4 * Copyright (c) 1988 University of Utah.
5 * Copyright (c) 1991, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
38 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
39 * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $
43 * Mapped file (mmap) interface to VM
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/sysproto.h>
50 #include <sys/filedesc.h>
51 #include <sys/kern_syscall.h>
54 #include <sys/resource.h>
55 #include <sys/resourcevar.h>
56 #include <sys/vnode.h>
57 #include <sys/fcntl.h>
62 #include <sys/vmmeter.h>
63 #include <sys/sysctl.h>
66 #include <vm/vm_param.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_pager.h>
73 #include <vm/vm_pageout.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_kern.h>
77 #include <sys/file2.h>
78 #include <sys/thread.h>
79 #include <sys/thread2.h>
80 #include <vm/vm_page2.h>
82 static int max_proc_mmap
= 1000000;
83 SYSCTL_INT(_vm
, OID_AUTO
, max_proc_mmap
, CTLFLAG_RW
, &max_proc_mmap
, 0, "");
85 SYSCTL_INT(_vm
, OID_AUTO
, vkernel_enable
, CTLFLAG_RW
, &vkernel_enable
, 0, "");
93 sys_sstk(struct sstk_args
*uap
)
95 /* Not yet implemented */
100 * mmap_args(void *addr, size_t len, int prot, int flags, int fd,
101 * long pad, off_t pos)
103 * Memory Map (mmap) system call. Note that the file offset
104 * and address are allowed to be NOT page aligned, though if
105 * the MAP_FIXED flag it set, both must have the same remainder
106 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
107 * page-aligned, the actual mapping starts at trunc_page(addr)
108 * and the return value is adjusted up by the page offset.
110 * Generally speaking, only character devices which are themselves
111 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
112 * there would be no cache coherency between a descriptor and a VM mapping
113 * both to the same character device.
115 * Block devices can be mmap'd no matter what they represent. Cache coherency
116 * is maintained as long as you do not write directly to the underlying
122 kern_mmap(struct vmspace
*vms
, caddr_t uaddr
, size_t ulen
,
123 int uprot
, int uflags
, int fd
, off_t upos
, void **res
)
125 struct thread
*td
= curthread
;
126 struct proc
*p
= td
->td_proc
;
127 struct file
*fp
= NULL
;
131 vm_size_t size
, pageoff
;
132 vm_prot_t prot
, maxprot
;
140 addr
= (vm_offset_t
) uaddr
;
142 prot
= uprot
& VM_PROT_ALL
;
147 * Make sure mapping fits into numeric range etc.
149 * NOTE: We support the full unsigned range for size now.
151 if (((flags
& MAP_ANON
) && (fd
!= -1 || pos
!= 0)))
157 if (flags
& MAP_STACK
) {
160 if ((prot
& (PROT_READ
|PROT_WRITE
)) != (PROT_READ
|PROT_WRITE
))
167 * Virtual page tables cannot be used with MAP_STACK. Apart from
168 * it not making any sense, the aux union is used by both
171 * Because the virtual page table is stored in the backing object
172 * and might be updated by the kernel, the mapping must be R+W.
174 if (flags
& MAP_VPAGETABLE
) {
175 if (vkernel_enable
== 0)
177 if (flags
& MAP_STACK
)
179 if ((prot
& (PROT_READ
|PROT_WRITE
)) != (PROT_READ
|PROT_WRITE
))
184 * Align the file position to a page boundary,
185 * and save its page offset component.
187 pageoff
= (pos
& PAGE_MASK
);
190 /* Adjust size for rounding (on both ends). */
191 size
+= pageoff
; /* low end... */
192 size
= (vm_size_t
) round_page(size
); /* hi end */
193 if (size
< ulen
) /* wrap */
197 * Check for illegal addresses. Watch out for address wrap... Note
198 * that VM_*_ADDRESS are not constants due to casts (argh).
200 if (flags
& (MAP_FIXED
| MAP_TRYFIXED
)) {
202 * The specified address must have the same remainder
203 * as the file offset taken modulo PAGE_SIZE, so it
204 * should be aligned after adjustment by pageoff.
207 if (addr
& PAGE_MASK
)
211 * Address range must be all in user VM space and not wrap.
213 tmpaddr
= addr
+ size
;
216 if (VM_MAX_USER_ADDRESS
> 0 && tmpaddr
> VM_MAX_USER_ADDRESS
)
218 if (VM_MIN_USER_ADDRESS
> 0 && addr
< VM_MIN_USER_ADDRESS
)
222 * Get a hint of where to map. It also provides mmap offset
223 * randomization if enabled.
