Optimize lwkt_rwlock.c a bit
[dragonfly.git] / sys / vm / vm_mmap.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1988 University of Utah.
3 * Copyright (c) 1991, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * the Systems Programming Group of the University of Utah Computer
8 * Science Department.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
39 *
40 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
41 * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $
1de703da 42 * $DragonFly: src/sys/vm/vm_mmap.c,v 1.2 2003/06/17 04:29:00 dillon Exp $
984263bc
MD
43 */
44
45/*
46 * Mapped file (mmap) interface to VM
47 */
48
49#include "opt_compat.h"
50
51#include <sys/param.h>
52#include <sys/kernel.h>
53#include <sys/systm.h>
54#include <sys/sysproto.h>
55#include <sys/filedesc.h>
56#include <sys/proc.h>
57#include <sys/resource.h>
58#include <sys/resourcevar.h>
59#include <sys/vnode.h>
60#include <sys/fcntl.h>
61#include <sys/file.h>
62#include <sys/mman.h>
63#include <sys/conf.h>
64#include <sys/stat.h>
65#include <sys/vmmeter.h>
66#include <sys/sysctl.h>
67
68#include <vm/vm.h>
69#include <vm/vm_param.h>
70#include <sys/lock.h>
71#include <vm/pmap.h>
72#include <vm/vm_map.h>
73#include <vm/vm_object.h>
74#include <vm/vm_page.h>
75#include <vm/vm_pager.h>
76#include <vm/vm_pageout.h>
77#include <vm/vm_extern.h>
78#include <vm/vm_page.h>
79#include <vm/vm_kern.h>
80
81#ifndef _SYS_SYSPROTO_H_
82struct sbrk_args {
83 int incr;
84};
85#endif
86
87static int max_proc_mmap;
88SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
89
90/*
91 * Set the maximum number of vm_map_entry structures per process. Roughly
92 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
93 * of our KVM malloc space still results in generous limits. We want a
94 * default that is good enough to prevent the kernel running out of resources
95 * if attacked from compromised user account but generous enough such that
96 * multi-threaded processes are not unduly inconvenienced.
97 */
98
99static void vmmapentry_rsrc_init __P((void *));
100SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
101
102static void
103vmmapentry_rsrc_init(dummy)
104 void *dummy;
105{
106 max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
107 max_proc_mmap /= 100;
108}
109
110/* ARGSUSED */
111int
112sbrk(p, uap)
113 struct proc *p;
114 struct sbrk_args *uap;
115{
116
117 /* Not yet implemented */
118 return (EOPNOTSUPP);
119}
120
121#ifndef _SYS_SYSPROTO_H_
122struct sstk_args {
123 int incr;
124};
125#endif
126
127/* ARGSUSED */
128int
129sstk(p, uap)
130 struct proc *p;
131 struct sstk_args *uap;
132{
133
134 /* Not yet implemented */
135 return (EOPNOTSUPP);
136}
137
138#if defined(COMPAT_43) || defined(COMPAT_SUNOS)
139#ifndef _SYS_SYSPROTO_H_
140struct getpagesize_args {
141 int dummy;
142};
143#endif
144
145/* ARGSUSED */
146int
147ogetpagesize(p, uap)
148 struct proc *p;
149 struct getpagesize_args *uap;
150{
151
152 p->p_retval[0] = PAGE_SIZE;
153 return (0);
154}
155#endif /* COMPAT_43 || COMPAT_SUNOS */
156
157
158/*
159 * Memory Map (mmap) system call. Note that the file offset
160 * and address are allowed to be NOT page aligned, though if
161 * the MAP_FIXED flag it set, both must have the same remainder
162 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
163 * page-aligned, the actual mapping starts at trunc_page(addr)
164 * and the return value is adjusted up by the page offset.
165 *
166 * Generally speaking, only character devices which are themselves
167 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
168 * there would be no cache coherency between a descriptor and a VM mapping
169 * both to the same character device.
170 *
171 * Block devices can be mmap'd no matter what they represent. Cache coherency
172 * is maintained as long as you do not write directly to the underlying
173 * character device.
