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: releng/8.4/sys/vm/vm_mmap.c 239592 2012-08-22 20:34:23Z kib $");
45
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/sysproto.h>
54 #include <sys/filedesc.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/resource.h>
58 #include <sys/resourcevar.h>
59 #include <sys/vnode.h>
60 #include <sys/fcntl.h>
61 #include <sys/file.h>
62 #include <sys/mman.h>
63 #include <sys/mount.h>
64 #include <sys/conf.h>
65 #include <sys/stat.h>
66 #include <sys/sysent.h>
67 #include <sys/vmmeter.h>
68
69 #include <security/mac/mac_framework.h>
70
71 #include <vm/vm.h>
72 #include <vm/vm_param.h>
73 #include <vm/pmap.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_pager.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_page.h>
81
82 #ifdef HWPMC_HOOKS
83 #include <sys/pmckern.h>
84 #endif
85
86 #ifndef _SYS_SYSPROTO_H_
87 struct sbrk_args {
88 int incr;
89 };
90 #endif
91
92 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
93 int *, struct vnode *, vm_ooffset_t *, vm_object_t *);
94 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
95 int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
96 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
97 int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
98
99 /*
100 * MPSAFE
101 */
102 /* ARGSUSED */
103 int
104 sbrk(td, uap)
105 struct thread *td;
106 struct sbrk_args *uap;
107 {
108 /* Not yet implemented */
109 return (EOPNOTSUPP);
110 }
111
112 #ifndef _SYS_SYSPROTO_H_
113 struct sstk_args {
114 int incr;
115 };
116 #endif
117
118 /*
119 * MPSAFE
120 */
121 /* ARGSUSED */
122 int
123 sstk(td, uap)
124 struct thread *td;
125 struct sstk_args *uap;
126 {
127 /* Not yet implemented */
128 return (EOPNOTSUPP);
129 }
130
131 #if defined(COMPAT_43)
132 #ifndef _SYS_SYSPROTO_H_
133 struct getpagesize_args {
134 int dummy;
135 };
136 #endif
137
138 /* ARGSUSED */
139 int
140 ogetpagesize(td, uap)
141 struct thread *td;
142 struct getpagesize_args *uap;
143 {
144 /* MP SAFE */
145 td->td_retval[0] = PAGE_SIZE;
146 return (0);
147 }
148 #endif /* COMPAT_43 */
149
150
151 /*
152 * Memory Map (mmap) system call. Note that the file offset
153 * and address are allowed to be NOT page aligned, though if
154 * the MAP_FIXED flag it set, both must have the same remainder
155 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
156 * page-aligned, the actual mapping starts at trunc_page(addr)
157 * and the return value is adjusted up by the page offset.
158 *
159 * Generally speaking, only character devices which are themselves
160 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
161 * there would be no cache coherency between a descriptor and a VM mapping
162 * both to the same character device.
163 */
164 #ifndef _SYS_SYSPROTO_H_
165 struct mmap_args {
166 void *addr;
167 size_t len;
168 int prot;
169 int flags;
170 int fd;
171 long pad;
172 off_t pos;
173 };
174 #endif
175
176 /*
177 * MPSAFE
178 */
179 int
180 mmap(td, uap)
181 struct thread *td;
182 struct mmap_args *uap;
183 {
184 #ifdef HWPMC_HOOKS
185 struct pmckern_map_in pkm;
186 #endif
187 struct file *fp;
188 struct vnode *vp;
189 vm_offset_t addr;
190 vm_size_t size, pageoff;
191 vm_prot_t prot, maxprot;
192 void *handle;
193 objtype_t handle_type;
194 int flags, error;
195 off_t pos;
196 struct vmspace *vms = td->td_proc->p_vmspace;
197
198 addr = (vm_offset_t) uap->addr;
199 size = uap->len;
200 prot = uap->prot & VM_PROT_ALL;
201 flags = uap->flags;
202 pos = uap->pos;
203
204 fp = NULL;
205
206 /*
207 * Enforce the constraints.
208 * Mapping of length 0 is only allowed for old binaries.
209 * Anonymous mapping shall specify -1 as filedescriptor and
210 * zero position for new code. Be nice to ancient a.out
211 * binaries and correct pos for anonymous mapping, since old
212 * ld.so sometimes issues anonymous map requests with non-zero
213 * pos.
214 */
215 if (!SV_CURPROC_FLAG(SV_AOUT)) {
216 if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
217 ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
218 return (EINVAL);
219 } else {
220 if ((flags & MAP_ANON) != 0)
221 pos = 0;
222 }
223
224 if (flags & MAP_STACK) {
225 if ((uap->fd != -1) ||
226 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
227 return (EINVAL);
228 flags |= MAP_ANON;
229 pos = 0;
230 }
231
232 /*
233 * Align the file position to a page boundary,
234 * and save its page offset component.
235 */
236 pageoff = (pos & PAGE_MASK);
237 pos -= pageoff;
238
239 /* Adjust size for rounding (on both ends). */
240 size += pageoff; /* low end... */
241 size = (vm_size_t) round_page(size); /* hi end */
242
243 /*
244 * Check for illegal addresses. Watch out for address wrap... Note
245 * that VM_*_ADDRESS are not constants due to casts (argh).