225 addr
= vm_map_hint(p
, addr
, prot
);
228 if (flags
& MAP_ANON
) {
230 * Mapping blank space is trivial.
233 maxprot
= VM_PROT_ALL
;
236 * Mapping file, get fp for validation. Obtain vnode and make
237 * sure it is of appropriate type.
239 fp
= holdfp(td
, fd
, -1);
242 if (fp
->f_type
!= DTYPE_VNODE
) {
247 * POSIX shared-memory objects are defined to have
248 * kernel persistence, and are not defined to support
249 * read(2)/write(2) -- or even open(2). Thus, we can
250 * use MAP_ASYNC to trade on-disk coherence for speed.
251 * The shm_open(3) library routine turns on the FPOSIXSHM
252 * flag to request this behavior.
254 if (fp
->f_flag
& FPOSIXSHM
)
256 vp
= (struct vnode
*) fp
->f_data
;
259 * Validate the vnode for the operation.
264 * Get the proper underlying object
266 if ((obj
= vp
->v_object
) == NULL
) {
270 KKASSERT((struct vnode
*)obj
->handle
== vp
);
274 * Make sure a device has not been revoked.
275 * Mappability is handled by the device layer.
277 if (vp
->v_rdev
== NULL
) {
284 * Nothing else is mappable.
291 * XXX hack to handle use of /dev/zero to map anon memory (ala
294 if (vp
->v_type
== VCHR
&& iszerodev(vp
->v_rdev
)) {
296 maxprot
= VM_PROT_ALL
;
301 * cdevs does not provide private mappings of any kind.
303 if (vp
->v_type
== VCHR
&&
304 (flags
& (MAP_PRIVATE
|MAP_COPY
))) {
309 * Ensure that file and memory protections are
310 * compatible. Note that we only worry about
311 * writability if mapping is shared; in this case,
312 * current and max prot are dictated by the open file.
313 * XXX use the vnode instead? Problem is: what
314 * credentials do we use for determination? What if
315 * proc does a setuid?
317 maxprot
= VM_PROT_EXECUTE
;
318 if (fp
->f_flag
& FREAD
) {
319 maxprot
|= VM_PROT_READ
;
320 } else if (prot
& PROT_READ
) {
325 * If we are sharing potential changes (either via
326 * MAP_SHARED or via the implicit sharing of character
327 * device mappings), and we are trying to get write
328 * permission although we opened it without asking
329 * for it, bail out. Check for superuser, only if
330 * we're at securelevel < 1, to allow the XIG X server
331 * to continue to work.
333 * PROT_WRITE + MAP_SHARED
335 if ((flags
& MAP_SHARED
) != 0 || vp
->v_type
== VCHR
) {
336 if ((fp
->f_flag
& FWRITE
) != 0) {
338 if ((error
= VOP_GETATTR(vp
, &va
))) {
342 (IMMUTABLE
|APPEND
)) == 0) {
343 maxprot
|= VM_PROT_WRITE
;
346 * SHARED+RW file mmap()
347 * updates v_lastwrite_ts.
349 if ((prot
& PROT_WRITE
) &&
350 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
) == 0) {
351 vfs_timestamp(&vp
->v_lastwrite_ts
);
352 vsetflags(vp
, VLASTWRITETS
);
355 } else if (prot
& PROT_WRITE
) {
359 } else if ((prot
& PROT_WRITE
) != 0) {
364 maxprot
|= VM_PROT_WRITE
;
370 lwkt_gettoken(&vms
->vm_map
.token
);
373 * Do not allow more then a certain number of vm_map_entry structures
374 * per process. 0 to disable.
376 if (max_proc_mmap
&& vms
->vm_map
.nentries
>= max_proc_mmap
) {
378 lwkt_reltoken(&vms
->vm_map
.token
);
382 error
= vm_mmap(&vms
->vm_map
, &addr
, size
, prot
, maxprot
,
385 *res
= (void *)(addr
+ pageoff
);
387 lwkt_reltoken(&vms
->vm_map
.token
);
396 * mmap system call handler
401 sys_mmap(struct mmap_args
*uap
)
405 error
= kern_mmap(curproc
->p_vmspace
, uap
->addr
, uap
->len
,
406 uap
->prot
, uap
->flags
,
407 uap
->fd
, uap
->pos
, &uap
->sysmsg_resultp
);
413 * msync system call handler
415 * msync_args(void *addr, size_t len, int flags)
420 sys_msync(struct msync_args
*uap
)
422 struct proc
*p
= curproc
;
425 vm_size_t size
, pageoff
;
430 addr
= (vm_offset_t
) uap
->addr
;
434 pageoff
= (addr
& PAGE_MASK
);
437 size
= (vm_size_t
) round_page(size
);
438 if (size
< uap
->len
) /* wrap */
440 tmpaddr
= addr
+ size
; /* workaround gcc4 opt */
441 if (tmpaddr
< addr
) /* wrap */
444 if ((flags
& (MS_ASYNC
|MS_INVALIDATE
)) == (MS_ASYNC
|MS_INVALIDATE
))
447 map
= &p
->p_vmspace
->vm_map
;
450 * map->token serializes extracting the address range for size == 0
451 * msyncs with the vm_map_clean call; if the token were not held
452 * across the two calls, an intervening munmap/mmap pair, for example,
453 * could cause msync to occur on a wrong region.
