proc->thread stage 4: rework the VFS and DEVICE subsystems to take thread
[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 $
dadab5e9 42 * $DragonFly: src/sys/vm/vm_mmap.c,v 1.4 2003/06/25 03:56:13 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
dadab5e9
MD
81#include <sys/file2.h>
82
984263bc
MD
83#ifndef _SYS_SYSPROTO_H_
84struct sbrk_args {
85 int incr;
86};
87#endif
88
89static int max_proc_mmap;
90SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
91
92/*
93 * Set the maximum number of vm_map_entry structures per process. Roughly
94 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
95 * of our KVM malloc space still results in generous limits. We want a
96 * default that is good enough to prevent the kernel running out of resources
97 * if attacked from compromised user account but generous enough such that
98 * multi-threaded processes are not unduly inconvenienced.
99 */
100
101static void vmmapentry_rsrc_init __P((void *));
102SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
103
104static void
105vmmapentry_rsrc_init(dummy)
106 void *dummy;
107{
108 max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
109 max_proc_mmap /= 100;
110}
111
112/* ARGSUSED */
113int
41c20dac 114sbrk(struct sbrk_args *uap)
984263bc 115{
984263bc
MD
116 /* Not yet implemented */
117 return (EOPNOTSUPP);
118}
119
41c20dac
MD
120/*
121 * sstk_args(int incr)
122 */
984263bc
MD
123/* ARGSUSED */
124int
41c20dac 125sstk(struct sstk_args *uap)
984263bc 126{
984263bc
MD
127 /* Not yet implemented */
128 return (EOPNOTSUPP);
129}
130
131#if defined(COMPAT_43) || defined(COMPAT_SUNOS)
984263bc 132
41c20dac
MD
133/*
134 * getpagesize_args(int dummy)
135 */
984263bc
MD
136/* ARGSUSED */
137int
41c20dac 138ogetpagesize(struct getpagesize_args *uap)
984263bc 139{
41c20dac 140 struct proc *p = curproc;
984263bc
MD
141 p->p_retval[0] = PAGE_SIZE;
142 return (0);
143}
144#endif /* COMPAT_43 || COMPAT_SUNOS */
145
146
147/*
41c20dac
MD
148 * mmap_args(void *addr, size_t len, int prot, int flags, int fd,
149 * long pad, off_t pos)
150 *
984263bc
MD
151 * Memory Map (mmap) system call. Note that the file offset
152 * and address are allowed to be NOT page aligned, though if
153 * the MAP_FIXED flag it set, both must have the same remainder
154 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
155 * page-aligned, the actual mapping starts at trunc_page(addr)
156 * and the return value is adjusted up by the page offset.
157 *
158 * Generally speaking, only character devices which are themselves
159 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
160 * there would be no cache coherency between a descriptor and a VM mapping
161 * both to the same character device.
162 *
163 * Block devices can be mmap'd no matter what they represent. Cache coherency
164 * is maintained as long as you do not write directly to the underlying
165 * character device.
166 */
984263bc
MD
167
168int
41c20dac 169mmap(struct mmap_args *uap)
984263bc 170{
dadab5e9
MD
171 struct thread *td = curthread;
172 struct proc *p = td->td_proc;
41c20dac
MD
173 struct filedesc *fdp = p->p_fd;
174 struct file *fp = NULL;
984263bc
MD
175 struct vnode *vp;
176 vm_offset_t addr;
177 vm_size_t size, pageoff;
178 vm_prot_t prot, maxprot;
179 void *handle;
180 int flags, error;
181 int disablexworkaround;
182 off_t pos;
183 struct vmspace *vms = p->p_vmspace;
184 vm_object_t obj;
185
dadab5e9
MD
186 KKASSERT(p);
187
984263bc
MD
188 addr = (vm_offset_t) uap->addr;
189 size = uap->len;
190 prot = uap->prot & VM_PROT_ALL;
191 flags = uap->flags;
192 pos = uap->pos;
193
194 /* make sure mapping fits into numeric range etc */
195 if ((ssize_t) uap->len < 0 ||
196 ((flags & MAP_ANON) && uap->fd != -1))
197 return (EINVAL);
198
199 if (flags & MAP_STACK) {
200 if ((uap->fd != -1) ||
201 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
202 return (EINVAL);
203 flags |= MAP_ANON;
204 pos = 0;
205 }
206
207 /*
208 * Align the file position to a page boundary,
209 * and save its page offset component.
