FreeBSD/Linux Kernel Cross Reference
sys/uvm/uvm_mmap.c
1 /* $NetBSD: uvm_mmap.c,v 1.82 2004/03/24 07:47:33 junyoung Exp $ */
2
3 /*
4 * Copyright (c) 1997 Charles D. Cranor and Washington University.
5 * Copyright (c) 1991, 1993 The Regents of the University of California.
6 * Copyright (c) 1988 University of Utah.
7 *
8 * All rights reserved.
9 *
10 * This code is derived from software contributed to Berkeley by
11 * the Systems Programming Group of the University of Utah Computer
12 * Science Department.
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. All advertising materials mentioning features or use of this software
23 * must display the following acknowledgement:
24 * This product includes software developed by the Charles D. Cranor,
25 * Washington University, University of California, Berkeley and
26 * its contributors.
27 * 4. Neither the name of the University nor the names of its contributors
28 * may be used to endorse or promote products derived from this software
29 * without specific prior written permission.
30 *
31 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41 * SUCH DAMAGE.
42 *
43 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
44 * @(#)vm_mmap.c 8.5 (Berkeley) 5/19/94
45 * from: Id: uvm_mmap.c,v 1.1.2.14 1998/01/05 21:04:26 chuck Exp
46 */
47
48 /*
49 * uvm_mmap.c: system call interface into VM system, plus kernel vm_mmap
50 * function.
51 */
52
53 #include <sys/cdefs.h>
54 __KERNEL_RCSID(0, "$NetBSD: uvm_mmap.c,v 1.82 2004/03/24 07:47:33 junyoung Exp $");
55
56 #include "opt_compat_netbsd.h"
57
58 #include <sys/param.h>
59 #include <sys/systm.h>
60 #include <sys/file.h>
61 #include <sys/filedesc.h>
62 #include <sys/resourcevar.h>
63 #include <sys/mman.h>
64 #include <sys/mount.h>
65 #include <sys/proc.h>
66 #include <sys/malloc.h>
67 #include <sys/vnode.h>
68 #include <sys/conf.h>
69 #include <sys/stat.h>
70
71 #include <miscfs/specfs/specdev.h>
72
73 #include <sys/sa.h>
74 #include <sys/syscallargs.h>
75
76 #include <uvm/uvm.h>
77 #include <uvm/uvm_device.h>
78
79 #ifndef COMPAT_ZERODEV
80 #define COMPAT_ZERODEV(dev) (0)
81 #endif
82
83 /*
84 * unimplemented VM system calls:
85 */
86
87 /*
88 * sys_sbrk: sbrk system call.
89 */
90
91 /* ARGSUSED */
92 int
93 sys_sbrk(l, v, retval)
94 struct lwp *l;
95 void *v;
96 register_t *retval;
97 {
98 #if 0
99 struct sys_sbrk_args /* {
100 syscallarg(intptr_t) incr;
101 } */ *uap = v;
102 #endif
103
104 return (ENOSYS);
105 }
106
107 /*
108 * sys_sstk: sstk system call.
109 */
110
111 /* ARGSUSED */
112 int
113 sys_sstk(l, v, retval)
114 struct lwp *l;
115 void *v;
116 register_t *retval;
117 {
118 #if 0
119 struct sys_sstk_args /* {
120 syscallarg(int) incr;
121 } */ *uap = v;
122 #endif
123
124 return (ENOSYS);
125 }
126
127 /*
128 * sys_mincore: determine if pages are in core or not.
129 */
130
131 /* ARGSUSED */
132 int
133 sys_mincore(l, v, retval)
134 struct lwp *l;
135 void *v;
136 register_t *retval;
137 {
138 struct sys_mincore_args /* {
139 syscallarg(void *) addr;
140 syscallarg(size_t) len;
141 syscallarg(char *) vec;
142 } */ *uap = v;
143 struct proc *p = l->l_proc;
144 struct vm_page *pg;
145 char *vec, pgi;
146 struct uvm_object *uobj;
147 struct vm_amap *amap;
148 struct vm_anon *anon;
149 struct vm_map_entry *entry;
150 vaddr_t start, end, lim;
151 struct vm_map *map;
152 vsize_t len;
153 int error = 0, npgs;
154
155 map = &p->p_vmspace->vm_map;
156
157 start = (vaddr_t)SCARG(uap, addr);
158 len = SCARG(uap, len);
159 vec = SCARG(uap, vec);
160
161 if (start & PAGE_MASK)
162 return (EINVAL);
163 len = round_page(len);
164 end = start + len;
165 if (end <= start)
166 return (EINVAL);
167
168 /*
169 * Lock down vec, so our returned status isn't outdated by
170 * storing the status byte for a page.