455 lwkt_gettoken(&map
->token
);
458 * XXX Gak! If size is zero we are supposed to sync "all modified
459 * pages with the region containing addr". Unfortunately, we don't
460 * really keep track of individual mmaps so we approximate by flushing
461 * the range of the map entry containing addr. This can be incorrect
462 * if the region splits or is coalesced with a neighbor.
465 vm_map_entry_t entry
;
467 vm_map_lock_read(map
);
468 rv
= vm_map_lookup_entry(map
, addr
, &entry
);
470 vm_map_unlock_read(map
);
471 rv
= KERN_INVALID_ADDRESS
;
475 size
= entry
->end
- entry
->start
;
476 vm_map_unlock_read(map
);
480 * Clean the pages and interpret the return value.
482 rv
= vm_map_clean(map
, addr
, addr
+ size
, (flags
& MS_ASYNC
) == 0,
483 (flags
& MS_INVALIDATE
) != 0);
485 lwkt_reltoken(&map
->token
);
490 case KERN_INVALID_ADDRESS
:
491 return (EINVAL
); /* Sun returns ENOMEM? */
502 * munmap system call handler
504 * munmap_args(void *addr, size_t len)
509 sys_munmap(struct munmap_args
*uap
)
511 struct proc
*p
= curproc
;
514 vm_size_t size
, pageoff
;
517 addr
= (vm_offset_t
) uap
->addr
;
520 pageoff
= (addr
& PAGE_MASK
);
523 size
= (vm_size_t
) round_page(size
);
524 if (size
< uap
->len
) /* wrap */
526 tmpaddr
= addr
+ size
; /* workaround gcc4 opt */
527 if (tmpaddr
< addr
) /* wrap */
534 * Check for illegal addresses. Watch out for address wrap... Note
535 * that VM_*_ADDRESS are not constants due to casts (argh).
537 if (VM_MAX_USER_ADDRESS
> 0 && tmpaddr
> VM_MAX_USER_ADDRESS
)
539 if (VM_MIN_USER_ADDRESS
> 0 && addr
< VM_MIN_USER_ADDRESS
)
542 map
= &p
->p_vmspace
->vm_map
;
544 /* map->token serializes between the map check and the actual unmap */
545 lwkt_gettoken(&map
->token
);
548 * Make sure entire range is allocated.
550 if (!vm_map_check_protection(map
, addr
, addr
+ size
,
551 VM_PROT_NONE
, FALSE
)) {
552 lwkt_reltoken(&map
->token
);
555 /* returns nothing but KERN_SUCCESS anyway */
556 vm_map_remove(map
, addr
, addr
+ size
);
557 lwkt_reltoken(&map
->token
);
562 * mprotect_args(const void *addr, size_t len, int prot)
567 sys_mprotect(struct mprotect_args
*uap
)
569 struct proc
*p
= curproc
;
572 vm_size_t size
, pageoff
;
576 addr
= (vm_offset_t
) uap
->addr
;
578 prot
= uap
->prot
& VM_PROT_ALL
;
580 pageoff
= (addr
& PAGE_MASK
);
583 size
= (vm_size_t
) round_page(size
);
584 if (size
< uap
->len
) /* wrap */
586 tmpaddr
= addr
+ size
; /* workaround gcc4 opt */
587 if (tmpaddr
< addr
) /* wrap */
590 switch (vm_map_protect(&p
->p_vmspace
->vm_map
, addr
, addr
+ size
,
595 case KERN_PROTECTION_FAILURE
:
606 * minherit system call handler
608 * minherit_args(void *addr, size_t len, int inherit)
613 sys_minherit(struct minherit_args
*uap
)
615 struct proc
*p
= curproc
;
618 vm_size_t size
, pageoff
;
619 vm_inherit_t inherit
;
622 addr
= (vm_offset_t
)uap
->addr
;
624 inherit
= uap
->inherit
;
626 pageoff
= (addr
& PAGE_MASK
);
629 size
= (vm_size_t
) round_page(size
);
630 if (size
< uap
->len
) /* wrap */
632 tmpaddr
= addr
+ size
; /* workaround gcc4 opt */
633 if (tmpaddr
< addr
) /* wrap */
636 switch (vm_map_inherit(&p
->p_vmspace
->vm_map
, addr
,
637 addr
+ size
, inherit
)) {
641 case KERN_PROTECTION_FAILURE
:
652 * madvise system call handler
654 * madvise_args(void *addr, size_t len, int behav)
659 sys_madvise(struct madvise_args
*uap
)
661 struct proc
*p
= curproc
;
662 vm_offset_t start
, end
;
663 vm_offset_t tmpaddr
= (vm_offset_t
)uap
->addr
+ uap
->len
;
667 * Check for illegal behavior
669 if (uap
->behav
< 0 || uap
->behav
>= MADV_CONTROL_END
)
672 * Check for illegal addresses. Watch out for address wrap... Note
673 * that VM_*_ADDRESS are not constants due to casts (argh).
