FreeBSD/Linux Kernel Cross Reference
sys/uvm/uvm_mmap.c
1 /* $NetBSD: uvm_mmap.c,v 1.88.4.3 2005/10/15 15:34:08 riz 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 1.88.4.3 2005/10/15 15:34:08 riz 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, defaddr;
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 /*
339 * now check (MAP_FIXED) or get (!MAP_FIXED) the "addr"
340 */
341
342 if (flags & MAP_FIXED) {
343
344 /* ensure address and file offset are aligned properly */
345 addr -= pageoff;
346 if (addr & PAGE_MASK)
347 return (EINVAL);
348
349 if (VM_MAXUSER_ADDRESS > 0 &&
350 (addr + size) > VM_MAXUSER_ADDRESS)
351 return (EFBIG);
352 if (vm_min_address > 0 && addr < vm_min_address)
353 return (EINVAL);
354 if (addr > addr + size)
355 return (EOVERFLOW); /* no wrapping! */
356
357 } else if (addr == 0 || !(flags & MAP_TRYFIXED)) {
358
359 /*
360 * not fixed: make sure we skip over the largest
361 * possible heap for non-topdown mapping arrangements.
362 * we will refine our guess later (e.g. to account for
363 * VAC, etc)
364 */
365
366 defaddr = p->p_emul->e_vm_default_addr(p,
367 (vaddr_t)p->p_vmspace->vm_daddr, size);
368
369 if (addr == 0 ||
370 !(p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN))
371 addr = MAX(addr, defaddr);
372 else
373 addr = MIN(addr, defaddr);
374 }
375
376 /*
377 * check for file mappings (i.e. not anonymous) and verify file.
378 */
379
380 if ((flags & MAP_ANON) == 0) {
381
382 if ((fp = fd_getfile(fdp, fd)) == NULL)
383 return (EBADF);
384
385 simple_unlock(&fp->f_slock);
386
387 if (fp->f_type != DTYPE_VNODE)
388 return (ENODEV); /* only mmap vnodes! */
389 vp = (struct vnode *)fp->f_data; /* convert to vnode */
390
391 if (vp->v_type != VREG && vp->v_type != VCHR &&
392 vp->v_type != VBLK)
393 return (ENODEV); /* only REG/CHR/BLK support mmap */
394
395 if (vp->v_type != VCHR && pos < 0)
396 return (EINVAL);
397
398 if (vp->v_type != VCHR && (pos + size) < pos)
399 return (EOVERFLOW); /* no offset wrapping */
400
401 /* special case: catch SunOS style /dev/zero */
402 if (vp->v_type == VCHR
403 && (vp->v_rdev == zerodev || COMPAT_ZERODEV(vp->v_rdev))) {
404 flags |= MAP_ANON;
405 goto is_anon;
406 }
407
408 /*
409 * Old programs may not select a specific sharing type, so
410 * default to an appropriate one.
411 *
412 * XXX: how does MAP_ANON fit in the picture?
413 */
414 if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
415 #if defined(DEBUG)
416 printf("WARNING: defaulted mmap() share type to "
417 "%s (pid %d command %s)\n", vp->v_type == VCHR ?
418 "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
419 p->p_comm);
420 #endif
421 if (vp->v_type == VCHR)
422 flags |= MAP_SHARED; /* for a device */
423 else
424 flags |= MAP_PRIVATE; /* for a file */
425 }
426
427 /*
428 * MAP_PRIVATE device mappings don't make sense (and aren't
429 * supported anyway). However, some programs rely on this,
430 * so just change it to MAP_SHARED.
431 */
432 if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
433 flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
434 }
435
436 /*
437 * now check protection
438 */
439
440 maxprot = VM_PROT_EXECUTE;
441
442 /* check read access */
443 if (fp->f_flag & FREAD)
444 maxprot |= VM_PROT_READ;
445 else if (prot & PROT_READ)
446 return (EACCES);
447
448 /* check write access, shared case first */
449 if (flags & MAP_SHARED) {
450 /*
451 * if the file is writable, only add PROT_WRITE to
452 * maxprot if the file is not immutable, append-only.
453 * otherwise, if we have asked for PROT_WRITE, return
454 * EPERM.
