FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_mmap.c
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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35 *
36 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
37 */
38
39 /*
40 * Mapped file (mmap) interface to VM
41 */
42
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45
46 #include "opt_compat.h"
47 #include "opt_mac.h"
48
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.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/mac.h>
63 #include <sys/mman.h>
64 #include <sys/mount.h>
65 #include <sys/conf.h>
66 #include <sys/stat.h>
67 #include <sys/vmmeter.h>
68 #include <sys/sysctl.h>
69
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <vm/pmap.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_pager.h>
77 #include <vm/vm_pageout.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_kern.h>
81
82 #ifndef _SYS_SYSPROTO_H_
83 struct sbrk_args {
84 int incr;
85 };
86 #endif
87
88 static int max_proc_mmap;
89 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
90
91 /*
92 * Set the maximum number of vm_map_entry structures per process. Roughly
93 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
94 * of our KVM malloc space still results in generous limits. We want a
95 * default that is good enough to prevent the kernel running out of resources
96 * if attacked from compromised user account but generous enough such that
97 * multi-threaded processes are not unduly inconvenienced.
98 */
99 static void vmmapentry_rsrc_init(void *);
100 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
101
102 static void
103 vmmapentry_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 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
111 int *, struct vnode *, vm_ooffset_t, vm_object_t *);
112 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
113 int *, struct cdev *, vm_ooffset_t, vm_object_t *);
114
115 /*
116 * MPSAFE
117 */
118 /* ARGSUSED */
119 int
120 sbrk(td, uap)
121 struct thread *td;
122 struct sbrk_args *uap;
123 {
124 /* Not yet implemented */
125 return (EOPNOTSUPP);
126 }
127
128 #ifndef _SYS_SYSPROTO_H_
129 struct sstk_args {
130 int incr;
131 };
132 #endif
133
134 /*
135 * MPSAFE
136 */
137 /* ARGSUSED */
138 int
139 sstk(td, uap)
140 struct thread *td;
141 struct sstk_args *uap;
142 {
143 /* Not yet implemented */
144 return (EOPNOTSUPP);
145 }
146
147 #if defined(COMPAT_43)
148 #ifndef _SYS_SYSPROTO_H_
149 struct getpagesize_args {
150 int dummy;
151 };
152 #endif
153
154 /* ARGSUSED */
155 int
156 ogetpagesize(td, uap)
157 struct thread *td;
158 struct getpagesize_args *uap;
159 {
160 /* MP SAFE */
161 td->td_retval[0] = PAGE_SIZE;
162 return (0);
163 }
164 #endif /* COMPAT_43 */
165
166
167 /*
168 * Memory Map (mmap) system call. Note that the file offset
169 * and address are allowed to be NOT page aligned, though if
170 * the MAP_FIXED flag it set, both must have the same remainder
171 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
172 * page-aligned, the actual mapping starts at trunc_page(addr)
173 * and the return value is adjusted up by the page offset.
174 *
175 * Generally speaking, only character devices which are themselves
176 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
177 * there would be no cache coherency between a descriptor and a VM mapping
178 * both to the same character device.
179 *
180 * Block devices can be mmap'd no matter what they represent. Cache coherency
181 * is maintained as long as you do not write directly to the underlying
182 * character device.
183 */
184 #ifndef _SYS_SYSPROTO_H_
185 struct mmap_args {
186 void *addr;
187 size_t len;
188 int prot;
189 int flags;
190 int fd;
191 long pad;
192 off_t pos;
193 };
194 #endif
195
196 /*
197 * MPSAFE
198 */
199 int
200 mmap(td, uap)
201 struct thread *td;
202 struct mmap_args *uap;
203 {
204 struct file *fp;
205 struct vnode *vp;
206 vm_offset_t addr;
207 vm_size_t size, pageoff;
208 vm_prot_t prot, maxprot;
209 void *handle;
210 objtype_t handle_type;
211 int flags, error;
212 off_t pos;
213 struct vmspace *vms = td->td_proc->p_vmspace;
214
215 addr = (vm_offset_t) uap->addr;
216 size = uap->len;
217 prot = uap->prot & VM_PROT_ALL;
218 flags = uap->flags;
219 pos = uap->pos;
220
221 fp = NULL;
222 /* make sure mapping fits into numeric range etc */
223 if ((ssize_t) uap->len < 0 ||
224 ((flags & MAP_ANON) && uap->fd != -1))
225 return (EINVAL);
226
227 if (flags & MAP_STACK) {
228 if ((uap->fd != -1) ||
229 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
230 return (EINVAL);
231 flags |= MAP_ANON;
232 pos = 0;
233 }
234
235 /*
236 * Align the file position to a page boundary,
237 * and save its page offset component.
