2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. 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: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
54 * Carnegie Mellon requests users of this software to return to
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
64 * $FreeBSD: src/sys/vm/vm_map.c,v 1.187.2.19 2003/05/27 00:47:02 alc Exp $
65 * $DragonFly: src/sys/vm/vm_map.c,v 1.56 2007/04/29 18:25:41 dillon Exp $
69 * Virtual memory mapping module.
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/kernel.h>
77 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/resourcevar.h>
83 #include <sys/malloc.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_kern.h>
93 #include <vm/vm_extern.h>
94 #include <vm/swap_pager.h>
95 #include <vm/vm_zone.h>
97 #include <sys/thread2.h>
98 #include <sys/sysref2.h>
101 * Virtual memory maps provide for the mapping, protection,
102 * and sharing of virtual memory objects. In addition,
103 * this module provides for an efficient virtual copy of
104 * memory from one map to another.
106 * Synchronization is required prior to most operations.
108 * Maps consist of an ordered doubly-linked list of simple
109 * entries; a single hint is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static void vmspace_terminate(struct vmspace
*vm
);
125 static void vmspace_dtor(void *obj
, void *private);
127 MALLOC_DEFINE(M_VMSPACE
, "vmspace", "vmspace objcache backingstore");
129 struct sysref_class vmspace_sysref_class
= {
132 .proto
= SYSREF_PROTO_VMSPACE
,
133 .offset
= offsetof(struct vmspace
, vm_sysref
),
134 .objsize
= sizeof(struct vmspace
),
136 .flags
= SRC_MANAGEDINIT
,
137 .dtor
= vmspace_dtor
,
139 .terminate
= (sysref_terminate_func_t
)vmspace_terminate
145 static struct vm_zone mapentzone_store
, mapzone_store
;
146 static vm_zone_t mapentzone
, mapzone
;
147 static struct vm_object mapentobj
, mapobj
;
149 static struct vm_map_entry map_entry_init
[MAX_MAPENT
];
150 static struct vm_map_entry cpu_map_entry_init
[MAXCPU
][VMEPERCPU
];
151 static struct vm_map map_init
[MAX_KMAP
];
153 static void vm_map_entry_shadow(vm_map_entry_t entry
);
154 static vm_map_entry_t
vm_map_entry_create(vm_map_t map
, int *);
155 static void vm_map_entry_dispose (vm_map_t map
, vm_map_entry_t entry
, int *);
156 static void _vm_map_clip_end (vm_map_t
, vm_map_entry_t
, vm_offset_t
, int *);
157 static void _vm_map_clip_start (vm_map_t
, vm_map_entry_t
, vm_offset_t
, int *);
158 static void vm_map_entry_delete (vm_map_t
, vm_map_entry_t
, int *);
159 static void vm_map_entry_unwire (vm_map_t
, vm_map_entry_t
);
160 static void vm_map_copy_entry (vm_map_t
, vm_map_t
, vm_map_entry_t
,
162 static void vm_map_split (vm_map_entry_t
);
163 static void vm_map_unclip_range (vm_map_t map
, vm_map_entry_t start_entry
, vm_offset_t start
, vm_offset_t end
, int *count
, int flags
);
168 * Initialize the vm_map module. Must be called before
169 * any other vm_map routines.
171 * Map and entry structures are allocated from the general
172 * purpose memory pool with some exceptions:
174 * - The kernel map and kmem submap are allocated statically.
175 * - Kernel map entries are allocated out of a static pool.
177 * These restrictions are necessary since malloc() uses the
178 * maps and requires map entries.
183 mapzone
= &mapzone_store
;
184 zbootinit(mapzone
, "MAP", sizeof (struct vm_map
),
186 mapentzone
= &mapentzone_store
;
187 zbootinit(mapentzone
, "MAP ENTRY", sizeof (struct vm_map_entry
),
188 map_entry_init
, MAX_MAPENT
);
192 * vm_init2 - called prior to any vmspace allocations
197 zinitna(mapentzone
, &mapentobj
, NULL
, 0, 0,
198 ZONE_USE_RESERVE
| ZONE_SPECIAL
, 1);
199 zinitna(mapzone
, &mapobj
, NULL
, 0, 0, 0, 1);
206 * Red black tree functions
208 static int rb_vm_map_compare(vm_map_entry_t a
, vm_map_entry_t b
);
209 RB_GENERATE(vm_map_rb_tree
, vm_map_entry
, rb_entry
, rb_vm_map_compare
);
211 /* a->start is address, and the only field has to be initialized */
213 rb_vm_map_compare(vm_map_entry_t a
, vm_map_entry_t b
)
215 if (a
->start
< b
->start
)
217 else if (a
->start
> b
->start
)
223 * Allocate a vmspace structure, including a vm_map and pmap.
224 * Initialize numerous fields. While the initial allocation is zerod,
225 * subsequence reuse from the objcache leaves elements of the structure
226 * intact (particularly the pmap), so portions must be zerod.
228 * The structure is not considered activated until we call sysref_activate().
231 vmspace_alloc(vm_offset_t min
, vm_offset_t max
)
235 vm
= sysref_alloc(&vmspace_sysref_class
);
236 bzero(&vm
->vm_startcopy
,
237 (char *)&vm
->vm_endcopy
- (char *)&vm
->vm_startcopy
);
238 vm_map_init(&vm
->vm_map
, min
, max
, NULL
);
239 pmap_pinit(vmspace_pmap(vm
)); /* (some fields reused) */
240 vm
->vm_map
.pmap
= vmspace_pmap(vm
); /* XXX */
242 vm
->vm_exitingcnt
= 0;
243 cpu_vmspace_alloc(vm
);
244 sysref_activate(&vm
->vm_sysref
);
249 * dtor function - Some elements of the pmap are retained in the
250 * free-cached vmspaces to improve performance. We have to clean them up
251 * here before returning the vmspace to the memory pool.
254 vmspace_dtor(void *obj
, void *private)
256 struct vmspace
*vm
= obj
;
258 pmap_puninit(vmspace_pmap(vm
));
262 * Called in two cases:
264 * (1) When the last sysref is dropped, but exitingcnt might still be
267 * (2) When there are no sysrefs (i.e. refcnt is negative) left and the
268 * exitingcnt becomes zero
270 * sysref will not scrap the object until we call sysref_put() once more
271 * after the last ref has been dropped.
274 vmspace_terminate(struct vmspace
*vm
)
279 * If exitingcnt is non-zero we can't get rid of the entire vmspace
280 * yet, but we can scrap user memory.
282 if (vm
->vm_exitingcnt
) {
284 pmap_remove_pages(vmspace_pmap(vm
), VM_MIN_USER_ADDRESS
,
285 VM_MAX_USER_ADDRESS
);
286 vm_map_remove(&vm
->vm_map
, VM_MIN_USER_ADDRESS
,
287 VM_MAX_USER_ADDRESS
);
291 cpu_vmspace_free(vm
);
294 * Make sure any SysV shm is freed, it might not have in
299 KKASSERT(vm
->vm_upcalls
== NULL
);
302 * Lock the map, to wait out all other references to it.
303 * Delete all of the mappings and pages they hold, then call
304 * the pmap module to reclaim anything left.
306 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
307 vm_map_lock(&vm
->vm_map
);
308 vm_map_delete(&vm
->vm_map
, vm
->vm_map
.min_offset
,
309 vm
->vm_map
.max_offset
, &count
);
310 vm_map_unlock(&vm
->vm_map
);
311 vm_map_entry_release(count
);
313 pmap_release(vmspace_pmap(vm
));
314 sysref_put(&vm
->vm_sysref
);
318 * This is called in the wait*() handling code. The vmspace can be terminated
319 * after the last wait is finished using it.
322 vmspace_exitfree(struct proc
*p
)
329 if (--vm
->vm_exitingcnt
== 0 && sysref_isinactive(&vm
->vm_sysref
))
330 vmspace_terminate(vm
);
334 * vmspace_swap_count()
336 * Swap useage is determined by taking the proportional swap used by
337 * VM objects backing the VM map. To make up for fractional losses,
338 * if the VM object has any swap use at all the associated map entries
339 * count for at least 1 swap page.