455 */
456 if (fp->f_flag & FWRITE) {
457 if ((error =
458 VOP_GETATTR(vp, &va, p->p_ucred, p)))
459 return (error);
460 if ((va.va_flags &
461 (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0)
462 maxprot |= VM_PROT_WRITE;
463 else if (prot & PROT_WRITE)
464 return (EPERM);
465 }
466 else if (prot & PROT_WRITE)
467 return (EACCES);
468 } else {
469 /* MAP_PRIVATE mappings can always write to */
470 maxprot |= VM_PROT_WRITE;
471 }
472 handle = vp;
473
474 } else { /* MAP_ANON case */
475 /*
476 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0?
477 */
478 if (fd != -1)
479 return (EINVAL);
480
481 is_anon: /* label for SunOS style /dev/zero */
482 handle = NULL;
483 maxprot = VM_PROT_ALL;
484 pos = 0;
485 }
486
487 /*
488 * XXX (in)sanity check. We don't do proper datasize checking
489 * XXX for anonymous (or private writable) mmap(). However,
490 * XXX know that if we're trying to allocate more than the amount
491 * XXX remaining under our current data size limit, _that_ should
492 * XXX be disallowed.
493 */
494 if ((flags & MAP_ANON) != 0 ||
495 ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) {
496 if (size >
497 (p->p_rlimit[RLIMIT_DATA].rlim_cur -
498 ctob(p->p_vmspace->vm_dsize))) {
499 return (ENOMEM);
500 }
501 }
502
503 /*
504 * now let kernel internal function uvm_mmap do the work.
505 */
506
507 error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot,
508 flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
509
510 if (error == 0)
511 /* remember to add offset */
512 *retval = (register_t)(addr + pageoff);
513
514 return (error);
515 }
516
517 /*
518 * sys___msync13: the msync system call (a front-end for flush)
519 */
520
521 int
522 sys___msync13(l, v, retval)
523 struct lwp *l;
524 void *v;
525 register_t *retval;
526 {
527 struct sys___msync13_args /* {
528 syscallarg(caddr_t) addr;
529 syscallarg(size_t) len;
530 syscallarg(int) flags;
531 } */ *uap = v;
532 struct proc *p = l->l_proc;
533 vaddr_t addr;
534 vsize_t size, pageoff;
535 struct vm_map *map;
536 int error, rv, flags, uvmflags;
537
538 /*
539 * extract syscall args from the uap
540 */
541
542 addr = (vaddr_t)SCARG(uap, addr);
543 size = (vsize_t)SCARG(uap, len);
544 flags = SCARG(uap, flags);
545
546 /* sanity check flags */
547 if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 ||
548 (flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 ||
549 (flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC))
550 return (EINVAL);
551 if ((flags & (MS_ASYNC | MS_SYNC)) == 0)
552 flags |= MS_SYNC;
553
554 /*
555 * align the address to a page boundary and adjust the size accordingly.
556 */
557
558 pageoff = (addr & PAGE_MASK);
559 addr -= pageoff;
560 size += pageoff;
561 size = (vsize_t)round_page(size);
562
563 /* disallow wrap-around. */
564 if (addr + size < addr)
565 return (EINVAL);
566
567 /*
568 * get map
569 */
570
571 map = &p->p_vmspace->vm_map;
572
573 /*
574 * XXXCDC: do we really need this semantic?
575 *
576 * XXX Gak! If size is zero we are supposed to sync "all modified
577 * pages with the region containing addr". Unfortunately, we
578 * don't really keep track of individual mmaps so we approximate
579 * by flushing the range of the map entry containing addr.
580 * This can be incorrect if the region splits or is coalesced
581 * with a neighbor.
582 */
583
584 if (size == 0) {
585 struct vm_map_entry *entry;
586
587 vm_map_lock_read(map);
588 rv = uvm_map_lookup_entry(map, addr, &entry);
589 if (rv == TRUE) {
590 addr = entry->start;
591 size = entry->end - entry->start;
592 }
593 vm_map_unlock_read(map);
594 if (rv == FALSE)
595 return (EINVAL);
596 }
597
598 /*
599 * translate MS_ flags into PGO_ flags
600 */
601
602 uvmflags = PGO_CLEANIT;
603 if (flags & MS_INVALIDATE)
604 uvmflags |= PGO_FREE;
605 if (flags & MS_SYNC)
606 uvmflags |= PGO_SYNCIO;
607
608 error = uvm_map_clean(map, addr, addr+size, uvmflags);
609 return error;
610 }
611
612 /*
613 * sys_munmap: unmap a users memory
614 */
615
616 int
617 sys_munmap(l, v, retval)
618 struct lwp *l;
619 void *v;
620 register_t *retval;
621 {
622 struct sys_munmap_args /* {
623 syscallarg(caddr_t) addr;
624 syscallarg(size_t) len;
625 } */ *uap = v;
626 struct proc *p = l->l_proc;
627 vaddr_t addr;
628 vsize_t size, pageoff;
629 struct vm_map *map;
630 vaddr_t vm_min_address = VM_MIN_ADDRESS;
631 struct vm_map_entry *dead_entries;
632
633 /*
634 * get syscall args.
