xref: /freebsd/sys/vm/vm_mmap.c (revision df114daef4c48548c3c2b86717612761185ae18f)
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"));
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, "prot is not subset of max_prot"));
206 	}
207 
208 	p = td->td_proc;
209 
210 	/*
211 	 * Always honor PROT_MAX if set.  If not, default to all
212 	 * permissions unless we're implying maximum permissions.
213 	 */
214 	if (max_prot == 0)
215 		max_prot = kern_mmap_maxprot(p, prot);
216 
217 	vms = p->p_vmspace;
218 	fp = NULL;
219 	AUDIT_ARG_FD(fd);
220 
221 	/*
222 	 * Ignore old flags that used to be defined but did not do anything.
223 	 */
224 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
225 
226 	/*
227 	 * Enforce the constraints.
228 	 * Mapping of length 0 is only allowed for old binaries.
229 	 * Anonymous mapping shall specify -1 as filedescriptor and
230 	 * zero position for new code. Be nice to ancient a.out
231 	 * binaries and correct pos for anonymous mapping, since old
232 	 * ld.so sometimes issues anonymous map requests with non-zero
233 	 * pos.
234 	 */
235 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
236 		if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
237 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0))) {
238 			return (EXTERROR(EINVAL,
239 			    "offset not zero/fd not -1 for MAP_ANON",
240 			    fd, pos));
241 		}
242 	} else {
243 		if ((flags & MAP_ANON) != 0)
244 			pos = 0;
245 	}
246 
247 	if (flags & MAP_STACK) {
248 		if ((fd != -1) || ((prot & (PROT_READ | PROT_WRITE)) !=
249 		    (PROT_READ | PROT_WRITE))) {
250 			return (EXTERROR(EINVAL, "MAP_STACK with prot < rw",
251 			    prot));
252 		}
253 		flags |= MAP_ANON;
254 		pos = 0;
255 	}
256 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
257 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
258 	    MAP_PREFAULT_READ | MAP_GUARD | MAP_32BIT |
259 	    MAP_ALIGNMENT_MASK)) != 0) {
260 		return (EXTERROR(EINVAL, "reserved flag set"));
261 	}
262 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) {
263 		return (EXTERROR(EINVAL, "EXCL without FIXED"));
264 	}
265 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED |
266 	    MAP_PRIVATE)) {
267 		return (EXTERROR(EINVAL, "both SHARED and PRIVATE set"));
268 	}
269 	if (prot != PROT_NONE &&
270 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) {
271 		return (EXTERROR(EINVAL, "invalid prot", prot));
272 	}
273 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
274 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
275 	    MAP_32BIT | MAP_ALIGNMENT_MASK)) != 0)) {
276 		return (EXTERROR(EINVAL, "GUARD with wrong parameters"));
277 	}
278 
279 	/*
280 	 * Align the file position to a page boundary,
281 	 * and save its page offset component.
282 	 */
283 	pageoff = (pos & PAGE_MASK);
284 	pos -= pageoff;
285 
286 	/* Compute size from len by rounding (on both ends). */
287 	size = len + pageoff;			/* low end... */
288 	size = round_page(size);		/* hi end */
289 	/* Check for rounding up to zero. */
290 	if (len > size)
291 		return (ENOMEM);
292 
293 	/* Ensure alignment is at least a page and fits in a pointer. */
294 	align = flags & MAP_ALIGNMENT_MASK;
295 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
296 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
297 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) {
298 		return (EXTERROR(EINVAL, "bad alignment", align));
299 	}
300 
301 	/*
302 	 * Check for illegal addresses.  Watch out for address wrap... Note
303 	 * that VM_*_ADDRESS are not constants due to casts (argh).
304 	 */
305 	if (flags & MAP_FIXED) {
306 		/*
307 		 * The specified address must have the same remainder
308 		 * as the file offset taken modulo PAGE_SIZE, so it
309 		 * should be aligned after adjustment by pageoff.
310 		 */
311 		addr -= pageoff;
312 		if ((addr & PAGE_MASK) != 0) {
313 			return (EXTERROR(EINVAL, "fixed mapping not aligned",
314 			    addr));
315 		}
316 
317 		/* Address range must be all in user VM space. */
318 		if (!vm_map_range_valid(&vms->vm_map, addr, addr + size)) {
319 			EXTERROR(EINVAL, "mapping outside vm_map");
320 			return (EINVAL);
321 		}
322 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) {
323 			return (EXTERROR(EINVAL,
324 			    "fixed 32bit mapping does not fit into 4G"));
325 		}
326 	} else if (flags & MAP_32BIT) {
327 		/*
328 		 * For MAP_32BIT, override the hint if it is too high and
329 		 * do not bother moving the mapping past the heap (since
330 		 * the heap is usually above 2GB).
331 		 */
332 		if (addr + size > MAP_32BIT_MAX_ADDR)
333 			addr = 0;
334 	} else {
335 		/*
336 		 * XXX for non-fixed mappings where no hint is provided or
337 		 * the hint would fall in the potential heap space,
338 		 * place it after the end of the largest possible heap.
339 		 *
340 		 * For anonymous mappings within the address space of the
341 		 * calling process, the absence of a hint is handled at a
342 		 * lower level in order to implement different clustering
343 		 * strategies for ASLR.
344 		 */
345 		if (((flags & MAP_ANON) == 0 && addr == 0) ||
346 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
347 		    addr < round_page((vm_offset_t)vms->vm_daddr +
348 		    lim_max(td, RLIMIT_DATA))))
349 			addr = round_page((vm_offset_t)vms->vm_daddr +
350 			    lim_max(td, RLIMIT_DATA));
351 	}
352 	if (len == 0) {
353 		/*
354 		 * Return success without mapping anything for old
355 		 * binaries that request a page-aligned mapping of
356 		 * length 0.  For modern binaries, this function
357 		 * returns an error earlier.
358 		 */
359 		error = 0;
360 	} else if ((flags & MAP_GUARD) != 0) {
361 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
362 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
363 	} else if ((flags & MAP_ANON) != 0) {
364 		/*
365 		 * Mapping blank space is trivial.
366 		 *
367 		 * This relies on VM_PROT_* matching PROT_*.