210 */
211 pageoff = (pos & PAGE_MASK);
212 pos -= pageoff;
213
214 /* Adjust size for rounding (on both ends). */
215 size += pageoff; /* low end... */
216 size = (vm_size_t) round_page(size); /* hi end */
217
218 /*
219 * Check for illegal addresses. Watch out for address wrap... Note
220 * that VM_*_ADDRESS are not constants due to casts (argh).
221 */
222 if (flags & MAP_FIXED) {
223 /*
224 * The specified address must have the same remainder
225 * as the file offset taken modulo PAGE_SIZE, so it
226 * should be aligned after adjustment by pageoff.
227 */
228 addr -= pageoff;
229 if (addr & PAGE_MASK)
230 return (EINVAL);
231 /* Address range must be all in user VM space. */
232 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
233 return (EINVAL);
234#ifndef i386
235 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
236 return (EINVAL);
237#endif
238 if (addr + size < addr)
239 return (EINVAL);
240 }
241 /*
242 * XXX for non-fixed mappings where no hint is provided or
243 * the hint would fall in the potential heap space,
244 * place it after the end of the largest possible heap.
245 *
246 * There should really be a pmap call to determine a reasonable
247 * location.
248 */
249 else if (addr == 0 ||
250 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
251 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
252 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
253
254 if (flags & MAP_ANON) {
255 /*
256 * Mapping blank space is trivial.
257 */
258 handle = NULL;
259 maxprot = VM_PROT_ALL;
260 pos = 0;
261 } else {
262 /*
263 * Mapping file, get fp for validation. Obtain vnode and make
264 * sure it is of appropriate type.
265 */
266 if (((unsigned) uap->fd) >= fdp->fd_nfiles ||
267 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
268 return (EBADF);
269 if (fp->f_type != DTYPE_VNODE)
270 return (EINVAL);
271 /*
272 * POSIX shared-memory objects are defined to have
273 * kernel persistence, and are not defined to support
274 * read(2)/write(2) -- or even open(2). Thus, we can
275 * use MAP_ASYNC to trade on-disk coherence for speed.
276 * The shm_open(3) library routine turns on the FPOSIXSHM
277 * flag to request this behavior.
278 */
279 if (fp->f_flag & FPOSIXSHM)
280 flags |= MAP_NOSYNC;
281 vp = (struct vnode *) fp->f_data;
282 if (vp->v_type != VREG && vp->v_type != VCHR)
283 return (EINVAL);
284 if (vp->v_type == VREG) {
285 /*
286 * Get the proper underlying object
287 */
288 if (VOP_GETVOBJECT(vp, &obj) != 0)
289 return (EINVAL);
290 vp = (struct vnode*)obj->handle;
291 }
292
293 /*
294 * don't let the descriptor disappear on us if we block
295 */
296 fhold(fp);
297
298 /*
299 * XXX hack to handle use of /dev/zero to map anon memory (ala
300 * SunOS).
301 */
302 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
303 handle = NULL;
304 maxprot = VM_PROT_ALL;
305 flags |= MAP_ANON;
306 pos = 0;
307 } else {
308 /*
309 * cdevs does not provide private mappings of any kind.
310 */
311 /*
312 * However, for XIG X server to continue to work,
313 * we should allow the superuser to do it anyway.
314 * We only allow it at securelevel < 1.
315 * (Because the XIG X server writes directly to video
316 * memory via /dev/mem, it should never work at any
317 * other securelevel.