635 */
636
637 addr = (vaddr_t)SCARG(uap, addr);
638 size = (vsize_t)SCARG(uap, len);
639
640 /*
641 * align the address to a page boundary and adjust the size accordingly.
642 */
643
644 pageoff = (addr & PAGE_MASK);
645 addr -= pageoff;
646 size += pageoff;
647 size = (vsize_t)round_page(size);
648
649 if ((int)size < 0)
650 return (EINVAL);
651 if (size == 0)
652 return (0);
653
654 /*
655 * Check for illegal addresses. Watch out for address wrap...
656 * Note that VM_*_ADDRESS are not constants due to casts (argh).
657 */
658 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
659 return (EINVAL);
660 if (vm_min_address > 0 && addr < vm_min_address)
661 return (EINVAL);
662 if (addr > addr + size)
663 return (EINVAL);
664 map = &p->p_vmspace->vm_map;
665
666 /*
667 * interesting system call semantic: make sure entire range is
668 * allocated before allowing an unmap.
669 */
670
671 vm_map_lock(map);
672 #if 0
673 if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
674 vm_map_unlock(map);
675 return (EINVAL);
676 }
677 #endif
678 uvm_unmap_remove(map, addr, addr + size, &dead_entries, NULL);
679 vm_map_unlock(map);
680 if (dead_entries != NULL)
681 uvm_unmap_detach(dead_entries, 0);
682 return (0);
683 }
684
685 /*
686 * sys_mprotect: the mprotect system call
687 */
688
689 int
690 sys_mprotect(l, v, retval)
691 struct lwp *l;
692 void *v;
693 register_t *retval;
694 {
695 struct sys_mprotect_args /* {
696 syscallarg(caddr_t) addr;
697 syscallarg(size_t) len;
698 syscallarg(int) prot;
699 } */ *uap = v;
700 struct proc *p = l->l_proc;
701 vaddr_t addr;
702 vsize_t size, pageoff;
703 vm_prot_t prot;
704 int error;
705
706 /*
707 * extract syscall args from uap
708 */
709
710 addr = (vaddr_t)SCARG(uap, addr);
711 size = (vsize_t)SCARG(uap, len);
712 prot = SCARG(uap, prot) & VM_PROT_ALL;
713
714 /*
715 * align the address to a page boundary and adjust the size accordingly.
716 */
717
718 pageoff = (addr & PAGE_MASK);
719 addr -= pageoff;
720 size += pageoff;
721 size = round_page(size);
722
723 error = uvm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
724 FALSE);
725 return error;
726 }
727
728 /*
729 * sys_minherit: the minherit system call
730 */
731
732 int
733 sys_minherit(l, v, retval)
734 struct lwp *l;
735 void *v;
736 register_t *retval;
737 {
738 struct sys_minherit_args /* {
739 syscallarg(caddr_t) addr;
740 syscallarg(int) len;
741 syscallarg(int) inherit;
742 } */ *uap = v;
743 struct proc *p = l->l_proc;
744 vaddr_t addr;
745 vsize_t size, pageoff;
746 vm_inherit_t inherit;
747 int error;
748
749 addr = (vaddr_t)SCARG(uap, addr);
750 size = (vsize_t)SCARG(uap, len);
751 inherit = SCARG(uap, inherit);
752
753 /*
754 * align the address to a page boundary and adjust the size accordingly.
755 */
756
757 pageoff = (addr & PAGE_MASK);
758 addr -= pageoff;
759 size += pageoff;
760 size = (vsize_t)round_page(size);
761
762 if ((int)size < 0)
763 return (EINVAL);
764 error = uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr + size,
765 inherit);
766 return error;
767 }
768
769 /*
770 * sys_madvise: give advice about memory usage.