368 		 */
369 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
370 		    max_prot, flags, NULL, pos, FALSE, td);
371 	} else {
372 		/*
373 		 * Mapping file, get fp for validation and don't let the
374 		 * descriptor disappear on us if we block. Check capability
375 		 * rights, but also return the maximum rights to be combined
376 		 * with maxprot later.
377 		 */
378 		cap_rights_init_one(&rights, CAP_MMAP);
379 		if (prot & PROT_READ)
380 			cap_rights_set_one(&rights, CAP_MMAP_R);
381 		if ((flags & MAP_SHARED) != 0) {
382 			if (prot & PROT_WRITE)
383 				cap_rights_set_one(&rights, CAP_MMAP_W);
384 		}
385 		if (prot & PROT_EXEC)
386 			cap_rights_set_one(&rights, CAP_MMAP_X);
387 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
388 		if (error != 0)
389 			goto done;
390 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
391 		    p->p_osrel >= P_OSREL_MAP_FSTRICT) {
392 			EXTERROR(EINVAL, "neither SHARED nor PRIVATE req");
393 			error = EINVAL;
394 			goto done;
395 		}
396 		if (check_fp_fn != NULL) {
397 			error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
398 			    flags);
399 			if (error != 0)
400 				goto done;
401 		}
402 		if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data))
403 			addr = orig_addr;
404 		/* This relies on VM_PROT_* matching PROT_*. */
405 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
406 		    max_prot & cap_maxprot, flags, pos, td);
407 	}
408 
409 	if (error == 0)
410 		td->td_retval[0] = addr + pageoff;
411 done:
412 	if (fp)
413 		fdrop(fp, td);
414 
415 	return (error);
416 }
417 
418 #if defined(COMPAT_FREEBSD6)
419 int
freebsd6_mmap(struct thread * td,struct freebsd6_mmap_args * uap)420 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
421 {
422 	return (kern_mmap(td, &(struct mmap_req){
423 		.mr_hint = (uintptr_t)uap->addr,
424 		.mr_len = uap->len,
425 		.mr_prot = uap->prot,
426 		.mr_flags = uap->flags,
427 		.mr_fd = uap->fd,
428 		.mr_pos = uap->pos,
429 	    }));
430 }
431 #endif
432 
433 #ifdef COMPAT_43
434 #ifndef _SYS_SYSPROTO_H_
435 struct ommap_args {
436 	caddr_t addr;
437 	int len;
438 	int prot;
439 	int flags;
440 	int fd;
441 	long pos;
442 };
443 #endif
444 int
ommap(struct thread * td,struct ommap_args * uap)445 ommap(struct thread *td, struct ommap_args *uap)
446 {
447 	return (kern_ommap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
448 	    uap->flags, uap->fd, uap->pos));
449 }
450 
451 int
kern_ommap(struct thread * td,uintptr_t hint,int len,int oprot,int oflags,int fd,long pos)452 kern_ommap(struct thread *td, uintptr_t hint, int len, int oprot,
453     int oflags, int fd, long pos)
454 {
455 	static const char cvtbsdprot[8] = {
456 		0,
457 		PROT_EXEC,
458 		PROT_WRITE,
459 		PROT_EXEC | PROT_WRITE,
460 		PROT_READ,
461 		PROT_EXEC | PROT_READ,
462 		PROT_WRITE | PROT_READ,
463 		PROT_EXEC | PROT_WRITE | PROT_READ,
464 	};
465 	int flags, prot;
466 
467 	if (len < 0)
468 		return (EINVAL);
469 
470 #define	OMAP_ANON	0x0002
471 #define	OMAP_COPY	0x0020
472 #define	OMAP_SHARED	0x0010
473 #define	OMAP_FIXED	0x0100
474 
475 	prot = cvtbsdprot[oprot & 0x7];
476 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
477 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
478 	    prot != 0)
479 		prot |= PROT_EXEC;
480 #endif
481 	flags = 0;
482 	if (oflags & OMAP_ANON)
483 		flags |= MAP_ANON;
484 	if (oflags & OMAP_COPY)
485 		flags |= MAP_COPY;
486 	if (oflags & OMAP_SHARED)
487 		flags |= MAP_SHARED;
488 	else
489 		flags |= MAP_PRIVATE;
490 	if (oflags & OMAP_FIXED)
491 		flags |= MAP_FIXED;
492 	return (kern_mmap(td, &(struct mmap_req){
493 		.mr_hint = hint,
494 		.mr_len = len,
495 		.mr_prot = prot,
496 		.mr_flags = flags,
497 		.mr_fd = fd,
498 		.mr_pos = pos,
499 	    }));
500 }
501 #endif				/* COMPAT_43 */
502 
503 #ifndef _SYS_SYSPROTO_H_
504 struct msync_args {
505 	void *addr;
506 	size_t len;
507 	int flags;
508 };
509 #endif
510 int
sys_msync(struct thread * td,struct msync_args * uap)511 sys_msync(struct thread *td, struct msync_args *uap)
512 {
513 
514 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
515 }
516 
517 int
kern_msync(struct thread * td,uintptr_t addr0,size_t size,int flags)518 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
519 {
520 	vm_offset_t addr;
521 	vm_size_t pageoff;
522 	vm_map_t map;
523 	int rv;
524 
525 	addr = addr0;
526 	pageoff = (addr & PAGE_MASK);
527 	addr -= pageoff;
528 	size += pageoff;
529 	size = (vm_size_t) round_page(size);
530 	if (addr + size < addr)
531 		return (EINVAL);
532 
533 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
534 		return (EINVAL);
535 
536 	map = &td->td_proc->p_vmspace->vm_map;
537 
538 	/*
539 	 * Clean the pages and interpret the return value.
