error == 0)
382 		td->td_retval[0] = addr + pageoff;
383 done:
384 	if (fp)
385 		fdrop(fp, td);
386 
387 	return (error);
388 }
389 
390 #if defined(COMPAT_FREEBSD6)
391 int
392 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
393 {
394 	return (kern_mmap(td, &(struct mmap_req){
395 		.mr_hint = (uintptr_t)uap->addr,
396 		.mr_len = uap->len,
397 		.mr_prot = uap->prot,
398 		.mr_flags = uap->flags,
399 		.mr_fd = uap->fd,
400 		.mr_pos = uap->pos,
401 	    }));
402 }
403 #endif
404 
405 #ifdef COMPAT_43
406 #ifndef _SYS_SYSPROTO_H_
407 struct ommap_args {
408 	caddr_t addr;
409 	int len;
410 	int prot;
411 	int flags;
412 	int fd;
413 	long pos;
414 };
415 #endif
416 int
417 ommap(struct thread *td, struct ommap_args *uap)
418 {
419 	return (kern_ommap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
420 	    uap->flags, uap->fd, uap->pos));
421 }
422 
423 int
424 kern_ommap(struct thread *td, uintptr_t hint, int len, int oprot,
425     int oflags, int fd, long pos)
426 {
427 	static const char cvtbsdprot[8] = {
428 		0,
429 		PROT_EXEC,
430 		PROT_WRITE,
431 		PROT_EXEC | PROT_WRITE,
432 		PROT_READ,
433 		PROT_EXEC | PROT_READ,
434 		PROT_WRITE | PROT_READ,
435 		PROT_EXEC | PROT_WRITE | PROT_READ,
436 	};
437 	int flags, prot;
438 
439 	if (len < 0)
440 		return (EINVAL);
441 
442 #define	OMAP_ANON	0x0002
443 #define	OMAP_COPY	0x0020
444 #define	OMAP_SHARED	0x0010
445 #define	OMAP_FIXED	0x0100
446 
447 	prot = cvtbsdprot[oprot & 0x7];
448 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
449 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
450 	    prot != 0)
451 		prot |= PROT_EXEC;
452 #endif
453 	flags = 0;
454 	if (oflags & OMAP_ANON)
455 		flags |= MAP_ANON;
456 	if (oflags & OMAP_COPY)
457 		flags |= MAP_COPY;
458 	if (oflags & OMAP_SHARED)
459 		flags |= MAP_SHARED;
460 	else
461 		flags |= MAP_PRIVATE;
462 	if (oflags & OMAP_FIXED)
463 		flags |= MAP_FIXED;
464 	return (kern_mmap(td, &(struct mmap_req){
465 		.mr_hint = hint,
466 		.mr_len = len,
467 		.mr_prot = prot,
468 		.mr_flags = flags,
469 		.mr_fd = fd,
470 		.mr_pos = pos,
471 	    }));
472 }
473 #endif				/* COMPAT_43 */
474 
475 #ifndef _SYS_SYSPROTO_H_
476 struct msync_args {
477 	void *addr;
478 	size_t len;
479 	int flags;
480 };
481 #endif
482 int
483 sys_msync(struct thread *td, struct msync_args *uap)
484 {
485 
486 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
487 }
488 
489 int
490 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
491 {
492 	vm_offset_t addr;
493 	vm_size_t pageoff;
494 	vm_map_t map;
495 	int rv;
496 
497 	addr = addr0;
498 	pageoff = (addr & PAGE_MASK);
499 	addr -= pageoff;
500 	size += pageoff;
501 	size = (vm_size_t) round_page(size);
502 	if (addr + size < addr)
503 		return (EINVAL);
504 
505 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
506 		return (EINVAL);
507 
508 	map = &td->td_proc->p_vmspace->vm_map;
509 
510 	/*
511 	 * Clean the pages and interpret the return value.
512 	 */
513 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
514 	    (flags & MS_INVALIDATE) != 0);
515 	switch (rv) {
516 	case KERN_SUCCESS:
517 		return (0);
518 	case KERN_INVALID_ADDRESS:
519 		return (ENOMEM);
520 	case KERN_INVALID_ARGUMENT:
521 		return (EBUSY);
522 	case KERN_FAILURE:
523 		return (EIO);
524 	default:
525 		return (EINVAL);
526 	}
527 }
528 
529 #ifndef _SYS_SYSPROTO_H_
530 struct munmap_args {
531 	void *addr;
532 	size_t len;
533 };
534 #endif
535 int
536 sys_munmap(struct thread *td, struct munmap_args *uap)
537 {
538 
539 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
540 }
541 
542 int
543 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
544 {
545 #ifdef HWPMC_HOOKS
546 	struct pmckern_map_out pkm;
547 	vm_map_entry_t entry;
548 	bool pmc_handled;
549 #endif
550 	vm_offset_t addr, end;
551 	vm_size_t pageoff;
552 	vm_map_t map;
553 	int rv;
554 
555 	if (size == 0)
556 		return (EINVAL);
557 
558 	addr = addr0;
559 	pageoff = (addr & PAGE_MASK);
560 	addr -= pageoff;
561 	size += pageoff;
562 	size = (vm_size_t) round_page(size);
563 	end = addr + size;
564 	map = &td->td_proc->p_vmspace->vm_map;
565 	if (!vm_map_range_valid(map, addr, end))
566 		return (EINVAL);
567 
568 	vm_map_lock(map);
569 #ifdef HWPMC_HOOKS
570 	pmc_handled = false;
571 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
572 		pmc_handled = true;
573 		/*
574 		 * Inform hwpmc if the address range being unmapped contains
575 		 * an executable region.