238 */
239 pageoff = (pos & PAGE_MASK);
240 pos -= pageoff;
241
242 /* Adjust size for rounding (on both ends). */
243 size += pageoff; /* low end... */
244 size = (vm_size_t) round_page(size); /* hi end */
245
246 /*
247 * Check for illegal addresses. Watch out for address wrap... Note
248 * that VM_*_ADDRESS are not constants due to casts (argh).
249 */
250 if (flags & MAP_FIXED) {
251 /*
252 * The specified address must have the same remainder
253 * as the file offset taken modulo PAGE_SIZE, so it
254 * should be aligned after adjustment by pageoff.
255 */
256 addr -= pageoff;
257 if (addr & PAGE_MASK)
258 return (EINVAL);
259 /* Address range must be all in user VM space. */
260 if (addr < vm_map_min(&vms->vm_map) ||
261 addr + size > vm_map_max(&vms->vm_map))
262 return (EINVAL);
263 if (addr + size < addr)
264 return (EINVAL);
265 } else {
266 /*
267 * XXX for non-fixed mappings where no hint is provided or
268 * the hint would fall in the potential heap space,
269 * place it after the end of the largest possible heap.
270 *
271 * There should really be a pmap call to determine a reasonable
272 * location.
273 */
274 PROC_LOCK(td->td_proc);
275 if (addr == 0 ||
276 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
277 addr < round_page((vm_offset_t)vms->vm_daddr +
278 lim_max(td->td_proc, RLIMIT_DATA))))
279 addr = round_page((vm_offset_t)vms->vm_daddr +
280 lim_max(td->td_proc, RLIMIT_DATA));
281 PROC_UNLOCK(td->td_proc);
282 }
283 if (flags & MAP_ANON) {
284 /*
285 * Mapping blank space is trivial.
286 */
287 handle = NULL;
288 handle_type = OBJT_DEFAULT;
289 maxprot = VM_PROT_ALL;
290 pos = 0;
291 } else {
292 /*
293 * Mapping file, get fp for validation. Obtain vnode and make
294 * sure it is of appropriate type.
295 * don't let the descriptor disappear on us if we block
296 */
297 if ((error = fget(td, uap->fd, &fp)) != 0)
298 goto done;
299 if (fp->f_type != DTYPE_VNODE) {
300 error = EINVAL;
301 goto done;
302 }
303 /*
304 * POSIX shared-memory objects are defined to have
305 * kernel persistence, and are not defined to support
306 * read(2)/write(2) -- or even open(2). Thus, we can
307 * use MAP_ASYNC to trade on-disk coherence for speed.
308 * The shm_open(3) library routine turns on the FPOSIXSHM
309 * flag to request this behavior.
310 */
311 if (fp->f_flag & FPOSIXSHM)
312 flags |= MAP_NOSYNC;
313 vp = fp->f_vnode;
314 /*
315 * Ensure that file and memory protections are
316 * compatible. Note that we only worry about
317 * writability if mapping is shared; in this case,
318 * current and max prot are dictated by the open file.
319 * XXX use the vnode instead? Problem is: what
320 * credentials do we use for determination? What if
321 * proc does a setuid?
322 */
323 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
324 maxprot = VM_PROT_NONE;
325 else
326 maxprot = VM_PROT_EXECUTE;
327 if (fp->f_flag & FREAD) {
328 maxprot |= VM_PROT_READ;
329 } else if (prot & PROT_READ) {
330 error = EACCES;
331 goto done;
332 }
333 /*
334 * If we are sharing potential changes (either via
335 * MAP_SHARED or via the implicit sharing of character
336 * device mappings), and we are trying to get write
337 * permission although we opened it without asking
338 * for it, bail out.
339 */
340 if ((flags & MAP_SHARED) != 0) {
341 if ((fp->f_flag & FWRITE) != 0) {
342 maxprot |= VM_PROT_WRITE;
343 } else if ((prot & PROT_WRITE) != 0) {
344 error = EACCES;
345 goto done;
346 }
347 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
348 maxprot |= VM_PROT_WRITE;
349 }
350 handle = (void *)vp;
351 handle_type = OBJT_VNODE;
352 }
353
354 /*
355 * Do not allow more then a certain number of vm_map_entry structures
356 * per process. Scale with the number of rforks sharing the map
357 * to make the limit reasonable for threads.