318 * XXX this will have to go
319 */
320 if (securelevel >= 1)
321 disablexworkaround = 1;
322 else
dadab5e9 323 disablexworkaround = suser(td);
984263bc
MD
324 if (vp->v_type == VCHR && disablexworkaround &&
325 (flags & (MAP_PRIVATE|MAP_COPY))) {
326 error = EINVAL;
327 goto done;
328 }
329 /*
330 * Ensure that file and memory protections are
331 * compatible. Note that we only worry about
332 * writability if mapping is shared; in this case,
333 * current and max prot are dictated by the open file.
334 * XXX use the vnode instead? Problem is: what
335 * credentials do we use for determination? What if
336 * proc does a setuid?
337 */
338 maxprot = VM_PROT_EXECUTE; /* ??? */
339 if (fp->f_flag & FREAD) {
340 maxprot |= VM_PROT_READ;
341 } else if (prot & PROT_READ) {
342 error = EACCES;
343 goto done;
344 }
345 /*
346 * If we are sharing potential changes (either via
347 * MAP_SHARED or via the implicit sharing of character
348 * device mappings), and we are trying to get write
349 * permission although we opened it without asking
350 * for it, bail out. Check for superuser, only if
351 * we're at securelevel < 1, to allow the XIG X server
352 * to continue to work.
353 */
354
355 if ((flags & MAP_SHARED) != 0 ||
356 (vp->v_type == VCHR && disablexworkaround)) {
357 if ((fp->f_flag & FWRITE) != 0) {
358 struct vattr va;
359 if ((error =
dadab5e9 360 VOP_GETATTR(vp, &va, p->p_ucred, td))) {
984263bc
MD
361 goto done;
362 }
363 if ((va.va_flags &
364 (IMMUTABLE|APPEND)) == 0) {
365 maxprot |= VM_PROT_WRITE;
366 } else if (prot & PROT_WRITE) {
367 error = EPERM;
368 goto done;
369 }
370 } else if ((prot & PROT_WRITE) != 0) {
371 error = EACCES;
372 goto done;
373 }
374 } else {
375 maxprot |= VM_PROT_WRITE;
376 }
377 handle = (void *)vp;
378 }
379 }
380
381 /*
382 * Do not allow more then a certain number of vm_map_entry structures
383 * per process. Scale with the number of rforks sharing the map
384 * to make the limit reasonable for threads.
385 */
386 if (max_proc_mmap &&
387 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
388 error = ENOMEM;
389 goto done;
390 }
391
392 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
393 flags, handle, pos);
394 if (error == 0)
395 p->p_retval[0] = (register_t) (addr + pageoff);
396done:
397 if (fp)
dadab5e9 398 fdrop(fp, td);
984263bc
MD
399 return (error);
400}
401
402#ifdef COMPAT_43
41c20dac
MD
403/*
404 * ommap_args(caddr_t addr, int len, int prot, int flags, int fd, long pos)
405 */
984263bc 406int
41c20dac 407ommap(struct ommap_args *uap)
984263bc
MD
408{
409 struct mmap_args nargs;
410 static const char cvtbsdprot[8] = {
411 0,
412 PROT_EXEC,
413 PROT_WRITE,
414 PROT_EXEC | PROT_WRITE,
415 PROT_READ,
416 PROT_EXEC | PROT_READ,
417 PROT_WRITE | PROT_READ,
418 PROT_EXEC | PROT_WRITE | PROT_READ,
419 };
420
421#define OMAP_ANON 0x0002
422#define OMAP_COPY 0x0020
423#define OMAP_SHARED 0x0010
424#define OMAP_FIXED 0x0100
425#define OMAP_INHERIT 0x0800
426
427 nargs.addr = uap->addr;
428 nargs.len = uap->len;
429 nargs.prot = cvtbsdprot[uap->prot & 0x7];
430 nargs.flags = 0;
431 if (uap->flags & OMAP_ANON)