174 */
175#ifndef _SYS_SYSPROTO_H_
176struct mmap_args {
177 void *addr;
178 size_t len;
179 int prot;
180 int flags;
181 int fd;
182 long pad;
183 off_t pos;
184};
185#endif
186
187int
188mmap(p, uap)
189 struct proc *p;
190 register struct mmap_args *uap;
191{
192 register struct filedesc *fdp = p->p_fd;
193 register struct file *fp = NULL;
194 struct vnode *vp;
195 vm_offset_t addr;
196 vm_size_t size, pageoff;
197 vm_prot_t prot, maxprot;
198 void *handle;
199 int flags, error;
200 int disablexworkaround;
201 off_t pos;
202 struct vmspace *vms = p->p_vmspace;
203 vm_object_t obj;
204
205 addr = (vm_offset_t) uap->addr;
206 size = uap->len;
207 prot = uap->prot & VM_PROT_ALL;
208 flags = uap->flags;
209 pos = uap->pos;
210
211 /* make sure mapping fits into numeric range etc */
212 if ((ssize_t) uap->len < 0 ||
213 ((flags & MAP_ANON) && uap->fd != -1))
214 return (EINVAL);
215
216 if (flags & MAP_STACK) {
217 if ((uap->fd != -1) ||
218 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
219 return (EINVAL);
220 flags |= MAP_ANON;
221 pos = 0;
222 }
223
224 /*
225 * Align the file position to a page boundary,
226 * and save its page offset component.
227 */
228 pageoff = (pos & PAGE_MASK);
229 pos -= pageoff;
230
231 /* Adjust size for rounding (on both ends). */
232 size += pageoff; /* low end... */
233 size = (vm_size_t) round_page(size); /* hi end */
234
235 /*
236 * Check for illegal addresses. Watch out for address wrap... Note
237 * that VM_*_ADDRESS are not constants due to casts (argh).
238 */
239 if (flags & MAP_FIXED) {
240 /*
241 * The specified address must have the same remainder
242 * as the file offset taken modulo PAGE_SIZE, so it
243 * should be aligned after adjustment by pageoff.
244 */
245 addr -= pageoff;
246 if (addr & PAGE_MASK)
247 return (EINVAL);
248 /* Address range must be all in user VM space. */
249 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
250 return (EINVAL);
251#ifndef i386
252 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
253 return (EINVAL);
254#endif
255 if (addr + size < addr)
256 return (EINVAL);
257 }
258 /*
259 * XXX for non-fixed mappings where no hint is provided or
260 * the hint would fall in the potential heap space,
261 * place it after the end of the largest possible heap.
262 *
263 * There should really be a pmap call to determine a reasonable
264 * location.
265 */
266 else if (addr == 0 ||
267 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
268 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
269 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
270
271 if (flags & MAP_ANON) {
272 /*
273 * Mapping blank space is trivial.
274 */
275 handle = NULL;
276 maxprot = VM_PROT_ALL;
277 pos = 0;
278 } else {
279 /*
280 * Mapping file, get fp for validation. Obtain vnode and make
281 * sure it is of appropriate type.
282 */
283 if (((unsigned) uap->fd) >= fdp->fd_nfiles ||
284 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
285 return (EBADF);
286 if (fp->f_type != DTYPE_VNODE)
287 return (EINVAL);
288 /*
289 * POSIX shared-memory objects are defined to have
290 * kernel persistence, and are not defined to support
291 * read(2)/write(2) -- or even open(2). Thus, we can
292 * use MAP_ASYNC to trade on-disk coherence for speed.
293 * The shm_open(3) library routine turns on the FPOSIXSHM
294 * flag to request this behavior.
295 */
296 if (fp->f_flag & FPOSIXSHM)
297 flags |= MAP_NOSYNC;
298 vp = (struct vnode *) fp->f_data;
299 if (vp->v_type != VREG && vp->v_type != VCHR)
300 return (EINVAL);
301 if (vp->v_type == VREG) {
302 /*
303 * Get the proper underlying object
304 */
305 if (VOP_GETVOBJECT(vp, &obj) != 0)
306 return (EINVAL);
307 vp = (struct vnode*)obj->handle;
308 }
309
310 /*
311 * don't let the descriptor disappear on us if we block
312 */
313 fhold(fp);
314
315 /*
316 * XXX hack to handle use of /dev/zero to map anon memory (ala
317 * SunOS).