358 */
359 if (max_proc_mmap &&
360 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
361 error = ENOMEM;
362 goto done;
363 }
364
365 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
366 flags, handle_type, handle, pos);
367 if (error == 0)
368 td->td_retval[0] = (register_t) (addr + pageoff);
369 done:
370 if (fp)
371 fdrop(fp, td);
372
373 return (error);
374 }
375
376 #ifdef COMPAT_43
377 #ifndef _SYS_SYSPROTO_H_
378 struct ommap_args {
379 caddr_t addr;
380 int len;
381 int prot;
382 int flags;
383 int fd;
384 long pos;
385 };
386 #endif
387 int
388 ommap(td, uap)
389 struct thread *td;
390 struct ommap_args *uap;
391 {
392 struct mmap_args nargs;
393 static const char cvtbsdprot[8] = {
394 0,
395 PROT_EXEC,
396 PROT_WRITE,
397 PROT_EXEC | PROT_WRITE,
398 PROT_READ,
399 PROT_EXEC | PROT_READ,
400 PROT_WRITE | PROT_READ,
401 PROT_EXEC | PROT_WRITE | PROT_READ,
402 };
403
404 #define OMAP_ANON 0x0002
405 #define OMAP_COPY 0x0020
406 #define OMAP_SHARED 0x0010
407 #define OMAP_FIXED 0x0100
408
409 nargs.addr = uap->addr;
410 nargs.len = uap->len;
411 nargs.prot = cvtbsdprot[uap->prot & 0x7];
412 nargs.flags = 0;
413 if (uap->flags & OMAP_ANON)
414 nargs.flags |= MAP_ANON;
415 if (uap->flags & OMAP_COPY)
416 nargs.flags |= MAP_COPY;
417 if (uap->flags & OMAP_SHARED)
418 nargs.flags |= MAP_SHARED;
419 else
420 nargs.flags |= MAP_PRIVATE;
421 if (uap->flags & OMAP_FIXED)
422 nargs.flags |= MAP_FIXED;
423 nargs.fd = uap->fd;
424 nargs.pos = uap->pos;
425 return (mmap(td, &nargs));
426 }
427 #endif /* COMPAT_43 */
428
429
430 #ifndef _SYS_SYSPROTO_H_
431 struct msync_args {
432 void *addr;
433 int len;
434 int flags;
435 };
436 #endif
437 /*
438 * MPSAFE
439 */
440 int
441 msync(td, uap)
442 struct thread *td;
443 struct msync_args *uap;
444 {
445 vm_offset_t addr;
446 vm_size_t size, pageoff;
447 int flags;
448 vm_map_t map;
449 int rv;
450
451 addr = (vm_offset_t) uap->addr;
452 size = uap->len;
453 flags = uap->flags;
454
455 pageoff = (addr & PAGE_MASK);
456 addr -= pageoff;
457 size += pageoff;
458 size = (vm_size_t) round_page(size);
459 if (addr + size < addr)
460 return (EINVAL);
461
462 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
463 return (EINVAL);
464
465 map = &td->td_proc->p_vmspace->vm_map;
466
467 /*
468 * Clean the pages and interpret the return value.
469 */
470 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
471 (flags & MS_INVALIDATE) != 0);
472 switch (rv) {
473 case KERN_SUCCESS:
474 return (0);
475 case KERN_INVALID_ADDRESS:
476 return (EINVAL); /* Sun returns ENOMEM? */
477 case KERN_INVALID_ARGUMENT:
478 return (EBUSY);
479 default:
480 return (EINVAL);
481 }
482 }
483
484 #ifndef _SYS_SYSPROTO_H_
485 struct munmap_args {
486 void *addr;
487 size_t len;
488 };
489 #endif
490 /*
491 * MPSAFE
492 */
493 int
494 munmap(td, uap)
495 struct thread *td;
496 struct munmap_args *uap;
497 {
498 vm_offset_t addr;
499 vm_size_t size, pageoff;
500 vm_map_t map;
501
502 addr = (vm_offset_t) uap->addr;
503 size = uap->len;
504 if (size == 0)
505 return (EINVAL);
506
507 pageoff = (addr & PAGE_MASK);
508 addr -= pageoff;
509 size += pageoff;
510 size = (vm_size_t) round_page(size);
511 if (addr + size < addr)
512 return (EINVAL);
513
514 /*
515 * Check for illegal addresses. Watch out for address wrap...