342 vmspace_swap_count(struct vmspace
*vmspace
)
344 vm_map_t map
= &vmspace
->vm_map
;
350 for (cur
= map
->header
.next
; cur
!= &map
->header
; cur
= cur
->next
) {
351 switch(cur
->maptype
) {
352 case VM_MAPTYPE_NORMAL
:
353 case VM_MAPTYPE_VPAGETABLE
:
354 if ((object
= cur
->object
.vm_object
) == NULL
)
356 if (object
->type
!= OBJT_SWAP
)
358 n
= (cur
->end
- cur
->start
) / PAGE_SIZE
;
359 if (object
->un_pager
.swp
.swp_bcount
) {
360 count
+= object
->un_pager
.swp
.swp_bcount
*
361 SWAP_META_PAGES
* n
/ object
->size
+ 1;
372 * vmspace_anonymous_count()
374 * Calculate the approximate number of anonymous pages in use by
375 * this vmspace. To make up for fractional losses, we count each
376 * VM object as having at least 1 anonymous page.
379 vmspace_anonymous_count(struct vmspace
*vmspace
)
381 vm_map_t map
= &vmspace
->vm_map
;
386 for (cur
= map
->header
.next
; cur
!= &map
->header
; cur
= cur
->next
) {
387 switch(cur
->maptype
) {
388 case VM_MAPTYPE_NORMAL
:
389 case VM_MAPTYPE_VPAGETABLE
:
390 if ((object
= cur
->object
.vm_object
) == NULL
)
392 if (object
->type
!= OBJT_DEFAULT
&&
393 object
->type
!= OBJT_SWAP
) {
396 count
+= object
->resident_page_count
;
411 * Creates and returns a new empty VM map with
412 * the given physical map structure, and having
413 * the given lower and upper address bounds.
416 vm_map_create(vm_map_t result
, pmap_t pmap
, vm_offset_t min
, vm_offset_t max
)
419 result
= zalloc(mapzone
);
420 vm_map_init(result
, min
, max
, pmap
);
425 * Initialize an existing vm_map structure
426 * such as that in the vmspace structure.
427 * The pmap is set elsewhere.
430 vm_map_init(struct vm_map
*map
, vm_offset_t min
, vm_offset_t max
, pmap_t pmap
)
432 map
->header
.next
= map
->header
.prev
= &map
->header
;
433 RB_INIT(&map
->rb_root
);
438 map
->min_offset
= min
;
439 map
->max_offset
= max
;
441 map
->first_free
= &map
->header
;
442 map
->hint
= &map
->header
;
444 lockinit(&map
->lock
, "thrd_sleep", 0, 0);
448 * Shadow the vm_map_entry's object. This typically needs to be done when
449 * a write fault is taken on an entry which had previously been cloned by
450 * fork(). The shared object (which might be NULL) must become private so
451 * we add a shadow layer above it.
453 * Object allocation for anonymous mappings is defered as long as possible.
454 * When creating a shadow, however, the underlying object must be instantiated
455 * so it can be shared.
457 * If the map segment is governed by a virtual page table then it is
458 * possible to address offsets beyond the mapped area. Just allocate
459 * a maximally sized object for this case.
463 vm_map_entry_shadow(vm_map_entry_t entry
)
465 if (entry
->maptype
== VM_MAPTYPE_VPAGETABLE
) {
466 vm_object_shadow(&entry
->object
.vm_object
, &entry
->offset
,
467 0x7FFFFFFF); /* XXX */
469 vm_object_shadow(&entry
->object
.vm_object
, &entry
->offset
,
470 atop(entry
->end
- entry
->start
));
472 entry
->eflags
&= ~MAP_ENTRY_NEEDS_COPY
;
476 * Allocate an object for a vm_map_entry.
478 * Object allocation for anonymous mappings is defered as long as possible.
479 * This function is called when we can defer no longer, generally when a map
480 * entry might be split or forked or takes a page fault.
482 * If the map segment is governed by a virtual page table then it is
483 * possible to address offsets beyond the mapped area. Just allocate
484 * a maximally sized object for this case.
487 vm_map_entry_allocate_object(vm_map_entry_t entry
)
491 if (entry
->maptype
== VM_MAPTYPE_VPAGETABLE
) {
492 obj
= vm_object_allocate(OBJT_DEFAULT
, 0x7FFFFFFF); /* XXX */
494 obj
= vm_object_allocate(OBJT_DEFAULT
,
495 atop(entry
->end
- entry
->start
));
497 entry
->object
.vm_object
= obj
;
502 * vm_map_entry_reserve_cpu_init:
504 * Set an initial negative count so the first attempt to reserve
505 * space preloads a bunch of vm_map_entry's for this cpu. Also
506 * pre-allocate 2 vm_map_entries which will be needed by zalloc() to
507 * map a new page for vm_map_entry structures. SMP systems are
508 * particularly sensitive.
510 * This routine is called in early boot so we cannot just call
511 * vm_map_entry_reserve().
513 * May be called for a gd other then mycpu, but may only be called
517 vm_map_entry_reserve_cpu_init(globaldata_t gd
)
519 vm_map_entry_t entry
;
522 gd
->gd_vme_avail
-= MAP_RESERVE_COUNT
* 2;
523 entry
= &cpu_map_entry_init
[gd
->gd_cpuid
][0];
524 for (i
= 0; i
< VMEPERCPU
; ++i
, ++entry
) {
525 entry
->next
= gd
->gd_vme_base
;
526 gd
->gd_vme_base
= entry
;
531 * vm_map_entry_reserve:
533 * Reserves vm_map_entry structures so code later on can manipulate
534 * map_entry structures within a locked map without blocking trying
535 * to allocate a new vm_map_entry.
538 vm_map_entry_reserve(int count
)
540 struct globaldata
*gd
= mycpu
;
541 vm_map_entry_t entry
;
546 * Make sure we have enough structures in gd_vme_base to handle
547 * the reservation request.
549 while (gd
->gd_vme_avail
< count
) {
550 entry
= zalloc(mapentzone
);
551 entry
->next
= gd
->gd_vme_base
;
552 gd
->gd_vme_base
= entry
;
555 gd
->gd_vme_avail
-= count
;
561 * vm_map_entry_release:
563 * Releases previously reserved vm_map_entry structures that were not
564 * used. If we have too much junk in our per-cpu cache clean some of
568 vm_map_entry_release(int count
)
570 struct globaldata
*gd
= mycpu
;
571 vm_map_entry_t entry
;
574 gd
->gd_vme_avail
+= count
;
575 while (gd
->gd_vme_avail
> MAP_RESERVE_SLOP
) {
576 entry
= gd
->gd_vme_base
;
577 KKASSERT(entry
!= NULL
);
578 gd
->gd_vme_base
= entry
->next
;
581 zfree(mapentzone
, entry
);
588 * vm_map_entry_kreserve:
590 * Reserve map entry structures for use in kernel_map itself. These
591 * entries have *ALREADY* been reserved on a per-cpu basis when the map
592 * was inited. This function is used by zalloc() to avoid a recursion
593 * when zalloc() itself needs to allocate additional kernel memory.
595 * This function works like the normal reserve but does not load the
596 * vm_map_entry cache (because that would result in an infinite
597 * recursion). Note that gd_vme_avail may go negative. This is expected.
599 * Any caller of this function must be sure to renormalize after
600 * potentially eating entries to ensure that the reserve supply
604 vm_map_entry_kreserve(int count
)
606 struct globaldata
*gd
= mycpu
;
609 gd
->gd_vme_avail
-= count
;
611 KASSERT(gd
->gd_vme_base
!= NULL
, ("no reserved entries left, gd_vme_avail = %d\n", gd
->gd_vme_avail
));
616 * vm_map_entry_krelease:
618 * Release previously reserved map entries for kernel_map. We do not
619 * attempt to clean up like the normal release function as this would
620 * cause an unnecessary (but probably not fatal) deep procedure call.