540 	 */
541 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
542 	    (flags & MS_INVALIDATE) != 0);
543 	switch (rv) {
544 	case KERN_SUCCESS:
545 		return (0);
546 	case KERN_INVALID_ADDRESS:
547 		return (ENOMEM);
548 	case KERN_INVALID_ARGUMENT:
549 		return (EBUSY);
550 	case KERN_FAILURE:
551 		return (EIO);
552 	default:
553 		return (EINVAL);
554 	}
555 }
556 
557 #ifndef _SYS_SYSPROTO_H_
558 struct munmap_args {
559 	void *addr;
560 	size_t len;
561 };
562 #endif
563 int
sys_munmap(struct thread * td,struct munmap_args * uap)564 sys_munmap(struct thread *td, struct munmap_args *uap)
565 {
566 
567 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
568 }
569 
570 int
kern_munmap(struct thread * td,uintptr_t addr0,size_t size)571 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
572 {
573 #ifdef HWPMC_HOOKS
574 	struct pmckern_map_out pkm;
575 	vm_map_entry_t entry;
576 	bool pmc_handled;
577 #endif
578 	vm_offset_t addr, end;
579 	vm_size_t pageoff;
580 	vm_map_t map;
581 	int rv;
582 
583 	if (size == 0)
584 		return (EINVAL);
585 
586 	addr = addr0;
587 	pageoff = (addr & PAGE_MASK);
588 	addr -= pageoff;
589 	size += pageoff;
590 	size = (vm_size_t) round_page(size);
591 	end = addr + size;
592 	map = &td->td_proc->p_vmspace->vm_map;
593 	if (!vm_map_range_valid(map, addr, end))
594 		return (EINVAL);
595 
596 	vm_map_lock(map);
597 #ifdef HWPMC_HOOKS
598 	pmc_handled = false;
599 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
600 		pmc_handled = true;
601 		/*
602 		 * Inform hwpmc if the address range being unmapped contains
603 		 * an executable region.
604 		 */
605 		pkm.pm_address = (uintptr_t) NULL;
606 		if (vm_map_lookup_entry(map, addr, &entry)) {
607 			for (; entry->start < end;
608 			    entry = vm_map_entry_succ(entry)) {
609 				if (vm_map_check_protection(map, entry->start,
610 					entry->end, VM_PROT_EXECUTE) == TRUE) {
611 					pkm.pm_address = (uintptr_t) addr;
612 					pkm.pm_size = (size_t) size;
613 					break;
614 				}
615 			}
616 		}
617 	}
618 #endif
619 	rv = vm_map_delete(map, addr, end);
620 
621 #ifdef HWT_HOOKS
622 	if (HWT_HOOK_INSTALLED && rv == KERN_SUCCESS) {
623 		struct hwt_record_entry ent;
624 
625 		ent.addr = (uintptr_t) addr;
626 		ent.fullpath = NULL;
627 		ent.record_type = HWT_RECORD_MUNMAP;
628 		HWT_CALL_HOOK(td, HWT_RECORD, &ent);
629 	}
630 #endif
631 
632 #ifdef HWPMC_HOOKS
633 	if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) {
634 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
635 		vm_map_lock_downgrade(map);
636 		if (pkm.pm_address != (uintptr_t) NULL)
637 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
638 		vm_map_unlock_read(map);
639 	} else
640 #endif
641 		vm_map_unlock(map);
642 
643 	return (vm_mmap_to_errno(rv));
644 }
645 
646 #ifndef _SYS_SYSPROTO_H_
647 struct mprotect_args {
648 	const void *addr;
649 	size_t len;
650 	int prot;
651 };
652 #endif
653 int
sys_mprotect(struct thread * td,struct mprotect_args * uap)654 sys_mprotect(struct thread *td, struct mprotect_args *uap)
655 {
656 
657 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len,
658 	    uap->prot, 0));
659 }
660 
661 int
kern_mprotect(struct thread * td,uintptr_t addr0,size_t size,int prot,int flags)662 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot,
663     int flags)
664 {
665 	vm_offset_t addr;
666 	vm_size_t pageoff;
667 	int vm_error, max_prot;
668 
669 	addr = addr0;
670 	if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
671 		return (EINVAL);
672 	max_prot = PROT_MAX_EXTRACT(prot);
673 	prot = PROT_EXTRACT(prot);
674 	pageoff = (addr & PAGE_MASK);
675 	addr -= pageoff;
676 	size += pageoff;
677 	size = (vm_size_t) round_page(size);
678 #ifdef COMPAT_FREEBSD32
679 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
680 		if (((addr + size) & 0xffffffff) < addr)
681 			return (EINVAL);
682 	} else
683 #endif
684 	if (addr + size < addr)
685 		return (EINVAL);
686 
687 	flags |= VM_MAP_PROTECT_SET_PROT;
688 	if (max_prot != 0)
689 		flags |= VM_MAP_PROTECT_SET_MAXPROT;
690 	vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
691 	    addr, addr + size, prot, max_prot, flags);
692 
693 	switch (vm_error) {
694 	case KERN_SUCCESS:
695 		return (0);
696 	case KERN_PROTECTION_FAILURE:
697 		return (EACCES);
698 	case KERN_RESOURCE_SHORTAGE:
699 		return (ENOMEM);
700 	case KERN_OUT_OF_BOUNDS:
701 		return (ENOTSUP);
702 	}
703 	return (EINVAL);
704 }
705 
706 #ifndef _SYS_SYSPROTO_H_
707 struct minherit_args {
708 	void *addr;
709 	size_t len;
710 	int inherit;
711 };
712 #endif
713 int
sys_minherit(struct thread * td,struct minherit_args * uap)714 sys_minherit(struct thread *td, struct minherit_args *uap)
715 {
716 
717 	return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
718 	    uap->inherit));
719 }
720 
721 int
kern_minherit(struct thread * td,uintptr_t addr0,size_t len,int inherit0)722 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
723 {
724 	vm_offset_t addr;
725 	vm_size_t size, pageoff;
726 	vm_inherit_t inherit;
727 
728 	addr = (vm_offset_t)addr0;
729 	size = len;
730 	inherit = inherit0;
731 
732 	pageoff = (addr & PAGE_MASK);
733 	addr -= pageoff;
734 	size += pageoff;
735 	size = (vm_size_t) round_page(size);
736 	if (addr + size < addr)
737 		return (EINVAL);
738 
739 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
740 	    addr + size, inherit)) {
741 	case KERN_SUCCESS:
742 		return (0);
743 	case KERN_PROTECTION_FAILURE:
744 		return (EACCES);
745 	}
746 	return (EINVAL);
747 }
748 
749 #ifndef _SYS_SYSPROTO_H_
750 struct madvise_args {
751 	void *addr;
752 	size_t len;
753 	int behav;
754 };
755 #endif
756 
757 int
sys_madvise(struct thread * td,struct madvise_args * uap)758 sys_madvise(struct thread *td, struct madvise_args *uap)
759 {
760 
761 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
762 }
763 
764 int
kern_madvise(struct thread * td,uintptr_t addr0,size_t len,int behav)765 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
766 {
767 	vm_map_t map;
768 	vm_offset_t addr, end, start;
769 	int flags;
770 
771 	/*
772 	 * Check for our special case, advising the swap pager we are
773 	 * "immortal."