432 nargs.flags |= MAP_ANON;
433 if (uap->flags & OMAP_COPY)
434 nargs.flags |= MAP_COPY;
435 if (uap->flags & OMAP_SHARED)
436 nargs.flags |= MAP_SHARED;
437 else
438 nargs.flags |= MAP_PRIVATE;
439 if (uap->flags & OMAP_FIXED)
440 nargs.flags |= MAP_FIXED;
441 if (uap->flags & OMAP_INHERIT)
442 nargs.flags |= MAP_INHERIT;
443 nargs.fd = uap->fd;
444 nargs.pos = uap->pos;
41c20dac 445 return (mmap(&nargs));
984263bc
MD
446}
447#endif /* COMPAT_43 */
448
449
41c20dac
MD
450/*
451 * msync_args(void *addr, int len, int flags)
452 */
984263bc 453int
41c20dac 454msync(struct msync_args *uap)
984263bc 455{
41c20dac 456 struct proc *p = curproc;
984263bc
MD
457 vm_offset_t addr;
458 vm_size_t size, pageoff;
459 int flags;
460 vm_map_t map;
461 int rv;
462
463 addr = (vm_offset_t) uap->addr;
464 size = uap->len;
465 flags = uap->flags;
466
467 pageoff = (addr & PAGE_MASK);
468 addr -= pageoff;
469 size += pageoff;
470 size = (vm_size_t) round_page(size);
471 if (addr + size < addr)
472 return(EINVAL);
473
474 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
475 return (EINVAL);
476
477 map = &p->p_vmspace->vm_map;
478
479 /*
480 * XXX Gak! If size is zero we are supposed to sync "all modified
481 * pages with the region containing addr". Unfortunately, we don't
482 * really keep track of individual mmaps so we approximate by flushing
483 * the range of the map entry containing addr. This can be incorrect
484 * if the region splits or is coalesced with a neighbor.
485 */
486 if (size == 0) {
487 vm_map_entry_t entry;
488
489 vm_map_lock_read(map);
490 rv = vm_map_lookup_entry(map, addr, &entry);
491 vm_map_unlock_read(map);
492 if (rv == FALSE)
493 return (EINVAL);
494 addr = entry->start;
495 size = entry->end - entry->start;
496 }
497
498 /*
499 * Clean the pages and interpret the return value.
500 */
501 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
502 (flags & MS_INVALIDATE) != 0);
503
504 switch (rv) {
505 case KERN_SUCCESS:
506 break;
507 case KERN_INVALID_ADDRESS:
508 return (EINVAL); /* Sun returns ENOMEM? */
509 case KERN_FAILURE:
510 return (EIO);
511 default:
512 return (EINVAL);
513 }
514
515 return (0);
516}
517
41c20dac
MD
518/*
519 * munmap_args(void *addr, size_t len)
520 */
984263bc 521int
41c20dac 522munmap(struct munmap_args *uap)
984263bc 523{
41c20dac 524 struct proc *p = curproc;
984263bc
MD
525 vm_offset_t addr;
526 vm_size_t size, pageoff;
527 vm_map_t map;
528
529 addr = (vm_offset_t) uap->addr;
530 size = uap->len;
531
532 pageoff = (addr & PAGE_MASK);
533 addr -= pageoff;
534 size += pageoff;
535 size = (vm_size_t) round_page(size);
536 if (addr + size < addr)
537 return(EINVAL);
538
539 if (size == 0)
540 return (0);
541
542 /*
543 * Check for illegal addresses. Watch out for address wrap... Note
544 * that VM_*_ADDRESS are not constants due to casts (argh).
545 */
546 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
547 return (EINVAL);
548#ifndef i386
549 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
550 return (EINVAL);
551#endif
552 map = &p->p_vmspace->vm_map;
553 /*
554 * Make sure entire range is allocated.