623 vm_map_entry_krelease(int count
)
625 struct globaldata
*gd
= mycpu
;
628 gd
->gd_vme_avail
+= count
;
633 * vm_map_entry_create: [ internal use only ]
635 * Allocates a VM map entry for insertion. No entry fields are filled
638 * This routine may be called from an interrupt thread but not a FAST
639 * interrupt. This routine may recurse the map lock.
641 static vm_map_entry_t
642 vm_map_entry_create(vm_map_t map
, int *countp
)
644 struct globaldata
*gd
= mycpu
;
645 vm_map_entry_t entry
;
647 KKASSERT(*countp
> 0);
650 entry
= gd
->gd_vme_base
;
651 KASSERT(entry
!= NULL
, ("gd_vme_base NULL! count %d", *countp
));
652 gd
->gd_vme_base
= entry
->next
;
658 * vm_map_entry_dispose: [ internal use only ]
660 * Dispose of a vm_map_entry that is no longer being referenced. This
661 * function may be called from an interrupt.
664 vm_map_entry_dispose(vm_map_t map
, vm_map_entry_t entry
, int *countp
)
666 struct globaldata
*gd
= mycpu
;
668 KKASSERT(map
->hint
!= entry
);
669 KKASSERT(map
->first_free
!= entry
);
673 entry
->next
= gd
->gd_vme_base
;
674 gd
->gd_vme_base
= entry
;
680 * vm_map_entry_{un,}link:
682 * Insert/remove entries from maps.
685 vm_map_entry_link(vm_map_t map
,
686 vm_map_entry_t after_where
,
687 vm_map_entry_t entry
)
690 entry
->prev
= after_where
;
691 entry
->next
= after_where
->next
;
692 entry
->next
->prev
= entry
;
693 after_where
->next
= entry
;
694 if (vm_map_rb_tree_RB_INSERT(&map
->rb_root
, entry
))
695 panic("vm_map_entry_link: dup addr map %p ent %p", map
, entry
);
699 vm_map_entry_unlink(vm_map_t map
,
700 vm_map_entry_t entry
)
705 if (entry
->eflags
& MAP_ENTRY_IN_TRANSITION
)
706 panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry
);
711 vm_map_rb_tree_RB_REMOVE(&map
->rb_root
, entry
);
716 * vm_map_lookup_entry: [ internal use only ]
718 * Finds the map entry containing (or
719 * immediately preceding) the specified address
720 * in the given map; the entry is returned
721 * in the "entry" parameter. The boolean
722 * result indicates whether the address is
723 * actually contained in the map.
726 vm_map_lookup_entry(vm_map_t map
, vm_offset_t address
,
727 vm_map_entry_t
*entry
/* OUT */)
734 * XXX TEMPORARILY DISABLED. For some reason our attempt to revive
735 * the hint code with the red-black lookup meets with system crashes
736 * and lockups. We do not yet know why.
738 * It is possible that the problem is related to the setting
739 * of the hint during map_entry deletion, in the code specified
740 * at the GGG comment later on in this file.
743 * Quickly check the cached hint, there's a good chance of a match.
745 if (map
->hint
!= &map
->header
) {
747 if (address
>= tmp
->start
&& address
< tmp
->end
) {
755 * Locate the record from the top of the tree. 'last' tracks the
756 * closest prior record and is returned if no match is found, which
757 * in binary tree terms means tracking the most recent right-branch
758 * taken. If there is no prior record, &map->header is returned.
761 tmp
= RB_ROOT(&map
->rb_root
);
764 if (address
>= tmp
->start
) {
765 if (address
< tmp
->end
) {
771 tmp
= RB_RIGHT(tmp
, rb_entry
);
773 tmp
= RB_LEFT(tmp
, rb_entry
);
783 * Inserts the given whole VM object into the target
784 * map at the specified address range. The object's
785 * size should match that of the address range.
787 * Requires that the map be locked, and leaves it so. Requires that
788 * sufficient vm_map_entry structures have been reserved and tracks
789 * the use via countp.
791 * If object is non-NULL, ref count must be bumped by caller
792 * prior to making call to account for the new entry.
795 vm_map_insert(vm_map_t map
, int *countp
,
796 vm_object_t object
, vm_ooffset_t offset
,
797 vm_offset_t start
, vm_offset_t end
,
798 vm_maptype_t maptype
,
799 vm_prot_t prot
, vm_prot_t max
,
802 vm_map_entry_t new_entry
;
803 vm_map_entry_t prev_entry
;
804 vm_map_entry_t temp_entry
;
805 vm_eflags_t protoeflags
;
808 * Check that the start and end points are not bogus.
811 if ((start
< map
->min_offset
) || (end
> map
->max_offset
) ||
813 return (KERN_INVALID_ADDRESS
);
816 * Find the entry prior to the proposed starting address; if it's part
817 * of an existing entry, this range is bogus.
820 if (vm_map_lookup_entry(map
, start
, &temp_entry
))
821 return (KERN_NO_SPACE
);
823 prev_entry
= temp_entry
;
826 * Assert that the next entry doesn't overlap the end point.
829 if ((prev_entry
->next
!= &map
->header
) &&
830 (prev_entry
->next
->start
< end
))
831 return (KERN_NO_SPACE
);
835 if (cow
& MAP_COPY_ON_WRITE
)
836 protoeflags
|= MAP_ENTRY_COW
|MAP_ENTRY_NEEDS_COPY
;
838 if (cow
& MAP_NOFAULT
) {
839 protoeflags
|= MAP_ENTRY_NOFAULT
;
841 KASSERT(object
== NULL
,
842 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
844 if (cow
& MAP_DISABLE_SYNCER
)
845 protoeflags
|= MAP_ENTRY_NOSYNC
;
846 if (cow
& MAP_DISABLE_COREDUMP
)
847 protoeflags
|= MAP_ENTRY_NOCOREDUMP
;
848 if (cow
& MAP_IS_STACK
)
849 protoeflags
|= MAP_ENTRY_STACK
;
853 * When object is non-NULL, it could be shared with another
854 * process. We have to set or clear OBJ_ONEMAPPING
857 if ((object
->ref_count
> 1) || (object
->shadow_count
!= 0)) {
858 vm_object_clear_flag(object
, OBJ_ONEMAPPING
);
861 else if ((prev_entry
!= &map
->header
) &&
862 (prev_entry
->eflags
== protoeflags
) &&
863 (prev_entry
->end
== start
) &&
864 (prev_entry
->wired_count
== 0) &&
865 prev_entry
->maptype
== maptype
&&
866 ((prev_entry
->object
.vm_object
== NULL
) ||
867 vm_object_coalesce(prev_entry
->object
.vm_object
,
868 OFF_TO_IDX(prev_entry
->offset
),
869 (vm_size_t
)(prev_entry
->end
- prev_entry
->start
),
870 (vm_size_t
)(end
- prev_entry
->end
)))) {
872 * We were able to extend the object. Determine if we
873 * can extend the previous map entry to include the
876 if ((prev_entry
->inheritance
== VM_INHERIT_DEFAULT
) &&
877 (prev_entry
->protection
== prot
) &&
878 (prev_entry
->max_protection
== max
)) {
879 map
->size
+= (end
- prev_entry
->end
);
880 prev_entry
->end
= end
;
881 vm_map_simplify_entry(map
, prev_entry
, countp
);
882 return (KERN_SUCCESS
);
886 * If we can extend the object but cannot extend the
887 * map entry, we have to create a new map entry. We
888 * must bump the ref count on the extended object to
889 * account for it. object may be NULL.