171 */
172
173 npgs = len >> PAGE_SHIFT;
174 error = uvm_vslock(p, vec, npgs, VM_PROT_WRITE);
175 if (error) {
176 return error;
177 }
178 vm_map_lock_read(map);
179
180 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
181 error = ENOMEM;
182 goto out;
183 }
184
185 for (/* nothing */;
186 entry != &map->header && entry->start < end;
187 entry = entry->next) {
188 KASSERT(!UVM_ET_ISSUBMAP(entry));
189 KASSERT(start >= entry->start);
190
191 /* Make sure there are no holes. */
192 if (entry->end < end &&
193 (entry->next == &map->header ||
194 entry->next->start > entry->end)) {
195 error = ENOMEM;
196 goto out;
197 }
198
199 lim = end < entry->end ? end : entry->end;
200
201 /*
202 * Special case for objects with no "real" pages. Those
203 * are always considered resident (mapped devices).
204 */
205
206 if (UVM_ET_ISOBJ(entry)) {
207 KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj));
208 if (UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) {
209 for (/* nothing */; start < lim;
210 start += PAGE_SIZE, vec++)
211 subyte(vec, 1);
212 continue;
213 }
214 }
215
216 amap = entry->aref.ar_amap; /* top layer */
217 uobj = entry->object.uvm_obj; /* bottom layer */
218
219 if (amap != NULL)
220 amap_lock(amap);
221 if (uobj != NULL)
222 simple_lock(&uobj->vmobjlock);
223
224 for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) {
225 pgi = 0;
226 if (amap != NULL) {
227 /* Check the top layer first. */
228 anon = amap_lookup(&entry->aref,
229 start - entry->start);
230 /* Don't need to lock anon here. */
231 if (anon != NULL && anon->u.an_page != NULL) {
232
233 /*
234 * Anon has the page for this entry
235 * offset.
236 */
237
238 pgi = 1;
239 }
240 }
241 if (uobj != NULL && pgi == 0) {
242 /* Check the bottom layer. */
243 pg = uvm_pagelookup(uobj,
244 entry->offset + (start - entry->start));
245 if (pg != NULL) {
246
247 /*
248 * Object has the page for this entry
249 * offset.
250 */
251
252 pgi = 1;
253 }
254 }
255 (void) subyte(vec, pgi);
256 }
257 if (uobj != NULL)
258 simple_unlock(&uobj->vmobjlock);
259 if (amap != NULL)
260 amap_unlock(amap);
261 }
262
263 out:
264 vm_map_unlock_read(map);
265 uvm_vsunlock(p, SCARG(uap, vec), npgs);
266 return (error);
267 }
268
269 /*
270 * sys_mmap: mmap system call.
271 *
272 * => file offset and address may not be page aligned
273 * - if MAP_FIXED, offset and address must have remainder mod PAGE_SIZE
274 * - if address isn't page aligned the mapping starts at trunc_page(addr)
275 * and the return value is adjusted up by the page offset.
276 */
277
278 int
279 sys_mmap(l, v, retval)
280 struct lwp *l;
281 void *v;
282 register_t *retval;
283 {
284 struct sys_mmap_args /* {
285 syscallarg(caddr_t) addr;
286 syscallarg(size_t) len;
287 syscallarg(int) prot;
288 syscallarg(int) flags;
289 syscallarg(int) fd;
290 syscallarg(long) pad;
291 syscallarg(off_t) pos;
292 } */ *uap = v;
293 struct proc *p = l->l_proc;
294 vaddr_t addr;
295 struct vattr va;
296 off_t pos;
297 vsize_t size, pageoff;
298 vm_prot_t prot, maxprot;
299 int flags, fd;
300 vaddr_t vm_min_address = VM_MIN_ADDRESS;
301 struct filedesc *fdp = p->p_fd;
302 struct file *fp;
303 struct vnode *vp;
304 void *handle;
305 int error;
306
307 /*
308 * first, extract syscall args from the uap.
309 */
310
311 addr = (vaddr_t)SCARG(uap, addr);
312 size = (vsize_t)SCARG(uap, len);
313 prot = SCARG(uap, prot) & VM_PROT_ALL;
314 flags = SCARG(uap, flags);
315 fd = SCARG(uap, fd);
316 pos = SCARG(uap, pos);
317
318 /*
319 * Fixup the old deprecated MAP_COPY into MAP_PRIVATE, and
320 * validate the flags.