516 */
517 map = &td->td_proc->p_vmspace->vm_map;
518 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
519 return (EINVAL);
520 vm_map_lock(map);
521 /* returns nothing but KERN_SUCCESS anyway */
522 vm_map_delete(map, addr, addr + size);
523 vm_map_unlock(map);
524 return (0);
525 }
526
527 #ifndef _SYS_SYSPROTO_H_
528 struct mprotect_args {
529 const void *addr;
530 size_t len;
531 int prot;
532 };
533 #endif
534 /*
535 * MPSAFE
536 */
537 int
538 mprotect(td, uap)
539 struct thread *td;
540 struct mprotect_args *uap;
541 {
542 vm_offset_t addr;
543 vm_size_t size, pageoff;
544 vm_prot_t prot;
545
546 addr = (vm_offset_t) uap->addr;
547 size = uap->len;
548 prot = uap->prot & VM_PROT_ALL;
549 #if defined(VM_PROT_READ_IS_EXEC)
550 if (prot & VM_PROT_READ)
551 prot |= VM_PROT_EXECUTE;
552 #endif
553
554 pageoff = (addr & PAGE_MASK);
555 addr -= pageoff;
556 size += pageoff;
557 size = (vm_size_t) round_page(size);
558 if (addr + size < addr)
559 return (EINVAL);
560
561 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
562 addr + size, prot, FALSE)) {
563 case KERN_SUCCESS:
564 return (0);
565 case KERN_PROTECTION_FAILURE:
566 return (EACCES);
567 }
568 return (EINVAL);
569 }
570
571 #ifndef _SYS_SYSPROTO_H_
572 struct minherit_args {
573 void *addr;
574 size_t len;
575 int inherit;
576 };
577 #endif
578 /*
579 * MPSAFE
580 */
581 int
582 minherit(td, uap)
583 struct thread *td;
584 struct minherit_args *uap;
585 {
586 vm_offset_t addr;
587 vm_size_t size, pageoff;
588 vm_inherit_t inherit;
589
590 addr = (vm_offset_t)uap->addr;
591 size = uap->len;
592 inherit = uap->inherit;
593
594 pageoff = (addr & PAGE_MASK);
595 addr -= pageoff;
596 size += pageoff;
597 size = (vm_size_t) round_page(size);
598 if (addr + size < addr)
599 return (EINVAL);
600
601 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
602 addr + size, inherit)) {
603 case KERN_SUCCESS:
604 return (0);
605 case KERN_PROTECTION_FAILURE:
606 return (EACCES);
607 }
608 return (EINVAL);
609 }
610
611 #ifndef _SYS_SYSPROTO_H_
612 struct madvise_args {
613 void *addr;
614 size_t len;
615 int behav;
616 };
617 #endif
618
619 /*
620 * MPSAFE
621 */
622 /* ARGSUSED */
623 int
624 madvise(td, uap)
625 struct thread *td;
626 struct madvise_args *uap;
627 {
628 vm_offset_t start, end;
629 vm_map_t map;
630 struct proc *p;
631 int error;
632
633 /*
634 * Check for our special case, advising the swap pager we are
635 * "immortal."
636 */
637 if (uap->behav == MADV_PROTECT) {
638 error = suser(td);
639 if (error == 0) {
640 p = td->td_proc;
641 PROC_LOCK(p);
642 p->p_flag |= P_PROTECTED;
643 PROC_UNLOCK(p);
644 }
645 return (error);
646 }
647 /*
648 * Check for illegal behavior
649 */
650 if (uap->behav < 0 || uap->behav > MADV_CORE)
651 return (EINVAL);
652 /*
653 * Check for illegal addresses. Watch out for address wrap... Note
654 * that VM_*_ADDRESS are not constants due to casts (argh).
655 */
656 map = &td->td_proc->p_vmspace->vm_map;
657 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
658 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
659 return (EINVAL);
660 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
661 return (EINVAL);
662
663 /*
664 * Since this routine is only advisory, we default to conservative
665 * behavior.
666 */
667 start = trunc_page((vm_offset_t) uap->addr);
668 end = round_page((vm_offset_t) uap->addr + uap->len);
669
670 if (vm_map_madvise(map, start, end, uap->behav))
671 return (EINVAL);
672 return (0);
673 }
674
675 #ifndef _SYS_SYSPROTO_H_
676 struct mincore_args {
677 const void *addr;
678 size_t len;
679 char *vec;
680 };
681 #endif
682
683 /*
684 * MPSAFE
685 */
686 /* ARGSUSED */
687 int
688 mincore(td, uap)
689 struct thread *td;
690 struct mincore_args *uap;
691 {
692 vm_offset_t addr, first_addr;
693 vm_offset_t end, cend;
694 pmap_t pmap;
695 vm_map_t map;
696 char *vec;
697 int error = 0;
698 int vecindex, lastvecindex;
699 vm_map_entry_t current;
700 vm_map_entry_t entry;
701 int mincoreinfo;
702 unsigned int timestamp;
703
704 /*
705 * Make sure that the addresses presented are valid for user
706 * mode.
707 */
708 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
709 end = addr + (vm_size_t)round_page(uap->len);
710 map = &td->td_proc->p_vmspace->vm_map;
711 if (end > vm_map_max(map) || end < addr)
712 return (ENOMEM);
713
714 /*
715 * Address of byte vector
716 */
717 vec = uap->vec;
718
719 pmap = vmspace_pmap(td->td_proc->p_vmspace);
720
721 vm_map_lock_read(map);
722 RestartScan:
723 timestamp = map->timestamp;
724
725 if (!vm_map_lookup_entry(map, addr, &entry)) {
726 vm_map_unlock_read(map);
727 return (ENOMEM);
728 }
729
730 /*
731 * Do this on a map entry basis so that if the pages are not
732 * in the current processes address space, we can easily look
733 * up the pages elsewhere.