771 */
772
773 /* ARGSUSED */
774 int
775 sys_madvise(l, v, retval)
776 struct lwp *l;
777 void *v;
778 register_t *retval;
779 {
780 struct sys_madvise_args /* {
781 syscallarg(caddr_t) addr;
782 syscallarg(size_t) len;
783 syscallarg(int) behav;
784 } */ *uap = v;
785 struct proc *p = l->l_proc;
786 vaddr_t addr;
787 vsize_t size, pageoff;
788 int advice, error;
789
790 addr = (vaddr_t)SCARG(uap, addr);
791 size = (vsize_t)SCARG(uap, len);
792 advice = SCARG(uap, behav);
793
794 /*
795 * align the address to a page boundary, and adjust the size accordingly
796 */
797
798 pageoff = (addr & PAGE_MASK);
799 addr -= pageoff;
800 size += pageoff;
801 size = (vsize_t)round_page(size);
802
803 if ((ssize_t)size <= 0)
804 return (EINVAL);
805
806 switch (advice) {
807 case MADV_NORMAL:
808 case MADV_RANDOM:
809 case MADV_SEQUENTIAL:
810 error = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size,
811 advice);
812 break;
813
814 case MADV_WILLNEED:
815
816 /*
817 * Activate all these pages, pre-faulting them in if
818 * necessary.
819 */
820 /*
821 * XXX IMPLEMENT ME.
822 * Should invent a "weak" mode for uvm_fault()
823 * which would only do the PGO_LOCKED pgo_get().
824 */
825
826 return (0);
827
828 case MADV_DONTNEED:
829
830 /*
831 * Deactivate all these pages. We don't need them
832 * any more. We don't, however, toss the data in
833 * the pages.
834 */
835
836 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
837 PGO_DEACTIVATE);
838 break;
839
840 case MADV_FREE:
841
842 /*
843 * These pages contain no valid data, and may be
844 * garbage-collected. Toss all resources, including
845 * any swap space in use.
846 */
847
848 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
849 PGO_FREE);
850 break;
851
852 case MADV_SPACEAVAIL:
853
854 /*
855 * XXXMRG What is this? I think it's:
856 *
857 * Ensure that we have allocated backing-store
858 * for these pages.
859 *
860 * This is going to require changes to the page daemon,
861 * as it will free swap space allocated to pages in core.
862 * There's also what to do for device/file/anonymous memory.
863 */
864
865 return (EINVAL);
866
867 default:
868 return (EINVAL);
869 }
870
871 return error;
872 }
873
874 /*
875 * sys_mlock: memory lock
876 */
877
878 int
879 sys_mlock(l, v, retval)
880 struct lwp *l;
881 void *v;
882 register_t *retval;
883 {
884 struct sys_mlock_args /* {
885 syscallarg(const void *) addr;
886 syscallarg(size_t) len;
887 } */ *uap = v;
888 struct proc *p = l->l_proc;
889 vaddr_t addr;
890 vsize_t size, pageoff;
891 int error;
892
893 /*
894 * extract syscall args from uap
895 */
896
897 addr = (vaddr_t)SCARG(uap, addr);
898 size = (vsize_t)SCARG(uap, len);
899
900 /*
901 * align the address to a page boundary and adjust the size accordingly
902 */
903
904 pageoff = (addr & PAGE_MASK);
905 addr -= pageoff;
906 size += pageoff;
907 size = (vsize_t)round_page(size);
908
909 /* disallow wrap-around. */
910 if (addr + size < addr)
911 return (EINVAL);
912
913 if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
914 return (EAGAIN);
915
916 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
917 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
918 return (EAGAIN);
919
920 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
921 0);
922 if (error == EFAULT)
923 error = ENOMEM;
924 return error;
925 }
926
927 /*
928 * sys_munlock: unlock wired pages
929 */
930
931 int
932 sys_munlock(l, v, retval)
933 struct lwp *l;
934 void *v;
935 register_t *retval;
936 {
937 struct sys_munlock_args /* {
938 syscallarg(const void *) addr;
939 syscallarg(size_t) len;
940 } */ *uap = v;
941 struct proc *p = l->l_proc;
942 vaddr_t addr;
943 vsize_t size, pageoff;
944 int error;
945
946 /*
947 * extract syscall args from uap
948 */
949
950 addr = (vaddr_t)SCARG(uap, addr);
951 size = (vsize_t)SCARG(uap, len);
952
953 /*
954 * align the address to a page boundary, and adjust the size accordingly
955 */
956
957 pageoff = (addr & PAGE_MASK);
958 addr -= pageoff;
959 size += pageoff;
960 size = (vsize_t)round_page(size);
961
962 /* disallow wrap-around. */
963 if (addr + size < addr)
964 return (EINVAL);
965
966 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
967 0);
968 if (error == EFAULT)
969 error = ENOMEM;
970 return error;
971 }
972
973 /*
974 * sys_mlockall: lock all pages mapped into an address space.