675 if (tmpaddr
< (vm_offset_t
)uap
->addr
)
677 if (VM_MAX_USER_ADDRESS
> 0 && tmpaddr
> VM_MAX_USER_ADDRESS
)
679 if (VM_MIN_USER_ADDRESS
> 0 && uap
->addr
< VM_MIN_USER_ADDRESS
)
683 * Since this routine is only advisory, we default to conservative
686 start
= trunc_page((vm_offset_t
)uap
->addr
);
687 end
= round_page(tmpaddr
);
689 error
= vm_map_madvise(&p
->p_vmspace
->vm_map
, start
, end
,
695 * mcontrol system call handler
697 * mcontrol_args(void *addr, size_t len, int behav, off_t value)
702 sys_mcontrol(struct mcontrol_args
*uap
)
704 struct proc
*p
= curproc
;
705 vm_offset_t start
, end
;
706 vm_offset_t tmpaddr
= (vm_offset_t
)uap
->addr
+ uap
->len
;
710 * Check for illegal behavior
712 if (uap
->behav
< 0 || uap
->behav
> MADV_CONTROL_END
)
715 * Check for illegal addresses. Watch out for address wrap... Note
716 * that VM_*_ADDRESS are not constants due to casts (argh).
718 if (tmpaddr
< (vm_offset_t
) uap
->addr
)
720 if (VM_MAX_USER_ADDRESS
> 0 && tmpaddr
> VM_MAX_USER_ADDRESS
)
722 if (VM_MIN_USER_ADDRESS
> 0 && uap
->addr
< VM_MIN_USER_ADDRESS
)
726 * Since this routine is only advisory, we default to conservative
729 start
= trunc_page((vm_offset_t
)uap
->addr
);
730 end
= round_page(tmpaddr
);
732 error
= vm_map_madvise(&p
->p_vmspace
->vm_map
, start
, end
,
733 uap
->behav
, uap
->value
);
739 * mincore system call handler
741 * mincore_args(const void *addr, size_t len, char *vec)
746 sys_mincore(struct mincore_args
*uap
)
748 struct proc
*p
= curproc
;
749 vm_offset_t addr
, first_addr
;
750 vm_offset_t end
, cend
;
755 int vecindex
, lastvecindex
;
756 vm_map_entry_t current
;
757 vm_map_entry_t entry
;
759 unsigned int timestamp
;
762 * Make sure that the addresses presented are valid for user
765 first_addr
= addr
= trunc_page((vm_offset_t
) uap
->addr
);
766 end
= addr
+ (vm_size_t
)round_page(uap
->len
);
769 if (VM_MAX_USER_ADDRESS
> 0 && end
> VM_MAX_USER_ADDRESS
)
773 * Address of byte vector
777 map
= &p
->p_vmspace
->vm_map
;
778 pmap
= vmspace_pmap(p
->p_vmspace
);
780 lwkt_gettoken(&map
->token
);
781 vm_map_lock_read(map
);
783 timestamp
= map
->timestamp
;
785 if (!vm_map_lookup_entry(map
, addr
, &entry
))
789 * Do this on a map entry basis so that if the pages are not
790 * in the current processes address space, we can easily look
791 * up the pages elsewhere.