734 */
735 lastvecindex = -1;
736 for (current = entry;
737 (current != &map->header) && (current->start < end);
738 current = current->next) {
739
740 /*
741 * check for contiguity
742 */
743 if (current->end < end &&
744 (entry->next == &map->header ||
745 current->next->start > current->end)) {
746 vm_map_unlock_read(map);
747 return (ENOMEM);
748 }
749
750 /*
751 * ignore submaps (for now) or null objects
752 */
753 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
754 current->object.vm_object == NULL)
755 continue;
756
757 /*
758 * limit this scan to the current map entry and the
759 * limits for the mincore call
760 */
761 if (addr < current->start)
762 addr = current->start;
763 cend = current->end;
764 if (cend > end)
765 cend = end;
766
767 /*
768 * scan this entry one page at a time
769 */
770 while (addr < cend) {
771 /*
772 * Check pmap first, it is likely faster, also
773 * it can provide info as to whether we are the
774 * one referencing or modifying the page.
775 */
776 mincoreinfo = pmap_mincore(pmap, addr);
777 if (!mincoreinfo) {
778 vm_pindex_t pindex;
779 vm_ooffset_t offset;
780 vm_page_t m;
781 /*
782 * calculate the page index into the object
783 */
784 offset = current->offset + (addr - current->start);
785 pindex = OFF_TO_IDX(offset);
786 VM_OBJECT_LOCK(current->object.vm_object);
787 m = vm_page_lookup(current->object.vm_object,
788 pindex);
789 /*
790 * if the page is resident, then gather information about
791 * it.
792 */
793 if (m != NULL && m->valid != 0) {
794 mincoreinfo = MINCORE_INCORE;
795 vm_page_lock_queues();
796 if (m->dirty ||
797 pmap_is_modified(m))
798 mincoreinfo |= MINCORE_MODIFIED_OTHER;
799 if ((m->flags & PG_REFERENCED) ||
800 pmap_ts_referenced(m)) {
801 vm_page_flag_set(m, PG_REFERENCED);
802 mincoreinfo |= MINCORE_REFERENCED_OTHER;
803 }
804 vm_page_unlock_queues();
805 }
806 VM_OBJECT_UNLOCK(current->object.vm_object);
807 }
808
809 /*
810 * subyte may page fault. In case it needs to modify
811 * the map, we release the lock.
812 */
813 vm_map_unlock_read(map);
814
815 /*
816 * calculate index into user supplied byte vector
817 */
818 vecindex = OFF_TO_IDX(addr - first_addr);
819
820 /*
821 * If we have skipped map entries, we need to make sure that
822 * the byte vector is zeroed for those skipped entries.
823 */
824 while ((lastvecindex + 1) < vecindex) {
825 error = subyte(vec + lastvecindex, 0);
826 if (error) {
827 error = EFAULT;
828 goto done2;
829 }
830 ++lastvecindex;
831 }
832
833 /*
834 * Pass the page information to the user
835 */
836 error = subyte(vec + vecindex, mincoreinfo);
837 if (error) {
838 error = EFAULT;
839 goto done2;
840 }
841
842 /*
843 * If the map has changed, due to the subyte, the previous
844 * output may be invalid.
845 */
846 vm_map_lock_read(map);
847 if (timestamp != map->timestamp)
848 goto RestartScan;
849
850 lastvecindex = vecindex;
851 addr += PAGE_SIZE;
852 }
853 }
854
855 /*
856 * subyte may page fault. In case it needs to modify
857 * the map, we release the lock.
858 */
859 vm_map_unlock_read(map);
860
861 /*
862 * Zero the last entries in the byte vector.
863 */
864 vecindex = OFF_TO_IDX(end - first_addr);
865 while ((lastvecindex + 1) < vecindex) {
866 error = subyte(vec + lastvecindex, 0);
867 if (error) {
868 error = EFAULT;
869 goto done2;
870 }
871 ++lastvecindex;
872 }
873
874 /*
875 * If the map has changed, due to the subyte, the previous
876 * output may be invalid.