891 object
= prev_entry
->object
.vm_object
;
892 offset
= prev_entry
->offset
+
893 (prev_entry
->end
- prev_entry
->start
);
894 vm_object_reference(object
);
898 * NOTE: if conditionals fail, object can be NULL here. This occurs
899 * in things like the buffer map where we manage kva but do not manage
907 new_entry
= vm_map_entry_create(map
, countp
);
908 new_entry
->start
= start
;
909 new_entry
->end
= end
;
911 new_entry
->maptype
= maptype
;
912 new_entry
->eflags
= protoeflags
;
913 new_entry
->object
.vm_object
= object
;
914 new_entry
->offset
= offset
;
915 new_entry
->aux
.master_pde
= 0;
917 new_entry
->inheritance
= VM_INHERIT_DEFAULT
;
918 new_entry
->protection
= prot
;
919 new_entry
->max_protection
= max
;
920 new_entry
->wired_count
= 0;
923 * Insert the new entry into the list
926 vm_map_entry_link(map
, prev_entry
, new_entry
);
927 map
->size
+= new_entry
->end
- new_entry
->start
;
930 * Update the free space hint
932 if ((map
->first_free
== prev_entry
) &&
933 (prev_entry
->end
>= new_entry
->start
)) {
934 map
->first_free
= new_entry
;
939 * Temporarily removed to avoid MAP_STACK panic, due to
940 * MAP_STACK being a huge hack. Will be added back in
941 * when MAP_STACK (and the user stack mapping) is fixed.
944 * It may be possible to simplify the entry
946 vm_map_simplify_entry(map
, new_entry
, countp
);
950 * Try to pre-populate the page table. Mappings governed by virtual
951 * page tables cannot be prepopulated without a lot of work, so
954 if ((cow
& (MAP_PREFAULT
|MAP_PREFAULT_PARTIAL
)) &&
955 maptype
!= VM_MAPTYPE_VPAGETABLE
) {
956 pmap_object_init_pt(map
->pmap
, start
, prot
,
957 object
, OFF_TO_IDX(offset
), end
- start
,
958 cow
& MAP_PREFAULT_PARTIAL
);
961 return (KERN_SUCCESS
);
965 * Find sufficient space for `length' bytes in the given map, starting at
966 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
968 * This function will returned an arbitrarily aligned pointer. If no
969 * particular alignment is required you should pass align as 1. Note that
970 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
971 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
974 * 'align' should be a power of 2 but is not required to be.
977 vm_map_findspace(vm_map_t map
, vm_offset_t start
, vm_size_t length
,
978 vm_offset_t align
, int flags
, vm_offset_t
*addr
)
980 vm_map_entry_t entry
, next
;
982 vm_offset_t align_mask
;
984 if (start
< map
->min_offset
)
985 start
= map
->min_offset
;
986 if (start
> map
->max_offset
)
990 * If the alignment is not a power of 2 we will have to use
991 * a mod/division, set align_mask to a special value.
993 if ((align
| (align
- 1)) + 1 != (align
<< 1))
994 align_mask
= (vm_offset_t
)-1;
996 align_mask
= align
- 1;
1000 * Look for the first possible address; if there's already something
1001 * at this address, we have to start after it.
1003 if (start
== map
->min_offset
) {
1004 if ((entry
= map
->first_free
) != &map
->header
)
1009 if (vm_map_lookup_entry(map
, start
, &tmp
))
1015 * Look through the rest of the map, trying to fit a new region in the
1016 * gap between existing regions, or after the very last region.
1018 for (;; start
= (entry
= next
)->end
) {
1020 * Adjust the proposed start by the requested alignment,
1021 * be sure that we didn't wrap the address.
1023 if (align_mask
== (vm_offset_t
)-1)
1024 end
= ((start
+ align
- 1) / align
) * align
;
1026 end
= (start
+ align_mask
) & ~align_mask
;
1031 * Find the end of the proposed new region. Be sure we didn't
1032 * go beyond the end of the map, or wrap around the address.
1033 * Then check to see if this is the last entry or if the
1034 * proposed end fits in the gap between this and the next
1037 end
= start
+ length
;
1038 if (end
> map
->max_offset
|| end
< start
)
1043 * If the next entry's start address is beyond the desired
1044 * end address we may have found a good entry.
1046 * If the next entry is a stack mapping we do not map into
1047 * the stack's reserved space.
1049 * XXX continue to allow mapping into the stack's reserved
1050 * space if doing a MAP_STACK mapping inside a MAP_STACK
1051 * mapping, for backwards compatibility. But the caller
1052 * really should use MAP_STACK | MAP_TRYFIXED if they
1055 if (next
== &map
->header
)
1057 if (next
->start
>= end
) {
1058 if ((next
->eflags
& MAP_ENTRY_STACK
) == 0)
1060 if (flags
& MAP_STACK
)
1062 if (next
->start
- next
->aux
.avail_ssize
>= end
)
1067 if (map
== &kernel_map
) {
1069 if ((ksize
= round_page(start
+ length
)) > kernel_vm_end
) {
1070 pmap_growkernel(ksize
);
1079 * vm_map_find finds an unallocated region in the target address
1080 * map with the given length. The search is defined to be
1081 * first-fit from the specified address; the region found is
1082 * returned in the same parameter.
1084 * If object is non-NULL, ref count must be bumped by caller
1085 * prior to making call to account for the new entry.
1088 vm_map_find(vm_map_t map
, vm_object_t object
, vm_ooffset_t offset
,
1089 vm_offset_t
*addr
, vm_size_t length
,
1091 vm_maptype_t maptype
,
1092 vm_prot_t prot
, vm_prot_t max
,
1101 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
1104 if (vm_map_findspace(map
, start
, length
, 1, 0, addr
)) {
1106 vm_map_entry_release(count
);
1107 return (KERN_NO_SPACE
);
1111 result
= vm_map_insert(map
, &count
, object
, offset
,
1112 start
, start
+ length
,
1117 vm_map_entry_release(count
);
1123 * vm_map_simplify_entry:
1125 * Simplify the given map entry by merging with either neighbor. This
1126 * routine also has the ability to merge with both neighbors.
1128 * The map must be locked.
1130 * This routine guarentees that the passed entry remains valid (though
1131 * possibly extended). When merging, this routine may delete one or
1132 * both neighbors. No action is taken on entries which have their
1133 * in-transition flag set.
1136 vm_map_simplify_entry(vm_map_t map
, vm_map_entry_t entry
, int *countp
)
1138 vm_map_entry_t next
, prev
;
1139 vm_size_t prevsize
, esize
;
1141 if (entry
->eflags
& MAP_ENTRY_IN_TRANSITION
) {
1142 ++mycpu
->gd_cnt
.v_intrans_coll
;
1146 if (entry
->maptype
== VM_MAPTYPE_SUBMAP
)
1150 if (prev
!= &map
->header
) {
1151 prevsize
= prev
->end
- prev
->start
;
1152 if ( (prev
->end
== entry
->start
) &&
1153 (prev
->maptype
== entry
->maptype
) &&
1154 (prev
->object
.vm_object
== entry
->object
.vm_object
) &&
1155 (!prev
->object
.vm_object
||
1156 (prev
->offset
+ prevsize
== entry
->offset
)) &&
1157 (prev
->eflags
== entry
->eflags
) &&
1158 (prev
->protection
== entry
->protection
) &&
1159 (prev
->max_protection
== entry
->max_protection
) &&
1160 (prev
->inheritance
== entry
->inheritance
) &&
1161 (prev
->wired_count
== entry
->wired_count
)) {
1162 if (map
->first_free
== prev
)
1163 map
->first_free
= entry
;
1164 if (map
->hint
== prev
)
1166 vm_map_entry_unlink(map
, prev
);
1167 entry
->start
= prev
->start
;
1168 entry
->offset
= prev
->offset
;
1169 if (prev
->object
.vm_object
)
1170 vm_object_deallocate(prev
->object
.vm_object
);
1171 vm_map_entry_dispose(map
, prev
, countp
);
1176 if (next
!= &map
->header
) {
1177 esize
= entry
->end
- entry
->start
;
1178 if ((entry
->end
== next
->start
) &&
1179 (next
->maptype
== entry
->maptype
) &&
1180 (next
->object
.vm_object
== entry
->object
.vm_object
) &&
1181 (!entry
->object
.vm_object
||
1182 (entry
->offset
+ esize
== next
->offset
)) &&
1183 (next
->eflags
== entry
->eflags
) &&
1184 (next
->protection
== entry
->protection
) &&
1185 (next
->max_protection
== entry
->max_protection
) &&
1186 (next
->inheritance
== entry
->inheritance
) &&
1187 (next
->wired_count
== entry
->wired_count
)) {
1188 if (map
->first_free
== next
)
1189 map
->first_free
= entry
;
1190 if (map
->hint
== next
)
1192 vm_map_entry_unlink(map
, next
);
1193 entry
->end
= next
->end
;
1194 if (next
->object
.vm_object
)
1195 vm_object_deallocate(next
->object
.vm_object
);
1196 vm_map_entry_dispose(map
, next
, countp
);
1201 * vm_map_clip_start: [ internal use only ]
1203 * Asserts that the given entry begins at or after
1204 * the specified address; if necessary,
1205 * it splits the entry into two.