555 */
556 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
557 return (EINVAL);
558 /* returns nothing but KERN_SUCCESS anyway */
559 (void) vm_map_remove(map, addr, addr + size);
560 return (0);
561}
562
563#if 0
564void
565munmapfd(p, fd)
566 struct proc *p;
567 int fd;
568{
569 /*
570 * XXX should unmap any regions mapped to this file
571 */
572 p->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
573}
574#endif
575
41c20dac
MD
576/*
577 * mprotect_args(const void *addr, size_t len, int prot)
578 */
984263bc 579int
41c20dac 580mprotect(struct mprotect_args *uap)
984263bc 581{
41c20dac 582 struct proc *p = curproc;
984263bc
MD
583 vm_offset_t addr;
584 vm_size_t size, pageoff;
585 register vm_prot_t prot;
586
587 addr = (vm_offset_t) uap->addr;
588 size = uap->len;
589 prot = uap->prot & VM_PROT_ALL;
590#if defined(VM_PROT_READ_IS_EXEC)
591 if (prot & VM_PROT_READ)
592 prot |= VM_PROT_EXECUTE;
593#endif
594
595 pageoff = (addr & PAGE_MASK);
596 addr -= pageoff;
597 size += pageoff;
598 size = (vm_size_t) round_page(size);
599 if (addr + size < addr)
600 return(EINVAL);
601
602 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
603 FALSE)) {
604 case KERN_SUCCESS:
605 return (0);
606 case KERN_PROTECTION_FAILURE:
607 return (EACCES);
608 }
609 return (EINVAL);
610}
611
41c20dac
MD
612/*
613 * minherit_args(void *addr, size_t len, int inherit)
614 */
984263bc 615int
41c20dac 616minherit(struct minherit_args *uap)
984263bc 617{
41c20dac 618 struct proc *p = curproc;
984263bc
MD
619 vm_offset_t addr;
620 vm_size_t size, pageoff;
621 register vm_inherit_t inherit;
622
623 addr = (vm_offset_t)uap->addr;
624 size = uap->len;
625 inherit = uap->inherit;
626
627 pageoff = (addr & PAGE_MASK);
628 addr -= pageoff;
629 size += pageoff;
630 size = (vm_size_t) round_page(size);
631 if (addr + size < addr)
632 return(EINVAL);
633
634 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
635 inherit)) {
636 case KERN_SUCCESS:
637 return (0);
638 case KERN_PROTECTION_FAILURE:
639 return (EACCES);
640 }
641 return (EINVAL);
642}
643
41c20dac
MD
644/*
645 * madvise_args(void *addr, size_t len, int behav)
646 */
984263bc
MD
647/* ARGSUSED */
648int
41c20dac 649madvise(struct madvise_args *uap)
984263bc 650{
41c20dac 651 struct proc *p = curproc;
984263bc
MD
652 vm_offset_t start, end;
653
654 /*
655 * Check for illegal behavior
656 */
657 if (uap->behav < 0 || uap->behav > MADV_CORE)
658 return (EINVAL);
659 /*
660 * Check for illegal addresses. Watch out for address wrap... Note
661 * that VM_*_ADDRESS are not constants due to casts (argh).
662 */
663 if (VM_MAXUSER_ADDRESS > 0 &&
664 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
665 return (EINVAL);
666#ifndef i386
667 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
668 return (EINVAL);
669#endif
670 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
671 return (EINVAL);
672
673 /*
674 * Since this routine is only advisory, we default to conservative
675 * behavior.
676 */
677 start = trunc_page((vm_offset_t) uap->addr);
678 end = round_page((vm_offset_t) uap->addr + uap->len);
679
680 if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav))
681 return (EINVAL);
682 return (0);
683}
684
41c20dac
MD
685/*
686 * mincore_args(const void *addr, size_t len, char *vec)
687 */
984263bc
MD
688/* ARGSUSED */
689int
41c20dac 690mincore(struct mincore_args *uap)
984263bc 691{
41c20dac 692 struct proc *p = curproc;
984263bc
MD
693 vm_offset_t addr, first_addr;
694 vm_offset_t end, cend;
695 pmap_t pmap;
696 vm_map_t map;
697 char *vec;
698 int error;
699 int vecindex, lastvecindex;
700 register vm_map_entry_t current;
701 vm_map_entry_t entry;
702 int mincoreinfo;
703 unsigned int timestamp;
704
705 /*
706 * Make sure that the addresses presented are valid for user
707 * mode.