318 */
319 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
320 handle = NULL;
321 maxprot = VM_PROT_ALL;
322 flags |= MAP_ANON;
323 pos = 0;
324 } else {
325 /*
326 * cdevs does not provide private mappings of any kind.
327 */
328 /*
329 * However, for XIG X server to continue to work,
330 * we should allow the superuser to do it anyway.
331 * We only allow it at securelevel < 1.
332 * (Because the XIG X server writes directly to video
333 * memory via /dev/mem, it should never work at any
334 * other securelevel.
335 * XXX this will have to go
336 */
337 if (securelevel >= 1)
338 disablexworkaround = 1;
339 else
340 disablexworkaround = suser(p);
341 if (vp->v_type == VCHR && disablexworkaround &&
342 (flags & (MAP_PRIVATE|MAP_COPY))) {
343 error = EINVAL;
344 goto done;
345 }
346 /*
347 * Ensure that file and memory protections are
348 * compatible. Note that we only worry about
349 * writability if mapping is shared; in this case,
350 * current and max prot are dictated by the open file.
351 * XXX use the vnode instead? Problem is: what
352 * credentials do we use for determination? What if
353 * proc does a setuid?
354 */
355 maxprot = VM_PROT_EXECUTE; /* ??? */
356 if (fp->f_flag & FREAD) {
357 maxprot |= VM_PROT_READ;
358 } else if (prot & PROT_READ) {
359 error = EACCES;
360 goto done;
361 }
362 /*
363 * If we are sharing potential changes (either via
364 * MAP_SHARED or via the implicit sharing of character
365 * device mappings), and we are trying to get write
366 * permission although we opened it without asking
367 * for it, bail out. Check for superuser, only if
368 * we're at securelevel < 1, to allow the XIG X server
369 * to continue to work.
370 */
371
372 if ((flags & MAP_SHARED) != 0 ||
373 (vp->v_type == VCHR && disablexworkaround)) {
374 if ((fp->f_flag & FWRITE) != 0) {
375 struct vattr va;
376 if ((error =
377 VOP_GETATTR(vp, &va,
378 p->p_ucred, p))) {
379 goto done;
380 }
381 if ((va.va_flags &
382 (IMMUTABLE|APPEND)) == 0) {
383 maxprot |= VM_PROT_WRITE;
384 } else if (prot & PROT_WRITE) {
385 error = EPERM;
386 goto done;
387 }
388 } else if ((prot & PROT_WRITE) != 0) {
389 error = EACCES;
390 goto done;
391 }
392 } else {
393 maxprot |= VM_PROT_WRITE;
394 }
395 handle = (void *)vp;
396 }
397 }
398
399 /*
400 * Do not allow more then a certain number of vm_map_entry structures
401 * per process. Scale with the number of rforks sharing the map
402 * to make the limit reasonable for threads.