795 (current
!= &map
->header
) && (current
->start
< end
);
796 current
= current
->next
) {
799 * ignore submaps (for now) or null objects
801 if (current
->maptype
!= VM_MAPTYPE_NORMAL
&&
802 current
->maptype
!= VM_MAPTYPE_VPAGETABLE
) {
805 if (current
->object
.vm_object
== NULL
)
809 * limit this scan to the current map entry and the
810 * limits for the mincore call
812 if (addr
< current
->start
)
813 addr
= current
->start
;
819 * scan this entry one page at a time
821 while (addr
< cend
) {
823 * Check pmap first, it is likely faster, also
824 * it can provide info as to whether we are the
825 * one referencing or modifying the page.
827 * If we have to check the VM object, only mess
828 * around with normal maps. Do not mess around
829 * with virtual page tables (XXX).
831 mincoreinfo
= pmap_mincore(pmap
, addr
);
832 if (mincoreinfo
== 0 &&
833 current
->maptype
== VM_MAPTYPE_NORMAL
) {
839 * calculate the page index into the object
841 offset
= current
->offset
+ (addr
- current
->start
);
842 pindex
= OFF_TO_IDX(offset
);
845 * if the page is resident, then gather
846 * information about it. spl protection is
847 * required to maintain the object
848 * association. And XXX what if the page is
849 * busy? What's the deal with that?
851 * XXX vm_token - legacy for pmap_ts_referenced
852 * in x86 and vkernel pmap code.
854 lwkt_gettoken(&vm_token
);
855 vm_object_hold(current
->object
.vm_object
);
856 m
= vm_page_lookup(current
->object
.vm_object
,
859 mincoreinfo
= MINCORE_INCORE
;
860 if (m
->dirty
|| pmap_is_modified(m
))
861 mincoreinfo
|= MINCORE_MODIFIED_OTHER
;
862 if ((m
->flags
& PG_REFERENCED
) ||
863 pmap_ts_referenced(m
)) {
864 vm_page_flag_set(m
, PG_REFERENCED
);
865 mincoreinfo
|= MINCORE_REFERENCED_OTHER
;
868 vm_object_drop(current
->object
.vm_object
);
869 lwkt_reltoken(&vm_token
);
873 * subyte may page fault. In case it needs to modify
874 * the map, we release the lock.
876 vm_map_unlock_read(map
);
879 * calculate index into user supplied byte vector
881 vecindex
= OFF_TO_IDX(addr
- first_addr
);
884 * If we have skipped map entries, we need to make sure that
885 * the byte vector is zeroed for those skipped entries.
887 while((lastvecindex
+ 1) < vecindex
) {
888 error
= subyte( vec
+ lastvecindex
, 0);
897 * Pass the page information to the user
899 error
= subyte(vec
+ vecindex
, mincoreinfo
);
906 * If the map has changed, due to the subyte,
907 * the previous output may be invalid.
909 vm_map_lock_read(map
);
910 if (timestamp
!= map
->timestamp
)
913 lastvecindex
= vecindex
;
919 * subyte may page fault. In case it needs to modify
920 * the map, we release the lock.
922 vm_map_unlock_read(map
);
925 * Zero the last entries in the byte vector.
927 vecindex
= OFF_TO_IDX(end
- first_addr
);
928 while((lastvecindex
+ 1) < vecindex
) {
929 error
= subyte( vec
+ lastvecindex
, 0);
938 * If the map has changed, due to the subyte, the previous
939 * output may be invalid.
941 vm_map_lock_read(map
);
942 if (timestamp
!= map
->timestamp
)
944 vm_map_unlock_read(map
);
948 lwkt_reltoken(&map
->token
);
953 * mlock system call handler
955 * mlock_args(const void *addr, size_t len)
960 sys_mlock(struct mlock_args
*uap
)
964 vm_size_t size
, pageoff
;
965 struct thread
*td
= curthread
;
966 struct proc
*p
= td
->td_proc
;
969 addr
= (vm_offset_t
) uap
->addr
;
972 pageoff
= (addr
& PAGE_MASK
);
975 size
= (vm_size_t
) round_page(size
);
976 if (size
< uap
->len
) /* wrap */
978 if (size
== 0) /* silently allow 0 size */
980 tmpaddr
= addr
+ size
; /* workaround gcc4 opt */
981 if (tmpaddr
< addr
) /* wrap */
984 if (atop(size
) + vmstats
.v_wire_count
> vm_page_max_wired
)
988 * We do not need to synchronize against other threads updating ucred;
989 * they update p->ucred, which is synchronized into td_ucred ourselves.