975 */
976
977 int
978 sys_mlockall(l, v, retval)
979 struct lwp *l;
980 void *v;
981 register_t *retval;
982 {
983 struct sys_mlockall_args /* {
984 syscallarg(int) flags;
985 } */ *uap = v;
986 struct proc *p = l->l_proc;
987 int error, flags;
988
989 flags = SCARG(uap, flags);
990
991 if (flags == 0 ||
992 (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
993 return (EINVAL);
994
995 error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
996 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
997 return (error);
998 }
999
1000 /*
1001 * sys_munlockall: unlock all pages mapped into an address space.
1002 */
1003
1004 int
1005 sys_munlockall(l, v, retval)
1006 struct lwp *l;
1007 void *v;
1008 register_t *retval;
1009 {
1010 struct proc *p = l->l_proc;
1011
1012 (void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
1013 return (0);
1014 }
1015
1016 /*
1017 * uvm_mmap: internal version of mmap
1018 *
1019 * - used by sys_mmap and various framebuffers
1020 * - handle is a vnode pointer or NULL for MAP_ANON
1021 * - caller must page-align the file offset
1022 */
1023
1024 int
1025 uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit)
1026 struct vm_map *map;
1027 vaddr_t *addr;
1028 vsize_t size;
1029 vm_prot_t prot, maxprot;
1030 int flags;
1031 void *handle;
1032 voff_t foff;
1033 vsize_t locklimit;
1034 {
1035 struct uvm_object *uobj;
1036 struct vnode *vp;
1037 vaddr_t align = 0;
1038 int error;
1039 int advice = UVM_ADV_NORMAL;
1040 uvm_flag_t uvmflag = 0;
1041
1042 /*
1043 * check params
1044 */
1045
1046 if (size == 0)
1047 return(0);
1048 if (foff & PAGE_MASK)
1049 return(EINVAL);
1050 if ((prot & maxprot) != prot)
1051 return(EINVAL);
1052
1053 /*
1054 * for non-fixed mappings, round off the suggested address.
1055 * for fixed mappings, check alignment and zap old mappings.
1056 */
1057
1058 if ((flags & MAP_FIXED) == 0) {
1059 *addr = round_page(*addr);
1060 } else {
1061 if (*addr & PAGE_MASK)
1062 return(EINVAL);
1063 uvmflag |= UVM_FLAG_FIXED;
1064 (void) uvm_unmap(map, *addr, *addr + size);
1065 }
1066
1067 /*
1068 * Try to see if any requested alignment can even be attemped.
1069 * Make sure we can express the alignment (asking for a >= 4GB
1070 * alignment on an ILP32 architecure make no sense) and the
1071 * alignment is at least for a page sized quanitiy. If the
1072 * request was for a fixed mapping, make sure supplied address
1073 * adheres to the request alignment.
1074 */
1075 align = (flags & MAP_ALIGNMENT_MASK) >> MAP_ALIGNMENT_SHIFT;
1076 if (align) {
1077 if (align >= sizeof(vaddr_t) * NBBY)
1078 return(EINVAL);
1079 align = 1L << align;
1080 if (align < PAGE_SIZE)
1081 return(EINVAL);
1082 if (align >= vm_map_max(map))
1083 return(ENOMEM);
1084 if (flags & MAP_FIXED) {
1085 if ((*addr & (align-1)) != 0)
1086 return(EINVAL);
1087 align = 0;
1088 }
1089 }
1090
1091 /*
1092 * handle anon vs. non-anon mappings. for non-anon mappings attach
1093 * to underlying vm object.