403 */
404 if (max_proc_mmap &&
405 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
406 error = ENOMEM;
407 goto done;
408 }
409
410 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
411 flags, handle, pos);
412 if (error == 0)
413 p->p_retval[0] = (register_t) (addr + pageoff);
414done:
415 if (fp)
416 fdrop(fp, p);
417 return (error);
418}
419
420#ifdef COMPAT_43
421#ifndef _SYS_SYSPROTO_H_
422struct ommap_args {
423 caddr_t addr;
424 int len;
425 int prot;
426 int flags;
427 int fd;
428 long pos;
429};
430#endif
431int
432ommap(p, uap)
433 struct proc *p;
434 register struct ommap_args *uap;
435{
436 struct mmap_args nargs;
437 static const char cvtbsdprot[8] = {
438 0,
439 PROT_EXEC,
440 PROT_WRITE,
441 PROT_EXEC | PROT_WRITE,
442 PROT_READ,
443 PROT_EXEC | PROT_READ,
444 PROT_WRITE | PROT_READ,
445 PROT_EXEC | PROT_WRITE | PROT_READ,
446 };
447
448#define OMAP_ANON 0x0002
449#define OMAP_COPY 0x0020
450#define OMAP_SHARED 0x0010
451#define OMAP_FIXED 0x0100
452#define OMAP_INHERIT 0x0800
453
454 nargs.addr = uap->addr;
455 nargs.len = uap->len;
456 nargs.prot = cvtbsdprot[uap->prot & 0x7];
457 nargs.flags = 0;
458 if (uap->flags & OMAP_ANON)
459 nargs.flags |= MAP_ANON;
460 if (uap->flags & OMAP_COPY)
461 nargs.flags |= MAP_COPY;
462 if (uap->flags & OMAP_SHARED)
463 nargs.flags |= MAP_SHARED;
464 else
465 nargs.flags |= MAP_PRIVATE;
466 if (uap->flags & OMAP_FIXED)
467 nargs.flags |= MAP_FIXED;
468 if (uap->flags & OMAP_INHERIT)
469 nargs.flags |= MAP_INHERIT;
470 nargs.fd = uap->fd;
471 nargs.pos = uap->pos;
472 return (mmap(p, &nargs));
473}
474#endif /* COMPAT_43 */
475
476
477#ifndef _SYS_SYSPROTO_H_
478struct msync_args {
479 void *addr;
480 int len;
481 int flags;
482};
483#endif
484int
485msync(p, uap)
486 struct proc *p;
487 struct msync_args *uap;
488{
489 vm_offset_t addr;
490 vm_size_t size, pageoff;
491 int flags;
492 vm_map_t map;
493 int rv;
494
495 addr = (vm_offset_t) uap->addr;
496 size = uap->len;
497 flags = uap->flags;
498
499 pageoff = (addr & PAGE_MASK);
500 addr -= pageoff;
501 size += pageoff;
502 size = (vm_size_t) round_page(size);
503 if (addr + size < addr)
504 return(EINVAL);
505
506 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
507 return (EINVAL);
508
509 map = &p->p_vmspace->vm_map;
510
511 /*
512 * XXX Gak! If size is zero we are supposed to sync "all modified
513 * pages with the region containing addr". Unfortunately, we don't
514 * really keep track of individual mmaps so we approximate by flushing
515 * the range of the map entry containing addr. This can be incorrect
516 * if the region splits or is coalesced with a neighbor.
517 */
518 if (size == 0) {
519 vm_map_entry_t entry;
520
521 vm_map_lock_read(map);
522 rv = vm_map_lookup_entry(map, addr, &entry);
523 vm_map_unlock_read(map);
524 if (rv == FALSE)
525 return (EINVAL);
526 addr = entry->start;
527 size = entry->end - entry->start;
528 }
529
530 /*
531 * Clean the pages and interpret the return value.
532 */
533 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
534 (flags & MS_INVALIDATE) != 0);
535
536 switch (rv) {
537 case KERN_SUCCESS:
538 break;
539 case KERN_INVALID_ADDRESS:
540 return (EINVAL); /* Sun returns ENOMEM? */
541 case KERN_FAILURE:
542 return (EIO);
543 default:
544 return (EINVAL);
545 }
546
547 return (0);
548}
549
550#ifndef _SYS_SYSPROTO_H_
551struct munmap_args {
552 void *addr;
553 size_t len;
554};
555#endif
556int
557munmap(p, uap)
558 register struct proc *p;
559 register struct munmap_args *uap;
560{
561 vm_offset_t addr;
562 vm_size_t size, pageoff;
563 vm_map_t map;
564
565 addr = (vm_offset_t) uap->addr;
566 size = uap->len;
567
568 pageoff = (addr & PAGE_MASK);
569 addr -= pageoff;
570 size += pageoff;
571 size = (vm_size_t) round_page(size);
572 if (addr + size < addr)
573 return(EINVAL);
574
575 if (size == 0)
576 return (0);
577
578 /*
579 * Check for illegal addresses. Watch out for address wrap... Note
580 * that VM_*_ADDRESS are not constants due to casts (argh).
581 */
582 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
583 return (EINVAL);
584#ifndef i386
585 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
586 return (EINVAL);
587#endif
588 map = &p->p_vmspace->vm_map;
589 /*
590 * Make sure entire range is allocated.