774 	 */
775 	if (behav == MADV_PROTECT) {
776 		flags = PPROT_SET;
777 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
778 		    PROC_SPROTECT, &flags));
779 	}
780 
781 	/*
782 	 * Check for illegal addresses.  Watch out for address wrap... Note
783 	 * that VM_*_ADDRESS are not constants due to casts (argh).
784 	 */
785 	map = &td->td_proc->p_vmspace->vm_map;
786 	addr = addr0;
787 	if (!vm_map_range_valid(map, addr, addr + len))
788 		return (EINVAL);
789 
790 	/*
791 	 * Since this routine is only advisory, we default to conservative
792 	 * behavior.
793 	 */
794 	start = trunc_page(addr);
795 	end = round_page(addr + len);
796 
797 	/*
798 	 * vm_map_madvise() checks for illegal values of behav.
799 	 */
800 	return (vm_map_madvise(map, start, end, behav));
801 }
802 
803 #ifndef _SYS_SYSPROTO_H_
804 struct mincore_args {
805 	const void *addr;
806 	size_t len;
807 	char *vec;
808 };
809 #endif
810 
811 int
sys_mincore(struct thread * td,struct mincore_args * uap)812 sys_mincore(struct thread *td, struct mincore_args *uap)
813 {
814 
815 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
816 }
817 
818 int
kern_mincore(struct thread * td,uintptr_t addr0,size_t len,char * vec)819 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
820 {
821 	pmap_t pmap;
822 	vm_map_t map;
823 	vm_map_entry_t current, entry;
824 	vm_object_t object;
825 	vm_offset_t addr, cend, end, first_addr;
826 	vm_paddr_t pa;
827 	vm_page_t m;
828 	vm_pindex_t pindex;
829 	int error, lastvecindex, mincoreinfo, vecindex;
830 	unsigned int timestamp;
831 
832 	/*
833 	 * Make sure that the addresses presented are valid for user
834 	 * mode.
835 	 */
836 	first_addr = addr = trunc_page(addr0);
837 	end = round_page(addr0 + len);
838 	map = &td->td_proc->p_vmspace->vm_map;
839 	if (end > vm_map_max(map) || end < addr)
840 		return (ENOMEM);
841 
842 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
843 
844 	vm_map_lock_read(map);
845 RestartScan:
846 	timestamp = map->timestamp;
847 
848 	if (!vm_map_lookup_entry(map, addr, &entry)) {
849 		vm_map_unlock_read(map);
850 		return (ENOMEM);
851 	}
852 
853 	/*
854 	 * Do this on a map entry basis so that if the pages are not
855 	 * in the current processes address space, we can easily look
856 	 * up the pages elsewhere.
857 	 */
858 	lastvecindex = -1;
859 	while (entry->start < end) {
860 		/*
861 		 * check for contiguity
862 		 */
863 		current = entry;
864 		entry = vm_map_entry_succ(current);
865 		if (current->end < end &&
866 		    entry->start > current->end) {
867 			vm_map_unlock_read(map);
868 			return (ENOMEM);
869 		}
870 
871 		/*
872 		 * ignore submaps (for now) or null objects
873 		 */
874 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
875 		    current->object.vm_object == NULL)
876 			continue;
877 
878 		/*
879 		 * limit this scan to the current map entry and the
880 		 * limits for the mincore call
881 		 */
882 		if (addr < current->start)
883 			addr = current->start;
884 		cend = current->end;
885 		if (cend > end)
886 			cend = end;
887 
888 		for (; addr < cend; addr += PAGE_SIZE) {
889 			/*
890 			 * Check pmap first, it is likely faster, also
891 			 * it can provide info as to whether we are the
892 			 * one referencing or modifying the page.
893 			 */
894 			m = NULL;
895 			object = NULL;
896 retry:
897 			pa = 0;
898 			mincoreinfo = pmap_mincore(pmap, addr, &pa);
899 			if (mincore_mapped) {
900 				/*
901 				 * We only care about this pmap's
902 				 * mapping of the page, if any.
903 				 */
904 				;
905 			} else if (pa != 0) {
906 				/*
907 				 * The page is mapped by this process but not
908 				 * both accessed and modified.  It is also
909 				 * managed.  Acquire the object lock so that
910 				 * other mappings might be examined.  The page's
911 				 * identity may change at any point before its
912 				 * object lock is acquired, so re-validate if
913 				 * necessary.
914 				 */
915 				m = PHYS_TO_VM_PAGE(pa);
916 				while (object == NULL || m->object != object) {
917 					if (object != NULL)
918 						VM_OBJECT_WUNLOCK(object);
919 					object = atomic_load_ptr(&m->object);
920 					if (object == NULL)
921 						goto retry;
922 					VM_OBJECT_WLOCK(object);
923 				}
924 				if (pa != pmap_extract(pmap, addr))
925 					goto retry;
926 				KASSERT(vm_page_all_valid(m),
927 				    ("mincore: page %p is mapped but invalid",
928 				    m));
929 			} else if (mincoreinfo == 0) {
930 				/*
931 				 * The page is not mapped by this process.  If
932 				 * the object implements managed pages, then
933 				 * determine if the page is resident so that
934 				 * the mappings might be examined.
935 				 */
936 				if (current->object.vm_object != object) {
937 					if (object != NULL)
938 						VM_OBJECT_WUNLOCK(object);
939 					object = current->object.vm_object;
940 					VM_OBJECT_WLOCK(object);
941 				}
942 				if ((object->flags & OBJ_SWAP) != 0 ||
943 				    object->type == OBJT_VNODE) {
944 					pindex = OFF_TO_IDX(current->offset +
945 					    (addr - current->start));
946 					m = vm_page_lookup(object, pindex);
947 					if (m != NULL && vm_page_none_valid(m))
948 						m = NULL;
949 					if (m != NULL)
950 						mincoreinfo = MINCORE_INCORE;
951 				}
952 			}
953 			if (m != NULL) {
954 				VM_OBJECT_ASSERT_WLOCKED(m->object);
955 
956 				/* Examine other mappings of the page. */
957 				if (m->dirty == 0 && pmap_is_modified(m))
958 					vm_page_dirty(m);
959 				if (m->dirty != 0)
960 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
961 
962 				/*
963 				 * The first test for PGA_REFERENCED is an
964 				 * optimization.  The second test is
965 				 * required because a concurrent pmap
966 				 * operation could clear the last reference
967 				 * and set PGA_REFERENCED before the call to
968 				 * pmap_is_referenced().