1207 #define vm_map_clip_start(map, entry, startaddr, countp) \
1209 if (startaddr > entry->start) \
1210 _vm_map_clip_start(map, entry, startaddr, countp); \
1214 * This routine is called only when it is known that
1215 * the entry must be split.
1218 _vm_map_clip_start(vm_map_t map
, vm_map_entry_t entry
, vm_offset_t start
, int *countp
)
1220 vm_map_entry_t new_entry
;
1223 * Split off the front portion -- note that we must insert the new
1224 * entry BEFORE this one, so that this entry has the specified
1228 vm_map_simplify_entry(map
, entry
, countp
);
1231 * If there is no object backing this entry, we might as well create
1232 * one now. If we defer it, an object can get created after the map
1233 * is clipped, and individual objects will be created for the split-up
1234 * map. This is a bit of a hack, but is also about the best place to
1235 * put this improvement.
1237 if (entry
->object
.vm_object
== NULL
&& !map
->system_map
) {
1238 vm_map_entry_allocate_object(entry
);
1241 new_entry
= vm_map_entry_create(map
, countp
);
1242 *new_entry
= *entry
;
1244 new_entry
->end
= start
;
1245 entry
->offset
+= (start
- entry
->start
);
1246 entry
->start
= start
;
1248 vm_map_entry_link(map
, entry
->prev
, new_entry
);
1250 switch(entry
->maptype
) {
1251 case VM_MAPTYPE_NORMAL
:
1252 case VM_MAPTYPE_VPAGETABLE
:
1253 vm_object_reference(new_entry
->object
.vm_object
);
1261 * vm_map_clip_end: [ internal use only ]
1263 * Asserts that the given entry ends at or before
1264 * the specified address; if necessary,
1265 * it splits the entry into two.
1268 #define vm_map_clip_end(map, entry, endaddr, countp) \
1270 if (endaddr < entry->end) \
1271 _vm_map_clip_end(map, entry, endaddr, countp); \
1275 * This routine is called only when it is known that
1276 * the entry must be split.
1279 _vm_map_clip_end(vm_map_t map
, vm_map_entry_t entry
, vm_offset_t end
, int *countp
)
1281 vm_map_entry_t new_entry
;
1284 * If there is no object backing this entry, we might as well create
1285 * one now. If we defer it, an object can get created after the map
1286 * is clipped, and individual objects will be created for the split-up
1287 * map. This is a bit of a hack, but is also about the best place to
1288 * put this improvement.
1291 if (entry
->object
.vm_object
== NULL
&& !map
->system_map
) {
1292 vm_map_entry_allocate_object(entry
);
1296 * Create a new entry and insert it AFTER the specified entry
1299 new_entry
= vm_map_entry_create(map
, countp
);
1300 *new_entry
= *entry
;
1302 new_entry
->start
= entry
->end
= end
;
1303 new_entry
->offset
+= (end
- entry
->start
);
1305 vm_map_entry_link(map
, entry
, new_entry
);
1307 switch(entry
->maptype
) {
1308 case VM_MAPTYPE_NORMAL
:
1309 case VM_MAPTYPE_VPAGETABLE
:
1310 vm_object_reference(new_entry
->object
.vm_object
);
1318 * VM_MAP_RANGE_CHECK: [ internal use only ]
1320 * Asserts that the starting and ending region
1321 * addresses fall within the valid range of the map.
1323 #define VM_MAP_RANGE_CHECK(map, start, end) \
1325 if (start < vm_map_min(map)) \
1326 start = vm_map_min(map); \
1327 if (end > vm_map_max(map)) \
1328 end = vm_map_max(map); \
1334 * vm_map_transition_wait: [ kernel use only ]
1336 * Used to block when an in-transition collison occurs. The map
1337 * is unlocked for the sleep and relocked before the return.
1341 vm_map_transition_wait(vm_map_t map
)
1344 tsleep(map
, 0, "vment", 0);
1352 * When we do blocking operations with the map lock held it is
1353 * possible that a clip might have occured on our in-transit entry,
1354 * requiring an adjustment to the entry in our loop. These macros
1355 * help the pageable and clip_range code deal with the case. The
1356 * conditional costs virtually nothing if no clipping has occured.
1359 #define CLIP_CHECK_BACK(entry, save_start) \
1361 while (entry->start != save_start) { \
1362 entry = entry->prev; \
1363 KASSERT(entry != &map->header, ("bad entry clip")); \
1367 #define CLIP_CHECK_FWD(entry, save_end) \
1369 while (entry->end != save_end) { \
1370 entry = entry->next; \
1371 KASSERT(entry != &map->header, ("bad entry clip")); \
1377 * vm_map_clip_range: [ kernel use only ]
1379 * Clip the specified range and return the base entry. The
1380 * range may cover several entries starting at the returned base
1381 * and the first and last entry in the covering sequence will be
1382 * properly clipped to the requested start and end address.
1384 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1387 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1388 * covered by the requested range.
1390 * The map must be exclusively locked on entry and will remain locked
1391 * on return. If no range exists or the range contains holes and you
1392 * specified that no holes were allowed, NULL will be returned. This
1393 * routine may temporarily unlock the map in order avoid a deadlock when
1398 vm_map_clip_range(vm_map_t map
, vm_offset_t start
, vm_offset_t end
,
1399 int *countp
, int flags
)
1401 vm_map_entry_t start_entry
;
1402 vm_map_entry_t entry
;
1405 * Locate the entry and effect initial clipping. The in-transition
1406 * case does not occur very often so do not try to optimize it.
1409 if (vm_map_lookup_entry(map
, start
, &start_entry
) == FALSE
)
1411 entry
= start_entry
;
1412 if (entry
->eflags
& MAP_ENTRY_IN_TRANSITION
) {
1413 entry
->eflags
|= MAP_ENTRY_NEEDS_WAKEUP
;
1414 ++mycpu
->gd_cnt
.v_intrans_coll
;
1415 ++mycpu
->gd_cnt
.v_intrans_wait
;
1416 vm_map_transition_wait(map
);
1418 * entry and/or start_entry may have been clipped while
1419 * we slept, or may have gone away entirely. We have
1420 * to restart from the lookup.
1425 * Since we hold an exclusive map lock we do not have to restart
1426 * after clipping, even though clipping may block in zalloc.
1428 vm_map_clip_start(map
, entry
, start
, countp
);
1429 vm_map_clip_end(map
, entry
, end
, countp
);
1430 entry
->eflags
|= MAP_ENTRY_IN_TRANSITION
;
1433 * Scan entries covered by the range. When working on the next
1434 * entry a restart need only re-loop on the current entry which
1435 * we have already locked, since 'next' may have changed. Also,
1436 * even though entry is safe, it may have been clipped so we
1437 * have to iterate forwards through the clip after sleeping.
1439 while (entry
->next
!= &map
->header
&& entry
->next
->start
< end
) {
1440 vm_map_entry_t next
= entry
->next
;
1442 if (flags
& MAP_CLIP_NO_HOLES
) {
1443 if (next
->start
> entry
->end
) {
1444 vm_map_unclip_range(map
, start_entry
,
1445 start
, entry
->end
, countp
, flags
);
1450 if (next
->eflags
& MAP_ENTRY_IN_TRANSITION
) {
1451 vm_offset_t save_end
= entry
->end
;
1452 next
->eflags
|= MAP_ENTRY_NEEDS_WAKEUP
;
1453 ++mycpu
->gd_cnt
.v_intrans_coll
;
1454 ++mycpu
->gd_cnt
.v_intrans_wait
;
1455 vm_map_transition_wait(map
);
1458 * clips might have occured while we blocked.