708 */
709 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
710 end = addr + (vm_size_t)round_page(uap->len);
711 if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
712 return (EINVAL);
713 if (end < addr)
714 return (EINVAL);
715
716 /*
717 * Address of byte vector
718 */
719 vec = uap->vec;
720
721 map = &p->p_vmspace->vm_map;
722 pmap = vmspace_pmap(p->p_vmspace);
723
724 vm_map_lock_read(map);
725RestartScan:
726 timestamp = map->timestamp;
727
728 if (!vm_map_lookup_entry(map, addr, &entry))
729 entry = entry->next;
730
731 /*
732 * Do this on a map entry basis so that if the pages are not
733 * in the current processes address space, we can easily look
734 * up the pages elsewhere.
735 */
736 lastvecindex = -1;
737 for(current = entry;
738 (current != &map->header) && (current->start < end);
739 current = current->next) {
740
741 /*
742 * ignore submaps (for now) or null objects
743 */
744 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
745 current->object.vm_object == NULL)
746 continue;
747
748 /*
749 * limit this scan to the current map entry and the
750 * limits for the mincore call
751 */
752 if (addr < current->start)
753 addr = current->start;
754 cend = current->end;
755 if (cend > end)
756 cend = end;
757
758 /*
759 * scan this entry one page at a time
760 */
761 while(addr < cend) {
762 /*
763 * Check pmap first, it is likely faster, also
764 * it can provide info as to whether we are the
765 * one referencing or modifying the page.
766 */
767 mincoreinfo = pmap_mincore(pmap, addr);
768 if (!mincoreinfo) {
769 vm_pindex_t pindex;
770 vm_ooffset_t offset;
771 vm_page_t m;
772 /*
773 * calculate the page index into the object
774 */
775 offset = current->offset + (addr - current->start);
776 pindex = OFF_TO_IDX(offset);
777 m = vm_page_lookup(current->object.vm_object,
778 pindex);
779 /*
780 * if the page is resident, then gather information about
781 * it.
782 */
783 if (m) {
784 mincoreinfo = MINCORE_INCORE;
785 if (m->dirty ||
786 pmap_is_modified(m))
787 mincoreinfo |= MINCORE_MODIFIED_OTHER;
788 if ((m->flags & PG_REFERENCED) ||
789 pmap_ts_referenced(m)) {
790 vm_page_flag_set(m, PG_REFERENCED);
791 mincoreinfo |= MINCORE_REFERENCED_OTHER;
792 }
793 }
794 }
795
796 /*
797 * subyte may page fault. In case it needs to modify
798 * the map, we release the lock.
799 */
800 vm_map_unlock_read(map);
801
802 /*
803 * calculate index into user supplied byte vector
804 */
805 vecindex = OFF_TO_IDX(addr - first_addr);
806
807 /*
808 * If we have skipped map entries, we need to make sure that
809 * the byte vector is zeroed for those skipped entries.
810 */
811 while((lastvecindex + 1) < vecindex) {
812 error = subyte( vec + lastvecindex, 0);
813 if (error) {
814 return (EFAULT);
815 }
816 ++lastvecindex;
817 }
818
819 /*
820 * Pass the page information to the user
821 */
822 error = subyte( vec + vecindex, mincoreinfo);
823 if (error) {
824 return (EFAULT);
825 }
826
827 /*
828 * If the map has changed, due to the subyte, the previous
829 * output may be invalid.
830 */
831 vm_map_lock_read(map);
832 if (timestamp != map->timestamp)
833 goto RestartScan;
834
835 lastvecindex = vecindex;
836 addr += PAGE_SIZE;
837 }
838 }
839
840 /*
841 * subyte may page fault. In case it needs to modify
842 * the map, we release the lock.
843 */
844 vm_map_unlock_read(map);
845
846 /*
847 * Zero the last entries in the byte vector.
848 */
849 vecindex = OFF_TO_IDX(end - first_addr);
850 while((lastvecindex + 1) < vecindex) {
851 error = subyte( vec + lastvecindex, 0);
852 if (error) {
853 return (EFAULT);
854 }
855 ++lastvecindex;
856 }
857
858 /*
859 * If the map has changed, due to the subyte, the previous
860 * output may be invalid.