591 */
592 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
593 return (EINVAL);
594 /* returns nothing but KERN_SUCCESS anyway */
595 (void) vm_map_remove(map, addr, addr + size);
596 return (0);
597}
598
599#if 0
600void
601munmapfd(p, fd)
602 struct proc *p;
603 int fd;
604{
605 /*
606 * XXX should unmap any regions mapped to this file
607 */
608 p->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
609}
610#endif
611
612#ifndef _SYS_SYSPROTO_H_
613struct mprotect_args {
614 const void *addr;
615 size_t len;
616 int prot;
617};
618#endif
619int
620mprotect(p, uap)
621 struct proc *p;
622 struct mprotect_args *uap;
623{
624 vm_offset_t addr;
625 vm_size_t size, pageoff;
626 register vm_prot_t prot;
627
628 addr = (vm_offset_t) uap->addr;
629 size = uap->len;
630 prot = uap->prot & VM_PROT_ALL;
631#if defined(VM_PROT_READ_IS_EXEC)
632 if (prot & VM_PROT_READ)
633 prot |= VM_PROT_EXECUTE;
634#endif
635
636 pageoff = (addr & PAGE_MASK);
637 addr -= pageoff;
638 size += pageoff;
639 size = (vm_size_t) round_page(size);
640 if (addr + size < addr)
641 return(EINVAL);
642
643 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
644 FALSE)) {
645 case KERN_SUCCESS:
646 return (0);
647 case KERN_PROTECTION_FAILURE:
648 return (EACCES);
649 }
650 return (EINVAL);
651}
652
653#ifndef _SYS_SYSPROTO_H_
654struct minherit_args {
655 void *addr;
656 size_t len;
657 int inherit;
658};
659#endif
660int
661minherit(p, uap)
662 struct proc *p;
663 struct minherit_args *uap;
664{
665 vm_offset_t addr;
666 vm_size_t size, pageoff;
667 register vm_inherit_t inherit;
668
669 addr = (vm_offset_t)uap->addr;
670 size = uap->len;
671 inherit = uap->inherit;
672
673 pageoff = (addr & PAGE_MASK);
674 addr -= pageoff;
675 size += pageoff;
676 size = (vm_size_t) round_page(size);
677 if (addr + size < addr)
678 return(EINVAL);
679
680 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
681 inherit)) {
682 case KERN_SUCCESS:
683 return (0);
684 case KERN_PROTECTION_FAILURE:
685 return (EACCES);
686 }
687 return (EINVAL);
688}
689
690#ifndef _SYS_SYSPROTO_H_
691struct madvise_args {
692 void *addr;
693 size_t len;
694 int behav;
695};
696#endif
697
698/* ARGSUSED */
699int
700madvise(p, uap)
701 struct proc *p;
702 struct madvise_args *uap;
703{
704 vm_offset_t start, end;
705
706 /*
707 * Check for illegal behavior
708 */
709 if (uap->behav < 0 || uap->behav > MADV_CORE)
710 return (EINVAL);
711 /*
712 * Check for illegal addresses. Watch out for address wrap... Note
713 * that VM_*_ADDRESS are not constants due to casts (argh).
714 */
715 if (VM_MAXUSER_ADDRESS > 0 &&
716 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
717 return (EINVAL);
718#ifndef i386
719 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
720 return (EINVAL);
721#endif
722 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
723 return (EINVAL);
724
725 /*
726 * Since this routine is only advisory, we default to conservative
727 * behavior.
728 */
729 start = trunc_page((vm_offset_t) uap->addr);
730 end = round_page((vm_offset_t) uap->addr + uap->len);
731
732 if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav))
733 return (EINVAL);
734 return (0);
735}
736
737#ifndef _SYS_SYSPROTO_H_
738struct mincore_args {
739 const void *addr;
740 size_t len;
741 char *vec;
742};
743#endif
744
745/* ARGSUSED */
746int
747mincore(p, uap)
748 struct proc *p;
749 struct mincore_args *uap;
750{
751 vm_offset_t addr, first_addr;
752 vm_offset_t end, cend;
753 pmap_t pmap;
754 vm_map_t map;
755 char *vec;
756 int error;
757 int vecindex, lastvecindex;
758 register vm_map_entry_t current;
759 vm_map_entry_t entry;
760 int mincoreinfo;
761 unsigned int timestamp;
762
763 /*
764 * Make sure that the addresses presented are valid for user
765 * mode.