246 */
247 if (flags & MAP_FIXED) {
248 /*
249 * The specified address must have the same remainder
250 * as the file offset taken modulo PAGE_SIZE, so it
251 * should be aligned after adjustment by pageoff.
252 */
253 addr -= pageoff;
254 if (addr & PAGE_MASK)
255 return (EINVAL);
256 /* Address range must be all in user VM space. */
257 if (addr < vm_map_min(&vms->vm_map) ||
258 addr + size > vm_map_max(&vms->vm_map))
259 return (EINVAL);
260 if (addr + size < addr)
261 return (EINVAL);
262 } else {
263 /*
264 * XXX for non-fixed mappings where no hint is provided or
265 * the hint would fall in the potential heap space,
266 * place it after the end of the largest possible heap.
267 *
268 * There should really be a pmap call to determine a reasonable
269 * location.
270 */
271 PROC_LOCK(td->td_proc);
272 if (addr == 0 ||
273 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
274 addr < round_page((vm_offset_t)vms->vm_daddr +
275 lim_max(td->td_proc, RLIMIT_DATA))))
276 addr = round_page((vm_offset_t)vms->vm_daddr +
277 lim_max(td->td_proc, RLIMIT_DATA));
278 PROC_UNLOCK(td->td_proc);
279 }
280 if (flags & MAP_ANON) {
281 /*
282 * Mapping blank space is trivial.
283 */
284 handle = NULL;
285 handle_type = OBJT_DEFAULT;
286 maxprot = VM_PROT_ALL;
287 } else {
288 /*
289 * Mapping file, get fp for validation and
290 * don't let the descriptor disappear on us if we block.
291 */
292 if ((error = fget(td, uap->fd, &fp)) != 0)
293 goto done;
294 if (fp->f_type == DTYPE_SHM) {
295 handle = fp->f_data;
296 handle_type = OBJT_SWAP;
297 maxprot = VM_PROT_NONE;
298
299 /* FREAD should always be set. */
300 if (fp->f_flag & FREAD)
301 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
302 if (fp->f_flag & FWRITE)
303 maxprot |= VM_PROT_WRITE;
304 goto map;
305 }
306 if (fp->f_type != DTYPE_VNODE) {
307 error = ENODEV;
308 goto done;
309 }
310 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
311 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
312 /*
313 * POSIX shared-memory objects are defined to have
314 * kernel persistence, and are not defined to support
315 * read(2)/write(2) -- or even open(2). Thus, we can
316 * use MAP_ASYNC to trade on-disk coherence for speed.
317 * The shm_open(3) library routine turns on the FPOSIXSHM
318 * flag to request this behavior.
319 */
320 if (fp->f_flag & FPOSIXSHM)
321 flags |= MAP_NOSYNC;
322 #endif
323 vp = fp->f_vnode;
324 /*
325 * Ensure that file and memory protections are
326 * compatible. Note that we only worry about
327 * writability if mapping is shared; in this case,
328 * current and max prot are dictated by the open file.
329 * XXX use the vnode instead? Problem is: what
330 * credentials do we use for determination? What if
331 * proc does a setuid?
332 */
333 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
334 maxprot = VM_PROT_NONE;
335 else
336 maxprot = VM_PROT_EXECUTE;
337 if (fp->f_flag & FREAD) {
338 maxprot |= VM_PROT_READ;
339 } else if (prot & PROT_READ) {
340 error = EACCES;
341 goto done;
342 }
343 /*
344 * If we are sharing potential changes (either via
345 * MAP_SHARED or via the implicit sharing of character
346 * device mappings), and we are trying to get write
347 * permission although we opened it without asking
348 * for it, bail out.