1460 CLIP_CHECK_FWD(entry
, save_end
);
1461 CLIP_CHECK_BACK(start_entry
, start
);
1465 * No restart necessary even though clip_end may block, we
1466 * are holding the map lock.
1468 vm_map_clip_end(map
, next
, end
, countp
);
1469 next
->eflags
|= MAP_ENTRY_IN_TRANSITION
;
1472 if (flags
& MAP_CLIP_NO_HOLES
) {
1473 if (entry
->end
!= end
) {
1474 vm_map_unclip_range(map
, start_entry
,
1475 start
, entry
->end
, countp
, flags
);
1479 return(start_entry
);
1483 * vm_map_unclip_range: [ kernel use only ]
1485 * Undo the effect of vm_map_clip_range(). You should pass the same
1486 * flags and the same range that you passed to vm_map_clip_range().
1487 * This code will clear the in-transition flag on the entries and
1488 * wake up anyone waiting. This code will also simplify the sequence
1489 * and attempt to merge it with entries before and after the sequence.
1491 * The map must be locked on entry and will remain locked on return.
1493 * Note that you should also pass the start_entry returned by
1494 * vm_map_clip_range(). However, if you block between the two calls
1495 * with the map unlocked please be aware that the start_entry may
1496 * have been clipped and you may need to scan it backwards to find
1497 * the entry corresponding with the original start address. You are
1498 * responsible for this, vm_map_unclip_range() expects the correct
1499 * start_entry to be passed to it and will KASSERT otherwise.
1503 vm_map_unclip_range(
1505 vm_map_entry_t start_entry
,
1511 vm_map_entry_t entry
;
1513 entry
= start_entry
;
1515 KASSERT(entry
->start
== start
, ("unclip_range: illegal base entry"));
1516 while (entry
!= &map
->header
&& entry
->start
< end
) {
1517 KASSERT(entry
->eflags
& MAP_ENTRY_IN_TRANSITION
, ("in-transition flag not set during unclip on: %p", entry
));
1518 KASSERT(entry
->end
<= end
, ("unclip_range: tail wasn't clipped"));
1519 entry
->eflags
&= ~MAP_ENTRY_IN_TRANSITION
;
1520 if (entry
->eflags
& MAP_ENTRY_NEEDS_WAKEUP
) {
1521 entry
->eflags
&= ~MAP_ENTRY_NEEDS_WAKEUP
;
1524 entry
= entry
->next
;
1528 * Simplification does not block so there is no restart case.
1530 entry
= start_entry
;
1531 while (entry
!= &map
->header
&& entry
->start
< end
) {
1532 vm_map_simplify_entry(map
, entry
, countp
);
1533 entry
= entry
->next
;
1538 * vm_map_submap: [ kernel use only ]
1540 * Mark the given range as handled by a subordinate map.
1542 * This range must have been created with vm_map_find,
1543 * and no other operations may have been performed on this
1544 * range prior to calling vm_map_submap.
1546 * Only a limited number of operations can be performed
1547 * within this rage after calling vm_map_submap:
1549 * [Don't try vm_map_copy!]
1551 * To remove a submapping, one must first remove the
1552 * range from the superior map, and then destroy the
1553 * submap (if desired). [Better yet, don't try it.]
1556 vm_map_submap(vm_map_t map
, vm_offset_t start
, vm_offset_t end
, vm_map_t submap
)
1558 vm_map_entry_t entry
;
1559 int result
= KERN_INVALID_ARGUMENT
;
1562 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
1565 VM_MAP_RANGE_CHECK(map
, start
, end
);
1567 if (vm_map_lookup_entry(map
, start
, &entry
)) {
1568 vm_map_clip_start(map
, entry
, start
, &count
);
1570 entry
= entry
->next
;
1573 vm_map_clip_end(map
, entry
, end
, &count
);
1575 if ((entry
->start
== start
) && (entry
->end
== end
) &&
1576 ((entry
->eflags
& MAP_ENTRY_COW
) == 0) &&
1577 (entry
->object
.vm_object
== NULL
)) {
1578 entry
->object
.sub_map
= submap
;
1579 entry
->maptype
= VM_MAPTYPE_SUBMAP
;
1580 result
= KERN_SUCCESS
;
1583 vm_map_entry_release(count
);
1591 * Sets the protection of the specified address region in the target map.
1592 * If "set_max" is specified, the maximum protection is to be set;
1593 * otherwise, only the current protection is affected.
1595 * The protection is not applicable to submaps, but is applicable to normal
1596 * maps and maps governed by virtual page tables. For example, when operating
1597 * on a virtual page table our protection basically controls how COW occurs
1598 * on the backing object, whereas the virtual page table abstraction itself
1599 * is an abstraction for userland.
1602 vm_map_protect(vm_map_t map
, vm_offset_t start
, vm_offset_t end
,
1603 vm_prot_t new_prot
, boolean_t set_max
)
1605 vm_map_entry_t current
;
1606 vm_map_entry_t entry
;
1609 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
1612 VM_MAP_RANGE_CHECK(map
, start
, end
);
1614 if (vm_map_lookup_entry(map
, start
, &entry
)) {
1615 vm_map_clip_start(map
, entry
, start
, &count
);
1617 entry
= entry
->next
;
1621 * Make a first pass to check for protection violations.
1624 while ((current
!= &map
->header
) && (current
->start
< end
)) {
1625 if (current
->maptype
== VM_MAPTYPE_SUBMAP
) {
1627 vm_map_entry_release(count
);
1628 return (KERN_INVALID_ARGUMENT
);
1630 if ((new_prot
& current
->max_protection
) != new_prot
) {
1632 vm_map_entry_release(count
);
1633 return (KERN_PROTECTION_FAILURE
);
1635 current
= current
->next
;
1639 * Go back and fix up protections. [Note that clipping is not
1640 * necessary the second time.]
1644 while ((current
!= &map
->header
) && (current
->start
< end
)) {
1647 vm_map_clip_end(map
, current
, end
, &count
);
1649 old_prot
= current
->protection
;
1651 current
->protection
=
1652 (current
->max_protection
= new_prot
) &
1655 current
->protection
= new_prot
;
1659 * Update physical map if necessary. Worry about copy-on-write
1660 * here -- CHECK THIS XXX
1663 if (current
->protection
!= old_prot
) {
1664 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1667 pmap_protect(map
->pmap
, current
->start
,
1669 current
->protection
& MASK(current
));
1673 vm_map_simplify_entry(map
, current
, &count
);
1675 current
= current
->next
;
1679 vm_map_entry_release(count
);
1680 return (KERN_SUCCESS
);
1686 * This routine traverses a processes map handling the madvise
1687 * system call. Advisories are classified as either those effecting
1688 * the vm_map_entry structure, or those effecting the underlying
1691 * The <value> argument is used for extended madvise calls.
1694 vm_map_madvise(vm_map_t map
, vm_offset_t start
, vm_offset_t end
,
1695 int behav
, off_t value
)
1697 vm_map_entry_t current
, entry
;
1703 * Some madvise calls directly modify the vm_map_entry, in which case
1704 * we need to use an exclusive lock on the map and we need to perform
1705 * various clipping operations. Otherwise we only need a read-lock
1709 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
1713 case MADV_SEQUENTIAL
:
1727 vm_map_lock_read(map
);
1730 vm_map_entry_release(count
);
1735 * Locate starting entry and clip if necessary.
1738 VM_MAP_RANGE_CHECK(map
, start
, end
);
1740 if (vm_map_lookup_entry(map
, start
, &entry
)) {
1742 vm_map_clip_start(map
, entry
, start
, &count
);
1744 entry
= entry
->next
;
1749 * madvise behaviors that are implemented in the vm_map_entry.
1751 * We clip the vm_map_entry so that behavioral changes are
1752 * limited to the specified address range.