576 		 */
577 		pkm.pm_address = (uintptr_t) NULL;
578 		if (vm_map_lookup_entry(map, addr, &entry)) {
579 			for (; entry->start < end;
580 			    entry = vm_map_entry_succ(entry)) {
581 				if (vm_map_check_protection(map, entry->start,
582 					entry->end, VM_PROT_EXECUTE) == TRUE) {
583 					pkm.pm_address = (uintptr_t) addr;
584 					pkm.pm_size = (size_t) size;
585 					break;
586 				}
587 			}
588 		}
589 	}
590 #endif
591 	rv = vm_map_delete(map, addr, end);
592 
593 #ifdef HWPMC_HOOKS
594 	if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) {
595 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
596 		vm_map_lock_downgrade(map);
597 		if (pkm.pm_address != (uintptr_t) NULL)
598 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
599 		vm_map_unlock_read(map);
600 	} else
601 #endif
602 		vm_map_unlock(map);
603 
604 	return (vm_mmap_to_errno(rv));
605 }
606 
607 #ifndef _SYS_SYSPROTO_H_
608 struct mprotect_args {
609 	const void *addr;
610 	size_t len;
611 	int prot;
612 };
613 #endif
614 int
615 sys_mprotect(struct thread *td, struct mprotect_args *uap)
616 {
617 
618 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len,
619 	    uap->prot, 0));
620 }
621 
622 int
623 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot,
624     int flags)
625 {
626 	vm_offset_t addr;
627 	vm_size_t pageoff;
628 	int vm_error, max_prot;
629 
630 	addr = addr0;
631 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
632 		return (EINVAL);
633 	max_prot = PROT_MAX_EXTRACT(prot);
634 	prot = PROT_EXTRACT(prot);
635 	pageoff = (addr & PAGE_MASK);
636 	addr -= pageoff;
637 	size += pageoff;
638 	size = (vm_size_t) round_page(size);
639 #ifdef COMPAT_FREEBSD32
640 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
641 		if (((addr + size) & 0xffffffff) < addr)
642 			return (EINVAL);
643 	} else
644 #endif
645 	if (addr + size < addr)
646 		return (EINVAL);
647 
648 	flags |= VM_MAP_PROTECT_SET_PROT;
649 	if (max_prot != 0)
650 		flags |= VM_MAP_PROTECT_SET_MAXPROT;
651 	vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
652 	    addr, addr + size, prot, max_prot, flags);
653 
654 	switch (vm_error) {
655 	case KERN_SUCCESS:
656 		return (0);
657 	case KERN_PROTECTION_FAILURE:
658 		return (EACCES);
659 	case KERN_RESOURCE_SHORTAGE:
660 		return (ENOMEM);
661 	case KERN_OUT_OF_BOUNDS:
662 		return (ENOTSUP);
663 	}
664 	return (EINVAL);
665 }
666 
667 #ifndef _SYS_SYSPROTO_H_
668 struct minherit_args {
669 	void *addr;
670 	size_t len;
671 	int inherit;
672 };
673 #endif
674 int
675 sys_minherit(struct thread *td, struct minherit_args *uap)
676 {
677 
678 	return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
679 	    uap->inherit));
680 }
681 
682 int
683 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
684 {
685 	vm_offset_t addr;
686 	vm_size_t size, pageoff;
687 	vm_inherit_t inherit;
688 
689 	addr = (vm_offset_t)addr0;
690 	size = len;
691 	inherit = inherit0;
692 
693 	pageoff = (addr & PAGE_MASK);
694 	addr -= pageoff;
695 	size += pageoff;
696 	size = (vm_size_t) round_page(size);
697 	if (addr + size < addr)
698 		return (EINVAL);
699 
700 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
701 	    addr + size, inherit)) {
702 	case KERN_SUCCESS:
703 		return (0);
704 	case KERN_PROTECTION_FAILURE:
705 		return (EACCES);
706 	}
707 	return (EINVAL);
708 }
709 
710 #ifndef _SYS_SYSPROTO_H_
711 struct madvise_args {
712 	void *addr;
713 	size_t len;
714 	int behav;
715 };
716 #endif
717 
718 int
719 sys_madvise(struct thread *td, struct madvise_args *uap)
720 {
721 
722 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
723 }
724 
725 int
726 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
727 {
728 	vm_map_t map;
729 	vm_offset_t addr, end, start;
730 	int flags;
731 
732 	/*
733 	 * Check for our special case, advising the swap pager we are
734 	 * "immortal."