861 */
862 vm_map_lock_read(map);
863 if (timestamp != map->timestamp)
864 goto RestartScan;
865 vm_map_unlock_read(map);
866
867 return (0);
868}
869
41c20dac
MD
870/*
871 * mlock_args(const void *addr, size_t len)
872 */
984263bc 873int
41c20dac 874mlock(struct mlock_args *uap)
984263bc
MD
875{
876 vm_offset_t addr;
877 vm_size_t size, pageoff;
878 int error;
41c20dac 879 struct proc *p = curproc;
984263bc
MD
880
881 addr = (vm_offset_t) uap->addr;
882 size = uap->len;
883
884 pageoff = (addr & PAGE_MASK);
885 addr -= pageoff;
886 size += pageoff;
887 size = (vm_size_t) round_page(size);
888
889 /* disable wrap around */
890 if (addr + size < addr)
891 return (EINVAL);
892
893 if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
894 return (EAGAIN);
895
896#ifdef pmap_wired_count
897 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
898 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
899 return (ENOMEM);
900#else
dadab5e9 901 error = suser_cred(p->p_ucred, 0);
984263bc
MD
902 if (error)
903 return (error);
904#endif
905
906 error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
907 return (error == KERN_SUCCESS ? 0 : ENOMEM);
908}
909
41c20dac
MD
910/*
911 * mlockall_args(int how)
912 */
984263bc 913int
41c20dac 914mlockall(struct mlockall_args *uap)
984263bc
MD
915{
916 return 0;
917}
918
41c20dac
MD
919/*
920 * mlockall_args(int how)
921 */
984263bc 922int
41c20dac 923munlockall(struct munlockall_args *uap)
984263bc
MD
924{
925 return 0;
926}
927
41c20dac
MD
928/*
929 * munlock_args(const void *addr, size_t len)
930 */
984263bc 931int
41c20dac 932munlock(struct munlock_args *uap)
984263bc 933{
dadab5e9
MD
934 struct thread *td = curthread;
935 struct proc *p = td->td_proc;
984263bc
MD
936 vm_offset_t addr;
937 vm_size_t size, pageoff;
938 int error;
939
940 addr = (vm_offset_t) uap->addr;
941 size = uap->len;
942
943 pageoff = (addr & PAGE_MASK);
944 addr -= pageoff;
945 size += pageoff;
946 size = (vm_size_t) round_page(size);
947
948 /* disable wrap around */
949 if (addr + size < addr)
950 return (EINVAL);
951
952#ifndef pmap_wired_count
dadab5e9 953 error = suser(td);
984263bc
MD
954 if (error)
955 return (error);
956#endif
957
958 error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
959 return (error == KERN_SUCCESS ? 0 : ENOMEM);
960}
961
962/*
963 * Internal version of mmap.
964 * Currently used by mmap, exec, and sys5 shared memory.
965 * Handle is either a vnode pointer or NULL for MAP_ANON.
966 */
967int
968vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
969 vm_prot_t maxprot, int flags,
970 void *handle,
971 vm_ooffset_t foff)
972{
973 boolean_t fitit;
974 vm_object_t object;
975 struct vnode *vp = NULL;
976 objtype_t type;
977 int rv = KERN_SUCCESS;
978 vm_ooffset_t objsize;
979 int docow;
dadab5e9
MD
980 struct thread *td = curthread; /* XXX */
981 struct proc *p = td->td_proc;
982
983 KKASSERT(p);
984263bc
MD
984
985 if (size == 0)
986 return (0);
987
988 objsize = size = round_page(size);
989
990 if (p->p_vmspace->vm_map.size + size >
991 p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
992 return(ENOMEM);
993 }
994
995 /*
996 * We currently can only deal with page aligned file offsets.
997 * The check is here rather than in the syscall because the
998 * kernel calls this function internally for other mmaping
999 * operations (such as in exec) and non-aligned offsets will
1000 * cause pmap inconsistencies...so we want to be sure to
1001 * disallow this in all cases.