1754 for (current
= entry
;
1755 (current
!= &map
->header
) && (current
->start
< end
);
1756 current
= current
->next
1758 if (current
->maptype
== VM_MAPTYPE_SUBMAP
)
1761 vm_map_clip_end(map
, current
, end
, &count
);
1765 vm_map_entry_set_behavior(current
, MAP_ENTRY_BEHAV_NORMAL
);
1767 case MADV_SEQUENTIAL
:
1768 vm_map_entry_set_behavior(current
, MAP_ENTRY_BEHAV_SEQUENTIAL
);
1771 vm_map_entry_set_behavior(current
, MAP_ENTRY_BEHAV_RANDOM
);
1774 current
->eflags
|= MAP_ENTRY_NOSYNC
;
1777 current
->eflags
&= ~MAP_ENTRY_NOSYNC
;
1780 current
->eflags
|= MAP_ENTRY_NOCOREDUMP
;
1783 current
->eflags
&= ~MAP_ENTRY_NOCOREDUMP
;
1787 * Invalidate the related pmap entries, used
1788 * to flush portions of the real kernel's
1789 * pmap when the caller has removed or
1790 * modified existing mappings in a virtual
1793 pmap_remove(map
->pmap
,
1794 current
->start
, current
->end
);
1798 * Set the page directory page for a map
1799 * governed by a virtual page table. Mark
1800 * the entry as being governed by a virtual
1801 * page table if it is not.
1803 * XXX the page directory page is stored
1804 * in the avail_ssize field if the map_entry.
1806 * XXX the map simplification code does not
1807 * compare this field so weird things may
1808 * happen if you do not apply this function
1809 * to the entire mapping governed by the
1810 * virtual page table.
1812 if (current
->maptype
!= VM_MAPTYPE_VPAGETABLE
) {
1816 current
->aux
.master_pde
= value
;
1817 pmap_remove(map
->pmap
,
1818 current
->start
, current
->end
);
1824 vm_map_simplify_entry(map
, current
, &count
);
1832 * madvise behaviors that are implemented in the underlying
1835 * Since we don't clip the vm_map_entry, we have to clip
1836 * the vm_object pindex and count.
1838 * NOTE! We currently do not support these functions on
1839 * virtual page tables.
1841 for (current
= entry
;
1842 (current
!= &map
->header
) && (current
->start
< end
);
1843 current
= current
->next
1845 vm_offset_t useStart
;
1847 if (current
->maptype
!= VM_MAPTYPE_NORMAL
)
1850 pindex
= OFF_TO_IDX(current
->offset
);
1851 count
= atop(current
->end
- current
->start
);
1852 useStart
= current
->start
;
1854 if (current
->start
< start
) {
1855 pindex
+= atop(start
- current
->start
);
1856 count
-= atop(start
- current
->start
);
1859 if (current
->end
> end
)
1860 count
-= atop(current
->end
- end
);
1865 vm_object_madvise(current
->object
.vm_object
,
1866 pindex
, count
, behav
);
1869 * Try to populate the page table. Mappings governed
1870 * by virtual page tables cannot be pre-populated
1871 * without a lot of work so don't try.
1873 if (behav
== MADV_WILLNEED
&&
1874 current
->maptype
!= VM_MAPTYPE_VPAGETABLE
) {
1875 pmap_object_init_pt(
1878 current
->protection
,
1879 current
->object
.vm_object
,
1881 (count
<< PAGE_SHIFT
),
1882 MAP_PREFAULT_MADVISE
1886 vm_map_unlock_read(map
);
1888 vm_map_entry_release(count
);
1896 * Sets the inheritance of the specified address
1897 * range in the target map. Inheritance
1898 * affects how the map will be shared with
1899 * child maps at the time of vm_map_fork.
1902 vm_map_inherit(vm_map_t map
, vm_offset_t start
, vm_offset_t end
,
1903 vm_inherit_t new_inheritance
)
1905 vm_map_entry_t entry
;
1906 vm_map_entry_t temp_entry
;
1909 switch (new_inheritance
) {
1910 case VM_INHERIT_NONE
:
1911 case VM_INHERIT_COPY
:
1912 case VM_INHERIT_SHARE
:
1915 return (KERN_INVALID_ARGUMENT
);
1918 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
1921 VM_MAP_RANGE_CHECK(map
, start
, end
);
1923 if (vm_map_lookup_entry(map
, start
, &temp_entry
)) {
1925 vm_map_clip_start(map
, entry
, start
, &count
);
1927 entry
= temp_entry
->next
;
1929 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
1930 vm_map_clip_end(map
, entry
, end
, &count
);
1932 entry
->inheritance
= new_inheritance
;
1934 vm_map_simplify_entry(map
, entry
, &count
);
1936 entry
= entry
->next
;
1939 vm_map_entry_release(count
);
1940 return (KERN_SUCCESS
);
1944 * Implement the semantics of mlock
1947 vm_map_unwire(vm_map_t map
, vm_offset_t start
, vm_offset_t real_end
,
1948 boolean_t new_pageable
)
1950 vm_map_entry_t entry
;
1951 vm_map_entry_t start_entry
;
1953 int rv
= KERN_SUCCESS
;
1956 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
1958 VM_MAP_RANGE_CHECK(map
, start
, real_end
);
1961 start_entry
= vm_map_clip_range(map
, start
, end
, &count
, MAP_CLIP_NO_HOLES
);
1962 if (start_entry
== NULL
) {
1964 vm_map_entry_release(count
);
1965 return (KERN_INVALID_ADDRESS
);
1968 if (new_pageable
== 0) {
1969 entry
= start_entry
;
1970 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
1971 vm_offset_t save_start
;
1972 vm_offset_t save_end
;
1975 * Already user wired or hard wired (trivial cases)
1977 if (entry
->eflags
& MAP_ENTRY_USER_WIRED
) {
1978 entry
= entry
->next
;
1981 if (entry
->wired_count
!= 0) {
1982 entry
->wired_count
++;
1983 entry
->eflags
|= MAP_ENTRY_USER_WIRED
;
1984 entry
= entry
->next
;
1989 * A new wiring requires instantiation of appropriate
1990 * management structures and the faulting in of the
1993 if (entry
->maptype
!= VM_MAPTYPE_SUBMAP
) {
1994 int copyflag
= entry
->eflags
& MAP_ENTRY_NEEDS_COPY
;
1995 if (copyflag
&& ((entry
->protection
& VM_PROT_WRITE
) != 0)) {
1996 vm_map_entry_shadow(entry
);
1997 } else if (entry
->object
.vm_object
== NULL
&&
1999 vm_map_entry_allocate_object(entry
);
2002 entry
->wired_count
++;
2003 entry
->eflags
|= MAP_ENTRY_USER_WIRED
;
2006 * Now fault in the area. Note that vm_fault_wire()
2007 * may release the map lock temporarily, it will be
2008 * relocked on return. The in-transition
2009 * flag protects the entries.
2011 save_start
= entry
->start
;
2012 save_end
= entry
->end
;
2013 rv
= vm_fault_wire(map
, entry
, TRUE
);
2015 CLIP_CHECK_BACK(entry
, save_start
);
2017 KASSERT(entry
->wired_count
== 1, ("bad wired_count on entry"));
2018 entry
->eflags
&= ~MAP_ENTRY_USER_WIRED
;
2019 entry
->wired_count
= 0;
2020 if (entry
->end
== save_end
)
2022 entry
= entry
->next
;
2023 KASSERT(entry
!= &map
->header
, ("bad entry clip during backout"));
2025 end
= save_start
; /* unwire the rest */
2029 * note that even though the entry might have been
2030 * clipped, the USER_WIRED flag we set prevents
2031 * duplication so we do not have to do a
2034 entry
= entry
->next
;
2038 * If we failed fall through to the unwiring section to
2039 * unwire what we had wired so far. 'end' has already
2046 * start_entry might have been clipped if we unlocked the
2047 * map and blocked. No matter how clipped it has gotten
2048 * there should be a fragment that is on our start boundary.
2050 CLIP_CHECK_BACK(start_entry
, start
);
2054 * Deal with the unwiring case.