877 */
878 vm_map_lock_read(map);
879 if (timestamp != map->timestamp)
880 goto RestartScan;
881 vm_map_unlock_read(map);
882 done2:
883 return (error);
884 }
885
886 #ifndef _SYS_SYSPROTO_H_
887 struct mlock_args {
888 const void *addr;
889 size_t len;
890 };
891 #endif
892 /*
893 * MPSAFE
894 */
895 int
896 mlock(td, uap)
897 struct thread *td;
898 struct mlock_args *uap;
899 {
900 struct proc *proc;
901 vm_offset_t addr, end, last, start;
902 vm_size_t npages, size;
903 int error;
904
905 error = suser(td);
906 if (error)
907 return (error);
908 addr = (vm_offset_t)uap->addr;
909 size = uap->len;
910 last = addr + size;
911 start = trunc_page(addr);
912 end = round_page(last);
913 if (last < addr || end < addr)
914 return (EINVAL);
915 npages = atop(end - start);
916 if (npages > vm_page_max_wired)
917 return (ENOMEM);
918 proc = td->td_proc;
919 PROC_LOCK(proc);
920 if (ptoa(npages +
921 pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
922 lim_cur(proc, RLIMIT_MEMLOCK)) {
923 PROC_UNLOCK(proc);
924 return (ENOMEM);
925 }
926 PROC_UNLOCK(proc);
927 if (npages + cnt.v_wire_count > vm_page_max_wired)
928 return (EAGAIN);
929 error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
930 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
931 return (error == KERN_SUCCESS ? 0 : ENOMEM);
932 }
933
934 #ifndef _SYS_SYSPROTO_H_
935 struct mlockall_args {
936 int how;
937 };
938 #endif
939
940 /*
941 * MPSAFE
942 */
943 int
944 mlockall(td, uap)
945 struct thread *td;
946 struct mlockall_args *uap;
947 {
948 vm_map_t map;
949 int error;
950
951 map = &td->td_proc->p_vmspace->vm_map;
952 error = 0;
953
954 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
955 return (EINVAL);
956
957 #if 0
958 /*
959 * If wiring all pages in the process would cause it to exceed
960 * a hard resource limit, return ENOMEM.
961 */
962 PROC_LOCK(td->td_proc);
963 if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
964 lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
965 PROC_UNLOCK(td->td_proc);
966 return (ENOMEM);
967 }
968 PROC_UNLOCK(td->td_proc);
969 #else
970 error = suser(td);
971 if (error)
972 return (error);
973 #endif
974
975 if (uap->how & MCL_FUTURE) {
976 vm_map_lock(map);
977 vm_map_modflags(map, MAP_WIREFUTURE, 0);
978 vm_map_unlock(map);
979 error = 0;
980 }
981
982 if (uap->how & MCL_CURRENT) {
983 /*
984 * P1003.1-2001 mandates that all currently mapped pages
985 * will be memory resident and locked (wired) upon return
986 * from mlockall(). vm_map_wire() will wire pages, by
987 * calling vm_fault_wire() for each page in the region.
988 */
989 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
990 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
991 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
992 }
993
994 return (error);
995 }
996
997 #ifndef _SYS_SYSPROTO_H_
998 struct munlockall_args {
999 register_t dummy;
1000 };
1001 #endif
1002
1003 /*
1004 * MPSAFE
1005 */
1006 int
1007 munlockall(td, uap)
1008 struct thread *td;
1009 struct munlockall_args *uap;
1010 {
1011 vm_map_t map;
1012 int error;
1013
1014 map = &td->td_proc->p_vmspace->vm_map;
1015 error = suser(td);
1016 if (error)
1017 return (error);
1018
1019 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1020 vm_map_lock(map);
1021 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1022 vm_map_unlock(map);
1023
1024 /* Forcibly unwire all pages. */
1025 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1026 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1027
1028 return (error);
1029 }
1030
1031 #ifndef _SYS_SYSPROTO_H_
1032 struct munlock_args {
1033 const void *addr;
1034 size_t len;
1035 };
1036 #endif
1037 /*
1038 * MPSAFE
1039 */
1040 int
1041 munlock(td, uap)
1042 struct thread *td;
1043 struct munlock_args *uap;
1044 {
1045 vm_offset_t addr, end, last, start;
1046 vm_size_t size;
1047 int error;
1048
1049 error = suser(td);
1050 if (error)
1051 return (error);
1052 addr = (vm_offset_t)uap->addr;
1053 size = uap->len;
1054 last = addr + size;
1055 start = trunc_page(addr);
1056 end = round_page(last);
1057 if (last < addr || end < addr)
1058 return (EINVAL);
1059 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1060 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1061 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1062 }
1063
1064 /*
1065 * vm_mmap_vnode()
1066 *
1067 * MPSAFE
1068 *
1069 * Helper function for vm_mmap. Perform sanity check specific for mmap
1070 * operations on vnodes.