991 #ifdef pmap_wired_count
992 if (size
+ ptoa(pmap_wired_count(vm_map_pmap(&p
->p_vmspace
->vm_map
))) >
993 p
->p_rlimit
[RLIMIT_MEMLOCK
].rlim_cur
) {
997 error
= priv_check_cred(td
->td_ucred
, PRIV_ROOT
, 0);
1002 error
= vm_map_unwire(&p
->p_vmspace
->vm_map
, addr
, addr
+ size
, FALSE
);
1003 return (error
== KERN_SUCCESS
? 0 : ENOMEM
);
1012 sys_mlockall(struct mlockall_args
*uap
)
1014 struct thread
*td
= curthread
;
1015 struct proc
*p
= td
->td_proc
;
1016 vm_map_t map
= &p
->p_vmspace
->vm_map
;
1017 vm_map_entry_t entry
;
1019 int rc
= KERN_SUCCESS
;
1021 if (((how
& MCL_CURRENT
) == 0) && ((how
& MCL_FUTURE
) == 0))
1024 rc
= priv_check_cred(td
->td_ucred
, PRIV_ROOT
, 0);
1030 if (how
& MCL_CURRENT
) {
1031 for(entry
= map
->header
.next
;
1032 entry
!= &map
->header
;
1033 entry
= entry
->next
);
1039 if (how
& MCL_FUTURE
)
1040 map
->flags
|= MAP_WIREFUTURE
;
1050 * Unwire all user-wired map entries, cancel MCL_FUTURE.
1055 sys_munlockall(struct munlockall_args
*uap
)
1057 struct thread
*td
= curthread
;
1058 struct proc
*p
= td
->td_proc
;
1059 vm_map_t map
= &p
->p_vmspace
->vm_map
;
1060 vm_map_entry_t entry
;
1061 int rc
= KERN_SUCCESS
;
1065 /* Clear MAP_WIREFUTURE to cancel mlockall(MCL_FUTURE) */
1066 map
->flags
&= ~MAP_WIREFUTURE
;
1069 for (entry
= map
->header
.next
;
1070 entry
!= &map
->header
;
1071 entry
= entry
->next
) {
1072 if ((entry
->eflags
& MAP_ENTRY_USER_WIRED
) == 0)
1076 * If we encounter an in-transition entry, we release the
1077 * map lock and retry the scan; we do not decrement any
1078 * wired_count more than once because we do not touch
1079 * any entries with MAP_ENTRY_USER_WIRED not set.
1081 * There is a potential interleaving with concurrent
1082 * mlockall()s here -- if we abort a scan, an mlockall()
1083 * could start, wire a number of entries before our
1084 * current position in, and then stall itself on this
1085 * or any other in-transition entry. If that occurs, when
1086 * we resume, we will unwire those entries.
1088 if (entry
->eflags
& MAP_ENTRY_IN_TRANSITION
) {
1089 entry
->eflags
|= MAP_ENTRY_NEEDS_WAKEUP
;
1090 ++mycpu
->gd_cnt
.v_intrans_coll
;
1091 ++mycpu
->gd_cnt
.v_intrans_wait
;
1092 vm_map_transition_wait(map
, 1);
1096 KASSERT(entry
->wired_count
> 0,
1097 ("wired_count was 0 with USER_WIRED set! %p", entry
));
1099 /* Drop wired count, if it hits zero, unwire the entry */
1100 entry
->eflags
&= ~MAP_ENTRY_USER_WIRED
;
1101 entry
->wired_count
--;
1102 if (entry
->wired_count
== 0)
1103 vm_fault_unwire(map
, entry
);
1112 * munlock system call handler
1114 * munlock_args(const void *addr, size_t len)
1119 sys_munlock(struct munlock_args
*uap
)
1121 struct thread
*td
= curthread
;
1122 struct proc
*p
= td
->td_proc
;
1124 vm_offset_t tmpaddr
;
1125 vm_size_t size
, pageoff
;
1128 addr
= (vm_offset_t
) uap
->addr
;
1131 pageoff
= (addr
& PAGE_MASK
);
1134 size
= (vm_size_t
) round_page(size
);
1136 tmpaddr
= addr
+ size
;
1137 if (tmpaddr
< addr
) /* wrap */
1139 if (size
== 0) /* silently allow 0 size */
1142 #ifndef pmap_wired_count
1143 error
= priv_check(td
, PRIV_ROOT
);
1148 error
= vm_map_unwire(&p
->p_vmspace
->vm_map
, addr
, addr
+ size
, TRUE
);
1149 return (error
== KERN_SUCCESS
? 0 : ENOMEM
);
1153 * Internal version of mmap.