735 	 */
736 	if (behav == MADV_PROTECT) {
737 		flags = PPROT_SET;
738 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
739 		    PROC_SPROTECT, &flags));
740 	}
741 
742 	/*
743 	 * Check for illegal addresses.  Watch out for address wrap... Note
744 	 * that VM_*_ADDRESS are not constants due to casts (argh).
745 	 */
746 	map = &td->td_proc->p_vmspace->vm_map;
747 	addr = addr0;
748 	if (!vm_map_range_valid(map, addr, addr + len))
749 		return (EINVAL);
750 
751 	/*
752 	 * Since this routine is only advisory, we default to conservative
753 	 * behavior.
754 	 */
755 	start = trunc_page(addr);
756 	end = round_page(addr + len);
757 
758 	/*
759 	 * vm_map_madvise() checks for illegal values of behav.
760 	 */
761 	return (vm_map_madvise(map, start, end, behav));
762 }
763 
764 #ifndef _SYS_SYSPROTO_H_
765 struct mincore_args {
766 	const void *addr;
767 	size_t len;
768 	char *vec;
769 };
770 #endif
771 
772 int
773 sys_mincore(struct thread *td, struct mincore_args *uap)
774 {
775 
776 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
777 }
778 
779 int
780 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
781 {
782 	pmap_t pmap;
783 	vm_map_t map;
784 	vm_map_entry_t current, entry;
785 	vm_object_t object;
786 	vm_offset_t addr, cend, end, first_addr;
787 	vm_paddr_t pa;
788 	vm_page_t m;
789 	vm_pindex_t pindex;
790 	int error, lastvecindex, mincoreinfo, vecindex;
791 	unsigned int timestamp;
792 
793 	/*
794 	 * Make sure that the addresses presented are valid for user
795 	 * mode.
796 	 */
797 	first_addr = addr = trunc_page(addr0);
798 	end = round_page(addr0 + len);
799 	map = &td->td_proc->p_vmspace->vm_map;
800 	if (end > vm_map_max(map) || end < addr)
801 		return (ENOMEM);
802 
803 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
804 
805 	vm_map_lock_read(map);
806 RestartScan:
807 	timestamp = map->timestamp;
808 
809 	if (!vm_map_lookup_entry(map, addr, &entry)) {
810 		vm_map_unlock_read(map);
811 		return (ENOMEM);
812 	}
813 
814 	/*
815 	 * Do this on a map entry basis so that if the pages are not
816 	 * in the current processes address space, we can easily look
817 	 * up the pages elsewhere.
818 	 */
819 	lastvecindex = -1;
820 	while (entry->start < end) {
821 		/*
822 		 * check for contiguity
823 		 */
824 		current = entry;
825 		entry = vm_map_entry_succ(current);
826 		if (current->end < end &&
827 		    entry->start > current->end) {
828 			vm_map_unlock_read(map);
829 			return (ENOMEM);
830 		}
831 
832 		/*
833 		 * ignore submaps (for now) or null objects
834 		 */
835 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
836 		    current->object.vm_object == NULL)
837 			continue;
838 
839 		/*
840 		 * limit this scan to the current map entry and the
841 		 * limits for the mincore call
842 		 */
843 		if (addr < current->start)
844 			addr = current->start;
845 		cend = current->end;
846 		if (cend > end)
847 			cend = end;
848 
849 		for (; addr < cend; addr += PAGE_SIZE) {
850 			/*
851 			 * Check pmap first, it is likely faster, also
852 			 * it can provide info as to whether we are the
853 			 * one referencing or modifying the page.
854 			 */
855 			m = NULL;
856 			object = NULL;
857 retry:
858 			pa = 0;
859 			mincoreinfo = pmap_mincore(pmap, addr, &pa);
860 			if (mincore_mapped) {
861 				/*
862 				 * We only care about this pmap's
863 				 * mapping of the page, if any.
864 				 */
865 				;
866 			} else if (pa != 0) {
867 				/*
868 				 * The page is mapped by this process but not
869 				 * both accessed and modified.  It is also
870 				 * managed.  Acquire the object lock so that
871 				 * other mappings might be examined.  The page's
872 				 * identity may change at any point before its
873 				 * object lock is acquired, so re-validate if
874 				 * necessary.
875 				 */
876 				m = PHYS_TO_VM_PAGE(pa);
877 				while (object == NULL || m->object != object) {
878 					if (object != NULL)
879 						VM_OBJECT_WUNLOCK(object);
880 					object = atomic_load_ptr(&m->object);
881 					if (object == NULL)
882 						goto retry;
883 					VM_OBJECT_WLOCK(object);
884 				}
885 				if (pa != pmap_extract(pmap, addr))
886 					goto retry;
887 				KASSERT(vm_page_all_valid(m),
888 				    ("mincore: page %p is mapped but invalid",
889 				    m));
890 			} else if (mincoreinfo == 0) {
891 				/*
892 				 * The page is not mapped by this process.  If
893 				 * the object implements managed pages, then
894 				 * determine if the page is resident so that
895 				 * the mappings might be examined.