969 				 */
970 				if ((m->a.flags & PGA_REFERENCED) != 0 ||
971 				    pmap_is_referenced(m) ||
972 				    (m->a.flags & PGA_REFERENCED) != 0)
973 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
974 			}
975 			if (object != NULL)
976 				VM_OBJECT_WUNLOCK(object);
977 
978 			/*
979 			 * subyte may page fault.  In case it needs to modify
980 			 * the map, we release the lock.
981 			 */
982 			vm_map_unlock_read(map);
983 
984 			/*
985 			 * calculate index into user supplied byte vector
986 			 */
987 			vecindex = atop(addr - first_addr);
988 
989 			/*
990 			 * If we have skipped map entries, we need to make sure that
991 			 * the byte vector is zeroed for those skipped entries.
992 			 */
993 			while ((lastvecindex + 1) < vecindex) {
994 				++lastvecindex;
995 				error = subyte(vec + lastvecindex, 0);
996 				if (error) {
997 					error = EFAULT;
998 					goto done2;
999 				}
1000 			}
1001 
1002 			/*
1003 			 * Pass the page information to the user
1004 			 */
1005 			error = subyte(vec + vecindex, mincoreinfo);
1006 			if (error) {
1007 				error = EFAULT;
1008 				goto done2;
1009 			}
1010 
1011 			/*
1012 			 * If the map has changed, due to the subyte, the previous
1013 			 * output may be invalid.
1014 			 */
1015 			vm_map_lock_read(map);
1016 			if (timestamp != map->timestamp)
1017 				goto RestartScan;
1018 
1019 			lastvecindex = vecindex;
1020 		}
1021 	}
1022 
1023 	/*
1024 	 * subyte may page fault.  In case it needs to modify
1025 	 * the map, we release the lock.
1026 	 */
1027 	vm_map_unlock_read(map);
1028 
1029 	/*
1030 	 * Zero the last entries in the byte vector.
1031 	 */
1032 	vecindex = atop(end - first_addr);
1033 	while ((lastvecindex + 1) < vecindex) {
1034 		++lastvecindex;
1035 		error = subyte(vec + lastvecindex, 0);
1036 		if (error) {
1037 			error = EFAULT;
1038 			goto done2;
1039 		}
1040 	}
1041 
1042 	/*
1043 	 * If the map has changed, due to the subyte, the previous
1044 	 * output may be invalid.
1045 	 */
1046 	vm_map_lock_read(map);
1047 	if (timestamp != map->timestamp)
1048 		goto RestartScan;
1049 	vm_map_unlock_read(map);
1050 done2:
1051 	return (error);
1052 }
1053 
1054 #ifndef _SYS_SYSPROTO_H_
1055 struct mlock_args {
1056 	const void *addr;
1057 	size_t len;
1058 };
1059 #endif
1060 int
sys_mlock(struct thread * td,struct mlock_args * uap)1061 sys_mlock(struct thread *td, struct mlock_args *uap)
1062 {
1063 
1064 	return (kern_mlock(td->td_proc, td->td_ucred,
1065 	    __DECONST(uintptr_t, uap->addr), uap->len));
1066 }
1067 
1068 int
kern_mlock(struct proc * proc,struct ucred * cred,uintptr_t addr0,size_t len)1069 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1070 {
1071 	vm_offset_t addr, end, last, start;
1072 	vm_size_t npages, size;
1073 	vm_map_t map;
1074 	unsigned long nsize;
1075 	int error;
1076 
1077 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
1078 	if (error)
1079 		return (error);
1080 	addr = addr0;
1081 	size = len;
1082 	last = addr + size;
1083 	start = trunc_page(addr);
1084 	end = round_page(last);
1085 	if (last < addr || end < addr)
1086 		return (EINVAL);
1087 	npages = atop(end - start);
1088 	if (npages > vm_page_max_user_wired)
1089 		return (ENOMEM);
1090 	map = &proc->p_vmspace->vm_map;
1091 	PROC_LOCK(proc);
1092 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1093 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1094 		PROC_UNLOCK(proc);
1095 		return (ENOMEM);
1096 	}
1097 	PROC_UNLOCK(proc);
1098 #ifdef RACCT
1099 	if (racct_enable) {
1100 		PROC_LOCK(proc);
1101 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1102 		PROC_UNLOCK(proc);
1103 		if (error != 0)
1104 			return (ENOMEM);
1105 	}
1106 #endif
1107 	error = vm_map_wire(map, start, end,
1108 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1109 #ifdef RACCT
1110 	if (racct_enable && error != KERN_SUCCESS) {
1111 		PROC_LOCK(proc);
1112 		racct_set(proc, RACCT_MEMLOCK,
1113 		    ptoa(pmap_wired_count(map->pmap)));
1114 		PROC_UNLOCK(proc);
1115 	}
1116 #endif
1117 	switch (error) {
1118 	case KERN_SUCCESS:
1119 		return (0);
1120 	case KERN_INVALID_ARGUMENT:
1121 		return (EINVAL);
1122 	default:
1123 		return (ENOMEM);
1124 	}
1125 }
1126 
1127 #ifndef _SYS_SYSPROTO_H_
1128 struct mlockall_args {
1129 	int	how;
1130 };
1131 #endif
1132 
1133 int
sys_mlockall(struct thread * td,struct mlockall_args * uap)1134 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1135 {
1136 	vm_map_t map;
1137 	int error;
1138 
1139 	map = &td->td_proc->p_vmspace->vm_map;
1140 	error = priv_check(td, PRIV_VM_MLOCK);
1141 	if (error)
1142 		return (error);
1143 
1144 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1145 		return (EINVAL);
1146 
1147 	/*
1148 	 * If wiring all pages in the process would cause it to exceed
1149 	 * a hard resource limit, return ENOMEM.