1002 */
1003 if (foff & PAGE_MASK)
1004 return (EINVAL);
1005
1006 if ((flags & MAP_FIXED) == 0) {
1007 fitit = TRUE;
1008 *addr = round_page(*addr);
1009 } else {
1010 if (*addr != trunc_page(*addr))
1011 return (EINVAL);
1012 fitit = FALSE;
1013 (void) vm_map_remove(map, *addr, *addr + size);
1014 }
1015
1016 /*
1017 * Lookup/allocate object.
1018 */
1019 if (flags & MAP_ANON) {
1020 type = OBJT_DEFAULT;
1021 /*
1022 * Unnamed anonymous regions always start at 0.
1023 */
1024 if (handle == 0)
1025 foff = 0;
1026 } else {
1027 vp = (struct vnode *) handle;
1028 if (vp->v_type == VCHR) {
1029 type = OBJT_DEVICE;
1030 handle = (void *)(intptr_t)vp->v_rdev;
1031 } else {
1032 struct vattr vat;
1033 int error;
1034
dadab5e9 1035 error = VOP_GETATTR(vp, &vat, p->p_ucred, td);
984263bc
MD
1036 if (error)
1037 return (error);
1038 objsize = round_page(vat.va_size);
1039 type = OBJT_VNODE;
1040 /*
1041 * if it is a regular file without any references
1042 * we do not need to sync it.
1043 */
1044 if (vp->v_type == VREG && vat.va_nlink == 0) {
1045 flags |= MAP_NOSYNC;
1046 }
1047 }
1048 }
1049
1050 if (handle == NULL) {
1051 object = NULL;
1052 docow = 0;
1053 } else {
1054 object = vm_pager_allocate(type,
1055 handle, objsize, prot, foff);
1056 if (object == NULL)
1057 return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1058 docow = MAP_PREFAULT_PARTIAL;
1059 }
1060
1061 /*
1062 * Force device mappings to be shared.
1063 */
1064 if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1065 flags &= ~(MAP_PRIVATE|MAP_COPY);
1066 flags |= MAP_SHARED;
1067 }
1068
1069 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1070 docow |= MAP_COPY_ON_WRITE;
1071 if (flags & MAP_NOSYNC)
1072 docow |= MAP_DISABLE_SYNCER;
1073 if (flags & MAP_NOCORE)
1074 docow |= MAP_DISABLE_COREDUMP;
1075
1076#if defined(VM_PROT_READ_IS_EXEC)
1077 if (prot & VM_PROT_READ)
1078 prot |= VM_PROT_EXECUTE;
1079
1080 if (maxprot & VM_PROT_READ)
1081 maxprot |= VM_PROT_EXECUTE;
1082#endif
1083
1084 if (fitit) {
1085 *addr = pmap_addr_hint(object, *addr, size);
1086 }
1087
1088 if (flags & MAP_STACK)
1089 rv = vm_map_stack (map, *addr, size, prot,
1090 maxprot, docow);
1091 else
1092 rv = vm_map_find(map, object, foff, addr, size, fitit,
1093 prot, maxprot, docow);
1094
1095 if (rv != KERN_SUCCESS) {
1096 /*
1097 * Lose the object reference. Will destroy the
1098 * object if it's an unnamed anonymous mapping
1099 * or named anonymous without other references.
1100 */
1101 vm_object_deallocate(object);
1102 goto out;
1103 }
1104
1105 /*
1106 * Shared memory is also shared with children.
1107 */
1108 if (flags & (MAP_SHARED|MAP_INHERIT)) {
1109 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1110 if (rv != KERN_SUCCESS) {
1111 (void) vm_map_remove(map, *addr, *addr + size);
1112 goto out;
1113 }
1114 }
1115out:
1116 switch (rv) {
1117 case KERN_SUCCESS:
1118 return (0);
1119 case KERN_INVALID_ADDRESS:
1120 case KERN_NO_SPACE:
1121 return (ENOMEM);
1122 case KERN_PROTECTION_FAILURE:
1123 return (EACCES);
1124 default:
1125 return (EINVAL);
1126 }
1127}