349 */
350 if ((flags & MAP_SHARED) != 0) {
351 if ((fp->f_flag & FWRITE) != 0) {
352 maxprot |= VM_PROT_WRITE;
353 } else if ((prot & PROT_WRITE) != 0) {
354 error = EACCES;
355 goto done;
356 }
357 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
358 maxprot |= VM_PROT_WRITE;
359 }
360 handle = (void *)vp;
361 handle_type = OBJT_VNODE;
362 }
363 map:
364 td->td_fpop = fp;
365 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
366 flags, handle_type, handle, pos);
367 td->td_fpop = NULL;
368 #ifdef HWPMC_HOOKS
369 /* inform hwpmc(4) if an executable is being mapped */
370 if (error == 0 && handle_type == OBJT_VNODE &&
371 (prot & PROT_EXEC)) {
372 pkm.pm_file = handle;
373 pkm.pm_address = (uintptr_t) addr;
374 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
375 }
376 #endif
377 if (error == 0)
378 td->td_retval[0] = (register_t) (addr + pageoff);
379 done:
380 if (fp)
381 fdrop(fp, td);
382
383 return (error);
384 }
385
386 int
387 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
388 {
389 struct mmap_args oargs;
390
391 oargs.addr = uap->addr;
392 oargs.len = uap->len;
393 oargs.prot = uap->prot;
394 oargs.flags = uap->flags;
395 oargs.fd = uap->fd;
396 oargs.pos = uap->pos;
397 return (mmap(td, &oargs));
398 }
399
400 #ifdef COMPAT_43
401 #ifndef _SYS_SYSPROTO_H_
402 struct ommap_args {
403 caddr_t addr;
404 int len;
405 int prot;
406 int flags;
407 int fd;
408 long pos;
409 };
410 #endif
411 int
412 ommap(td, uap)
413 struct thread *td;
414 struct ommap_args *uap;
415 {
416 struct mmap_args nargs;
417 static const char cvtbsdprot[8] = {
418 0,
419 PROT_EXEC,
420 PROT_WRITE,
421 PROT_EXEC | PROT_WRITE,
422 PROT_READ,
423 PROT_EXEC | PROT_READ,
424 PROT_WRITE | PROT_READ,
425 PROT_EXEC | PROT_WRITE | PROT_READ,
426 };
427
428 #define OMAP_ANON 0x0002
429 #define OMAP_COPY 0x0020
430 #define OMAP_SHARED 0x0010
431 #define OMAP_FIXED 0x0100
432
433 nargs.addr = uap->addr;
434 nargs.len = uap->len;
435 nargs.prot = cvtbsdprot[uap->prot & 0x7];
436 nargs.flags = 0;
437 if (uap->flags & OMAP_ANON)
438 nargs.flags |= MAP_ANON;
439 if (uap->flags & OMAP_COPY)
440 nargs.flags |= MAP_COPY;
441 if (uap->flags & OMAP_SHARED)
442 nargs.flags |= MAP_SHARED;
443 else
444 nargs.flags |= MAP_PRIVATE;
445 if (uap->flags & OMAP_FIXED)
446 nargs.flags |= MAP_FIXED;
447 nargs.fd = uap->fd;
448 nargs.pos = uap->pos;
449 return (mmap(td, &nargs));
450 }
451 #endif /* COMPAT_43 */
452
453
454 #ifndef _SYS_SYSPROTO_H_
455 struct msync_args {
456 void *addr;
457 size_t len;
458 int flags;
459 };
460 #endif
461 /*
462 * MPSAFE
463 */
464 int
465 msync(td, uap)
466 struct thread *td;
467 struct msync_args *uap;
468 {
469 vm_offset_t addr;
470 vm_size_t size, pageoff;
471 int flags;
472 vm_map_t map;
473 int rv;
474
475 addr = (vm_offset_t) uap->addr;
476 size = uap->len;
477 flags = uap->flags;
478
479 pageoff = (addr & PAGE_MASK);
480 addr -= pageoff;
481 size += pageoff;
482 size = (vm_size_t) round_page(size);
483 if (addr + size < addr)
484 return (EINVAL);
485
486 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
487 return (EINVAL);
488
489 map = &td->td_proc->p_vmspace->vm_map;
490
491 /*
492 * Clean the pages and interpret the return value.
493 */
494 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
495 (flags & MS_INVALIDATE) != 0);
496 switch (rv) {
497 case KERN_SUCCESS:
498 return (0);
499 case KERN_INVALID_ADDRESS:
500 return (EINVAL); /* Sun returns ENOMEM? */
501 case KERN_INVALID_ARGUMENT:
502 return (EBUSY);
503 case KERN_FAILURE:
504 return (EIO);
505 default:
506 return (EINVAL);
507 }
508 }
509
510 #ifndef _SYS_SYSPROTO_H_
511 struct munmap_args {
512 void *addr;
513 size_t len;
514 };
515 #endif
516 /*
517 * MPSAFE
518 */
519 int
520 munmap(td, uap)
521 struct thread *td;
522 struct munmap_args *uap;
523 {
524 #ifdef HWPMC_HOOKS
525 struct pmckern_map_out pkm;
526 vm_map_entry_t entry;
527 #endif
528 vm_offset_t addr;
529 vm_size_t size, pageoff;
530 vm_map_t map;
531
532 addr = (vm_offset_t) uap->addr;
533 size = uap->len;
534 if (size == 0)
535 return (EINVAL);
536
537 pageoff = (addr & PAGE_MASK);
538 addr -= pageoff;
539 size += pageoff;
540 size = (vm_size_t) round_page(size);
541 if (addr + size < addr)
542 return (EINVAL);
543
544 /*
545 * Check for illegal addresses. Watch out for address wrap...
546 */
547 map = &td->td_proc->p_vmspace->vm_map;
548 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
549 return (EINVAL);
550 vm_map_lock(map);
551 #ifdef HWPMC_HOOKS
552 /*
553 * Inform hwpmc if the address range being unmapped contains
554 * an executable region.