1150 	 */
1151 	if (!old_mlock && uap->how & MCL_CURRENT) {
1152 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1153 			return (ENOMEM);
1154 	}
1155 #ifdef RACCT
1156 	if (racct_enable) {
1157 		PROC_LOCK(td->td_proc);
1158 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1159 		PROC_UNLOCK(td->td_proc);
1160 		if (error != 0)
1161 			return (ENOMEM);
1162 	}
1163 #endif
1164 
1165 	if (uap->how & MCL_FUTURE) {
1166 		vm_map_lock(map);
1167 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1168 		vm_map_unlock(map);
1169 		error = 0;
1170 	}
1171 
1172 	if (uap->how & MCL_CURRENT) {
1173 		/*
1174 		 * P1003.1-2001 mandates that all currently mapped pages
1175 		 * will be memory resident and locked (wired) upon return
1176 		 * from mlockall(). vm_map_wire() will wire pages, by
1177 		 * calling vm_fault_wire() for each page in the region.
1178 		 */
1179 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1180 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1181 		if (error == KERN_SUCCESS)
1182 			error = 0;
1183 		else if (error == KERN_RESOURCE_SHORTAGE)
1184 			error = ENOMEM;
1185 		else
1186 			error = EAGAIN;
1187 	}
1188 #ifdef RACCT
1189 	if (racct_enable && error != KERN_SUCCESS) {
1190 		PROC_LOCK(td->td_proc);
1191 		racct_set(td->td_proc, RACCT_MEMLOCK,
1192 		    ptoa(pmap_wired_count(map->pmap)));
1193 		PROC_UNLOCK(td->td_proc);
1194 	}
1195 #endif
1196 
1197 	return (error);
1198 }
1199 
1200 #ifndef _SYS_SYSPROTO_H_
1201 struct munlockall_args {
1202 	register_t dummy;
1203 };
1204 #endif
1205 
1206 int
sys_munlockall(struct thread * td,struct munlockall_args * uap)1207 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1208 {
1209 	vm_map_t map;
1210 	int error;
1211 
1212 	map = &td->td_proc->p_vmspace->vm_map;
1213 	error = priv_check(td, PRIV_VM_MUNLOCK);
1214 	if (error)
1215 		return (error);
1216 
1217 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1218 	vm_map_lock(map);
1219 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1220 	vm_map_unlock(map);
1221 
1222 	/* Forcibly unwire all pages. */
1223 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1224 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1225 #ifdef RACCT
1226 	if (racct_enable && error == KERN_SUCCESS) {
1227 		PROC_LOCK(td->td_proc);
1228 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1229 		PROC_UNLOCK(td->td_proc);
1230 	}
1231 #endif
1232 
1233 	return (error);
1234 }
1235 
1236 #ifndef _SYS_SYSPROTO_H_
1237 struct munlock_args {
1238 	const void *addr;
1239 	size_t len;
1240 };
1241 #endif
1242 int
sys_munlock(struct thread * td,struct munlock_args * uap)1243 sys_munlock(struct thread *td, struct munlock_args *uap)
1244 {
1245 
1246 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1247 }
1248 
1249 int
kern_munlock(struct thread * td,uintptr_t addr0,size_t size)1250 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1251 {
1252 	vm_offset_t addr, end, last, start;
1253 #ifdef RACCT
1254 	vm_map_t map;
1255 #endif
1256 	int error;
1257 
1258 	error = priv_check(td, PRIV_VM_MUNLOCK);
1259 	if (error)
1260 		return (error);
1261 	addr = addr0;
1262 	last = addr + size;
1263 	start = trunc_page(addr);
1264 	end = round_page(last);
1265 	if (last < addr || end < addr)
1266 		return (EINVAL);
1267 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1268 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1269 #ifdef RACCT
1270 	if (racct_enable && error == KERN_SUCCESS) {
1271 		PROC_LOCK(td->td_proc);
1272 		map = &td->td_proc->p_vmspace->vm_map;
1273 		racct_set(td->td_proc, RACCT_MEMLOCK,
1274 		    ptoa(pmap_wired_count(map->pmap)));
1275 		PROC_UNLOCK(td->td_proc);
1276 	}
1277 #endif
1278 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1279 }
1280 
1281 /*
1282  * vm_mmap_vnode()
1283  *
1284  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1285  * operations on vnodes.
1286  */
1287 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)1288 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1289     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1290     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1291     boolean_t *writecounted)
1292 {
1293 	struct vattr va;
1294 	vm_object_t obj;
1295 	vm_ooffset_t foff;
1296 	struct ucred *cred;
1297 	int error, flags;
1298 	bool writex;
1299 
1300 	cred = td->td_ucred;
1301 	writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1302 	    (*flagsp & MAP_SHARED) != 0;
1303 	if ((error = vget(vp, LK_SHARED)) != 0)
1304 		return (error);
1305 	AUDIT_ARG_VNODE1(vp);
1306 	foff = *foffp;
1307 	flags = *flagsp;
1308 	obj = vp->v_object;
1309 	if (vp->v_type == VREG) {
1310 		/*
1311 		 * Get the proper underlying object
1312 		 */
1313 		if (obj == NULL) {
1314 			error = EINVAL;
1315 			goto done;
1316 		}
1317 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1318 			vput(vp);
1319 			vp = (struct vnode *)obj->handle;
1320 			/*
1321 			 * Bypass filesystems obey the mpsafety of the
1322 			 * underlying fs.  Tmpfs never bypasses.
1323 			 */
1324 			error = vget(vp, LK_SHARED);
1325 			if (error != 0)
1326 				return (error);
1327 		}
1328 		if (writex) {
1329 			*writecounted = TRUE;
1330 			vm_pager_update_writecount(obj, 0, objsize);
1331 		}
1332 	} else {
1333 		error = EXTERROR(EINVAL, "non-reg file");
1334 		goto done;
1335 	}
1336 	if ((error = VOP_GETATTR(vp, &va, cred)))
1337 		goto done;
1338 #ifdef MAC
1339 	/* This relies on VM_PROT_* matching PROT_*. */
1340 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1341 	if (error != 0)
1342 		goto done;
1343 #endif
1344 	if ((flags & MAP_SHARED) != 0) {
1345 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1346 			if (prot & VM_PROT_WRITE) {
1347 				error = EPERM;
1348 				goto done;
1349 			}
1350 			*maxprotp &= ~VM_PROT_WRITE;
1351 		}
1352 	}
1353 	/*
1354 	 * If it is a regular file without any references
1355 	 * we do not need to sync it.