766 */
767 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
768 end = addr + (vm_size_t)round_page(uap->len);
769 if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
770 return (EINVAL);
771 if (end < addr)
772 return (EINVAL);
773
774 /*
775 * Address of byte vector
776 */
777 vec = uap->vec;
778
779 map = &p->p_vmspace->vm_map;
780 pmap = vmspace_pmap(p->p_vmspace);
781
782 vm_map_lock_read(map);
783RestartScan:
784 timestamp = map->timestamp;
785
786 if (!vm_map_lookup_entry(map, addr, &entry))
787 entry = entry->next;
788
789 /*
790 * Do this on a map entry basis so that if the pages are not
791 * in the current processes address space, we can easily look
792 * up the pages elsewhere.
793 */
794 lastvecindex = -1;
795 for(current = entry;
796 (current != &map->header) && (current->start < end);
797 current = current->next) {
798
799 /*
800 * ignore submaps (for now) or null objects
801 */
802 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
803 current->object.vm_object == NULL)
804 continue;
805
806 /*
807 * limit this scan to the current map entry and the
808 * limits for the mincore call
809 */
810 if (addr < current->start)
811 addr = current->start;
812 cend = current->end;
813 if (cend > end)
814 cend = end;
815
816 /*
817 * scan this entry one page at a time
818 */
819 while(addr < cend) {
820 /*
821 * Check pmap first, it is likely faster, also
822 * it can provide info as to whether we are the
823 * one referencing or modifying the page.
824 */
825 mincoreinfo = pmap_mincore(pmap, addr);
826 if (!mincoreinfo) {
827 vm_pindex_t pindex;
828 vm_ooffset_t offset;
829 vm_page_t m;
830 /*
831 * calculate the page index into the object
832 */
833 offset = current->offset + (addr - current->start);
834 pindex = OFF_TO_IDX(offset);
835 m = vm_page_lookup(current->object.vm_object,
836 pindex);
837 /*
838 * if the page is resident, then gather information about
839 * it.
840 */
841 if (m) {
842 mincoreinfo = MINCORE_INCORE;
843 if (m->dirty ||
844 pmap_is_modified(m))
845 mincoreinfo |= MINCORE_MODIFIED_OTHER;
846 if ((m->flags & PG_REFERENCED) ||
847 pmap_ts_referenced(m)) {
848 vm_page_flag_set(m, PG_REFERENCED);
849 mincoreinfo |= MINCORE_REFERENCED_OTHER;
850 }
851 }
852 }
853
854 /*
855 * subyte may page fault. In case it needs to modify
856 * the map, we release the lock.
857 */
858 vm_map_unlock_read(map);
859
860 /*
861 * calculate index into user supplied byte vector
862 */
863 vecindex = OFF_TO_IDX(addr - first_addr);
864
865 /*
866 * If we have skipped map entries, we need to make sure that
867 * the byte vector is zeroed for those skipped entries.
868 */
869 while((lastvecindex + 1) < vecindex) {
870 error = subyte( vec + lastvecindex, 0);
871 if (error) {
872 return (EFAULT);
873 }
874 ++lastvecindex;
875 }
876
877 /*
878 * Pass the page information to the user
879 */
880 error = subyte( vec + vecindex, mincoreinfo);
881 if (error) {
882 return (EFAULT);
883 }
884
885 /*
886 * If the map has changed, due to the subyte, the previous
887 * output may be invalid.
888 */
889 vm_map_lock_read(map);
890 if (timestamp != map->timestamp)
891 goto RestartScan;
892
893 lastvecindex = vecindex;
894 addr += PAGE_SIZE;
895 }
896 }
897
898 /*
899 * subyte may page fault. In case it needs to modify
900 * the map, we release the lock.
901 */
902 vm_map_unlock_read(map);
903
904 /*
905 * Zero the last entries in the byte vector.