321 */
322 if (flags & MAP_COPY)
323 flags = (flags & ~MAP_COPY) | MAP_PRIVATE;
324 if ((flags & (MAP_SHARED|MAP_PRIVATE)) == (MAP_SHARED|MAP_PRIVATE))
325 return (EINVAL);
326
327 /*
328 * align file position and save offset. adjust size.
329 */
330
331 pageoff = (pos & PAGE_MASK);
332 pos -= pageoff;
333 size += pageoff; /* add offset */
334 size = (vsize_t)round_page(size); /* round up */
335 if ((ssize_t) size < 0)
336 return (EINVAL); /* don't allow wrap */
337
338 #ifndef pmap_wired_count
339 /*
340 * if we're going to wire the mapping, restrict it to superuser.
341 */
342
343 if ((flags & MAP_WIRED) != 0 &&
344 (error = suser(p->p_ucred, &p->p_acflag)) != 0)
345 return (error);
346 #endif
347
348 /*
349 * now check (MAP_FIXED) or get (!MAP_FIXED) the "addr"
350 */
351
352 if (flags & MAP_FIXED) {
353
354 /* ensure address and file offset are aligned properly */
355 addr -= pageoff;
356 if (addr & PAGE_MASK)
357 return (EINVAL);
358
359 if (VM_MAXUSER_ADDRESS > 0 &&
360 (addr + size) > VM_MAXUSER_ADDRESS)
361 return (EFBIG);
362 if (vm_min_address > 0 && addr < vm_min_address)
363 return (EINVAL);
364 if (addr > addr + size)
365 return (EOVERFLOW); /* no wrapping! */
366
367 } else if (addr == 0 || !(flags & MAP_TRYFIXED)) {
368
369 /*
370 * not fixed: make sure we skip over the largest
371 * possible heap for non-topdown mapping arrangements.
372 * we will refine our guess later (e.g. to account for
373 * VAC, etc)
374 */
375
376 if (addr == 0 ||
377 !(p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN))
378 addr = MAX(addr,
379 VM_DEFAULT_ADDRESS(p->p_vmspace->vm_daddr, size));
380 else
381 addr = MIN(addr,
382 VM_DEFAULT_ADDRESS(p->p_vmspace->vm_daddr, size));
383 }
384
385 /*
386 * check for file mappings (i.e. not anonymous) and verify file.
387 */
388
389 if ((flags & MAP_ANON) == 0) {
390
391 if ((fp = fd_getfile(fdp, fd)) == NULL)
392 return (EBADF);
393
394 simple_unlock(&fp->f_slock);
395
396 if (fp->f_type != DTYPE_VNODE)
397 return (ENODEV); /* only mmap vnodes! */
398 vp = (struct vnode *)fp->f_data; /* convert to vnode */
399
400 if (vp->v_type != VREG && vp->v_type != VCHR &&
401 vp->v_type != VBLK)
402 return (ENODEV); /* only REG/CHR/BLK support mmap */
403
404 if (vp->v_type != VCHR && pos < 0)
405 return (EINVAL);
406
407 if (vp->v_type != VCHR && (pos + size) < pos)
408 return (EOVERFLOW); /* no offset wrapping */
409
410 /* special case: catch SunOS style /dev/zero */
411 if (vp->v_type == VCHR
412 && (vp->v_rdev == zerodev || COMPAT_ZERODEV(vp->v_rdev))) {
413 flags |= MAP_ANON;
414 goto is_anon;
415 }
416
417 /*
418 * Old programs may not select a specific sharing type, so
419 * default to an appropriate one.
420 *
421 * XXX: how does MAP_ANON fit in the picture?
422 */
423 if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
424 #if defined(DEBUG)
425 printf("WARNING: defaulted mmap() share type to "
426 "%s (pid %d command %s)\n", vp->v_type == VCHR ?
427 "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
428 p->p_comm);
429 #endif
430 if (vp->v_type == VCHR)
431 flags |= MAP_SHARED; /* for a device */
432 else
433 flags |= MAP_PRIVATE; /* for a file */
434 }
435
436 /*
437 * MAP_PRIVATE device mappings don't make sense (and aren't
438 * supported anyway). However, some programs rely on this,
439 * so just change it to MAP_SHARED.