1154 * Currently used by mmap, exec, and sys5 shared memory.
1155 * Handle is either a vnode pointer or NULL for MAP_ANON.
1160 vm_mmap(vm_map_t map
, vm_offset_t
*addr
, vm_size_t size
, vm_prot_t prot
,
1161 vm_prot_t maxprot
, int flags
, void *handle
, vm_ooffset_t foff
)
1168 int (*uksmap
)(cdev_t dev
, vm_page_t fake
);
1170 struct thread
*td
= curthread
;
1172 int rv
= KERN_SUCCESS
;
1180 objsize
= round_page(size
);
1185 lwkt_gettoken(&map
->token
);
1188 * XXX messy code, fixme
1190 * NOTE: Overflow checks require discrete statements or GCC4
1191 * will optimize it out.
1193 if ((p
= curproc
) != NULL
&& map
== &p
->p_vmspace
->vm_map
) {
1194 esize
= map
->size
+ size
; /* workaround gcc4 opt */
1195 if (esize
< map
->size
||
1196 esize
> p
->p_rlimit
[RLIMIT_VMEM
].rlim_cur
) {
1197 lwkt_reltoken(&map
->token
);
1203 * We currently can only deal with page aligned file offsets.
1204 * The check is here rather than in the syscall because the
1205 * kernel calls this function internally for other mmaping
1206 * operations (such as in exec) and non-aligned offsets will
1207 * cause pmap inconsistencies...so we want to be sure to
1208 * disallow this in all cases.
1210 * NOTE: Overflow checks require discrete statements or GCC4
1211 * will optimize it out.
1213 if (foff
& PAGE_MASK
) {
1214 lwkt_reltoken(&map
->token
);
1219 * Handle alignment. For large memory maps it is possible
1220 * that the MMU can optimize the page table so align anything
1221 * that is a multiple of SEG_SIZE to SEG_SIZE.
1223 * Also align any large mapping (bigger than 16x SG_SIZE) to a
1224 * SEG_SIZE address boundary.
1226 if (flags
& MAP_SIZEALIGN
) {
1228 if ((align
^ (align
- 1)) != (align
<< 1) - 1) {
1229 lwkt_reltoken(&map
->token
);
1232 } else if ((flags
& MAP_FIXED
) == 0 &&
1233 ((size
& SEG_MASK
) == 0 || size
> SEG_SIZE
* 16)) {
1239 if ((flags
& (MAP_FIXED
| MAP_TRYFIXED
)) == 0) {
1241 *addr
= round_page(*addr
);
1243 if (*addr
!= trunc_page(*addr
)) {
1244 lwkt_reltoken(&map
->token
);
1247 eaddr
= *addr
+ size
;
1248 if (eaddr
< *addr
) {
1249 lwkt_reltoken(&map
->token
);
1253 if ((flags
& MAP_TRYFIXED
) == 0)
1254 vm_map_remove(map
, *addr
, *addr
+ size
);
1260 * Lookup/allocate object.
1262 if (flags
& MAP_ANON
) {
1264 * Unnamed anonymous regions always start at 0.
1268 * Default memory object
1270 object
= default_pager_alloc(handle
, objsize
,
1272 if (object
== NULL
) {
1273 lwkt_reltoken(&map
->token
);
1276 docow
= MAP_PREFAULT_PARTIAL
;
1279 * Implicit single instance of a default memory
1280 * object, so we don't need a VM object yet.
1288 vp
= (struct vnode
*)handle
;
1291 * Non-anonymous mappings of VCHR (aka not /dev/zero)
1292 * cannot specify MAP_STACK or MAP_VPAGETABLE.
1294 if (vp
->v_type
== VCHR
) {
1295 if (flags
& (MAP_STACK
| MAP_VPAGETABLE
)) {
1296 lwkt_reltoken(&map
->token
);
1301 if (vp
->v_type
== VCHR
&& vp
->v_rdev
->si_ops
->d_uksmap
) {
1303 * Device mappings without a VM object, typically
1304 * sharing permanently allocated kernel memory or
1305 * process-context-specific (per-process) data.
1307 * Force them to be shared.
1309 uksmap
= vp
->v_rdev
->si_ops
->d_uksmap
;
1311 docow
= MAP_PREFAULT_PARTIAL
;
1312 flags
&= ~(MAP_PRIVATE
|MAP_COPY
);
1313 flags
|= MAP_SHARED
;
1314 } else if (vp
->v_type
== VCHR
) {
1316 * Device mappings (device size unknown?).
1317 * Force them to be shared.