906 */
907 vecindex = OFF_TO_IDX(end - first_addr);
908 while((lastvecindex + 1) < vecindex) {
909 error = subyte( vec + lastvecindex, 0);
910 if (error) {
911 return (EFAULT);
912 }
913 ++lastvecindex;
914 }
915
916 /*
917 * If the map has changed, due to the subyte, the previous
918 * output may be invalid.
919 */
920 vm_map_lock_read(map);
921 if (timestamp != map->timestamp)
922 goto RestartScan;
923 vm_map_unlock_read(map);
924
925 return (0);
926}
927
928#ifndef _SYS_SYSPROTO_H_
929struct mlock_args {
930 const void *addr;
931 size_t len;
932};
933#endif
934int
935mlock(p, uap)
936 struct proc *p;
937 struct mlock_args *uap;
938{
939 vm_offset_t addr;
940 vm_size_t size, pageoff;
941 int error;
942
943 addr = (vm_offset_t) uap->addr;
944 size = uap->len;
945
946 pageoff = (addr & PAGE_MASK);
947 addr -= pageoff;
948 size += pageoff;
949 size = (vm_size_t) round_page(size);
950
951 /* disable wrap around */
952 if (addr + size < addr)
953 return (EINVAL);
954
955 if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
956 return (EAGAIN);
957
958#ifdef pmap_wired_count
959 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
960 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
961 return (ENOMEM);
962#else
963 error = suser(p);
964 if (error)
965 return (error);
966#endif
967
968 error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
969 return (error == KERN_SUCCESS ? 0 : ENOMEM);
970}
971
972#ifndef _SYS_SYSPROTO_H_
973struct mlockall_args {
974 int how;
975};
976#endif
977
978int
979mlockall(p, uap)
980 struct proc *p;
981 struct mlockall_args *uap;
982{
983 return 0;
984}
985
986#ifndef _SYS_SYSPROTO_H_
987struct mlockall_args {
988 int how;
989};
990#endif
991
992int
993munlockall(p, uap)
994 struct proc *p;
995 struct munlockall_args *uap;
996{
997 return 0;
998}
999
1000#ifndef _SYS_SYSPROTO_H_
1001struct munlock_args {
1002 const void *addr;
1003 size_t len;
1004};
1005#endif
1006int
1007munlock(p, uap)
1008 struct proc *p;
1009 struct munlock_args *uap;
1010{
1011 vm_offset_t addr;
1012 vm_size_t size, pageoff;
1013 int error;
1014
1015 addr = (vm_offset_t) uap->addr;
1016 size = uap->len;
1017
1018 pageoff = (addr & PAGE_MASK);
1019 addr -= pageoff;
1020 size += pageoff;
1021 size = (vm_size_t) round_page(size);
1022
1023 /* disable wrap around */
1024 if (addr + size < addr)
1025 return (EINVAL);
1026
1027#ifndef pmap_wired_count
1028 error = suser(p);
1029 if (error)
1030 return (error);
1031#endif
1032
1033 error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
1034 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1035}
1036
1037/*
1038 * Internal version of mmap.
1039 * Currently used by mmap, exec, and sys5 shared memory.
1040 * Handle is either a vnode pointer or NULL for MAP_ANON.
1041 */
1042int
1043vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1044 vm_prot_t maxprot, int flags,
1045 void *handle,
1046 vm_ooffset_t foff)
1047{
1048 boolean_t fitit;
1049 vm_object_t object;
1050 struct vnode *vp = NULL;
1051 objtype_t type;
1052 int rv = KERN_SUCCESS;
1053 vm_ooffset_t objsize;
1054 int docow;
1055 struct proc *p = curproc;
1056
1057 if (size == 0)
1058 return (0);
1059
1060 objsize = size = round_page(size);
1061
1062 if (p->p_vmspace->vm_map.size + size >
1063 p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1064 return(ENOMEM);
1065 }
1066
1067 /*
1068 * We currently can only deal with page aligned file offsets.
1069 * The check is here rather than in the syscall because the
1070 * kernel calls this function internally for other mmaping
1071 * operations (such as in exec) and non-aligned offsets will
1072 * cause pmap inconsistencies...so we want to be sure to
1073 * disallow this in all cases.
1074 */
1075 if (foff & PAGE_MASK)
1076 return (EINVAL);
1077
1078 if ((flags & MAP_FIXED) == 0) {
1079 fitit = TRUE;
1080 *addr = round_page(*addr);
1081 } else {
1082 if (*addr != trunc_page(*addr))
1083 return (EINVAL);
1084 fitit = FALSE;
1085 (void) vm_map_remove(map, *addr, *addr + size);
1086 }
1087
1088 /*
1089 * Lookup/allocate object.
1090 */
1091 if (flags & MAP_ANON) {
1092 type = OBJT_DEFAULT;
1093 /*
1094 * Unnamed anonymous regions always start at 0.
1095 */
1096 if (handle == 0)
1097 foff = 0;
1098 } else {
1099 vp = (struct vnode *) handle;
1100 if (vp->v_type == VCHR) {
1101 type = OBJT_DEVICE;
1102 handle = (void *)(intptr_t)vp->v_rdev;
1103 } else {
1104 struct vattr vat;
1105 int error;
1106
1107 error = VOP_GETATTR(vp, &vat, p->p_ucred, p);
1108 if (error)
1109 return (error);
1110 objsize = round_page(vat.va_size);
1111 type = OBJT_VNODE;
1112 /*
1113 * if it is a regular file without any references
1114 * we do not need to sync it.
1115 */
1116 if (vp->v_type == VREG && vat.va_nlink == 0) {
1117 flags |= MAP_NOSYNC;
1118 }
1119 }
1120 }
1121
1122 if (handle == NULL) {
1123 object = NULL;
1124 docow = 0;
1125 } else {
1126 object = vm_pager_allocate(type,
1127 handle, objsize, prot, foff);
1128 if (object == NULL)
1129 return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1130 docow = MAP_PREFAULT_PARTIAL;
1131 }
1132
1133 /*
1134 * Force device mappings to be shared.
1135 */
1136 if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1137 flags &= ~(MAP_PRIVATE|MAP_COPY);
1138 flags |= MAP_SHARED;
1139 }
1140
1141 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1142 docow |= MAP_COPY_ON_WRITE;
1143 if (flags & MAP_NOSYNC)
1144 docow |= MAP_DISABLE_SYNCER;
1145 if (flags & MAP_NOCORE)
1146 docow |= MAP_DISABLE_COREDUMP;
1147
1148#if defined(VM_PROT_READ_IS_EXEC)
1149 if (prot & VM_PROT_READ)
1150 prot |= VM_PROT_EXECUTE;
1151
1152 if (maxprot & VM_PROT_READ)
1153 maxprot |= VM_PROT_EXECUTE;
1154#endif
1155
1156 if (fitit) {
1157 *addr = pmap_addr_hint(object, *addr, size);
1158 }
1159
1160 if (flags & MAP_STACK)
1161 rv = vm_map_stack (map, *addr, size, prot,
1162 maxprot, docow);
1163 else
1164 rv = vm_map_find(map, object, foff, addr, size, fitit,
1165 prot, maxprot, docow);
1166
1167 if (rv != KERN_SUCCESS) {
1168 /*
1169 * Lose the object reference. Will destroy the
1170 * object if it's an unnamed anonymous mapping
1171 * or named anonymous without other references.
1172 */
1173 vm_object_deallocate(object);
1174 goto out;
1175 }
1176
1177 /*
1178 * Shared memory is also shared with children.
1179 */
1180 if (flags & (MAP_SHARED|MAP_INHERIT)) {
1181 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1182 if (rv != KERN_SUCCESS) {
1183 (void) vm_map_remove(map, *addr, *addr + size);
1184 goto out;
1185 }
1186 }
1187out:
1188 switch (rv) {
1189 case KERN_SUCCESS:
1190 return (0);
1191 case KERN_INVALID_ADDRESS:
1192 case KERN_NO_SPACE:
1193 return (ENOMEM);
1194 case KERN_PROTECTION_FAILURE:
1195 return (EACCES);
1196 default:
1197 return (EINVAL);
1198 }
1199}