440 */
441 if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
442 flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
443 }
444
445 /*
446 * now check protection
447 */
448
449 maxprot = VM_PROT_EXECUTE;
450
451 /* check read access */
452 if (fp->f_flag & FREAD)
453 maxprot |= VM_PROT_READ;
454 else if (prot & PROT_READ)
455 return (EACCES);
456
457 /* check write access, shared case first */
458 if (flags & MAP_SHARED) {
459 /*
460 * if the file is writable, only add PROT_WRITE to
461 * maxprot if the file is not immutable, append-only.
462 * otherwise, if we have asked for PROT_WRITE, return
463 * EPERM.
464 */
465 if (fp->f_flag & FWRITE) {
466 if ((error =
467 VOP_GETATTR(vp, &va, p->p_ucred, p)))
468 return (error);
469 if ((va.va_flags & (IMMUTABLE|APPEND)) == 0)
470 maxprot |= VM_PROT_WRITE;
471 else if (prot & PROT_WRITE)
472 return (EPERM);
473 }
474 else if (prot & PROT_WRITE)
475 return (EACCES);
476 } else {
477 /* MAP_PRIVATE mappings can always write to */
478 maxprot |= VM_PROT_WRITE;
479 }
480 handle = vp;
481
482 } else { /* MAP_ANON case */
483 /*
484 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0?
485 */
486 if (fd != -1)
487 return (EINVAL);
488
489 is_anon: /* label for SunOS style /dev/zero */
490 handle = NULL;
491 maxprot = VM_PROT_ALL;
492 pos = 0;
493 }
494
495 /*
496 * XXX (in)sanity check. We don't do proper datasize checking
497 * XXX for anonymous (or private writable) mmap(). However,
498 * XXX know that if we're trying to allocate more than the amount
499 * XXX remaining under our current data size limit, _that_ should
500 * XXX be disallowed.
501 */
502 if ((flags & MAP_ANON) != 0 ||
503 ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) {
504 if (size >
505 (p->p_rlimit[RLIMIT_DATA].rlim_cur -
506 ctob(p->p_vmspace->vm_dsize))) {
507 return (ENOMEM);
508 }
509 }
510
511 /*
512 * now let kernel internal function uvm_mmap do the work.
513 */
514
515 error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot,
516 flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
517
518 if (error == 0)
519 /* remember to add offset */
520 *retval = (register_t)(addr + pageoff);
521
522 return (error);
523 }
524
525 /*
526 * sys___msync13: the msync system call (a front-end for flush)
527 */
528
529 int
530 sys___msync13(l, v, retval)
531 struct lwp *l;
532 void *v;
533 register_t *retval;
534 {
535 struct sys___msync13_args /* {
536 syscallarg(caddr_t) addr;
537 syscallarg(size_t) len;
538 syscallarg(int) flags;
539 } */ *uap = v;
540 struct proc *p = l->l_proc;
541 vaddr_t addr;
542 vsize_t size, pageoff;
543 struct vm_map *map;
544 int error, rv, flags, uvmflags;
545
546 /*
547 * extract syscall args from the uap
548 */
549
550 addr = (vaddr_t)SCARG(uap, addr);
551 size = (vsize_t)SCARG(uap, len);
552 flags = SCARG(uap, flags);
553
554 /* sanity check flags */
555 if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 ||
556 (flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 ||
557 (flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC))
558 return (EINVAL);
559 if ((flags & (MS_ASYNC | MS_SYNC)) == 0)
560 flags |= MS_SYNC;
561
562 /*
563 * align the address to a page boundary and adjust the size accordingly.
564 */
565
566 pageoff = (addr & PAGE_MASK);
567 addr -= pageoff;
568 size += pageoff;
569 size = (vsize_t)round_page(size);
570
571 /* disallow wrap-around. */
572 if (addr + size < addr)
573 return (EINVAL);
574
575 /*
576 * get map
577 */
578
579 map = &p->p_vmspace->vm_map;
580
581 /*
582 * XXXCDC: do we really need this semantic?
583 *
584 * XXX Gak! If size is zero we are supposed to sync "all modified
585 * pages with the region containing addr". Unfortunately, we
586 * don't really keep track of individual mmaps so we approximate
587 * by flushing the range of the map entry containing addr.
588 * This can be incorrect if the region splits or is coalesced
589 * with a neighbor.