1094 */
1095
1096 if (flags & MAP_ANON) {
1097 foff = UVM_UNKNOWN_OFFSET;
1098 uobj = NULL;
1099 if ((flags & MAP_SHARED) == 0)
1100 /* XXX: defer amap create */
1101 uvmflag |= UVM_FLAG_COPYONW;
1102 else
1103 /* shared: create amap now */
1104 uvmflag |= UVM_FLAG_OVERLAY;
1105
1106 } else {
1107 vp = (struct vnode *)handle;
1108
1109 /*
1110 * Don't allow mmap for EXEC if the file system
1111 * is mounted NOEXEC.
1112 */
1113 if ((prot & PROT_EXEC) != 0 &&
1114 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0)
1115 return (EACCES);
1116
1117 if (vp->v_type != VCHR) {
1118 error = VOP_MMAP(vp, 0, curproc->p_ucred, curproc);
1119 if (error) {
1120 return error;
1121 }
1122
1123 uobj = uvn_attach((void *)vp, (flags & MAP_SHARED) ?
1124 maxprot : (maxprot & ~VM_PROT_WRITE));
1125
1126 /* XXX for now, attach doesn't gain a ref */
1127 VREF(vp);
1128
1129 /*
1130 * If the vnode is being mapped with PROT_EXEC,
1131 * then mark it as text.
1132 */
1133 if (prot & PROT_EXEC)
1134 vn_markexec(vp);
1135 } else {
1136 int i = maxprot;
1137
1138 /*
1139 * XXX Some devices don't like to be mapped with
1140 * XXX PROT_EXEC or PROT_WRITE, but we don't really
1141 * XXX have a better way of handling this, right now
1142 */
1143 do {
1144 uobj = udv_attach((void *) &vp->v_rdev,
1145 (flags & MAP_SHARED) ? i :
1146 (i & ~VM_PROT_WRITE), foff, size);
1147 i--;
1148 } while ((uobj == NULL) && (i > 0));
1149 advice = UVM_ADV_RANDOM;
1150 }
1151 if (uobj == NULL)
1152 return((vp->v_type == VREG) ? ENOMEM : EINVAL);
1153 if ((flags & MAP_SHARED) == 0) {
1154 uvmflag |= UVM_FLAG_COPYONW;
1155 } else if ((maxprot & VM_PROT_WRITE) != 0) {
1156 simple_lock(&vp->v_interlock);
1157 vp->v_flag |= VWRITEMAP;
1158 simple_unlock(&vp->v_interlock);
1159 }
1160 }
1161
1162 uvmflag = UVM_MAPFLAG(prot, maxprot,
1163 (flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
1164 advice, uvmflag);
1165 error = uvm_map(map, addr, size, uobj, foff, align, uvmflag);
1166 if (error) {
1167 if (uobj)
1168 uobj->pgops->pgo_detach(uobj);
1169 return error;
1170 }
1171
1172 /*
1173 * POSIX 1003.1b -- if our address space was configured
1174 * to lock all future mappings, wire the one we just made.
1175 *
1176 * Also handle the MAP_WIRED flag here.
1177 */
1178
1179 if (prot == VM_PROT_NONE) {
1180
1181 /*
1182 * No more work to do in this case.
1183 */
1184
1185 return (0);
1186 }
1187 vm_map_lock(map);
1188 if ((flags & MAP_WIRED) != 0 || (map->flags & VM_MAP_WIREFUTURE) != 0) {
1189 if (atop(size) + uvmexp.wired > uvmexp.wiredmax ||
1190 (locklimit != 0 &&
1191 size + ptoa(pmap_wired_count(vm_map_pmap(map))) >
1192 locklimit)) {
1193 vm_map_unlock(map);
1194 uvm_unmap(map, *addr, *addr + size);
1195 return ENOMEM;
1196 }
1197
1198 /*
1199 * uvm_map_pageable() always returns the map unlocked.
1200 */
1201
1202 error = uvm_map_pageable(map, *addr, *addr + size,
1203 FALSE, UVM_LK_ENTER);
1204 if (error) {
1205 uvm_unmap(map, *addr, *addr + size);
1206 return error;
1207 }
1208 return (0);
1209 }
1210 vm_map_unlock(map);
1211 return 0;
1212 }
1213
1214 vaddr_t
1215 uvm_default_mapaddr(struct proc *p, vaddr_t base, vsize_t sz)
1216 {
1217 return VM_DEFAULT_ADDRESS(base, sz);
1218 }
Cache object: 950fa7e6ba098207323727d9a523bccb
|