1071 */
1072 int
1073 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1074 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1075 struct vnode *vp, vm_ooffset_t foff, vm_object_t *objp)
1076 {
1077 struct vattr va;
1078 void *handle;
1079 vm_object_t obj;
1080 struct mount *mp;
1081 struct cdevsw *dsw;
1082 int error, flags, type;
1083 int vfslocked;
1084
1085 mp = vp->v_mount;
1086 vfslocked = VFS_LOCK_GIANT(mp);
1087 if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0) {
1088 VFS_UNLOCK_GIANT(vfslocked);
1089 return (error);
1090 }
1091 flags = *flagsp;
1092 obj = vp->v_object;
1093 if (vp->v_type == VREG) {
1094 /*
1095 * Get the proper underlying object
1096 */
1097 if (obj == NULL) {
1098 error = EINVAL;
1099 goto done;
1100 }
1101 if (obj->handle != vp) {
1102 vput(vp);
1103 vp = (struct vnode*)obj->handle;
1104 vget(vp, LK_EXCLUSIVE, td);
1105 }
1106 type = OBJT_VNODE;
1107 handle = vp;
1108 } else if (vp->v_type == VCHR) {
1109 type = OBJT_DEVICE;
1110 handle = vp->v_rdev;
1111
1112 dsw = dev_refthread(handle);
1113 if (dsw == NULL) {
1114 error = ENXIO;
1115 goto done;
1116 }
1117 if (dsw->d_flags & D_MMAP_ANON) {
1118 dev_relthread(handle);
1119 *maxprotp = VM_PROT_ALL;
1120 *flagsp |= MAP_ANON;
1121 error = 0;
1122 goto done;
1123 }
1124 dev_relthread(handle);
1125 /*
1126 * cdevs does not provide private mappings of any kind.
1127 */
1128 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1129 (prot & PROT_WRITE) != 0) {
1130 error = EACCES;
1131 goto done;
1132 }
1133 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1134 error = EINVAL;
1135 goto done;
1136 }
1137 /*
1138 * Force device mappings to be shared.
1139 */
1140 flags |= MAP_SHARED;
1141 } else {
1142 error = EINVAL;
1143 goto done;
1144 }
1145 if ((error = VOP_GETATTR(vp, &va, td->td_ucred, td))) {
1146 goto done;
1147 }
1148 #ifdef MAC
1149 error = mac_check_vnode_mmap(td->td_ucred, vp, prot, flags);
1150 if (error != 0)
1151 goto done;
1152 #endif
1153 if ((flags & MAP_SHARED) != 0) {
1154 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1155 if (prot & PROT_WRITE) {
1156 error = EPERM;
1157 goto done;
1158 }
1159 *maxprotp &= ~VM_PROT_WRITE;
1160 }
1161 }
1162 /*
1163 * If it is a regular file without any references
1164 * we do not need to sync it.
1165 * Adjust object size to be the size of actual file.
1166 */
1167 if (vp->v_type == VREG) {
1168 objsize = round_page(va.va_size);
1169 if (va.va_nlink == 0)
1170 flags |= MAP_NOSYNC;
1171 }
1172 obj = vm_pager_allocate(type, handle, objsize, prot, foff);
1173 if (obj == NULL) {
1174 error = (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1175 goto done;
1176 }
1177 *objp = obj;
1178 *flagsp = flags;
1179 vfs_mark_atime(vp, td);
1180
1181 done:
1182 vput(vp);
1183 VFS_UNLOCK_GIANT(vfslocked);
1184 return (error);
1185 }
1186
1187 /*
1188 * vm_mmap_cdev()
1189 *
1190 * MPSAFE
1191 *
1192 * Helper function for vm_mmap. Perform sanity check specific for mmap
1193 * operations on cdevs.
1194 */
1195 int
1196 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1197 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1198 struct cdev *cdev, vm_ooffset_t foff, vm_object_t *objp)
1199 {
1200 vm_object_t obj;
1201 struct cdevsw *dsw;
1202 int flags;
1203
1204 flags = *flagsp;
1205
1206 dsw = dev_refthread(cdev);
1207 if (dsw == NULL)
1208 return (ENXIO);
1209 if (dsw->d_flags & D_MMAP_ANON) {
1210 dev_relthread(cdev);
1211 *maxprotp = VM_PROT_ALL;
1212 *flagsp |= MAP_ANON;
1213 return (0);
1214 }
1215 dev_relthread(cdev);
1216 /*
1217 * cdevs does not provide private mappings of any kind.
1218 */
1219 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1220 (prot & PROT_WRITE) != 0)
1221 return (EACCES);
1222 if (flags & (MAP_PRIVATE|MAP_COPY))
1223 return (EINVAL);
1224 /*
1225 * Force device mappings to be shared.