1356 	 * Adjust object size to be the size of actual file.
1357 	 */
1358 	objsize = round_page(va.va_size);
1359 	if (va.va_nlink == 0)
1360 		flags |= MAP_NOSYNC;
1361 	if (obj->type == OBJT_VNODE) {
1362 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1363 		    cred);
1364 		if (obj == NULL) {
1365 			error = ENOMEM;
1366 			goto done;
1367 		}
1368 	} else {
1369 		KASSERT((obj->flags & OBJ_SWAP) != 0, ("wrong object type"));
1370 		vm_object_reference(obj);
1371 #if VM_NRESERVLEVEL > 0
1372 		if ((obj->flags & OBJ_COLORED) == 0) {
1373 			VM_OBJECT_WLOCK(obj);
1374 			vm_object_color(obj, 0);
1375 			VM_OBJECT_WUNLOCK(obj);
1376 		}
1377 #endif
1378 	}
1379 	*objp = obj;
1380 	*flagsp = flags;
1381 
1382 	VOP_MMAPPED(vp);
1383 
1384 done:
1385 	if (error != 0 && *writecounted) {
1386 		*writecounted = FALSE;
1387 		vm_pager_update_writecount(obj, objsize, 0);
1388 	}
1389 	vput(vp);
1390 	return (error);
1391 }
1392 
1393 /*
1394  * vm_mmap_cdev()
1395  *
1396  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1397  * operations on cdevs.
1398  */
1399 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)1400 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1401     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1402     vm_ooffset_t *foff, vm_object_t *objp)
1403 {
1404 	vm_object_t obj;
1405 	int error, flags;
1406 
1407 	flags = *flagsp;
1408 
1409 	if (dsw->d_flags & D_MMAP_ANON) {
1410 		*objp = NULL;
1411 		*foff = 0;
1412 		*maxprotp = VM_PROT_ALL;
1413 		*flagsp |= MAP_ANON;
1414 		return (0);
1415 	}
1416 
1417 	/*
1418 	 * cdevs do not provide private mappings of any kind.
1419 	 */
1420 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1421 	    (prot & VM_PROT_WRITE) != 0)
1422 		return (EACCES);
1423 	if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0) {
1424 		return (EXTERROR(EINVAL, "cdev mapping must be shared"));
1425 	}
1426 
1427 	/*
1428 	 * Force device mappings to be shared.
1429 	 */
1430 	flags |= MAP_SHARED;
1431 #ifdef MAC_XXX
1432 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1433 	if (error != 0)
1434 		return (error);
1435 #endif
1436 	/*
1437 	 * First, try d_mmap_single().  If that is not implemented
1438 	 * (returns ENODEV), fall back to using the device pager.
1439 	 * Note that d_mmap_single() must return a reference to the
1440 	 * object (it needs to bump the reference count of the object
1441 	 * it returns somehow).
1442 	 *
1443 	 * XXX assumes VM_PROT_* == PROT_*
1444 	 */
1445 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1446 	if (error != ENODEV)
1447 		return (error);
1448 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1449 	    td->td_ucred);
1450 	if (obj == NULL) {
1451 		return (EXTERROR(EINVAL,
1452 		    "cdev driver does not support mmap"));
1453 	}
1454 	*objp = obj;
1455 	*flagsp = flags;
1456 	return (0);
1457 }
1458 
1459 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)1460 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1461 	vm_prot_t maxprot, int flags,
1462 	objtype_t handle_type, void *handle,
1463 	vm_ooffset_t foff)
1464 {
1465 	vm_object_t object;
1466 	struct thread *td = curthread;
1467 	int error;
1468 	boolean_t writecounted;
1469 
1470 	if (size == 0) {
1471 		return (EXTERROR(EINVAL, "zero-sized req"));
1472 	}
1473 
1474 	size = round_page(size);
1475 	object = NULL;
1476 	writecounted = FALSE;
1477 
1478 	switch (handle_type) {
1479 	case OBJT_DEVICE: {
1480 		struct cdevsw *dsw;
1481 		struct cdev *cdev;
1482 		int ref;
1483 
1484 		cdev = handle;
1485 		dsw = dev_refthread(cdev, &ref);
1486 		if (dsw == NULL)
1487 			return (ENXIO);
1488 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1489 		    dsw, &foff, &object);
1490 		dev_relthread(cdev, ref);
1491 		break;
1492 	}
1493 	case OBJT_VNODE:
1494 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1495 		    handle, &foff, &object, &writecounted);
1496 		break;
1497 	default:
1498 		error = EXTERROR(EINVAL, "unsupported backing obj type",
1499 		    handle_type);
1500 		break;
1501 	}
1502 	if (error)
1503 		return (error);
1504 
1505 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1506 	    foff, writecounted, td);
1507 	if (error != 0 && object != NULL) {
1508 		/*
1509 		 * If this mapping was accounted for in the vnode's
1510 		 * writecount, then undo that now.
1511 		 */
1512 		if (writecounted)
1513 			vm_pager_release_writecount(object, 0, size);
1514 		vm_object_deallocate(object);
1515 	}
1516 	return (error);
1517 }
1518 
1519 int
kern_mmap_racct_check(struct thread * td,vm_map_t map,vm_size_t size)1520 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size)
1521 {
1522 	int error;
1523 
1524 	RACCT_PROC_LOCK(td->td_proc);
1525 	if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1526 		RACCT_PROC_UNLOCK(td->td_proc);
1527 		return (ENOMEM);
1528 	}
1529 	if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1530 		RACCT_PROC_UNLOCK(td->td_proc);
1531 		return (ENOMEM);
1532 	}
1533 	if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1534 		if (ptoa(pmap_wired_count(map->pmap)) + size >
1535 		    lim_cur(td, RLIMIT_MEMLOCK)) {
1536 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1537 			RACCT_PROC_UNLOCK(td->td_proc);
1538 			return (ENOMEM);
1539 		}
1540 		error = racct_set(td->td_proc, RACCT_MEMLOCK,
1541 		    ptoa(pmap_wired_count(map->pmap)) + size);
1542 		if (error != 0) {
1543 			racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1544 			RACCT_PROC_UNLOCK(td->td_proc);
1545 			return (error);
1546 		}
1547 	}
1548 	RACCT_PROC_UNLOCK(td->td_proc);
1549 	return (0);
1550 }
1551 
1552 /*
1553  * Internal version of mmap that maps a specific VM object into an
1554  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1555  */
1556 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)1557 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1558     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1559     boolean_t writecounted, struct thread *td)
1560 {
1561 	vm_offset_t default_addr, max_addr;
1562 	int docow, error, findspace, rv;
1563 	bool curmap, fitit;
1564 
1565 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1566 	if (curmap) {
1567 		error = kern_mmap_racct_check(td, map, size);
1568 		if (error != 0)
1569 			return (error);
1570 	}
1571 
1572 	/*
1573 	 * We currently can only deal with page aligned file offsets.
1574 	 * The mmap() system call already enforces this by subtracting
1575 	 * the page offset from the file offset, but checking here
1576 	 * catches errors in device drivers (e.g. d_single_mmap()
1577 	 * callbacks) and other internal mapping requests (such as in
1578 	 * exec).
1579 	 */
1580 	if ((foff & PAGE_MASK) != 0) {
1581 		return (EXTERROR(EINVAL, "offset not page-aligned", foff));
1582 	}
1583 
1584 	if ((flags & MAP_FIXED) == 0) {
1585 		fitit = true;
1586 		*addr = round_page(*addr);
1587 	} else {
1588 		if (*addr != trunc_page(*addr)) {
1589 			return (EXTERROR(EINVAL,
1590 			    "non-fixed mapping address not aligned", *addr));
1591 		}
1592 		fitit = false;
1593 	}
1594 
1595 	if (flags & MAP_ANON) {
1596 		if (object != NULL) {
1597 			return (EXTERROR(EINVAL,
1598 			    "anon mapping backed by an object"));
1599 		}
1600 		if (foff != 0) {
1601 			return (EXTERROR(EINVAL,
1602 			    "anon mapping with non-zero offset"));
1603 		}
1604 		docow = 0;
1605 	} else if (flags & MAP_PREFAULT_READ)
1606 		docow = MAP_PREFAULT;
1607 	else
1608 		docow = MAP_PREFAULT_PARTIAL;
1609 
1610 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1611 		docow |= MAP_COPY_ON_WRITE;
1612 	if (flags & MAP_NOSYNC)
1613 		docow |= MAP_DISABLE_SYNCER;
1614 	if (flags & MAP_NOCORE)
1615 		docow |= MAP_DISABLE_COREDUMP;
1616 	/* Shared memory is also shared with children. */
1617 	if (flags & MAP_SHARED)
1618 		docow |= MAP_INHERIT_SHARE;
1619 	if (writecounted)
1620 		docow |= MAP_WRITECOUNT;
1621 	if (flags & MAP_STACK) {
1622 		if (object != NULL) {
1623 			return (EXTERROR(EINVAL,
1624 			    "stack mapping backed by an object"));
1625 		}
1626 		docow |= MAP_STACK_AREA;
1627 	}
1628 	if ((flags & MAP_EXCL) != 0)
1629 		docow |= MAP_CHECK_EXCL;
1630 	if ((flags & MAP_GUARD) != 0)
1631 		docow |= MAP_CREATE_GUARD;
1632 
1633 	if (fitit) {
1634 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1635 			findspace = VMFS_SUPER_SPACE;
1636 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1637 			findspace = VMFS_ALIGNED_SPACE(flags >>
1638 			    MAP_ALIGNMENT_SHIFT);
1639 		else
1640 			findspace = VMFS_OPTIMAL_SPACE;
1641 		max_addr = 0;
1642 		if ((flags & MAP_32BIT) != 0)
1643 			max_addr = MAP_32BIT_MAX_ADDR;
1644 		if (curmap) {
1645 			default_addr =
1646 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1647 			    vm_daddr + lim_max(td, RLIMIT_DATA));
1648 			if ((flags & MAP_32BIT) != 0)
1649 				default_addr = 0;
1650 			rv = vm_map_find_min(map, object, foff, addr, size,
1651 			    default_addr, max_addr, findspace, prot, maxprot,
1652 			    docow);
1653 		} else {
1654 			rv = vm_map_find(map, object, foff, addr, size,
1655 			    max_addr, findspace, prot, maxprot, docow);
1656 		}
1657 	} else {
1658 		rv = vm_map_fixed(map, object, foff, *addr, size,
1659 		    prot, maxprot, docow);
1660 	}
1661 
1662 	if (rv == KERN_SUCCESS) {
1663 		/*
1664 		 * If the process has requested that all future mappings
1665 		 * be wired, then heed this.
1666 		 */
1667 		if ((map->flags & MAP_WIREFUTURE) != 0) {
1668 			vm_map_lock(map);
1669 			if ((map->flags & MAP_WIREFUTURE) != 0)
1670 				(void)vm_map_wire_locked(map, *addr,
1671 				    *addr + size, VM_MAP_WIRE_USER |
1672 				    ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1673 				    VM_MAP_WIRE_NOHOLES));
1674 			vm_map_unlock(map);
1675 		}
1676 	}
1677 	return (vm_mmap_to_errno(rv));
1678 }
1679 
1680 /*
1681  * Translate a Mach VM return code to zero on success or the appropriate errno
1682  * on failure.
1683  */
1684 int
vm_mmap_to_errno(int rv)1685 vm_mmap_to_errno(int rv)
1686 {
1687 	int error;
1688 
1689 	switch (rv) {
1690 	case KERN_SUCCESS:
1691 		return (0);
1692 	case KERN_INVALID_ADDRESS:
1693 	case KERN_NO_SPACE:
1694 		error = ENOMEM;
1695 		break;
1696 	case KERN_PROTECTION_FAILURE:
1697 		error = EACCES;
1698 		break;
1699 	default:
1700 		error = EINVAL;
1701 		break;
1702 	}
1703 	if ((curthread->td_pflags2 & (TDP2_UEXTERR | TDP2_EXTERR)) ==
1704 	    TDP2_UEXTERR)
1705 		EXTERROR(error, "mach error", rv);
1706 	return (error);
1707 }
1708