1319 error
= dev_dmmap_single(vp
->v_rdev
, &foff
, objsize
,
1320 &object
, prot
, NULL
);
1322 if (error
== ENODEV
) {
1323 handle
= (void *)(intptr_t)vp
->v_rdev
;
1324 object
= dev_pager_alloc(handle
, objsize
, prot
, foff
);
1325 if (object
== NULL
) {
1326 lwkt_reltoken(&map
->token
);
1330 lwkt_reltoken(&map
->token
);
1334 docow
= MAP_PREFAULT_PARTIAL
;
1335 flags
&= ~(MAP_PRIVATE
|MAP_COPY
);
1336 flags
|= MAP_SHARED
;
1339 * Regular file mapping (typically). The attribute
1340 * check is for the link count test only. mmapable
1341 * vnodes must already have a VM object assigned.
1346 error
= VOP_GETATTR(vp
, &vat
);
1348 lwkt_reltoken(&map
->token
);
1351 docow
= MAP_PREFAULT_PARTIAL
;
1352 object
= vnode_pager_reference(vp
);
1353 if (object
== NULL
&& vp
->v_type
== VREG
) {
1354 lwkt_reltoken(&map
->token
);
1355 kprintf("Warning: cannot mmap vnode %p, no "
1361 * If it is a regular file without any references
1362 * we do not need to sync it.
1364 if (vp
->v_type
== VREG
&& vat
.va_nlink
== 0) {
1365 flags
|= MAP_NOSYNC
;
1371 * Deal with the adjusted flags
1373 if ((flags
& (MAP_ANON
|MAP_SHARED
)) == 0)
1374 docow
|= MAP_COPY_ON_WRITE
;
1375 if (flags
& MAP_NOSYNC
)
1376 docow
|= MAP_DISABLE_SYNCER
;
1377 if (flags
& MAP_NOCORE
)
1378 docow
|= MAP_DISABLE_COREDUMP
;
1381 * This may place the area in its own page directory if (size) is
1382 * large enough, otherwise it typically returns its argument.
1384 * (object can be NULL)
1387 *addr
= pmap_addr_hint(object
, *addr
, size
);
1391 * Stack mappings need special attention.
1393 * Mappings that use virtual page tables will default to storing
1394 * the page table at offset 0.
1397 rv
= vm_map_find(map
, uksmap
, vp
->v_rdev
,
1400 VM_MAPTYPE_UKSMAP
, VM_SUBSYS_MMAP
,
1401 prot
, maxprot
, docow
);
1402 } else if (flags
& MAP_STACK
) {
1403 rv
= vm_map_stack(map
, *addr
, size
, flags
,
1404 prot
, maxprot
, docow
);
1405 } else if (flags
& MAP_VPAGETABLE
) {
1406 rv
= vm_map_find(map
, object
, NULL
,
1409 VM_MAPTYPE_VPAGETABLE
, VM_SUBSYS_MMAP
,
1410 prot
, maxprot
, docow
);
1412 rv
= vm_map_find(map
, object
, NULL
,
1415 VM_MAPTYPE_NORMAL
, VM_SUBSYS_MMAP
,
1416 prot
, maxprot
, docow
);
1419 if (rv
!= KERN_SUCCESS
) {
1421 * Lose the object reference. Will destroy the
1422 * object if it's an unnamed anonymous mapping
1423 * or named anonymous without other references.
1425 * (NOTE: object can be NULL)
1427 vm_object_deallocate(object
);
1432 * Shared memory is also shared with children.
1434 if (flags
& (MAP_SHARED
|MAP_INHERIT
)) {
1435 rv
= vm_map_inherit(map
, *addr
, *addr
+ size
, VM_INHERIT_SHARE
);
1436 if (rv
!= KERN_SUCCESS
) {
1437 vm_map_remove(map
, *addr
, *addr
+ size
);
1442 /* If a process has marked all future mappings for wiring, do so */
1443 if ((rv
== KERN_SUCCESS
) && (map
->flags
& MAP_WIREFUTURE
))
1444 vm_map_unwire(map
, *addr
, *addr
+ size
, FALSE
);
1447 * Set the access time on the vnode
1450 vn_mark_atime(vp
, td
);
1452 lwkt_reltoken(&map
->token
);
1457 case KERN_INVALID_ADDRESS
:
1460 case KERN_PROTECTION_FAILURE
:
1468 * Translate a Mach VM return code to zero on success or the appropriate errno
1472 vm_mmap_to_errno(int rv
)
1478 case KERN_INVALID_ADDRESS
:
1481 case KERN_PROTECTION_FAILURE
: