xref: /src/sys/vm/vm_mmap.c (revision dad6e6fc1ea4b737e9f1661ebd30da5d551e3d4a)
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 "opt_hwpmc_hooks.h"
44 #include "opt_hwt_hooks.h"
45 #include "opt_vm.h"
46 
47 #define	EXTERR_CATEGORY	EXTERR_CAT_MMAP
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/capsicum.h>
51 #include <sys/exterrvar.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/elf.h>
57 #include <sys/filedesc.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procctl.h>
61 #include <sys/racct.h>
62 #include <sys/resource.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
68 #include <sys/file.h>
69 #include <sys/mman.h>
70 #include <sys/mount.h>
71 #include <sys/conf.h>
72 #include <sys/stat.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysent.h>
75 #include <sys/vmmeter.h>
76 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
77 #include <machine/md_var.h>
78 #endif
79 
80 #include <security/audit/audit.h>
81 #include <security/mac/mac_framework.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_param.h>
85 #include <vm/pmap.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vm_pageout.h>
91 #include <vm/vm_extern.h>
92 #include <vm/vm_page.h>
93 #include <vm/vnode_pager.h>
94 
95 #ifdef HWPMC_HOOKS
96 #include <sys/pmckern.h>
97 #endif
98 
99 #ifdef HWT_HOOKS
100 #include <dev/hwt/hwt_hook.h>
101 #endif
102 
103 int old_mlock = 0;
104 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
105     "Do not apply RLIMIT_MEMLOCK on mlockall");
106 static int mincore_mapped = 1;
107 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
108     "mincore reports mappings, not residency");
109 static int imply_prot_max = 0;
110 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0,
111     "Imply maximum page protections in mmap() when none are specified");
112 
113 _Static_assert(MAXPAGESIZES <= 4, "MINCORE_SUPER too narrow");
114 
115 #if defined(COMPAT_43)
116 int
ogetpagesize(struct thread * td,struct ogetpagesize_args * uap)117 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
118 {
119 
120 	td->td_retval[0] = PAGE_SIZE;
121 	return (0);
122 }
123 #endif				/* COMPAT_43 */
124 
125 /*
126  * Memory Map (mmap) system call.  Note that the file offset
127  * and address are allowed to be NOT page aligned, though if
128  * the MAP_FIXED flag it set, both must have the same remainder
129  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
130  * page-aligned, the actual mapping starts at trunc_page(addr)
131  * and the return value is adjusted up by the page offset.
132  *
133  * Generally speaking, only character devices which are themselves
134  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
135  * there would be no cache coherency between a descriptor and a VM mapping
136  * both to the same character device.
137  */
138 #ifndef _SYS_SYSPROTO_H_
139 struct mmap_args {
140 	void *addr;
141 	size_t len;
142 	int prot;
143 	int flags;
144 	int fd;
145 	long pad;
146 	off_t pos;
147 };
148 #endif
149 
150 int
sys_mmap(struct thread * td,struct mmap_args * uap)151 sys_mmap(struct thread *td, struct mmap_args *uap)
152 {
153 
154 	return (kern_mmap(td, &(struct mmap_req){
155 		.mr_hint = (uintptr_t)uap->addr,
156 		.mr_len = uap->len,
157 		.mr_prot = uap->prot,
158 		.mr_flags = uap->flags,
159 		.mr_fd = uap->fd,
160 		.mr_pos = uap->pos,
161 	    }));
162 }
163 
164 int
kern_mmap_maxprot(struct proc * p,int prot)165 kern_mmap_maxprot(struct proc *p, int prot)
166 {
167 
168 	if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
169 	    (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
170 		return (_PROT_ALL);
171 	if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
172 	    prot != PROT_NONE)
173 		 return (prot);
174 	return (_PROT_ALL);
175 }
176 
177 int
kern_mmap(struct thread * td,const struct mmap_req * mrp)178 kern_mmap(struct thread *td, const struct mmap_req *mrp)
179 {
180 	struct vmspace *vms;
181 	struct file *fp;
182 	struct proc *p;
183 	off_t pos;
184 	vm_offset_t addr, orig_addr;
185 	vm_size_t len, pageoff, size;
186 	vm_prot_t cap_maxprot;
187 	int align, error, fd, flags, max_prot, prot;
188 	cap_rights_t rights;
189 	mmap_check_fp_fn check_fp_fn;
190 
191 	orig_addr = addr = mrp->mr_hint;
192 	len = mrp->mr_len;
193 	prot = mrp->mr_prot;
194 	flags = mrp->mr_flags;
195 	fd = mrp->mr_fd;
196 	pos = mrp->mr_pos;
197 	check_fp_fn = mrp->mr_check_fp_fn;
198 
199 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0) {
200 		return (EXTERROR(EINVAL, "unknown PROT bits %#jx", prot));
201 	}
202 	max_prot = PROT_MAX_EXTRACT(prot);
203 	prot = PROT_EXTRACT(prot);
204 	if (max_prot != 0 && (max_prot & prot) != prot) {
205 		return (EXTERROR(ENOTSUP,
206 		    "prot %#jx is not subset of max_prot %#jx",
207 		    prot, max_prot));
208 	}
209 
210 	p = td->td_proc;
211 
212 	/*
213 	 * Always honor PROT_MAX if set.  If not, default to all
214 	 * permissions unless we're implying maximum permissions.
215 	 */
216 	if (max_prot == 0)
217 		max_prot = kern_mmap_maxprot(p, prot);
218 
219 	vms = p->p_vmspace;
220 	fp = NULL;
221 	AUDIT_ARG_FD(fd);
222 
223 	/*
224 	 * Ignore old flags that used to be defined but did not do anything.
225 	 */
226 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
227 
228 	/*
229 	 * Enforce the constraints.
230 	 * Mapping of length 0 is only allowed for old binaries.
231 	 * Anonymous mapping shall specify -1 as filedescriptor and
232 	 * zero position for new code. Be nice to ancient a.out
233 	 * binaries and correct pos for anonymous mapping, since old
234 	 * ld.so sometimes issues anonymous map requests with non-zero
235 	 * pos.
236 	 */
237 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
238 		if (len == 0 && p->p_osrel >= P_OSREL_MAP_ANON)
239 			return (EXTERROR(EINVAL, "mapping with zero length"));
240 		if ((flags & MAP_ANON) != 0) {
241 			if (fd != -1)
242 				return (EXTERROR(EINVAL,
243 				    "fd %#jd not -1 for MAP_ANON", fd));
244 			if (pos != 0)
245 				return (EXTERROR(EINVAL,
246 				    "offset %#jd not zero for MAP_ANON", pos));
247 		}
248 	} else {
249 		if ((flags & MAP_ANON) != 0)
250 			pos = 0;
251 	}
252 
253 	if (flags & MAP_STACK) {
254 		if ((fd != -1) || ((prot & (PROT_READ | PROT_WRITE)) !=
255 		    (PROT_READ | PROT_WRITE))) {
256 			return (EXTERROR(EINVAL,
257 			    "MAP_STACK with prot %#jx < rw", prot));
258 		}
259 		flags |= MAP_ANON;
260 		pos = 0;
261 	}
262 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
263 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
264 	    MAP_PREFAULT_READ | MAP_GUARD | MAP_32BIT |
265 	    MAP_ALIGNMENT_MASK)) != 0) {
266 		return (EXTERROR(EINVAL, "reserved flag set (flags %#jx)",
267 		    flags));
268 	}
269 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) {
270 		return (EXTERROR(EINVAL, "EXCL without FIXED (flags %#jx)",
271 		    flags));
272 	}
273 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED |
274 	    MAP_PRIVATE)) {
275 		return (EXTERROR(EINVAL,
276 		    "both SHARED and PRIVATE set (flags %#jx)", flags));
277 	}
278 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
279 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
280 	    MAP_32BIT | MAP_ALIGNMENT_MASK)) != 0)) {
281 		return (EXTERROR(EINVAL, "GUARD with wrong parameters"));
282 	}
283 
284 	/*
285 	 * Align the file position to a page boundary,
286 	 * and save its page offset component.
287 	 */
288 	pageoff = (pos & PAGE_MASK);
289 	pos -= pageoff;
290 
291 	/* Compute size from len by rounding (on both ends). */
292 	size = len + pageoff;			/* low end... */
293 	size = round_page(size);		/* hi end */
294 	/* Check for rounding up to zero. */
295 	if (len > size)
296 		return (ENOMEM);
297 
298 	/* Ensure alignment is at least a page and fits in a pointer. */
299 	align = flags & MAP_ALIGNMENT_MASK;
300 	if (align != 0 && align != MAP_ALIGNED_SUPER) {
301 		if (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY)
302 			return (EXTERROR(EINVAL, "bad alignment %#jx >= %#jx",
303 			    align >> MAP_ALIGNMENT_SHIFT,
304 			    sizeof(void *) * NBBY));
305 		else if (align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)
306 			return (EXTERROR(EINVAL, "bad alignment %#jx < %#jx",
307 			    align >> MAP_ALIGNMENT_SHIFT, PAGE_SHIFT));
308 	}
309 
310 	/*
311 	 * Check for illegal addresses.  Watch out for address wrap... Note
312 	 * that VM_*_ADDRESS are not constants due to casts (argh).
313 	 */
314 	if (flags & MAP_FIXED) {
315 		/*
316 		 * The specified address must have the same remainder
317 		 * as the file offset taken modulo PAGE_SIZE, so it
318 		 * should be aligned after adjustment by pageoff.
319 		 */
320 		addr -= pageoff;
321 		if ((addr & PAGE_MASK) != 0) {
322 			return (EXTERROR(EINVAL,
323 			    "fixed mapping at %#jx not page aligned %#jx", addr,
324 			    PAGE_SIZE));
325 		}
326 
327 		/* Address range must be all in user VM space. */
328 		if (!vm_map_range_valid(&vms->vm_map, addr, addr + size)) {
329 			return (EXTERROR(EINVAL,
330 			    "mapping %#jx-%#jx outside vm_map", addr,
331 			    addr + size));
332 		}
333 		if ((flags & MAP_32BIT) && addr + size > MAP_32BIT_MAX_ADDR) {
334 			return (EXTERROR(EINVAL,
335 		    "fixed 32bit mapping of [%#jx %#jx] does not fit into 4G",
336 			    addr, addr + size));
337 		}
338 	} else if (flags & MAP_32BIT) {
339 		/*
340 		 * For MAP_32BIT, override the hint if it is too high and
341 		 * do not bother moving the mapping past the heap (since
342 		 * the heap is usually above 2GB).
343 		 */
344 		if (addr + size > MAP_32BIT_MAX_ADDR)
345 			addr = 0;
346 	} else {
347 		/*
348 		 * XXX for non-fixed mappings where no hint is provided or
349 		 * the hint would fall in the potential heap space,
350 		 * place it after the end of the largest possible heap.
351 		 *
352 		 * For anonymous mappings within the address space of the
353 		 * calling process, the absence of a hint is handled at a
354 		 * lower level in order to implement different clustering
355 		 * strategies for ASLR.
356 		 */
357 		if (((flags & MAP_ANON) == 0 && addr == 0) ||
358 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
359 		    addr < round_page((vm_offset_t)vms->vm_daddr +
360 		    lim_max(td, RLIMIT_DATA))))
361 			addr = round_page((vm_offset_t)vms->vm_daddr +
362 			    lim_max(td, RLIMIT_DATA));
363 	}
364 	if (len == 0) {
365 		/*
366 		 * Return success without mapping anything for old
367 		 * binaries that request a page-aligned mapping of
368 		 * length 0.  For modern binaries, this function
369 		 * returns an error earlier.
370 		 */
371 		error = 0;
372 	} else if ((flags & MAP_GUARD) != 0) {
373 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
374 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
375 	} else if ((flags & MAP_ANON) != 0) {
376 		/*
377 		 * Mapping blank space is trivial.
378 		 *
379 		 * This relies on VM_PROT_* matching PROT_*.
380 		 */
381 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
382 		    max_prot, flags, NULL, pos, FALSE, td);
383 	} else {
384 		/*
385 		 * Mapping file, get fp for validation and don't let the
386 		 * descriptor disappear on us if we block. Check capability
387 		 * rights, but also return the maximum rights to be combined
388 		 * with maxprot later.
389 		 */
390 		cap_rights_init_one(&rights, CAP_MMAP);
391 		if (prot & PROT_READ)
392 			cap_rights_set_one(&rights, CAP_MMAP_R);
393 		if ((flags & MAP_SHARED) != 0) {
394 			if (prot & PROT_WRITE)
395 				cap_rights_set_one(&rights, CAP_MMAP_W);
396 		}
397 		if (prot & PROT_EXEC)
398 			cap_rights_set_one(&rights, CAP_MMAP_X);
399 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
400 		if (error != 0)
401 			goto done;
402 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
403 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
404 			EXTERROR(EINVAL, "neither SHARED nor PRIVATE req");
405 			error = EINVAL;
406 			goto done;
407 		}
408 		if (check_fp_fn != NULL) {
409 			error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
410 			    flags);
411 			if (error != 0)
412 				goto done;
413 		}
414 		if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data))
415 			addr = orig_addr;
416 		/* This relies on VM_PROT_* matching PROT_*. */
417 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
418 		    max_prot & cap_maxprot, flags, pos, td);
419 	}
420 
421 	if (error == 0)
422 		td->td_retval[0] = addr + pageoff;
423 done:
424 	if (fp)
425 		fdrop(fp, td);
426 
427 	return (error);
428 }
429 
430 #if defined(COMPAT_FREEBSD6)
431 int
freebsd6_mmap(struct thread * td,struct freebsd6_mmap_args * uap)432 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
433 {
434 	return (kern_mmap(td, &(struct mmap_req){
435 		.mr_hint = (uintptr_t)uap->addr,
436 		.mr_len = uap->len,
437 		.mr_prot = uap->prot,
438 		.mr_flags = uap->flags,
439 		.mr_fd = uap->fd,
440 		.mr_pos = uap->pos,
441 	    }));
442 }
443 #endif
444 
445 #ifdef COMPAT_43
446 #ifndef _SYS_SYSPROTO_H_
447 struct ommap_args {
448 	caddr_t addr;
449 	int len;
450 	int prot;
451 	int flags;
452 	int fd;
453 	long pos;
454 };
455 #endif
456 int
ommap(struct thread * td,struct ommap_args * uap)457 ommap(struct thread *td, struct ommap_args *uap)
458 {
459 	return (kern_ommap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
460 	    uap->flags, uap->fd, uap->pos));
461 }
462 
463 int
kern_ommap(struct thread * td,uintptr_t hint,int len,int oprot,int oflags,int fd,long pos)464 kern_ommap(struct thread *td, uintptr_t hint, int len, int oprot,
465     int oflags, int fd, long pos)
466 {
467 	static const char cvtbsdprot[8] = {
468 		0,
469 		PROT_EXEC,
470 		PROT_WRITE,
471 		PROT_EXEC | PROT_WRITE,
472 		PROT_READ,
473 		PROT_EXEC | PROT_READ,
474 		PROT_WRITE | PROT_READ,
475 		PROT_EXEC | PROT_WRITE | PROT_READ,
476 	};
477 	int flags, prot;
478 
479 	if (len < 0)
480 		return (EINVAL);
481 
482 #define	OMAP_ANON	0x0002
483 #define	OMAP_COPY	0x0020
484 #define	OMAP_SHARED	0x0010
485 #define	OMAP_FIXED	0x0100
486 
487 	prot = cvtbsdprot[oprot & 0x7];
488 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
489 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
490 	    prot != 0)
491 		prot |= PROT_EXEC;
492 #endif
493 	flags = 0;
494 	if (oflags & OMAP_ANON)
495 		flags |= MAP_ANON;
496 	if (oflags & OMAP_COPY)
497 		flags |= MAP_COPY;
498 	if (oflags & OMAP_SHARED)
499 		flags |= MAP_SHARED;
500 	else
501 		flags |= MAP_PRIVATE;
502 	if (oflags & OMAP_FIXED)
503 		flags |= MAP_FIXED;
504 	return (kern_mmap(td, &(struct mmap_req){
505 		.mr_hint = hint,
506 		.mr_len = len,
507 		.mr_prot = prot,
508 		.mr_flags = flags,
509 		.mr_fd = fd,
510 		.mr_pos = pos,
511 	    }));
512 }
513 #endif				/* COMPAT_43 */
514 
515 #ifndef _SYS_SYSPROTO_H_
516 struct msync_args {
517 	void *addr;
518 	size_t len;
519 	int flags;
520 };
521 #endif
522 int
sys_msync(struct thread * td,struct msync_args * uap)523 sys_msync(struct thread *td, struct msync_args *uap)
524 {
525 
526 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
527 }
528 
529 int
kern_msync(struct thread * td,uintptr_t addr0,size_t size,int flags)530 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
531 {
532 	vm_offset_t addr;
533 	vm_size_t pageoff;
534 	vm_map_t map;
535 	int rv;
536 
537 	addr = addr0;
538 	pageoff = (addr & PAGE_MASK);
539 	addr -= pageoff;
540 	size += pageoff;
541 	size = (vm_size_t) round_page(size);
542 	if (addr + size < addr)
543 		return (EINVAL);
544 
545 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
546 		return (EINVAL);
547 
548 	map = &td->td_proc->p_vmspace->vm_map;
549 
550 	/*
551 	 * Clean the pages and interpret the return value.
552 	 */
553 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
554 	    (flags & MS_INVALIDATE) != 0);
555 	switch (rv) {
556 	case KERN_SUCCESS:
557 		return (0);
558 	case KERN_INVALID_ADDRESS:
559 		return (ENOMEM);
560 	case KERN_INVALID_ARGUMENT:
561 		return (EBUSY);
562 	case KERN_FAILURE:
563 		return (EIO);
564 	default:
565 		return (EINVAL);
566 	}
567 }
568 
569 #ifndef _SYS_SYSPROTO_H_
570 struct munmap_args {
571 	void *addr;
572 	size_t len;
573 };
574 #endif
575 int
sys_munmap(struct thread * td,struct munmap_args * uap)576 sys_munmap(struct thread *td, struct munmap_args *uap)
577 {
578 
579 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
580 }
581 
582 int
kern_munmap(struct thread * td,uintptr_t addr0,size_t size)583 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
584 {
585 #ifdef HWPMC_HOOKS
586 	struct pmckern_map_out pkm;
587 	vm_map_entry_t entry;
588 	bool pmc_handled;
589 #endif
590 	vm_offset_t addr, end;
591 	vm_size_t pageoff;
592 	vm_map_t map;
593 	int rv;
594 
595 	if (size == 0)
596 		return (EINVAL);
597 
598 	addr = addr0;
599 	pageoff = (addr & PAGE_MASK);
600 	addr -= pageoff;
601 	size += pageoff;
602 	size = (vm_size_t) round_page(size);
603 	end = addr + size;
604 	map = &td->td_proc->p_vmspace->vm_map;
605 	if (!vm_map_range_valid(map, addr, end))
606 		return (EINVAL);
607 
608 	vm_map_lock(map);
609 #ifdef HWPMC_HOOKS
610 	pmc_handled = false;
611 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
612 		pmc_handled = true;
613 		/*
614 		 * Inform hwpmc if the address range being unmapped contains
615 		 * an executable region.
616 		 */
617 		pkm.pm_address = (uintptr_t) NULL;
618 		if (vm_map_lookup_entry(map, addr, &entry)) {
619 			for (; entry->start < end;
620 			    entry = vm_map_entry_succ(entry)) {
621 				if (vm_map_check_protection(map, entry->start,
622 					entry->end, VM_PROT_EXECUTE) == TRUE) {
623 					pkm.pm_address = (uintptr_t) addr;
624 					pkm.pm_size = (size_t) size;
625 					break;
626 				}
627 			}
628 		}
629 	}
630 #endif
631 	rv = vm_map_delete(map, addr, end);
632 
633 #ifdef HWT_HOOKS
634 	if (HWT_HOOK_INSTALLED && rv == KERN_SUCCESS) {
635 		struct hwt_record_entry ent;
636 
637 		ent.addr = (uintptr_t) addr;
638 		ent.fullpath = NULL;
639 		ent.record_type = HWT_RECORD_MUNMAP;
640 		HWT_CALL_HOOK(td, HWT_RECORD, &ent);
641 	}
642 #endif
643 
644 #ifdef HWPMC_HOOKS
645 	if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) {
646 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
647 		vm_map_lock_downgrade(map);
648 		if (pkm.pm_address != (uintptr_t) NULL)
649 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
650 		vm_map_unlock_read(map);
651 	} else
652 #endif
653 		vm_map_unlock(map);
654 
655 	return (vm_mmap_to_errno(rv));
656 }
657 
658 #ifndef _SYS_SYSPROTO_H_
659 struct mprotect_args {
660 	const void *addr;
661 	size_t len;
662 	int prot;
663 };
664 #endif
665 int
sys_mprotect(struct thread * td,struct mprotect_args * uap)666 sys_mprotect(struct thread *td, struct mprotect_args *uap)
667 {
668 
669 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len,
670 	    uap->prot, 0));
671 }
672 
673 int
kern_mprotect(struct thread * td,uintptr_t addr0,size_t size,int prot,int flags)674 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot,
675     int flags)
676 {
677 	vm_offset_t addr;
678 	vm_size_t pageoff;
679 	int vm_error, max_prot;
680 
681 	addr = addr0;
682 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
683 		return (EINVAL);
684 	max_prot = PROT_MAX_EXTRACT(prot);
685 	prot = PROT_EXTRACT(prot);
686 	pageoff = (addr & PAGE_MASK);
687 	addr -= pageoff;
688 	size += pageoff;
689 	size = (vm_size_t) round_page(size);
690 #ifdef COMPAT_FREEBSD32
691 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
692 		if (((addr + size) & 0xffffffff) < addr)
693 			return (EINVAL);
694 	} else
695 #endif
696 	if (addr + size < addr)
697 		return (EINVAL);
698 
699 	flags |= VM_MAP_PROTECT_SET_PROT;
700 	if (max_prot != 0)
701 		flags |= VM_MAP_PROTECT_SET_MAXPROT;
702 	vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
703 	    addr, addr + size, prot, max_prot, flags);
704 
705 	switch (vm_error) {
706 	case KERN_SUCCESS:
707 		return (0);
708 	case KERN_PROTECTION_FAILURE:
709 		return (EACCES);
710 	case KERN_RESOURCE_SHORTAGE:
711 		return (ENOMEM);
712 	case KERN_OUT_OF_BOUNDS:
713 		return (ENOTSUP);
714 	}
715 	return (EINVAL);
716 }
717 
718 #ifndef _SYS_SYSPROTO_H_
719 struct minherit_args {
720 	void *addr;
721 	size_t len;
722 	int inherit;
723 };
724 #endif
725 int
sys_minherit(struct thread * td,struct minherit_args * uap)726 sys_minherit(struct thread *td, struct minherit_args *uap)
727 {
728 
729 	return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
730 	    uap->inherit));
731 }
732 
733 int
kern_minherit(struct thread * td,uintptr_t addr0,size_t len,int inherit0)734 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
735 {
736 	vm_offset_t addr;
737 	vm_size_t size, pageoff;
738 	vm_inherit_t inherit;
739 
740 	addr = (vm_offset_t)addr0;
741 	size = len;
742 	inherit = inherit0;
743 
744 	pageoff = (addr & PAGE_MASK);
745 	addr -= pageoff;
746 	size += pageoff;
747 	size = (vm_size_t) round_page(size);
748 	if (addr + size < addr)
749 		return (EINVAL);
750 
751 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
752 	    addr + size, inherit)) {
753 	case KERN_SUCCESS:
754 		return (0);
755 	case KERN_PROTECTION_FAILURE:
756 		return (EACCES);
757 	}
758 	return (EINVAL);
759 }
760 
761 #ifndef _SYS_SYSPROTO_H_
762 struct madvise_args {
763 	void *addr;
764 	size_t len;
765 	int behav;
766 };
767 #endif
768 
769 int
sys_madvise(struct thread * td,struct madvise_args * uap)770 sys_madvise(struct thread *td, struct madvise_args *uap)
771 {
772 
773 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
774 }
775 
776 int
kern_madvise(struct thread * td,uintptr_t addr0,size_t len,int behav)777 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
778 {
779 	vm_map_t map;
780 	vm_offset_t addr, end, start;
781 	int flags;
782 
783 	/*
784 	 * Check for our special case, advising the swap pager we are
785 	 * "immortal."
786 	 */
787 	if (behav == MADV_PROTECT) {
788 		flags = PPROT_SET;
789 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
790 		    PROC_SPROTECT, &flags));
791 	}
792 
793 	/*
794 	 * Check for illegal addresses.  Watch out for address wrap... Note
795 	 * that VM_*_ADDRESS are not constants due to casts (argh).
796 	 */
797 	map = &td->td_proc->p_vmspace->vm_map;
798 	addr = addr0;
799 	if (!vm_map_range_valid(map, addr, addr + len))
800 		return (EINVAL);
801 
802 	/*
803 	 * Since this routine is only advisory, we default to conservative
804 	 * behavior.
805 	 */
806 	start = trunc_page(addr);
807 	end = round_page(addr + len);
808 
809 	/*
810 	 * vm_map_madvise() checks for illegal values of behav.
811 	 */
812 	return (vm_map_madvise(map, start, end, behav));
813 }
814 
815 #ifndef _SYS_SYSPROTO_H_
816 struct mincore_args {
817 	const void *addr;
818 	size_t len;
819 	char *vec;
820 };
821 #endif
822 
823 int
sys_mincore(struct thread * td,struct mincore_args * uap)824 sys_mincore(struct thread *td, struct mincore_args *uap)
825 {
826 
827 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
828 }
829 
830 int
kern_mincore(struct thread * td,uintptr_t addr0,size_t len,char * vec)831 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
832 {
833 	pmap_t pmap;
834 	vm_map_t map;
835 	vm_map_entry_t current, entry;
836 	vm_object_t object;
837 	vm_offset_t addr, cend, end, first_addr;
838 	vm_paddr_t pa;
839 	vm_page_t m;
840 	vm_pindex_t pindex;
841 	int error, lastvecindex, mincoreinfo, vecindex;
842 	unsigned int timestamp;
843 
844 	/*
845 	 * Make sure that the addresses presented are valid for user
846 	 * mode.
847 	 */
848 	first_addr = addr = trunc_page(addr0);
849 	end = round_page(addr0 + len);
850 	map = &td->td_proc->p_vmspace->vm_map;
851 	if (end > vm_map_max(map) || end < addr)
852 		return (ENOMEM);
853 
854 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
855 
856 	vm_map_lock_read(map);
857 RestartScan:
858 	timestamp = map->timestamp;
859 
860 	if (!vm_map_lookup_entry(map, addr, &entry)) {
861 		vm_map_unlock_read(map);
862 		return (ENOMEM);
863 	}
864 
865 	/*
866 	 * Do this on a map entry basis so that if the pages are not
867 	 * in the current processes address space, we can easily look
868 	 * up the pages elsewhere.
869 	 */
870 	lastvecindex = -1;
871 	while (entry->start < end) {
872 		/*
873 		 * check for contiguity
874 		 */
875 		current = entry;
876 		entry = vm_map_entry_succ(current);
877 		if (current->end < end &&
878 		    entry->start > current->end) {
879 			vm_map_unlock_read(map);
880 			return (ENOMEM);
881 		}
882 
883 		/*
884 		 * ignore submaps (for now) or null objects
885 		 */
886 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
887 		    current->object.vm_object == NULL)
888 			continue;
889 
890 		/*
891 		 * limit this scan to the current map entry and the
892 		 * limits for the mincore call
893 		 */
894 		if (addr < current->start)
895 			addr = current->start;
896 		cend = current->end;
897 		if (cend > end)
898 			cend = end;
899 
900 		for (; addr < cend; addr += PAGE_SIZE) {
901 			/*
902 			 * Check pmap first, it is likely faster, also
903 			 * it can provide info as to whether we are the
904 			 * one referencing or modifying the page.
905 			 */
906 			m = NULL;
907 			object = NULL;
908 retry:
909 			pa = 0;
910 			mincoreinfo = pmap_mincore(pmap, addr, &pa);
911 			if (mincore_mapped) {
912 				/*
913 				 * We only care about this pmap's
914 				 * mapping of the page, if any.
915 				 */
916 				;
917 			} else if (pa != 0) {
918 				/*
919 				 * The page is mapped by this process but not
920 				 * both accessed and modified.  It is also
921 				 * managed.  Acquire the object lock so that
922 				 * other mappings might be examined.  The page's
923 				 * identity may change at any point before its
924 				 * object lock is acquired, so re-validate if
925 				 * necessary.
926 				 */
927 				m = PHYS_TO_VM_PAGE(pa);
928 				while (object == NULL || m->object != object) {
929 					if (object != NULL)
930 						VM_OBJECT_WUNLOCK(object);
931 					object = atomic_load_ptr(&m->object);
932 					if (object == NULL)
933 						goto retry;
934 					VM_OBJECT_WLOCK(object);
935 				}
936 				if (pa != pmap_extract(pmap, addr))
937 					goto retry;
938 				KASSERT(vm_page_all_valid(m),
939 				    ("mincore: page %p is mapped but invalid",
940 				    m));
941 			} else if (mincoreinfo == 0) {
942 				/*
943 				 * The page is not mapped by this process.  If
944 				 * the object implements managed pages, then
945 				 * determine if the page is resident so that
946 				 * the mappings might be examined.
947 				 */
948 				if (current->object.vm_object != object) {
949 					if (object != NULL)
950 						VM_OBJECT_WUNLOCK(object);
951 					object = current->object.vm_object;
952 					VM_OBJECT_WLOCK(object);
953 				}
954 				if ((object->flags & OBJ_SWAP) != 0 ||
955 				    object->type == OBJT_VNODE) {
956 					pindex = OFF_TO_IDX(current->offset +
957 					    (addr - current->start));
958 					m = vm_page_lookup(object, pindex);
959 					if (m != NULL && vm_page_none_valid(m))
960 						m = NULL;
961 					if (m != NULL)
962 						mincoreinfo = MINCORE_INCORE;
963 				}
964 			}
965 			if (m != NULL) {
966 				VM_OBJECT_ASSERT_WLOCKED(m->object);
967 
968 				/* Examine other mappings of the page. */
969 				if (m->dirty == 0 && pmap_is_modified(m))
970 					vm_page_dirty(m);
971 				if (m->dirty != 0)
972 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
973 
974 				/*
975 				 * The first test for PGA_REFERENCED is an
976 				 * optimization.  The second test is
977 				 * required because a concurrent pmap
978 				 * operation could clear the last reference
979 				 * and set PGA_REFERENCED before the call to
980 				 * pmap_is_referenced().
981 				 */
982 				if ((m->a.flags & PGA_REFERENCED) != 0 ||
983 				    pmap_is_referenced(m) ||
984 				    (m->a.flags & PGA_REFERENCED) != 0)
985 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
986 			}
987 			if (object != NULL)
988 				VM_OBJECT_WUNLOCK(object);
989 
990 			/*
991 			 * subyte may page fault.  In case it needs to modify
992 			 * the map, we release the lock.
993 			 */
994 			vm_map_unlock_read(map);
995 
996 			/*
997 			 * calculate index into user supplied byte vector
998 			 */
999 			vecindex = atop(addr - first_addr);
1000 
1001 			/*
1002 			 * If we have skipped map entries, we need to make sure that
1003 			 * the byte vector is zeroed for those skipped entries.
1004 			 */
1005 			while ((lastvecindex + 1) < vecindex) {
1006 				++lastvecindex;
1007 				error = subyte(vec + lastvecindex, 0);
1008 				if (error) {
1009 					error = EFAULT;
1010 					goto done2;
1011 				}
1012 			}
1013 
1014 			/*
1015 			 * Pass the page information to the user
1016 			 */
1017 			error = subyte(vec + vecindex, mincoreinfo);
1018 			if (error) {
1019 				error = EFAULT;
1020 				goto done2;
1021 			}
1022 
1023 			/*
1024 			 * If the map has changed, due to the subyte, the previous
1025 			 * output may be invalid.
1026 			 */
1027 			vm_map_lock_read(map);
1028 			if (timestamp != map->timestamp)
1029 				goto RestartScan;
1030 
1031 			lastvecindex = vecindex;
1032 		}
1033 	}
1034 
1035 	/*
1036 	 * subyte may page fault.  In case it needs to modify
1037 	 * the map, we release the lock.
1038 	 */
1039 	vm_map_unlock_read(map);
1040 
1041 	/*
1042 	 * Zero the last entries in the byte vector.
1043 	 */
1044 	vecindex = atop(end - first_addr);
1045 	while ((lastvecindex + 1) < vecindex) {
1046 		++lastvecindex;
1047 		error = subyte(vec + lastvecindex, 0);
1048 		if (error) {
1049 			error = EFAULT;
1050 			goto done2;
1051 		}
1052 	}
1053 
1054 	/*
1055 	 * If the map has changed, due to the subyte, the previous
1056 	 * output may be invalid.
1057 	 */
1058 	vm_map_lock_read(map);
1059 	if (timestamp != map->timestamp)
1060 		goto RestartScan;
1061 	vm_map_unlock_read(map);
1062 done2:
1063 	return (error);
1064 }
1065 
1066 #ifndef _SYS_SYSPROTO_H_
1067 struct mlock_args {
1068 	const void *addr;
1069 	size_t len;
1070 };
1071 #endif
1072 int
sys_mlock(struct thread * td,struct mlock_args * uap)1073 sys_mlock(struct thread *td, struct mlock_args *uap)
1074 {
1075 
1076 	return (kern_mlock(td->td_proc, td->td_ucred,
1077 	    __DECONST(uintptr_t, uap->addr), uap->len));
1078 }
1079 
1080 int
kern_mlock(struct proc * proc,struct ucred * cred,uintptr_t addr0,size_t len)1081 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1082 {
1083 	vm_offset_t addr, end, last, start;
1084 	vm_size_t npages, size;
1085 	vm_map_t map;
1086 	unsigned long nsize;
1087 	int error;
1088 
1089 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1090 	if (error)
1091 		return (error);
1092 	addr = addr0;
1093 	size = len;
1094 	last = addr + size;
1095 	start = trunc_page(addr);
1096 	end = round_page(last);
1097 	if (last < addr || end < addr)
1098 		return (EINVAL);
1099 	npages = atop(end - start);
1100 	if (npages > vm_page_max_user_wired)
1101 		return (ENOMEM);
1102 	map = &proc->p_vmspace->vm_map;
1103 	PROC_LOCK(proc);
1104 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1105 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1106 		PROC_UNLOCK(proc);
1107 		return (ENOMEM);
1108 	}
1109 	PROC_UNLOCK(proc);
1110 #ifdef RACCT
1111 	if (racct_enable) {
1112 		PROC_LOCK(proc);
1113 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1114 		PROC_UNLOCK(proc);
1115 		if (error != 0)
1116 			return (ENOMEM);
1117 	}
1118 #endif
1119 	error = vm_map_wire(map, start, end,
1120 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1121 #ifdef RACCT
1122 	if (racct_enable && error != KERN_SUCCESS) {
1123 		PROC_LOCK(proc);
1124 		racct_set(proc, RACCT_MEMLOCK,
1125 		    ptoa(pmap_wired_count(map->pmap)));
1126 		PROC_UNLOCK(proc);
1127 	}
1128 #endif
1129 	switch (error) {
1130 	case KERN_SUCCESS:
1131 		return (0);
1132 	case KERN_INVALID_ARGUMENT:
1133 		return (EINVAL);
1134 	default:
1135 		return (ENOMEM);
1136 	}
1137 }
1138 
1139 #ifndef _SYS_SYSPROTO_H_
1140 struct mlockall_args {
1141 	int	how;
1142 };
1143 #endif
1144 
1145 int
sys_mlockall(struct thread * td,struct mlockall_args * uap)1146 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1147 {
1148 	vm_map_t map;
1149 	int error;
1150 
1151 	map = &td->td_proc->p_vmspace->vm_map;
1152 	error = priv_check(td, PRIV_VM_MLOCK);
1153 	if (error)
1154 		return (error);
1155 
1156 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1157 		return (EINVAL);
1158 
1159 	/*
1160 	 * If wiring all pages in the process would cause it to exceed
1161 	 * a hard resource limit, return ENOMEM.
1162 	 */
1163 	if (!old_mlock && uap->how & MCL_CURRENT) {
1164 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1165 			return (ENOMEM);
1166 	}
1167 #ifdef RACCT
1168 	if (racct_enable) {
1169 		PROC_LOCK(td->td_proc);
1170 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1171 		PROC_UNLOCK(td->td_proc);
1172 		if (error != 0)
1173 			return (ENOMEM);
1174 	}
1175 #endif
1176 
1177 	if (uap->how & MCL_FUTURE) {
1178 		vm_map_lock(map);
1179 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1180 		vm_map_unlock(map);
1181 		error = 0;
1182 	}
1183 
1184 	if (uap->how & MCL_CURRENT) {
1185 		/*
1186 		 * P1003.1-2001 mandates that all currently mapped pages
1187 		 * will be memory resident and locked (wired) upon return
1188 		 * from mlockall(). vm_map_wire() will wire pages, by
1189 		 * calling vm_fault_wire() for each page in the region.
1190 		 */
1191 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1192 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1193 		if (error == KERN_SUCCESS)
1194 			error = 0;
1195 		else if (error == KERN_RESOURCE_SHORTAGE)
1196 			error = ENOMEM;
1197 		else
1198 			error = EAGAIN;
1199 	}
1200 #ifdef RACCT
1201 	if (racct_enable && error != KERN_SUCCESS) {
1202 		PROC_LOCK(td->td_proc);
1203 		racct_set(td->td_proc, RACCT_MEMLOCK,
1204 		    ptoa(pmap_wired_count(map->pmap)));
1205 		PROC_UNLOCK(td->td_proc);
1206 	}
1207 #endif
1208 
1209 	return (error);
1210 }
1211 
1212 #ifndef _SYS_SYSPROTO_H_
1213 struct munlockall_args {
1214 	register_t dummy;
1215 };
1216 #endif
1217 
1218 int
sys_munlockall(struct thread * td,struct munlockall_args * uap)1219 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1220 {
1221 	vm_map_t map;
1222 	int error;
1223 
1224 	map = &td->td_proc->p_vmspace->vm_map;
1225 	error = priv_check(td, PRIV_VM_MUNLOCK);
1226 	if (error)
1227 		return (error);
1228 
1229 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1230 	vm_map_lock(map);
1231 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1232 	vm_map_unlock(map);
1233 
1234 	/* Forcibly unwire all pages. */
1235 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1236 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1237 #ifdef RACCT
1238 	if (racct_enable && error == KERN_SUCCESS) {
1239 		PROC_LOCK(td->td_proc);
1240 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1241 		PROC_UNLOCK(td->td_proc);
1242 	}
1243 #endif
1244 
1245 	return (error);
1246 }
1247 
1248 #ifndef _SYS_SYSPROTO_H_
1249 struct munlock_args {
1250 	const void *addr;
1251 	size_t len;
1252 };
1253 #endif
1254 int
sys_munlock(struct thread * td,struct munlock_args * uap)1255 sys_munlock(struct thread *td, struct munlock_args *uap)
1256 {
1257 
1258 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1259 }
1260 
1261 int
kern_munlock(struct thread * td,uintptr_t addr0,size_t size)1262 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1263 {
1264 	vm_offset_t addr, end, last, start;
1265 #ifdef RACCT
1266 	vm_map_t map;
1267 #endif
1268 	int error;
1269 
1270 	error = priv_check(td, PRIV_VM_MUNLOCK);
1271 	if (error)
1272 		return (error);
1273 	addr = addr0;
1274 	last = addr + size;
1275 	start = trunc_page(addr);
1276 	end = round_page(last);
1277 	if (last < addr || end < addr)
1278 		return (EINVAL);
1279 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1280 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1281 #ifdef RACCT
1282 	if (racct_enable && error == KERN_SUCCESS) {
1283 		PROC_LOCK(td->td_proc);
1284 		map = &td->td_proc->p_vmspace->vm_map;
1285 		racct_set(td->td_proc, RACCT_MEMLOCK,
1286 		    ptoa(pmap_wired_count(map->pmap)));
1287 		PROC_UNLOCK(td->td_proc);
1288 	}
1289 #endif
1290 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1291 }
1292 
1293 /*
1294  * vm_mmap_vnode()
1295  *
1296  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1297  * operations on vnodes.
1298  */
1299 int
vm_mmap_vnode(struct thread * td,vm_size_t objsize,vm_prot_t prot,vm_prot_t * maxprotp,int * flagsp,struct vnode * vp,vm_ooffset_t * foffp,vm_object_t * objp,boolean_t * writecounted)1300 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1301     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1302     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1303     boolean_t *writecounted)
1304 {
1305 	struct vattr va;
1306 	vm_object_t obj;
1307 	vm_ooffset_t foff;
1308 	struct ucred *cred;
1309 	int error, flags;
1310 	bool writex;
1311 
1312 	cred = td->td_ucred;
1313 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1314 	    (*flagsp & MAP_SHARED) != 0;
1315 	if ((error = vget(vp, LK_SHARED)) != 0)
1316 		return (error);
1317 	AUDIT_ARG_VNODE1(vp);
1318 	foff = *foffp;
1319 	flags = *flagsp;
1320 	obj = vp->v_object;
1321 	if (vp->v_type == VREG) {
1322 		/*
1323 		 * Get the proper underlying object
1324 		 */
1325 		if (obj == NULL) {
1326 			error = EINVAL;
1327 			goto done;
1328 		}
1329 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1330 			vput(vp);
1331 			vp = (struct vnode *)obj->handle;
1332 			/*
1333 			 * Bypass filesystems obey the mpsafety of the
1334 			 * underlying fs.  Tmpfs never bypasses.
1335 			 */
1336 			error = vget(vp, LK_SHARED);
1337 			if (error != 0)
1338 				return (error);
1339 		}
1340 		if (writex) {
1341 			*writecounted = TRUE;
1342 			vm_pager_update_writecount(obj, 0, objsize);
1343 		}
1344 	} else {
1345 		error = EXTERROR(EINVAL, "non-reg file");
1346 		goto done;
1347 	}
1348 	if ((error = VOP_GETATTR(vp, &va, cred)))
1349 		goto done;
1350 #ifdef MAC
1351 	/* This relies on VM_PROT_* matching PROT_*. */
1352 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1353 	if (error != 0)
1354 		goto done;
1355 #endif
1356 	if ((flags & MAP_SHARED) != 0) {
1357 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1358 			if (prot & VM_PROT_WRITE) {
1359 				error = EPERM;
1360 				goto done;
1361 			}
1362 			*maxprotp &= ~VM_PROT_WRITE;
1363 		}
1364 	}
1365 	/*
1366 	 * If it is a regular file without any references
1367 	 * we do not need to sync it.
1368 	 * Adjust object size to be the size of actual file.
1369 	 */
1370 	objsize = round_page(va.va_size);
1371 	if (va.va_nlink == 0)
1372 		flags |= MAP_NOSYNC;
1373 	if (obj->type == OBJT_VNODE) {
1374 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1375 		    cred);
1376 		if (obj == NULL) {
1377 			error = ENOMEM;
1378 			goto done;
1379 		}
1380 	} else {
1381 		KASSERT((obj->flags & OBJ_SWAP) != 0, ("wrong object type"));
1382 		vm_object_reference(obj);
1383 #if VM_NRESERVLEVEL > 0
1384 		if ((obj->flags & OBJ_COLORED) == 0) {
1385 			VM_OBJECT_WLOCK(obj);
1386 			vm_object_color(obj, 0);
1387 			VM_OBJECT_WUNLOCK(obj);
1388 		}
1389 #endif
1390 	}
1391 	*objp = obj;
1392 	*flagsp = flags;
1393 
1394 	VOP_MMAPPED(vp);
1395 
1396 done:
1397 	if (error != 0 && *writecounted) {
1398 		*writecounted = FALSE;
1399 		vm_pager_update_writecount(obj, objsize, 0);
1400 	}
1401 	vput(vp);
1402 	return (error);
1403 }
1404 
1405 /*
1406  * vm_mmap_cdev()
1407  *
1408  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1409  * operations on cdevs.
1410  */
1411 int
vm_mmap_cdev(struct thread * td,vm_size_t objsize,vm_prot_t prot,vm_prot_t * maxprotp,int * flagsp,struct cdev * cdev,struct cdevsw * dsw,vm_ooffset_t * foff,vm_object_t * objp)1412 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1413     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1414     vm_ooffset_t *foff, vm_object_t *objp)
1415 {
1416 	vm_object_t obj;
1417 	int error, flags;
1418 
1419 	flags = *flagsp;
1420 
1421 	if (dsw->d_flags & D_MMAP_ANON) {
1422 		*objp = NULL;
1423 		*foff = 0;
1424 		*maxprotp = VM_PROT_ALL;
1425 		*flagsp |= MAP_ANON;
1426 		return (0);
1427 	}
1428 
1429 	/*
1430 	 * cdevs do not provide private mappings of any kind.
1431 	 */
1432 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1433 	    (prot & VM_PROT_WRITE) != 0)
1434 		return (EACCES);
1435 	if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0) {
1436 		return (EXTERROR(EINVAL, "cdev mapping must be shared"));
1437 	}
1438 
1439 	/*
1440 	 * Force device mappings to be shared.
1441 	 */
1442 	flags |= MAP_SHARED;
1443 #ifdef MAC_XXX
1444 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1445 	if (error != 0)
1446 		return (error);
1447 #endif
1448 	/*
1449 	 * First, try d_mmap_single().  If that is not implemented
1450 	 * (returns ENODEV), fall back to using the device pager.
1451 	 * Note that d_mmap_single() must return a reference to the
1452 	 * object (it needs to bump the reference count of the object
1453 	 * it returns somehow).
1454 	 *
1455 	 * XXX assumes VM_PROT_* == PROT_*
1456 	 */
1457 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1458 	if (error != ENODEV)
1459 		return (error);
1460 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1461 	    td->td_ucred);
1462 	if (obj == NULL) {
1463 		return (EXTERROR(EINVAL,
1464 		    "cdev driver does not support mmap"));
1465 	}
1466 	*objp = obj;
1467 	*flagsp = flags;
1468 	return (0);
1469 }
1470 
1471 int
vm_mmap(vm_map_t map,vm_offset_t * addr,vm_size_t size,vm_prot_t prot,vm_prot_t maxprot,int flags,objtype_t handle_type,void * handle,vm_ooffset_t foff)1472 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1473 	vm_prot_t maxprot, int flags,
1474 	objtype_t handle_type, void *handle,
1475 	vm_ooffset_t foff)
1476 {
1477 	vm_object_t object;
1478 	struct thread *td = curthread;
1479 	int error;
1480 	boolean_t writecounted;
1481 
1482 	if (size == 0) {
1483 		return (EXTERROR(EINVAL, "zero-sized req"));
1484 	}
1485 
1486 	size = round_page(size);
1487 	object = NULL;
1488 	writecounted = FALSE;
1489 
1490 	switch (handle_type) {
1491 	case OBJT_DEVICE: {
1492 		struct cdevsw *dsw;
1493 		struct cdev *cdev;
1494 		int ref;
1495 
1496 		cdev = handle;
1497 		dsw = dev_refthread(cdev, &ref);
1498 		if (dsw == NULL)
1499 			return (ENXIO);
1500 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1501 		    dsw, &foff, &object);
1502 		dev_relthread(cdev, ref);
1503 		break;
1504 	}
1505 	case OBJT_VNODE:
1506 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1507 		    handle, &foff, &object, &writecounted);
1508 		break;
1509 	default:
1510 		error = EXTERROR(EINVAL, "unsupported backing obj type %jd",
1511 		    handle_type);
1512 		break;
1513 	}
1514 	if (error)
1515 		return (error);
1516 
1517 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1518 	    foff, writecounted, td);
1519 	if (error != 0 && object != NULL) {
1520 		/*
1521 		 * If this mapping was accounted for in the vnode's
1522 		 * writecount, then undo that now.
1523 		 */
1524 		if (writecounted)
1525 			vm_pager_release_writecount(object, 0, size);
1526 		vm_object_deallocate(object);
1527 	}
1528 	return (error);
1529 }
1530 
1531 int
kern_mmap_racct_check(struct thread * td,vm_map_t map,vm_size_t size)1532 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size)
1533 {
1534 	int error;
1535 
1536 	RACCT_PROC_LOCK(td->td_proc);
1537 	if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1538 		RACCT_PROC_UNLOCK(td->td_proc);
1539 		return (ENOMEM);
1540 	}
1541 	if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1542 		RACCT_PROC_UNLOCK(td->td_proc);
1543 		return (ENOMEM);
1544 	}
1545 	if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1546 		if (ptoa(pmap_wired_count(map->pmap)) + size >
1547 		    lim_cur(td, RLIMIT_MEMLOCK)) {
1548 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1549 			RACCT_PROC_UNLOCK(td->td_proc);
1550 			return (ENOMEM);
1551 		}
1552 		error = racct_set(td->td_proc, RACCT_MEMLOCK,
1553 		    ptoa(pmap_wired_count(map->pmap)) + size);
1554 		if (error != 0) {
1555 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1556 			RACCT_PROC_UNLOCK(td->td_proc);
1557 			return (error);
1558 		}
1559 	}
1560 	RACCT_PROC_UNLOCK(td->td_proc);
1561 	return (0);
1562 }
1563 
1564 /*
1565  * Internal version of mmap that maps a specific VM object into an
1566  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1567  */
1568 int
vm_mmap_object(vm_map_t map,vm_offset_t * addr,vm_size_t size,vm_prot_t prot,vm_prot_t maxprot,int flags,vm_object_t object,vm_ooffset_t foff,boolean_t writecounted,struct thread * td)1569 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1570     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1571     boolean_t writecounted, struct thread *td)
1572 {
1573 	vm_offset_t default_addr, max_addr;
1574 	int docow, error, findspace, rv;
1575 	bool curmap, fitit;
1576 
1577 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1578 	if (curmap) {
1579 		error = kern_mmap_racct_check(td, map, size);
1580 		if (error != 0)
1581 			return (error);
1582 	}
1583 
1584 	/*
1585 	 * We currently can only deal with page aligned file offsets.
1586 	 * The mmap() system call already enforces this by subtracting
1587 	 * the page offset from the file offset, but checking here
1588 	 * catches errors in device drivers (e.g. d_single_mmap()
1589 	 * callbacks) and other internal mapping requests (such as in
1590 	 * exec).
1591 	 */
1592 	if ((foff & PAGE_MASK) != 0) {
1593 		return (EXTERROR(EINVAL, "offset %#jx not page-aligned", foff));
1594 	}
1595 
1596 	if ((flags & MAP_FIXED) == 0) {
1597 		fitit = true;
1598 		*addr = round_page(*addr);
1599 	} else {
1600 		if (*addr != trunc_page(*addr)) {
1601 			return (EXTERROR(EINVAL,
1602 			    "non-fixed mapping address %#jx not aligned",
1603 			    *addr));
1604 		}
1605 		fitit = false;
1606 	}
1607 
1608 	if (flags & MAP_ANON) {
1609 		if (object != NULL) {
1610 			return (EXTERROR(EINVAL,
1611 			    "anon mapping backed by an object"));
1612 		}
1613 		if (foff != 0) {
1614 			return (EXTERROR(EINVAL,
1615 			    "anon mapping with non-zero offset %#jx", foff));
1616 		}
1617 		docow = 0;
1618 	} else if (flags & MAP_PREFAULT_READ)
1619 		docow = MAP_PREFAULT;
1620 	else
1621 		docow = MAP_PREFAULT_PARTIAL;
1622 
1623 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1624 		docow |= MAP_COPY_ON_WRITE;
1625 	if (flags & MAP_NOSYNC)
1626 		docow |= MAP_DISABLE_SYNCER;
1627 	if (flags & MAP_NOCORE)
1628 		docow |= MAP_DISABLE_COREDUMP;
1629 	/* Shared memory is also shared with children. */
1630 	if (flags & MAP_SHARED)
1631 		docow |= MAP_INHERIT_SHARE;
1632 	if (writecounted)
1633 		docow |= MAP_WRITECOUNT;
1634 	if (flags & MAP_STACK) {
1635 		if (object != NULL) {
1636 			return (EXTERROR(EINVAL,
1637 			    "stack mapping backed by an object"));
1638 		}
1639 		docow |= MAP_STACK_AREA;
1640 	}
1641 	if ((flags & MAP_EXCL) != 0)
1642 		docow |= MAP_CHECK_EXCL;
1643 	if ((flags & MAP_GUARD) != 0)
1644 		docow |= MAP_CREATE_GUARD;
1645 
1646 	if (fitit) {
1647 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1648 			findspace = VMFS_SUPER_SPACE;
1649 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1650 			findspace = VMFS_ALIGNED_SPACE(flags >>
1651 			    MAP_ALIGNMENT_SHIFT);
1652 		else
1653 			findspace = VMFS_OPTIMAL_SPACE;
1654 		max_addr = 0;
1655 		if ((flags & MAP_32BIT) != 0)
1656 			max_addr = MAP_32BIT_MAX_ADDR;
1657 		if (curmap) {
1658 			default_addr =
1659 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1660 			    vm_daddr + lim_max(td, RLIMIT_DATA));
1661 			if ((flags & MAP_32BIT) != 0)
1662 				default_addr = 0;
1663 			rv = vm_map_find_min(map, object, foff, addr, size,
1664 			    default_addr, max_addr, findspace, prot, maxprot,
1665 			    docow);
1666 		} else {
1667 			rv = vm_map_find(map, object, foff, addr, size,
1668 			    max_addr, findspace, prot, maxprot, docow);
1669 		}
1670 	} else {
1671 		rv = vm_map_fixed(map, object, foff, *addr, size,
1672 		    prot, maxprot, docow);
1673 	}
1674 
1675 	if (rv == KERN_SUCCESS) {
1676 		/*
1677 		 * If the process has requested that all future mappings
1678 		 * be wired, then heed this.
1679 		 */
1680 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1681 			vm_map_lock(map);
1682 			if ((map->flags & MAP_WIREFUTURE) != 0)
1683 				(void)vm_map_wire_locked(map, *addr,
1684 				    *addr + size, VM_MAP_WIRE_USER |
1685 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1686 				    VM_MAP_WIRE_NOHOLES));
1687 			vm_map_unlock(map);
1688 		}
1689 	}
1690 	return (vm_mmap_to_errno(rv));
1691 }
1692 
1693 /*
1694  * Translate a Mach VM return code to zero on success or the appropriate errno
1695  * on failure.
1696  */
1697 int
vm_mmap_to_errno(int rv)1698 vm_mmap_to_errno(int rv)
1699 {
1700 	int error;
1701 
1702 	switch (rv) {
1703 	case KERN_SUCCESS:
1704 		return (0);
1705 	case KERN_INVALID_ADDRESS:
1706 	case KERN_NO_SPACE:
1707 		error = ENOMEM;
1708 		break;
1709 	case KERN_PROTECTION_FAILURE:
1710 		error = EACCES;
1711 		break;
1712 	default:
1713 		error = EINVAL;
1714 		break;
1715 	}
1716 	if ((curthread->td_pflags2 & (TDP2_UEXTERR | TDP2_EXTERR)) ==
1717 	    TDP2_UEXTERR)
1718 		EXTERROR(error, "mach error %jd", rv);
1719 	return (error);
1720 }
1721