590 */
591
592 if (size == 0) {
593 struct vm_map_entry *entry;
594
595 vm_map_lock_read(map);
596 rv = uvm_map_lookup_entry(map, addr, &entry);
597 if (rv == TRUE) {
598 addr = entry->start;
599 size = entry->end - entry->start;
600 }
601 vm_map_unlock_read(map);
602 if (rv == FALSE)
603 return (EINVAL);
604 }
605
606 /*
607 * translate MS_ flags into PGO_ flags
608 */
609
610 uvmflags = PGO_CLEANIT;
611 if (flags & MS_INVALIDATE)
612 uvmflags |= PGO_FREE;
613 if (flags & MS_SYNC)
614 uvmflags |= PGO_SYNCIO;
615 else
616 uvmflags |= PGO_SYNCIO; /* XXXCDC: force sync for now! */
617
618 error = uvm_map_clean(map, addr, addr+size, uvmflags);
619 return error;
620 }
621
622 /*
623 * sys_munmap: unmap a users memory
624 */
625
626 int
627 sys_munmap(l, v, retval)
628 struct lwp *l;
629 void *v;
630 register_t *retval;
631 {
632 struct sys_munmap_args /* {
633 syscallarg(caddr_t) addr;
634 syscallarg(size_t) len;
635 } */ *uap = v;
636 struct proc *p = l->l_proc;
637 vaddr_t addr;
638 vsize_t size, pageoff;
639 struct vm_map *map;
640 vaddr_t vm_min_address = VM_MIN_ADDRESS;
641 struct vm_map_entry *dead_entries;
642
643 /*
644 * get syscall args.
645 */
646
647 addr = (vaddr_t)SCARG(uap, addr);
648 size = (vsize_t)SCARG(uap, len);
649
650 /*
651 * align the address to a page boundary and adjust the size accordingly.
652 */
653
654 pageoff = (addr & PAGE_MASK);
655 addr -= pageoff;
656 size += pageoff;
657 size = (vsize_t)round_page(size);
658
659 if ((int)size < 0)
660 return (EINVAL);
661 if (size == 0)
662 return (0);
663
664 /*
665 * Check for illegal addresses. Watch out for address wrap...
666 * Note that VM_*_ADDRESS are not constants due to casts (argh).
667 */
668 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
669 return (EINVAL);
670 if (vm_min_address > 0 && addr < vm_min_address)
671 return (EINVAL);
672 if (addr > addr + size)
673 return (EINVAL);
674 map = &p->p_vmspace->vm_map;
675
676 /*
677 * interesting system call semantic: make sure entire range is
678 * allocated before allowing an unmap.
679 */
680
681 vm_map_lock(map);
682 #if 0
683 if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
684 vm_map_unlock(map);
685 return (EINVAL);
686 }
687 #endif
688 uvm_unmap_remove(map, addr, addr + size, &dead_entries);
689 vm_map_unlock(map);
690 if (dead_entries != NULL)
691 uvm_unmap_detach(dead_entries, 0);
692 return (0);
693 }
694
695 /*
696 * sys_mprotect: the mprotect system call
697 */
698
699 int
700 sys_mprotect(l, v, retval)
701 struct lwp *l;
702 void *v;
703 register_t *retval;
704 {
705 struct sys_mprotect_args /* {
706 syscallarg(caddr_t) addr;
707 syscallarg(size_t) len;
708 syscallarg(int) prot;
709 } */ *uap = v;
710 struct proc *p = l->l_proc;
711 vaddr_t addr;
712 vsize_t size, pageoff;
713 vm_prot_t prot;
714 int error;
715
716 /*
717 * extract syscall args from uap
718 */
719
720 addr = (vaddr_t)SCARG(uap, addr);
721 size = (vsize_t)SCARG(uap, len);
722 prot = SCARG(uap, prot) & VM_PROT_ALL;
723
724 /*
725 * align the address to a page boundary and adjust the size accordingly.
726 */
727
728 pageoff = (addr & PAGE_MASK);
729 addr -= pageoff;
730 size += pageoff;
731 size = round_page(size);
732
733 error = uvm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
734 FALSE);
735 return error;
736 }
737
738 /*
739 * sys_minherit: the minherit system call
740 */
741
742 int
743 sys_minherit(l, v, retval)
744 struct lwp *l;
745 void *v;
746 register_t *retval;
747 {
748 struct sys_minherit_args /* {
749 syscallarg(caddr_t) addr;
750 syscallarg(int) len;
751 syscallarg(int) inherit;
752 } */ *uap = v;
753 struct proc *p = l->l_proc;
754 vaddr_t addr;
755 vsize_t size, pageoff;
756 vm_inherit_t inherit;
757 int error;
758
759 addr = (vaddr_t)SCARG(uap, addr);
760 size = (vsize_t)SCARG(uap, len);
761 inherit = SCARG(uap, inherit);
762
763 /*
764 * align the address to a page boundary and adjust the size accordingly.