896 				 */
897 				if (current->object.vm_object != object) {
898 					if (object != NULL)
899 						VM_OBJECT_WUNLOCK(object);
900 					object = current->object.vm_object;
901 					VM_OBJECT_WLOCK(object);
902 				}
903 				if ((object->flags & OBJ_SWAP) != 0 ||
904 				    object->type == OBJT_VNODE) {
905 					pindex = OFF_TO_IDX(current->offset +
906 					    (addr - current->start));
907 					m = vm_page_lookup(object, pindex);
908 					if (m != NULL && vm_page_none_valid(m))
909 						m = NULL;
910 					if (m != NULL)
911 						mincoreinfo = MINCORE_INCORE;
912 				}
913 			}
914 			if (m != NULL) {
915 				VM_OBJECT_ASSERT_WLOCKED(m->object);
916 
917 				/* Examine other mappings of the page. */
918 				if (m->dirty == 0 && pmap_is_modified(m))
919 					vm_page_dirty(m);
920 				if (m->dirty != 0)
921 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
922 
923 				/*
924 				 * The first test for PGA_REFERENCED is an
925 				 * optimization.  The second test is
926 				 * required because a concurrent pmap
927 				 * operation could clear the last reference
928 				 * and set PGA_REFERENCED before the call to
929 				 * pmap_is_referenced().
930 				 */
931 				if ((m->a.flags & PGA_REFERENCED) != 0 ||
932 				    pmap_is_referenced(m) ||
933 				    (m->a.flags & PGA_REFERENCED) != 0)
934 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
935 			}
936 			if (object != NULL)
937 				VM_OBJECT_WUNLOCK(object);
938 
939 			/*
940 			 * subyte may page fault.  In case it needs to modify
941 			 * the map, we release the lock.
942 			 */
943 			vm_map_unlock_read(map);
944 
945 			/*
946 			 * calculate index into user supplied byte vector
947 			 */
948 			vecindex = atop(addr - first_addr);
949 
950 			/*
951 			 * If we have skipped map entries, we need to make sure that
952 			 * the byte vector is zeroed for those skipped entries.
953 			 */
954 			while ((lastvecindex + 1) < vecindex) {
955 				++lastvecindex;
956 				error = subyte(vec + lastvecindex, 0);
957 				if (error) {
958 					error = EFAULT;
959 					goto done2;
960 				}
961 			}
962 
963 			/*
964 			 * Pass the page information to the user
965 			 */
966 			error = subyte(vec + vecindex, mincoreinfo);
967 			if (error) {
968 				error = EFAULT;
969 				goto done2;
970 			}
971 
972 			/*
973 			 * If the map has changed, due to the subyte, the previous
974 			 * output may be invalid.
975 			 */
976 			vm_map_lock_read(map);
977 			if (timestamp != map->timestamp)
978 				goto RestartScan;
979 
980 			lastvecindex = vecindex;
981 		}
982 	}
983 
984 	/*
985 	 * subyte may page fault.  In case it needs to modify
986 	 * the map, we release the lock.
987 	 */
988 	vm_map_unlock_read(map);
989 
990 	/*
991 	 * Zero the last entries in the byte vector.
992 	 */
993 	vecindex = atop(end - first_addr);
994 	while ((lastvecindex + 1) < vecindex) {
995 		++lastvecindex;
996 		error = subyte(vec + lastvecindex, 0);
997 		if (error) {
998 			error = EFAULT;
999 			goto done2;
1000 		}
1001 	}
1002 
1003 	/*
1004 	 * If the map has changed, due to the subyte, the previous
1005 	 * output may be invalid.
1006 	 */
1007 	vm_map_lock_read(map);
1008 	if (timestamp != map->timestamp)
1009 		goto RestartScan;
1010 	vm_map_unlock_read(map);
1011 done2:
1012 	return (error);
1013 }
1014 
1015 #ifndef _SYS_SYSPROTO_H_
1016 struct mlock_args {
1017 	const void *addr;
1018 	size_t len;
1019 };
1020 #endif
1021 int
1022 sys_mlock(struct thread *td, struct mlock_args *uap)
1023 {
1024 
1025 	return (kern_mlock(td->td_proc, td->td_ucred,
1026 	    __DECONST(uintptr_t, uap->addr), uap->len));
1027 }
1028 
1029 int
1030 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1031 {
1032 	vm_offset_t addr, end, last, start;
1033 	vm_size_t npages, size;
1034 	vm_map_t map;
1035 	unsigned long nsize;
1036 	int error;
1037 
1038 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1039 	if (error)
1040 		return (error);
1041 	addr = addr0;
1042 	size = len;
1043 	last = addr + size;
1044 	start = trunc_page(addr);
1045 	end = round_page(last);
1046 	if (last < addr || end < addr)
1047 		return (EINVAL);
1048 	npages = atop(end - start);
1049 	if (npages > vm_page_max_user_wired)
1050 		return (ENOMEM);
1051 	map = &proc->p_vmspace->vm_map;
1052 	PROC_LOCK(proc);
1053 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1054 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1055 		PROC_UNLOCK(proc);
1056 		return (ENOMEM);
1057 	}
1058 	PROC_UNLOCK(proc);
1059 #ifdef RACCT
1060 	if (racct_enable) {
1061 		PROC_LOCK(proc);
1062 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1063 		PROC_UNLOCK(proc);
1064 		if (error != 0)
1065 			return (ENOMEM);
1066 	}
1067 #endif
1068 	error = vm_map_wire(map, start, end,
1069 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1070 #ifdef RACCT
1071 	if (racct_enable && error != KERN_SUCCESS) {
1072 		PROC_LOCK(proc);
1073 		racct_set(proc, RACCT_MEMLOCK,
1074 		    ptoa(pmap_wired_count(map->pmap)));
1075 		PROC_UNLOCK(proc);
1076 	}
1077 #endif
1078 	switch (error) {
1079 	case KERN_SUCCESS:
1080 		return (0);
1081 	case KERN_INVALID_ARGUMENT:
1082 		return (EINVAL);
1083 	default:
1084 		return (ENOMEM);
1085 	}
1086 }
1087 
1088 #ifndef _SYS_SYSPROTO_H_
1089 struct mlockall_args {
1090 	int	how;
1091 };
1092 #endif
1093 
1094 int
1095 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1096 {
1097 	vm_map_t map;
1098 	int error;
1099 
1100 	map = &td->td_proc->p_vmspace->vm_map;
1101 	error = priv_check(td, PRIV_VM_MLOCK);
1102 	if (error)
1103 		return (error);
1104 
1105 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1106 		return (EINVAL);
1107 
1108 	/*
1109 	 * If wiring all pages in the process would cause it to exceed
1110 	 * a hard resource limit, return ENOMEM.
1111 	 */
1112 	if (!old_mlock && uap->how & MCL_CURRENT) {
1113 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1114 			return (ENOMEM);
1115 	}
1116 #ifdef RACCT
1117 	if (racct_enable) {
1118 		PROC_LOCK(td->td_proc);
1119 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1120 		PROC_UNLOCK(td->td_proc);
1121 		if (error != 0)
1122 			return (ENOMEM);
1123 	}
1124 #endif
1125 
1126 	if (uap->how & MCL_FUTURE) {
1127 		vm_map_lock(map);
1128 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1129 		vm_map_unlock(map);
1130 		error = 0;
1131 	}
1132 
1133 	if (uap->how & MCL_CURRENT) {
1134 		/*
1135 		 * P1003.1-2001 mandates that all currently mapped pages
1136 		 * will be memory resident and locked (wired) upon return
1137 		 * from mlockall(). vm_map_wire() will wire pages, by
1138 		 * calling vm_fault_wire() for each page in the region.
1139 		 */
1140 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1141 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1142 		if (error == KERN_SUCCESS)
1143 			error = 0;
1144 		else if (error == KERN_RESOURCE_SHORTAGE)
1145 			error = ENOMEM;
1146 		else
1147 			error = EAGAIN;
1148 	}
1149 #ifdef RACCT
1150 	if (racct_enable && error != KERN_SUCCESS) {
1151 		PROC_LOCK(td->td_proc);
1152 		racct_set(td->td_proc, RACCT_MEMLOCK,
1153 		    ptoa(pmap_wired_count(map->pmap)));
1154 		PROC_UNLOCK(td->td_proc);
1155 	}
1156 #endif
1157 
1158 	return (error);
1159 }
1160 
1161 #ifndef _SYS_SYSPROTO_H_
1162 struct munlockall_args {
1163 	register_t dummy;
1164 };
1165 #endif
1166 
1167 int
1168 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1169 {
1170 	vm_map_t map;
1171 	int error;
1172 
1173 	map = &td->td_proc->p_vmspace->vm_map;
1174 	error = priv_check(td, PRIV_VM_MUNLOCK);
1175 	if (error)
1176 		return (error);
1177 
1178 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1179 	vm_map_lock(map);
1180 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1181 	vm_map_unlock(map);
1182 
1183 	/* Forcibly unwire all pages. */
1184 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1185 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1186 #ifdef RACCT
1187 	if (racct_enable && error == KERN_SUCCESS) {
1188 		PROC_LOCK(td->td_proc);
1189 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1190 		PROC_UNLOCK(td->td_proc);
1191 	}
1192 #endif
1193 
1194 	return (error);
1195 }
1196 
1197 #ifndef _SYS_SYSPROTO_H_
1198 struct munlock_args {
1199 	const void *addr;
1200 	size_t len;
1201 };
1202 #endif
1203 int
1204 sys_munlock(struct thread *td, struct munlock_args *uap)
1205 {
1206 
1207 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1208 }
1209 
1210 int
1211 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1212 {
1213 	vm_offset_t addr, end, last, start;
1214 #ifdef RACCT
1215 	vm_map_t map;
1216 #endif
1217 	int error;
1218 
1219 	error = priv_check(td, PRIV_VM_MUNLOCK);
1220 	if (error)
1221 		return (error);
1222 	addr = addr0;
1223 	last = addr + size;
1224 	start = trunc_page(addr);
1225 	end = round_page(last);
1226 	if (last < addr || end < addr)
1227 		return (EINVAL);
1228 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1229 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1230 #ifdef RACCT
1231 	if (racct_enable && error == KERN_SUCCESS) {
1232 		PROC_LOCK(td->td_proc);
1233 		map = &td->td_proc->p_vmspace->vm_map;
1234 		racct_set(td->td_proc, RACCT_MEMLOCK,
1235 		    ptoa(pmap_wired_count(map->pmap)));
1236 		PROC_UNLOCK(td->td_proc);
1237 	}
1238 #endif
1239 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1240 }
1241 
1242 /*
1243  * vm_mmap_vnode()
1244  *
1245  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1246  * operations on vnodes.
1247  */
1248 int
1249 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1250     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1251     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1252     boolean_t *writecounted)
1253 {
1254 	struct vattr va;
1255 	vm_object_t obj;
1256 	vm_ooffset_t foff;
1257 	struct ucred *cred;
1258 	int error, flags;
1259 	bool writex;
1260 
1261 	cred = td->td_ucred;
1262 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1263 	    (*flagsp & MAP_SHARED) != 0;
1264 	if ((error = vget(vp, LK_SHARED)) != 0)
1265 		return (error);
1266 	AUDIT_ARG_VNODE1(vp);
1267 	foff = *foffp;
1268 	flags = *flagsp;
1269 	obj = vp->v_object;
1270 	if (vp->v_type == VREG) {
1271 		/*
1272 		 * Get the proper underlying object
1273 		 */
1274 		if (obj == NULL) {
1275 			error = EINVAL;
1276 			goto done;
1277 		}
1278 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1279 			vput(vp);
1280 			vp = (struct vnode *)obj->handle;
1281 			/*
1282 			 * Bypass filesystems obey the mpsafety of the
1283 			 * underlying fs.  Tmpfs never bypasses.
1284 			 */
1285 			error = vget(vp, LK_SHARED);
1286 			if (error != 0)
1287 				return (error);
1288 		}
1289 		if (writex) {
1290 			*writecounted = TRUE;
1291 			vm_pager_update_writecount(obj, 0, objsize);
1292 		}
1293 	} else {
1294 		error = EINVAL;
1295 		goto done;
1296 	}
1297 	if ((error = VOP_GETATTR(vp, &va, cred)))
1298 		goto done;
1299 #ifdef MAC
1300 	/* This relies on VM_PROT_* matching PROT_*. */
1301 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1302 	if (error != 0)
1303 		goto done;
1304 #endif
1305 	if ((flags & MAP_SHARED) != 0) {
1306 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1307 			if (prot & VM_PROT_WRITE) {
1308 				error = EPERM;
1309 				goto done;
1310 			}
1311 			*maxprotp &= ~VM_PROT_WRITE;
1312 		}
1313 	}
1314 	/*
1315 	 * If it is a regular file without any references
1316 	 * we do not need to sync it.
1317 	 * Adjust object size to be the size of actual file.
1318 	 */
1319 	objsize = round_page(va.va_size);
1320 	if (va.va_nlink == 0)
1321 		flags |= MAP_NOSYNC;
1322 	if (obj->type == OBJT_VNODE) {
1323 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1324 		    cred);
1325 		if (obj == NULL) {
1326 			error = ENOMEM;
1327 			goto done;
1328 		}
1329 	} else {
1330 		KASSERT((obj->flags & OBJ_SWAP) != 0, ("wrong object type"));
1331 		vm_object_reference(obj);
1332 #if VM_NRESERVLEVEL > 0
1333 		if ((obj->flags & OBJ_COLORED) == 0) {
1334 			VM_OBJECT_WLOCK(obj);
1335 			vm_object_color(obj, 0);
1336 			VM_OBJECT_WUNLOCK(obj);
1337 		}
1338 #endif
1339 	}
1340 	*objp = obj;
1341 	*flagsp = flags;
1342 
1343 	VOP_MMAPPED(vp);
1344 
1345 done:
1346 	if (error != 0 && *writecounted) {
1347 		*writecounted = FALSE;
1348 		vm_pager_update_writecount(obj, objsize, 0);
1349 	}
1350 	vput(vp);
1351 	return (error);
1352 }
1353 
1354 /*
1355  * vm_mmap_cdev()
1356  *
1357  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1358  * operations on cdevs.
1359  */
1360 int
1361 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1362     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1363     vm_ooffset_t *foff, vm_object_t *objp)
1364 {
1365 	vm_object_t obj;
1366 	int error, flags;
1367 
1368 	flags = *flagsp;
1369 
1370 	if (dsw->d_flags & D_MMAP_ANON) {
1371 		*objp = NULL;
1372 		*foff = 0;
1373 		*maxprotp = VM_PROT_ALL;
1374 		*flagsp |= MAP_ANON;
1375 		return (0);
1376 	}
1377 	/*
1378 	 * cdevs do not provide private mappings of any kind.
1379 	 */
1380 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1381 	    (prot & VM_PROT_WRITE) != 0)
1382 		return (EACCES);
1383 	if (flags & (MAP_PRIVATE|MAP_COPY))
1384 		return (EINVAL);
1385 	/*
1386 	 * Force device mappings to be shared.
1387 	 */
1388 	flags |= MAP_SHARED;
1389 #ifdef MAC_XXX
1390 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1391 	if (error != 0)
1392 		return (error);
1393 #endif
1394 	/*
1395 	 * First, try d_mmap_single().  If that is not implemented
1396 	 * (returns ENODEV), fall back to using the device pager.
1397 	 * Note that d_mmap_single() must return a reference to the
1398 	 * object (it needs to bump the reference count of the object
1399 	 * it returns somehow).
1400 	 *
1401 	 * XXX assumes VM_PROT_* == PROT_*
1402 	 */
1403 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1404 	if (error != ENODEV)
1405 		return (error);
1406 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1407 	    td->td_ucred);
1408 	if (obj == NULL)
1409 		return (EINVAL);
1410 	*objp = obj;
1411 	*flagsp = flags;
1412 	return (0);
1413 }
1414 
1415 int
1416 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1417 	vm_prot_t maxprot, int flags,
1418 	objtype_t handle_type, void *handle,
1419 	vm_ooffset_t foff)
1420 {
1421 	vm_object_t object;
1422 	struct thread *td = curthread;
1423 	int error;
1424 	boolean_t writecounted;
1425 
1426 	if (size == 0)
1427 		return (EINVAL);
1428 
1429 	size = round_page(size);
1430 	object = NULL;
1431 	writecounted = FALSE;
1432 
1433 	switch (handle_type) {
1434 	case OBJT_DEVICE: {
1435 		struct cdevsw *dsw;
1436 		struct cdev *cdev;
1437 		int ref;
1438 
1439 		cdev = handle;
1440 		dsw = dev_refthread(cdev, &ref);
1441 		if (dsw == NULL)
1442 			return (ENXIO);
1443 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1444 		    dsw, &foff, &object);
1445 		dev_relthread(cdev, ref);
1446 		break;
1447 	}
1448 	case OBJT_VNODE:
1449 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1450 		    handle, &foff, &object, &writecounted);
1451 		break;
1452 	default:
1453 		error = EINVAL;
1454 		break;
1455 	}
1456 	if (error)
1457 		return (error);
1458 
1459 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1460 	    foff, writecounted, td);
1461 	if (error != 0 && object != NULL) {
1462 		/*
1463 		 * If this mapping was accounted for in the vnode's
1464 		 * writecount, then undo that now.
1465 		 */
1466 		if (writecounted)
1467 			vm_pager_release_writecount(object, 0, size);
1468 		vm_object_deallocate(object);
1469 	}
1470 	return (error);
1471 }
1472 
1473 int
1474 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size)
1475 {
1476 	int error;
1477 
1478 	RACCT_PROC_LOCK(td->td_proc);
1479 	if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1480 		RACCT_PROC_UNLOCK(td->td_proc);
1481 		return (ENOMEM);
1482 	}
1483 	if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1484 		RACCT_PROC_UNLOCK(td->td_proc);
1485 		return (ENOMEM);
1486 	}
1487 	if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1488 		if (ptoa(pmap_wired_count(map->pmap)) + size >
1489 		    lim_cur(td, RLIMIT_MEMLOCK)) {
1490 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1491 			RACCT_PROC_UNLOCK(td->td_proc);
1492 			return (ENOMEM);
1493 		}
1494 		error = racct_set(td->td_proc, RACCT_MEMLOCK,
1495 		    ptoa(pmap_wired_count(map->pmap)) + size);
1496 		if (error != 0) {
1497 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1498 			RACCT_PROC_UNLOCK(td->td_proc);
1499 			return (error);
1500 		}
1501 	}
1502 	RACCT_PROC_UNLOCK(td->td_proc);
1503 	return (0);
1504 }
1505 
1506 /*
1507  * Internal version of mmap that maps a specific VM object into an
1508  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1509  */
1510 int
1511 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1512     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1513     boolean_t writecounted, struct thread *td)
1514 {
1515 	vm_offset_t default_addr, max_addr;
1516 	int docow, error, findspace, rv;
1517 	bool curmap, fitit;
1518 
1519 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1520 	if (curmap) {
1521 		error = kern_mmap_racct_check(td, map, size);
1522 		if (error != 0)
1523 			return (error);
1524 	}
1525 
1526 	/*
1527 	 * We currently can only deal with page aligned file offsets.
1528 	 * The mmap() system call already enforces this by subtracting
1529 	 * the page offset from the file offset, but checking here
1530 	 * catches errors in device drivers (e.g. d_single_mmap()
1531 	 * callbacks) and other internal mapping requests (such as in
1532 	 * exec).
1533 	 */
1534 	if (foff & PAGE_MASK)
1535 		return (EINVAL);
1536 
1537 	if ((flags & MAP_FIXED) == 0) {
1538 		fitit = true;
1539 		*addr = round_page(*addr);
1540 	} else {
1541 		if (*addr != trunc_page(*addr))
1542 			return (EINVAL);
1543 		fitit = false;
1544 	}
1545 
1546 	if (flags & MAP_ANON) {
1547 		if (object != NULL || foff != 0)
1548 			return (EINVAL);
1549 		docow = 0;
1550 	} else if (flags & MAP_PREFAULT_READ)
1551 		docow = MAP_PREFAULT;
1552 	else
1553 		docow = MAP_PREFAULT_PARTIAL;
1554 
1555 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1556 		docow |= MAP_COPY_ON_WRITE;
1557 	if (flags & MAP_NOSYNC)
1558 		docow |= MAP_DISABLE_SYNCER;
1559 	if (flags & MAP_NOCORE)
1560 		docow |= MAP_DISABLE_COREDUMP;
1561 	/* Shared memory is also shared with children. */
1562 	if (flags & MAP_SHARED)
1563 		docow |= MAP_INHERIT_SHARE;
1564 	if (writecounted)
1565 		docow |= MAP_WRITECOUNT;
1566 	if (flags & MAP_STACK) {
1567 		if (object != NULL)
1568 			return (EINVAL);
1569 		docow |= MAP_STACK_GROWS_DOWN;
1570 	}
1571 	if ((flags & MAP_EXCL) != 0)
1572 		docow |= MAP_CHECK_EXCL;
1573 	if ((flags & MAP_GUARD) != 0)
1574 		docow |= MAP_CREATE_GUARD;
1575 
1576 	if (fitit) {
1577 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1578 			findspace = VMFS_SUPER_SPACE;
1579 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1580 			findspace = VMFS_ALIGNED_SPACE(flags >>
1581 			    MAP_ALIGNMENT_SHIFT);
1582 		else
1583 			findspace = VMFS_OPTIMAL_SPACE;
1584 		max_addr = 0;
1585 		if ((flags & MAP_32BIT) != 0)
1586 			max_addr = MAP_32BIT_MAX_ADDR;
1587 		if (curmap) {
1588 			default_addr =
1589 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1590 			    vm_daddr + lim_max(td, RLIMIT_DATA));
1591 			if ((flags & MAP_32BIT) != 0)
1592 				default_addr = 0;
1593 			rv = vm_map_find_min(map, object, foff, addr, size,
1594 			    default_addr, max_addr, findspace, prot, maxprot,
1595 			    docow);
1596 		} else {
1597 			rv = vm_map_find(map, object, foff, addr, size,
1598 			    max_addr, findspace, prot, maxprot, docow);
1599 		}
1600 	} else {
1601 		rv = vm_map_fixed(map, object, foff, *addr, size,
1602 		    prot, maxprot, docow);
1603 	}
1604 
1605 	if (rv == KERN_SUCCESS) {
1606 		/*
1607 		 * If the process has requested that all future mappings
1608 		 * be wired, then heed this.
1609 		 */
1610 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1611 			vm_map_lock(map);
1612 			if ((map->flags & MAP_WIREFUTURE) != 0)
1613 				(void)vm_map_wire_locked(map, *addr,
1614 				    *addr + size, VM_MAP_WIRE_USER |
1615 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1616 				    VM_MAP_WIRE_NOHOLES));
1617 			vm_map_unlock(map);
1618 		}
1619 	}
1620 	return (vm_mmap_to_errno(rv));
1621 }
1622 
1623 /*
1624  * Translate a Mach VM return code to zero on success or the appropriate errno
1625  * on failure.
1626  */
1627 int
1628 vm_mmap_to_errno(int rv)
1629 {
1630 
1631 	switch (rv) {
1632 	case KERN_SUCCESS:
1633 		return (0);
1634 	case KERN_INVALID_ADDRESS:
1635 	case KERN_NO_SPACE:
1636 		return (ENOMEM);
1637 	case KERN_PROTECTION_FAILURE:
1638 		return (EACCES);
1639 	default:
1640 		return (EINVAL);
1641 	}
1642 }
1643