555 */
556 pkm.pm_address = (uintptr_t) NULL;
557 if (vm_map_lookup_entry(map, addr, &entry)) {
558 for (;
559 entry != &map->header && entry->start < addr + size;
560 entry = entry->next) {
561 if (vm_map_check_protection(map, entry->start,
562 entry->end, VM_PROT_EXECUTE) == TRUE) {
563 pkm.pm_address = (uintptr_t) addr;
564 pkm.pm_size = (size_t) size;
565 break;
566 }
567 }
568 }
569 #endif
570 vm_map_delete(map, addr, addr + size);
571
572 #ifdef HWPMC_HOOKS
573 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
574 vm_map_lock_downgrade(map);
575 if (pkm.pm_address != (uintptr_t) NULL)
576 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
577 vm_map_unlock_read(map);
578 #else
579 vm_map_unlock(map);
580 #endif
581 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
582 return (0);
583 }
584
585 #ifndef _SYS_SYSPROTO_H_
586 struct mprotect_args {
587 const void *addr;
588 size_t len;
589 int prot;
590 };
591 #endif
592 /*
593 * MPSAFE
594 */
595 int
596 mprotect(td, uap)
597 struct thread *td;
598 struct mprotect_args *uap;
599 {
600 vm_offset_t addr;
601 vm_size_t size, pageoff;
602 vm_prot_t prot;
603
604 addr = (vm_offset_t) uap->addr;
605 size = uap->len;
606 prot = uap->prot & VM_PROT_ALL;
607
608 pageoff = (addr & PAGE_MASK);
609 addr -= pageoff;
610 size += pageoff;
611 size = (vm_size_t) round_page(size);
612 if (addr + size < addr)
613 return (EINVAL);
614
615 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
616 addr + size, prot, FALSE)) {
617 case KERN_SUCCESS:
618 return (0);
619 case KERN_PROTECTION_FAILURE:
620 return (EACCES);
621 case KERN_RESOURCE_SHORTAGE:
622 return (ENOMEM);
623 }
624 return (EINVAL);
625 }
626
627 #ifndef _SYS_SYSPROTO_H_
628 struct minherit_args {
629 void *addr;
630 size_t len;
631 int inherit;
632 };
633 #endif
634 /*
635 * MPSAFE
636 */
637 int
638 minherit(td, uap)
639 struct thread *td;
640 struct minherit_args *uap;
641 {
642 vm_offset_t addr;
643 vm_size_t size, pageoff;
644 vm_inherit_t inherit;
645
646 addr = (vm_offset_t)uap->addr;
647 size = uap->len;
648 inherit = uap->inherit;
649
650 pageoff = (addr & PAGE_MASK);
651 addr -= pageoff;
652 size += pageoff;
653 size = (vm_size_t) round_page(size);
654 if (addr + size < addr)
655 return (EINVAL);
656
657 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
658 addr + size, inherit)) {
659 case KERN_SUCCESS:
660 return (0);
661 case KERN_PROTECTION_FAILURE:
662 return (EACCES);
663 }
664 return (EINVAL);
665 }
666
667 #ifndef _SYS_SYSPROTO_H_
668 struct madvise_args {
669 void *addr;
670 size_t len;
671 int behav;
672 };
673 #endif
674
675 /*
676 * MPSAFE
677 */
678 /* ARGSUSED */
679 int
680 madvise(td, uap)
681 struct thread *td;
682 struct madvise_args *uap;
683 {
684 vm_offset_t start, end;
685 vm_map_t map;
686 struct proc *p;
687 int error;
688
689 /*
690 * Check for our special case, advising the swap pager we are
691 * "immortal."
692 */
693 if (uap->behav == MADV_PROTECT) {
694 error = priv_check(td, PRIV_VM_MADV_PROTECT);
695 if (error == 0) {
696 p = td->td_proc;
697 PROC_LOCK(p);
698 p->p_flag |= P_PROTECTED;
699 PROC_UNLOCK(p);
700 }
701 return (error);
702 }
703 /*
704 * Check for illegal behavior
705 */
706 if (uap->behav < 0 || uap->behav > MADV_CORE)
707 return (EINVAL);
708 /*
709 * Check for illegal addresses. Watch out for address wrap... Note
710 * that VM_*_ADDRESS are not constants due to casts (argh).
711 */
712 map = &td->td_proc->p_vmspace->vm_map;
713 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
714 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
715 return (EINVAL);
716 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
717 return (EINVAL);
718
719 /*
720 * Since this routine is only advisory, we default to conservative
721 * behavior.
722 */
723 start = trunc_page((vm_offset_t) uap->addr);
724 end = round_page((vm_offset_t) uap->addr + uap->len);
725
726 if (vm_map_madvise(map, start, end, uap->behav))
727 return (EINVAL);
728 return (0);
729 }
730
731 #ifndef _SYS_SYSPROTO_H_
732 struct mincore_args {
733 const void *addr;
734 size_t len;
735 char *vec;
736 };
737 #endif
738
739 /*
740 * MPSAFE
741 */
742 /* ARGSUSED */
743 int
744 mincore(td, uap)
745 struct thread *td;
746 struct mincore_args *uap;
747 {
748 vm_offset_t addr, first_addr;
749 vm_offset_t end, cend;
750 pmap_t pmap;
751 vm_map_t map;
752 char *vec;
753 int error = 0;
754 int vecindex, lastvecindex;
755 vm_map_entry_t current;
756 vm_map_entry_t entry;
757 int mincoreinfo;
758 unsigned int timestamp;
759
760 /*
761 * Make sure that the addresses presented are valid for user
762 * mode.
763 */
764 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
765 end = addr + (vm_size_t)round_page(uap->len);
766 map = &td->td_proc->p_vmspace->vm_map;
767 if (end > vm_map_max(map) || end < addr)
768 return (ENOMEM);
769
770 /*
771 * Address of byte vector
772 */
773 vec = uap->vec;
774
775 pmap = vmspace_pmap(td->td_proc->p_vmspace);
776
777 vm_map_lock_read(map);
778 RestartScan:
779 timestamp = map->timestamp;
780
781 if (!vm_map_lookup_entry(map, addr, &entry)) {
782 vm_map_unlock_read(map);
783 return (ENOMEM);
784 }
785
786 /*
787 * Do this on a map entry basis so that if the pages are not
788 * in the current processes address space, we can easily look
789 * up the pages elsewhere.
790 */
791 lastvecindex = -1;
792 for (current = entry;
793 (current != &map->header) && (current->start < end);
794 current = current->next) {
795
796 /*
797 * check for contiguity
798 */
799 if (current->end < end &&
800 (entry->next == &map->header ||
801 current->next->start > current->end)) {
802 vm_map_unlock_read(map);
803 return (ENOMEM);
804 }
805
806 /*
807 * ignore submaps (for now) or null objects
808 */
809 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
810 current->object.vm_object == NULL)
811 continue;
812
813 /*
814 * limit this scan to the current map entry and the
815 * limits for the mincore call
816 */
817 if (addr < current->start)
818 addr = current->start;
819 cend = current->end;
820 if (cend > end)
821 cend = end;
822
823 /*
824 * scan this entry one page at a time
825 */
826 while (addr < cend) {
827 /*
828 * Check pmap first, it is likely faster, also
829 * it can provide info as to whether we are the
830 * one referencing or modifying the page.
831 */
832 mincoreinfo = pmap_mincore(pmap, addr);
833 if (!mincoreinfo) {
834 vm_pindex_t pindex;
835 vm_ooffset_t offset;
836 vm_page_t m;
837 /*
838 * calculate the page index into the object
839 */
840 offset = current->offset + (addr - current->start);
841 pindex = OFF_TO_IDX(offset);
842 VM_OBJECT_LOCK(current->object.vm_object);
843 m = vm_page_lookup(current->object.vm_object,
844 pindex);
845 /*
846 * if the page is resident, then gather information about
847 * it.
848 */
849 if (m != NULL && m->valid != 0) {
850 mincoreinfo = MINCORE_INCORE;
851 vm_page_lock_queues();
852 if (m->dirty ||
853 pmap_is_modified(m))
854 mincoreinfo |= MINCORE_MODIFIED_OTHER;
855 if ((m->flags & PG_REFERENCED) ||
856 pmap_ts_referenced(m)) {
857 vm_page_flag_set(m, PG_REFERENCED);
858 mincoreinfo |= MINCORE_REFERENCED_OTHER;
859 }
860 vm_page_unlock_queues();
861 }
862 VM_OBJECT_UNLOCK(current->object.vm_object);
863 }
864
865 /*
866 * subyte may page fault. In case it needs to modify
867 * the map, we release the lock.
868 */
869 vm_map_unlock_read(map);
870
871 /*
872 * calculate index into user supplied byte vector
873 */
874 vecindex = OFF_TO_IDX(addr - first_addr);
875
876 /*
877 * If we have skipped map entries, we need to make sure that
878 * the byte vector is zeroed for those skipped entries.
879 */
880 while ((lastvecindex + 1) < vecindex) {
881 error = subyte(vec + lastvecindex, 0);
882 if (error) {
883 error = EFAULT;
884 goto done2;
885 }
886 ++lastvecindex;
887 }
888
889 /*
890 * Pass the page information to the user
891 */
892 error = subyte(vec + vecindex, mincoreinfo);
893 if (error) {
894 error = EFAULT;
895 goto done2;
896 }
897
898 /*
899 * If the map has changed, due to the subyte, the previous
900 * output may be invalid.
901 */
902 vm_map_lock_read(map);
903 if (timestamp != map->timestamp)
904 goto RestartScan;
905
906 lastvecindex = vecindex;
907 addr += PAGE_SIZE;
908 }
909 }
910
911 /*
912 * subyte may page fault. In case it needs to modify
913 * the map, we release the lock.
914 */
915 vm_map_unlock_read(map);
916
917 /*
918 * Zero the last entries in the byte vector.
919 */
920 vecindex = OFF_TO_IDX(end - first_addr);
921 while ((lastvecindex + 1) < vecindex) {
922 error = subyte(vec + lastvecindex, 0);
923 if (error) {
924 error = EFAULT;
925 goto done2;
926 }
927 ++lastvecindex;
928 }
929
930 /*
931 * If the map has changed, due to the subyte, the previous
932 * output may be invalid.
933 */
934 vm_map_lock_read(map);
935 if (timestamp != map->timestamp)
936 goto RestartScan;
937 vm_map_unlock_read(map);
938 done2:
939 return (error);
940 }
941
942 #ifndef _SYS_SYSPROTO_H_
943 struct mlock_args {
944 const void *addr;
945 size_t len;
946 };
947 #endif
948 /*
949 * MPSAFE
950 */
951 int
952 mlock(td, uap)
953 struct thread *td;
954 struct mlock_args *uap;
955 {
956 struct proc *proc;
957 vm_offset_t addr, end, last, start;
958 vm_size_t npages, size;
959 int error;
960
961 error = priv_check(td, PRIV_VM_MLOCK);
962 if (error)
963 return (error);
964 addr = (vm_offset_t)uap->addr;
965 size = uap->len;
966 last = addr + size;
967 start = trunc_page(addr);
968 end = round_page(last);
969 if (last < addr || end < addr)
970 return (EINVAL);
971 npages = atop(end - start);
972 if (npages > vm_page_max_wired)
973 return (ENOMEM);
974 proc = td->td_proc;
975 PROC_LOCK(proc);
976 if (ptoa(npages +
977 pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
978 lim_cur(proc, RLIMIT_MEMLOCK)) {
979 PROC_UNLOCK(proc);
980 return (ENOMEM);
981 }
982 PROC_UNLOCK(proc);
983 if (npages + cnt.v_wire_count > vm_page_max_wired)
984 return (EAGAIN);
985 error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
986 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
987 return (error == KERN_SUCCESS ? 0 : ENOMEM);
988 }
989
990 #ifndef _SYS_SYSPROTO_H_
991 struct mlockall_args {
992 int how;
993 };
994 #endif
995
996 /*
997 * MPSAFE
998 */
999 int
1000 mlockall(td, uap)
1001 struct thread *td;
1002 struct mlockall_args *uap;
1003 {
1004 vm_map_t map;
1005 int error;
1006
1007 map = &td->td_proc->p_vmspace->vm_map;
1008 error = 0;
1009
1010 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1011 return (EINVAL);
1012
1013 #if 0
1014 /*
1015 * If wiring all pages in the process would cause it to exceed
1016 * a hard resource limit, return ENOMEM.
1017 */
1018 PROC_LOCK(td->td_proc);
1019 if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1020 lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1021 PROC_UNLOCK(td->td_proc);
1022 return (ENOMEM);
1023 }
1024 PROC_UNLOCK(td->td_proc);
1025 #else
1026 error = priv_check(td, PRIV_VM_MLOCK);
1027 if (error)
1028 return (error);
1029 #endif
1030
1031 if (uap->how & MCL_FUTURE) {
1032 vm_map_lock(map);
1033 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1034 vm_map_unlock(map);
1035 error = 0;
1036 }
1037
1038 if (uap->how & MCL_CURRENT) {
1039 /*
1040 * P1003.1-2001 mandates that all currently mapped pages
1041 * will be memory resident and locked (wired) upon return
1042 * from mlockall(). vm_map_wire() will wire pages, by
1043 * calling vm_fault_wire() for each page in the region.
1044 */
1045 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1046 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1047 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1048 }
1049
1050 return (error);
1051 }
1052
1053 #ifndef _SYS_SYSPROTO_H_
1054 struct munlockall_args {
1055 register_t dummy;
1056 };
1057 #endif
1058
1059 /*
1060 * MPSAFE
1061 */
1062 int
1063 munlockall(td, uap)
1064 struct thread *td;
1065 struct munlockall_args *uap;
1066 {
1067 vm_map_t map;
1068 int error;
1069
1070 map = &td->td_proc->p_vmspace->vm_map;
1071 error = priv_check(td, PRIV_VM_MUNLOCK);
1072 if (error)
1073 return (error);
1074
1075 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1076 vm_map_lock(map);
1077 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1078 vm_map_unlock(map);
1079
1080 /* Forcibly unwire all pages. */
1081 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1082 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1083
1084 return (error);
1085 }
1086
1087 #ifndef _SYS_SYSPROTO_H_
1088 struct munlock_args {
1089 const void *addr;
1090 size_t len;
1091 };
1092 #endif
1093 /*
1094 * MPSAFE
1095 */
1096 int
1097 munlock(td, uap)
1098 struct thread *td;
1099 struct munlock_args *uap;
1100 {
1101 vm_offset_t addr, end, last, start;
1102 vm_size_t size;
1103 int error;
1104
1105 error = priv_check(td, PRIV_VM_MUNLOCK);
1106 if (error)
1107 return (error);
1108 addr = (vm_offset_t)uap->addr;
1109 size = uap->len;
1110 last = addr + size;
1111 start = trunc_page(addr);
1112 end = round_page(last);
1113 if (last < addr || end < addr)
1114 return (EINVAL);
1115 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1116 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1117 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1118 }
1119
1120 /*
1121 * vm_mmap_vnode()
1122 *
1123 * MPSAFE
1124 *
1125 * Helper function for vm_mmap. Perform sanity check specific for mmap
1126 * operations on vnodes.
1127 */
1128 int
1129 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1130 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1131 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp)
1132 {
1133 struct vattr va;
1134 vm_object_t obj;
1135 vm_offset_t foff;
1136 struct mount *mp;
1137 struct ucred *cred;
1138 int error, flags;
1139 int vfslocked;
1140
1141 mp = vp->v_mount;
1142 cred = td->td_ucred;
1143 vfslocked = VFS_LOCK_GIANT(mp);
1144 if ((error = vget(vp, LK_SHARED, td)) != 0) {
1145 VFS_UNLOCK_GIANT(vfslocked);
1146 return (error);
1147 }
1148 foff = *foffp;
1149 flags = *flagsp;
1150 obj = vp->v_object;
1151 if (vp->v_type == VREG) {
1152 /*
1153 * Get the proper underlying object
1154 */
1155 if (obj == NULL) {
1156 error = EINVAL;
1157 goto done;
1158 }
1159 if (obj->handle != vp) {
1160 vput(vp);
1161 vp = (struct vnode*)obj->handle;
1162 vget(vp, LK_SHARED, td);
1163 }
1164 } else if (vp->v_type == VCHR) {
1165 error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1166 vp->v_rdev, foffp, objp);
1167 if (error == 0)
1168 goto mark_atime;
1169 goto done;
1170 } else {
1171 error = EINVAL;
1172 goto done;
1173 }
1174 if ((error = VOP_GETATTR(vp, &va, cred)))
1175 goto done;
1176 #ifdef MAC
1177 error = mac_vnode_check_mmap(cred, vp, prot, flags);
1178 if (error != 0)
1179 goto done;
1180 #endif
1181 if ((flags & MAP_SHARED) != 0) {
1182 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1183 if (prot & PROT_WRITE) {
1184 error = EPERM;
1185 goto done;
1186 }
1187 *maxprotp &= ~VM_PROT_WRITE;
1188 }
1189 }
1190 /*
1191 * If it is a regular file without any references
1192 * we do not need to sync it.
1193 * Adjust object size to be the size of actual file.
1194 */
1195 objsize = round_page(va.va_size);
1196 if (va.va_nlink == 0)
1197 flags |= MAP_NOSYNC;
1198 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, td->td_ucred);
1199 if (obj == NULL) {
1200 error = ENOMEM;
1201 goto done;
1202 }
1203 *objp = obj;
1204 *flagsp = flags;
1205
1206 mark_atime:
1207 vfs_mark_atime(vp, cred);
1208
1209 done:
1210 vput(vp);
1211 VFS_UNLOCK_GIANT(vfslocked);
1212 return (error);
1213 }
1214
1215 /*
1216 * vm_mmap_cdev()
1217 *
1218 * MPSAFE
1219 *
1220 * Helper function for vm_mmap. Perform sanity check specific for mmap
1221 * operations on cdevs.
1222 */
1223 int
1224 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1225 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1226 struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1227 {
1228 vm_object_t obj;
1229 struct cdevsw *dsw;
1230 int error, flags, ref;
1231
1232 flags = *flagsp;
1233
1234 dsw = dev_refthread(cdev, &ref);
1235 if (dsw == NULL)
1236 return (ENXIO);
1237 if (dsw->d_flags & D_MMAP_ANON) {
1238 dev_relthread(cdev, ref);
1239 *maxprotp = VM_PROT_ALL;
1240 *flagsp |= MAP_ANON;
1241 return (0);
1242 }
1243 /*
1244 * cdevs do not provide private mappings of any kind.
1245 */
1246 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1247 (prot & PROT_WRITE) != 0) {
1248 dev_relthread(cdev, ref);
1249 return (EACCES);
1250 }
1251 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1252 dev_relthread(cdev, ref);
1253 return (EINVAL);
1254 }
1255 /*
1256 * Force device mappings to be shared.
1257 */
1258 flags |= MAP_SHARED;
1259 #ifdef MAC_XXX
1260 error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1261 if (error != 0) {
1262 dev_relthread(cdev, ref);
1263 return (error);
1264 }
1265 #endif
1266 /*
1267 * First, try d_mmap_single(). If that is not implemented
1268 * (returns ENODEV), fall back to using the device pager.
1269 * Note that d_mmap_single() must return a reference to the
1270 * object (it needs to bump the reference count of the object
1271 * it returns somehow).
1272 *
1273 * XXX assumes VM_PROT_* == PROT_*
1274 */
1275 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1276 dev_relthread(cdev, ref);
1277 if (error != ENODEV)
1278 return (error);
1279 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1280 td->td_ucred);
1281 if (obj == NULL)
1282 return (EINVAL);
1283 *objp = obj;
1284 *flagsp = flags;
1285 return (0);
1286 }
1287
1288 /*
1289 * vm_mmap_shm()
1290 *
1291 * MPSAFE
1292 *
1293 * Helper function for vm_mmap. Perform sanity check specific for mmap
1294 * operations on shm file descriptors.
1295 */
1296 int
1297 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1298 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1299 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1300 {
1301 int error;
1302
1303 if ((*flagsp & MAP_SHARED) != 0 &&
1304 (*maxprotp & VM_PROT_WRITE) == 0 &&
1305 (prot & PROT_WRITE) != 0)
1306 return (EACCES);
1307 #ifdef MAC
1308 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1309 if (error != 0)
1310 return (error);
1311 #endif
1312 error = shm_mmap(shmfd, objsize, foff, objp);
1313 if (error)
1314 return (error);
1315 return (0);
1316 }
1317
1318 /*
1319 * vm_mmap()
1320 *
1321 * MPSAFE
1322 *
1323 * Internal version of mmap. Currently used by mmap, exec, and sys5
1324 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1325 */
1326 int
1327 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1328 vm_prot_t maxprot, int flags,
1329 objtype_t handle_type, void *handle,
1330 vm_ooffset_t foff)
1331 {
1332 boolean_t fitit;
1333 vm_object_t object = NULL;
1334 int rv = KERN_SUCCESS;
1335 int docow, error;
1336 struct thread *td = curthread;
1337
1338 if (size == 0)
1339 return (0);
1340
1341 size = round_page(size);
1342
1343 PROC_LOCK(td->td_proc);
1344 if (td->td_proc->p_vmspace->vm_map.size + size >
1345 lim_cur(td->td_proc, RLIMIT_VMEM)) {
1346 PROC_UNLOCK(td->td_proc);
1347 return (ENOMEM);
1348 }
1349 PROC_UNLOCK(td->td_proc);
1350
1351 /*
1352 * We currently can only deal with page aligned file offsets.
1353 * The check is here rather than in the syscall because the
1354 * kernel calls this function internally for other mmaping
1355 * operations (such as in exec) and non-aligned offsets will
1356 * cause pmap inconsistencies...so we want to be sure to
1357 * disallow this in all cases.
1358 */
1359 if (foff & PAGE_MASK)
1360 return (EINVAL);
1361
1362 if ((flags & MAP_FIXED) == 0) {
1363 fitit = TRUE;
1364 *addr = round_page(*addr);
1365 } else {
1366 if (*addr != trunc_page(*addr))
1367 return (EINVAL);
1368 fitit = FALSE;
1369 }
1370 /*
1371 * Lookup/allocate object.
1372 */
1373 switch (handle_type) {
1374 case OBJT_DEVICE:
1375 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1376 handle, &foff, &object);
1377 break;
1378 case OBJT_VNODE:
1379 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1380 handle, &foff, &object);
1381 break;
1382 case OBJT_SWAP:
1383 error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1384 handle, foff, &object);
1385 break;
1386 case OBJT_DEFAULT:
1387 if (handle == NULL) {
1388 error = 0;
1389 break;
1390 }
1391 /* FALLTHROUGH */
1392 default:
1393 error = EINVAL;
1394 break;
1395 }
1396 if (error)
1397 return (error);
1398 if (flags & MAP_ANON) {
1399 object = NULL;
1400 docow = 0;
1401 /*
1402 * Unnamed anonymous regions always start at 0.
1403 */
1404 if (handle == 0)
1405 foff = 0;
1406 } else if (flags & MAP_PREFAULT_READ)
1407 docow = MAP_PREFAULT;
1408 else
1409 docow = MAP_PREFAULT_PARTIAL;
1410
1411 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1412 docow |= MAP_COPY_ON_WRITE;
1413 if (flags & MAP_NOSYNC)
1414 docow |= MAP_DISABLE_SYNCER;
1415 if (flags & MAP_NOCORE)
1416 docow |= MAP_DISABLE_COREDUMP;
1417
1418 if (flags & MAP_STACK)
1419 rv = vm_map_stack(map, *addr, size, prot, maxprot,
1420 docow | MAP_STACK_GROWS_DOWN);
1421 else if (fitit)
1422 rv = vm_map_find(map, object, foff, addr, size,
1423 object != NULL && object->type == OBJT_DEVICE ?
1424 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1425 else
1426 rv = vm_map_fixed(map, object, foff, *addr, size,
1427 prot, maxprot, docow);
1428
1429 if (rv != KERN_SUCCESS) {
1430 /*
1431 * Lose the object reference. Will destroy the
1432 * object if it's an unnamed anonymous mapping
1433 * or named anonymous without other references.
1434 */
1435 vm_object_deallocate(object);
1436 } else if (flags & MAP_SHARED) {
1437 /*
1438 * Shared memory is also shared with children.
1439 */
1440 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1441 if (rv != KERN_SUCCESS)
1442 (void) vm_map_remove(map, *addr, *addr + size);
1443 }
1444
1445 /*
1446 * If the process has requested that all future mappings
1447 * be wired, then heed this.
1448 */
1449 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1450 vm_map_wire(map, *addr, *addr + size,
1451 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1452 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1453
1454 return (vm_mmap_to_errno(rv));
1455 }
1456
1457 int
1458 vm_mmap_to_errno(int rv)
1459 {
1460
1461 switch (rv) {
1462 case KERN_SUCCESS:
1463 return (0);
1464 case KERN_INVALID_ADDRESS:
1465 case KERN_NO_SPACE:
1466 return (ENOMEM);
1467 case KERN_PROTECTION_FAILURE:
1468 return (EACCES);
1469 default:
1470 return (EINVAL);
1471 }
1472 }
Cache object: 235b0bd648b1804b4509431424ad7393
|