765 */
766
767 pageoff = (addr & PAGE_MASK);
768 addr -= pageoff;
769 size += pageoff;
770 size = (vsize_t)round_page(size);
771
772 if ((int)size < 0)
773 return (EINVAL);
774 error = uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr + size,
775 inherit);
776 return error;
777 }
778
779 /*
780 * sys_madvise: give advice about memory usage.
781 */
782
783 /* ARGSUSED */
784 int
785 sys_madvise(l, v, retval)
786 struct lwp *l;
787 void *v;
788 register_t *retval;
789 {
790 struct sys_madvise_args /* {
791 syscallarg(caddr_t) addr;
792 syscallarg(size_t) len;
793 syscallarg(int) behav;
794 } */ *uap = v;
795 struct proc *p = l->l_proc;
796 vaddr_t addr;
797 vsize_t size, pageoff;
798 int advice, error;
799
800 addr = (vaddr_t)SCARG(uap, addr);
801 size = (vsize_t)SCARG(uap, len);
802 advice = SCARG(uap, behav);
803
804 /*
805 * align the address to a page boundary, and adjust the size accordingly
806 */
807
808 pageoff = (addr & PAGE_MASK);
809 addr -= pageoff;
810 size += pageoff;
811 size = (vsize_t)round_page(size);
812
813 if ((ssize_t)size <= 0)
814 return (EINVAL);
815
816 switch (advice) {
817 case MADV_NORMAL:
818 case MADV_RANDOM:
819 case MADV_SEQUENTIAL:
820 error = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size,
821 advice);
822 break;
823
824 case MADV_WILLNEED:
825
826 /*
827 * Activate all these pages, pre-faulting them in if
828 * necessary.
829 */
830 /*
831 * XXX IMPLEMENT ME.
832 * Should invent a "weak" mode for uvm_fault()
833 * which would only do the PGO_LOCKED pgo_get().
834 */
835
836 return (0);
837
838 case MADV_DONTNEED:
839
840 /*
841 * Deactivate all these pages. We don't need them
842 * any more. We don't, however, toss the data in
843 * the pages.
844 */
845
846 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
847 PGO_DEACTIVATE);
848 break;
849
850 case MADV_FREE:
851
852 /*
853 * These pages contain no valid data, and may be
854 * garbage-collected. Toss all resources, including
855 * any swap space in use.
856 */
857
858 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
859 PGO_FREE);
860 break;
861
862 case MADV_SPACEAVAIL:
863
864 /*
865 * XXXMRG What is this? I think it's:
866 *
867 * Ensure that we have allocated backing-store
868 * for these pages.
869 *
870 * This is going to require changes to the page daemon,
871 * as it will free swap space allocated to pages in core.
872 * There's also what to do for device/file/anonymous memory.
873 */
874
875 return (EINVAL);
876
877 default:
878 return (EINVAL);
879 }
880
881 return error;
882 }
883
884 /*
885 * sys_mlock: memory lock
886 */
887
888 int
889 sys_mlock(l, v, retval)
890 struct lwp *l;
891 void *v;
892 register_t *retval;
893 {
894 struct sys_mlock_args /* {
895 syscallarg(const void *) addr;
896 syscallarg(size_t) len;
897 } */ *uap = v;
898 struct proc *p = l->l_proc;
899 vaddr_t addr;
900 vsize_t size, pageoff;
901 int error;
902
903 /*
904 * extract syscall args from uap
905 */
906
907 addr = (vaddr_t)SCARG(uap, addr);
908 size = (vsize_t)SCARG(uap, len);
909
910 /*
911 * align the address to a page boundary and adjust the size accordingly
912 */
913
914 pageoff = (addr & PAGE_MASK);
915 addr -= pageoff;
916 size += pageoff;
917 size = (vsize_t)round_page(size);
918
919 /* disallow wrap-around. */
920 if (addr + size < addr)
921 return (EINVAL);
922
923 if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
924 return (EAGAIN);
925
926 #ifdef pmap_wired_count
927 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
928 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
929 return (EAGAIN);
930 #else
931 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
932 return (error);
933 #endif
934
935 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
936 0);
937 return error;
938 }
939
940 /*
941 * sys_munlock: unlock wired pages
942 */
943
944 int
945 sys_munlock(l, v, retval)
946 struct lwp *l;
947 void *v;
948 register_t *retval;
949 {
950 struct sys_munlock_args /* {
951 syscallarg(const void *) addr;
952 syscallarg(size_t) len;
953 } */ *uap = v;
954 struct proc *p = l->l_proc;
955 vaddr_t addr;
956 vsize_t size, pageoff;
957 int error;
958
959 /*
960 * extract syscall args from uap
961 */
962
963 addr = (vaddr_t)SCARG(uap, addr);
964 size = (vsize_t)SCARG(uap, len);
965
966 /*
967 * align the address to a page boundary, and adjust the size accordingly
968 */
969
970 pageoff = (addr & PAGE_MASK);
971 addr -= pageoff;
972 size += pageoff;
973 size = (vsize_t)round_page(size);
974
975 /* disallow wrap-around. */
976 if (addr + size < addr)
977 return (EINVAL);
978
979 #ifndef pmap_wired_count
980 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
981 return (error);
982 #endif
983
984 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
985 0);
986 return error;
987 }
988
989 /*
990 * sys_mlockall: lock all pages mapped into an address space.
991 */
992
993 int
994 sys_mlockall(l, v, retval)
995 struct lwp *l;
996 void *v;
997 register_t *retval;
998 {
999 struct sys_mlockall_args /* {
1000 syscallarg(int) flags;
1001 } */ *uap = v;
1002 struct proc *p = l->l_proc;
1003 int error, flags;
1004
1005 flags = SCARG(uap, flags);
1006
1007 if (flags == 0 ||
1008 (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
1009 return (EINVAL);
1010
1011 #ifndef pmap_wired_count
1012 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
1013 return (error);
1014 #endif
1015
1016 error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
1017 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
1018 return (error);
1019 }
1020
1021 /*
1022 * sys_munlockall: unlock all pages mapped into an address space.
1023 */
1024
1025 int
1026 sys_munlockall(l, v, retval)
1027 struct lwp *l;
1028 void *v;
1029 register_t *retval;
1030 {
1031 struct proc *p = l->l_proc;
1032
1033 (void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
1034 return (0);
1035 }
1036
1037 /*
1038 * uvm_mmap: internal version of mmap
1039 *
1040 * - used by sys_mmap and various framebuffers
1041 * - handle is a vnode pointer or NULL for MAP_ANON
1042 * - caller must page-align the file offset
1043 */
1044
1045 int
1046 uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit)
1047 struct vm_map *map;
1048 vaddr_t *addr;
1049 vsize_t size;
1050 vm_prot_t prot, maxprot;
1051 int flags;
1052 void *handle;
1053 voff_t foff;
1054 vsize_t locklimit;
1055 {
1056 struct uvm_object *uobj;
1057 struct vnode *vp;
1058 vaddr_t align = 0;
1059 int error;
1060 int advice = UVM_ADV_NORMAL;
1061 uvm_flag_t uvmflag = 0;
1062
1063 /*
1064 * check params
1065 */
1066
1067 if (size == 0)
1068 return(0);
1069 if (foff & PAGE_MASK)
1070 return(EINVAL);
1071 if ((prot & maxprot) != prot)
1072 return(EINVAL);
1073
1074 /*
1075 * for non-fixed mappings, round off the suggested address.
1076 * for fixed mappings, check alignment and zap old mappings.
1077 */
1078
1079 if ((flags & MAP_FIXED) == 0) {
1080 *addr = round_page(*addr);
1081 } else {
1082 if (*addr & PAGE_MASK)
1083 return(EINVAL);
1084 uvmflag |= UVM_FLAG_FIXED;
1085 (void) uvm_unmap(map, *addr, *addr + size);
1086 }
1087
1088 /*
1089 * Try to see if any requested alignment can even be attemped.
1090 * Make sure we can express the alignment (asking for a >= 4GB
1091 * alignment on an ILP32 architecure make no sense) and the
1092 * alignment is at least for a page sized quanitiy. If the
1093 * request was for a fixed mapping, make sure supplied address
1094 * adheres to the request alignment.
1095 */
1096 align = (flags & MAP_ALIGNMENT_MASK) >> MAP_ALIGNMENT_SHIFT;
1097 if (align) {
1098 if (align >= sizeof(vaddr_t) * NBBY)
1099 return(EINVAL);
1100 align = 1L << align;
1101 if (align < PAGE_SIZE)
1102 return(EINVAL);
1103 if (align >= map->max_offset)
1104 return(ENOMEM);
1105 if (flags & MAP_FIXED) {
1106 if ((*addr & (align-1)) != 0)
1107 return(EINVAL);
1108 align = 0;
1109 }
1110 }
1111
1112 /*
1113 * handle anon vs. non-anon mappings. for non-anon mappings attach
1114 * to underlying vm object.
1115 */
1116
1117 if (flags & MAP_ANON) {
1118 foff = UVM_UNKNOWN_OFFSET;
1119 uobj = NULL;
1120 if ((flags & MAP_SHARED) == 0)
1121 /* XXX: defer amap create */
1122 uvmflag |= UVM_FLAG_COPYONW;
1123 else
1124 /* shared: create amap now */
1125 uvmflag |= UVM_FLAG_OVERLAY;
1126
1127 } else {
1128 vp = (struct vnode *)handle;
1129
1130 /*
1131 * Don't allow mmap for EXEC if the file system
1132 * is mounted NOEXEC.
1133 */
1134 if ((prot & PROT_EXEC) != 0 &&
1135 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0)
1136 return (EACCES);
1137
1138 if (vp->v_type != VCHR) {
1139 error = VOP_MMAP(vp, 0, curproc->p_ucred, curproc);
1140 if (error) {
1141 return error;
1142 }
1143
1144 uobj = uvn_attach((void *)vp, (flags & MAP_SHARED) ?
1145 maxprot : (maxprot & ~VM_PROT_WRITE));
1146
1147 /* XXX for now, attach doesn't gain a ref */
1148 VREF(vp);
1149
1150 /*
1151 * If the vnode is being mapped with PROT_EXEC,
1152 * then mark it as text.
1153 */
1154 if (prot & PROT_EXEC)
1155 vn_markexec(vp);
1156 } else {
1157 uobj = udv_attach((void *) &vp->v_rdev,
1158 (flags & MAP_SHARED) ? maxprot :
1159 (maxprot & ~VM_PROT_WRITE), foff, size);
1160 /*
1161 * XXX Some devices don't like to be mapped with
1162 * XXX PROT_EXEC, but we don't really have a
1163 * XXX better way of handling this, right now
1164 */
1165 if (uobj == NULL && (prot & PROT_EXEC) == 0) {
1166 maxprot &= ~VM_PROT_EXECUTE;
1167 uobj = udv_attach((void *)&vp->v_rdev,
1168 (flags & MAP_SHARED) ? maxprot :
1169 (maxprot & ~VM_PROT_WRITE), foff, size);
1170 }
1171 advice = UVM_ADV_RANDOM;
1172 }
1173 if (uobj == NULL)
1174 return((vp->v_type == VREG) ? ENOMEM : EINVAL);
1175 if ((flags & MAP_SHARED) == 0)
1176 uvmflag |= UVM_FLAG_COPYONW;
1177 }
1178
1179 uvmflag = UVM_MAPFLAG(prot, maxprot,
1180 (flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
1181 advice, uvmflag);
1182 error = uvm_map(map, addr, size, uobj, foff, align, uvmflag);
1183 if (error) {
1184 if (uobj)
1185 uobj->pgops->pgo_detach(uobj);
1186 return error;
1187 }
1188
1189 /*
1190 * POSIX 1003.1b -- if our address space was configured
1191 * to lock all future mappings, wire the one we just made.
1192 *
1193 * Also handle the MAP_WIRED flag here.
1194 */
1195
1196 if (prot == VM_PROT_NONE) {
1197
1198 /*
1199 * No more work to do in this case.
1200 */
1201
1202 return (0);
1203 }
1204 vm_map_lock(map);
1205 if ((flags & MAP_WIRED) != 0 || (map->flags & VM_MAP_WIREFUTURE) != 0) {
1206 if ((atop(size) + uvmexp.wired) > uvmexp.wiredmax
1207 #ifdef pmap_wired_count
1208 || (locklimit != 0 && (size +
1209 ptoa(pmap_wired_count(vm_map_pmap(map)))) >
1210 locklimit)
1211 #endif
1212 ) {
1213 vm_map_unlock(map);
1214 uvm_unmap(map, *addr, *addr + size);
1215 return ENOMEM;
1216 }
1217
1218 /*
1219 * uvm_map_pageable() always returns the map unlocked.
1220 */
1221
1222 error = uvm_map_pageable(map, *addr, *addr + size,
1223 FALSE, UVM_LK_ENTER);
1224 if (error) {
1225 uvm_unmap(map, *addr, *addr + size);
1226 return error;
1227 }
1228 return (0);
1229 }
1230 vm_map_unlock(map);
1231 return 0;
1232 }
Cache object: b4a2ebe01f6c52570ed3c8b1bf95ab51
|