1226 */
1227 flags |= MAP_SHARED;
1228 #ifdef MAC_XXX
1229 error = mac_check_cdev_mmap(td->td_ucred, cdev, prot);
1230 if (error != 0)
1231 return (error);
1232 #endif
1233 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, foff);
1234 if (obj == NULL)
1235 return (EINVAL);
1236 *objp = obj;
1237 *flagsp = flags;
1238 return (0);
1239 }
1240
1241 /*
1242 * vm_mmap()
1243 *
1244 * MPSAFE
1245 *
1246 * Internal version of mmap. Currently used by mmap, exec, and sys5
1247 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1248 */
1249 int
1250 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1251 vm_prot_t maxprot, int flags,
1252 objtype_t handle_type, void *handle,
1253 vm_ooffset_t foff)
1254 {
1255 boolean_t fitit;
1256 vm_object_t object;
1257 int rv = KERN_SUCCESS;
1258 int docow, error;
1259 struct thread *td = curthread;
1260
1261 if (size == 0)
1262 return (0);
1263
1264 size = round_page(size);
1265
1266 PROC_LOCK(td->td_proc);
1267 if (td->td_proc->p_vmspace->vm_map.size + size >
1268 lim_cur(td->td_proc, RLIMIT_VMEM)) {
1269 PROC_UNLOCK(td->td_proc);
1270 return(ENOMEM);
1271 }
1272 PROC_UNLOCK(td->td_proc);
1273
1274 /*
1275 * We currently can only deal with page aligned file offsets.
1276 * The check is here rather than in the syscall because the
1277 * kernel calls this function internally for other mmaping
1278 * operations (such as in exec) and non-aligned offsets will
1279 * cause pmap inconsistencies...so we want to be sure to
1280 * disallow this in all cases.
1281 */
1282 if (foff & PAGE_MASK)
1283 return (EINVAL);
1284
1285 if ((flags & MAP_FIXED) == 0) {
1286 fitit = TRUE;
1287 *addr = round_page(*addr);
1288 } else {
1289 if (*addr != trunc_page(*addr))
1290 return (EINVAL);
1291 fitit = FALSE;
1292 }
1293 /*
1294 * Lookup/allocate object.
1295 */
1296 switch (handle_type) {
1297 case OBJT_DEVICE:
1298 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1299 handle, foff, &object);
1300 break;
1301 case OBJT_VNODE:
1302 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1303 handle, foff, &object);
1304 break;
1305 case OBJT_DEFAULT:
1306 if (handle == NULL) {
1307 error = 0;
1308 break;
1309 }
1310 /* FALLTHROUGH */
1311 default:
1312 error = EINVAL;
1313 }
1314 if (error)
1315 return (error);
1316 if (flags & MAP_ANON) {
1317 object = NULL;
1318 docow = 0;
1319 /*
1320 * Unnamed anonymous regions always start at 0.
1321 */
1322 if (handle == 0)
1323 foff = 0;
1324 } else {
1325 docow = MAP_PREFAULT_PARTIAL;
1326 }
1327
1328 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1329 docow |= MAP_COPY_ON_WRITE;
1330 if (flags & MAP_NOSYNC)
1331 docow |= MAP_DISABLE_SYNCER;
1332 if (flags & MAP_NOCORE)
1333 docow |= MAP_DISABLE_COREDUMP;
1334
1335 #if defined(VM_PROT_READ_IS_EXEC)
1336 if (prot & VM_PROT_READ)
1337 prot |= VM_PROT_EXECUTE;
1338
1339 if (maxprot & VM_PROT_READ)
1340 maxprot |= VM_PROT_EXECUTE;
1341 #endif
1342
1343 if (fitit)
1344 *addr = pmap_addr_hint(object, *addr, size);
1345
1346 if (flags & MAP_STACK)
1347 rv = vm_map_stack(map, *addr, size, prot, maxprot,
1348 docow | MAP_STACK_GROWS_DOWN);
1349 else if (fitit)
1350 rv = vm_map_find(map, object, foff, addr, size, TRUE,
1351 prot, maxprot, docow);
1352 else
1353 rv = vm_map_fixed(map, object, foff, addr, size,
1354 prot, maxprot, docow);
1355
1356 if (rv != KERN_SUCCESS) {
1357 /*
1358 * Lose the object reference. Will destroy the
1359 * object if it's an unnamed anonymous mapping
1360 * or named anonymous without other references.
1361 */
1362 vm_object_deallocate(object);
1363 } else if (flags & MAP_SHARED) {
1364 /*
1365 * Shared memory is also shared with children.
1366 */
1367 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1368 if (rv != KERN_SUCCESS)
1369 (void) vm_map_remove(map, *addr, *addr + size);
1370 }
1371
1372 /*
1373 * If the process has requested that all future mappings
1374 * be wired, then heed this.
1375 */
1376 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1377 vm_map_wire(map, *addr, *addr + size,
1378 VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1379
1380 switch (rv) {
1381 case KERN_SUCCESS:
1382 return (0);
1383 case KERN_INVALID_ADDRESS:
1384 case KERN_NO_SPACE:
1385 return (ENOMEM);
1386 case KERN_PROTECTION_FAILURE:
1387 return (EACCES);
1388 default:
1389 return (EINVAL);
1390 }
1391 }
Cache object: bdbb7097ff6feee8554a0b32d7ec4190
|