2058 * This is the unwiring case. We must first ensure that the
2059 * range to be unwired is really wired down. We know there
2062 entry
= start_entry
;
2063 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
2064 if ((entry
->eflags
& MAP_ENTRY_USER_WIRED
) == 0) {
2065 rv
= KERN_INVALID_ARGUMENT
;
2068 KASSERT(entry
->wired_count
!= 0, ("wired count was 0 with USER_WIRED set! %p", entry
));
2069 entry
= entry
->next
;
2073 * Now decrement the wiring count for each region. If a region
2074 * becomes completely unwired, unwire its physical pages and
2078 * The map entries are processed in a loop, checking to
2079 * make sure the entry is wired and asserting it has a wired
2080 * count. However, another loop was inserted more-or-less in
2081 * the middle of the unwiring path. This loop picks up the
2082 * "entry" loop variable from the first loop without first
2083 * setting it to start_entry. Naturally, the secound loop
2084 * is never entered and the pages backing the entries are
2085 * never unwired. This can lead to a leak of wired pages.
2087 entry
= start_entry
;
2088 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
2089 KASSERT(entry
->eflags
& MAP_ENTRY_USER_WIRED
,
2090 ("expected USER_WIRED on entry %p", entry
));
2091 entry
->eflags
&= ~MAP_ENTRY_USER_WIRED
;
2092 entry
->wired_count
--;
2093 if (entry
->wired_count
== 0)
2094 vm_fault_unwire(map
, entry
);
2095 entry
= entry
->next
;
2099 vm_map_unclip_range(map
, start_entry
, start
, real_end
, &count
,
2103 vm_map_entry_release(count
);
2110 * Sets the pageability of the specified address
2111 * range in the target map. Regions specified
2112 * as not pageable require locked-down physical
2113 * memory and physical page maps.
2115 * The map must not be locked, but a reference
2116 * must remain to the map throughout the call.
2118 * This function may be called via the zalloc path and must properly
2119 * reserve map entries for kernel_map.
2122 vm_map_wire(vm_map_t map
, vm_offset_t start
, vm_offset_t real_end
, int kmflags
)
2124 vm_map_entry_t entry
;
2125 vm_map_entry_t start_entry
;
2127 int rv
= KERN_SUCCESS
;
2130 if (kmflags
& KM_KRESERVE
)
2131 count
= vm_map_entry_kreserve(MAP_RESERVE_COUNT
);
2133 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
2135 VM_MAP_RANGE_CHECK(map
, start
, real_end
);
2138 start_entry
= vm_map_clip_range(map
, start
, end
, &count
, MAP_CLIP_NO_HOLES
);
2139 if (start_entry
== NULL
) {
2141 rv
= KERN_INVALID_ADDRESS
;
2144 if ((kmflags
& KM_PAGEABLE
) == 0) {
2148 * 1. Holding the write lock, we create any shadow or zero-fill
2149 * objects that need to be created. Then we clip each map
2150 * entry to the region to be wired and increment its wiring
2151 * count. We create objects before clipping the map entries
2152 * to avoid object proliferation.
2154 * 2. We downgrade to a read lock, and call vm_fault_wire to
2155 * fault in the pages for any newly wired area (wired_count is
2158 * Downgrading to a read lock for vm_fault_wire avoids a
2159 * possible deadlock with another process that may have faulted
2160 * on one of the pages to be wired (it would mark the page busy,
2161 * blocking us, then in turn block on the map lock that we
2162 * hold). Because of problems in the recursive lock package,
2163 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
2164 * any actions that require the write lock must be done
2165 * beforehand. Because we keep the read lock on the map, the
2166 * copy-on-write status of the entries we modify here cannot
2170 entry
= start_entry
;
2171 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
2173 * Trivial case if the entry is already wired
2175 if (entry
->wired_count
) {
2176 entry
->wired_count
++;
2177 entry
= entry
->next
;
2182 * The entry is being newly wired, we have to setup
2183 * appropriate management structures. A shadow
2184 * object is required for a copy-on-write region,
2185 * or a normal object for a zero-fill region. We
2186 * do not have to do this for entries that point to sub
2187 * maps because we won't hold the lock on the sub map.
2189 if (entry
->maptype
!= VM_MAPTYPE_SUBMAP
) {
2190 int copyflag
= entry
->eflags
& MAP_ENTRY_NEEDS_COPY
;
2192 ((entry
->protection
& VM_PROT_WRITE
) != 0)) {
2193 vm_map_entry_shadow(entry
);
2194 } else if (entry
->object
.vm_object
== NULL
&&
2196 vm_map_entry_allocate_object(entry
);
2200 entry
->wired_count
++;
2201 entry
= entry
->next
;
2209 * HACK HACK HACK HACK
2211 * Unlock the map to avoid deadlocks. The in-transit flag
2212 * protects us from most changes but note that
2213 * clipping may still occur. To prevent clipping from
2214 * occuring after the unlock, except for when we are
2215 * blocking in vm_fault_wire, we must run in a critical
2216 * section, otherwise our accesses to entry->start and
2217 * entry->end could be corrupted. We have to enter the
2218 * critical section prior to unlocking so start_entry does
2219 * not change out from under us at the very beginning of the
2222 * HACK HACK HACK HACK
2227 entry
= start_entry
;
2228 while (entry
!= &map
->header
&& entry
->start
< end
) {
2230 * If vm_fault_wire fails for any page we need to undo
2231 * what has been done. We decrement the wiring count
2232 * for those pages which have not yet been wired (now)
2233 * and unwire those that have (later).
2235 vm_offset_t save_start
= entry
->start
;
2236 vm_offset_t save_end
= entry
->end
;
2238 if (entry
->wired_count
== 1)
2239 rv
= vm_fault_wire(map
, entry
, FALSE
);
2241 CLIP_CHECK_BACK(entry
, save_start
);
2243 KASSERT(entry
->wired_count
== 1, ("wired_count changed unexpectedly"));
2244 entry
->wired_count
= 0;
2245 if (entry
->end
== save_end
)
2247 entry
= entry
->next
;
2248 KASSERT(entry
!= &map
->header
, ("bad entry clip during backout"));
2253 CLIP_CHECK_FWD(entry
, save_end
);
2254 entry
= entry
->next
;
2259 * If a failure occured undo everything by falling through
2260 * to the unwiring code. 'end' has already been adjusted
2264 kmflags
|= KM_PAGEABLE
;
2267 * start_entry is still IN_TRANSITION but may have been
2268 * clipped since vm_fault_wire() unlocks and relocks the
2269 * map. No matter how clipped it has gotten there should
2270 * be a fragment that is on our start boundary.
2272 CLIP_CHECK_BACK(start_entry
, start
);
2275 if (kmflags
& KM_PAGEABLE
) {
2277 * This is the unwiring case. We must first ensure that the
2278 * range to be unwired is really wired down. We know there
2281 entry
= start_entry
;
2282 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
2283 if (entry
->wired_count
== 0) {
2284 rv
= KERN_INVALID_ARGUMENT
;
2287 entry
= entry
->next
;
2291 * Now decrement the wiring count for each region. If a region
2292 * becomes completely unwired, unwire its physical pages and
2295 entry
= start_entry
;
2296 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
2297 entry
->wired_count
--;
2298 if (entry
->wired_count
== 0)
2299 vm_fault_unwire(map
, entry
);
2300 entry
= entry
->next
;
2304 vm_map_unclip_range(map
, start_entry
, start
, real_end
, &count
,
2309 if (kmflags
& KM_KRESERVE
)
2310 vm_map_entry_krelease(count
);
2312 vm_map_entry_release(count
);
2317 * vm_map_set_wired_quick()
2319 * Mark a newly allocated address range as wired but do not fault in
2320 * the pages. The caller is expected to load the pages into the object.
2322 * The map must be locked on entry and will remain locked on return.
2325 vm_map_set_wired_quick(vm_map_t map
, vm_offset_t addr
, vm_size_t size
, int *countp
)
2327 vm_map_entry_t scan
;
2328 vm_map_entry_t entry
;
2330 entry
= vm_map_clip_range(map
, addr
, addr
+ size
, countp
, MAP_CLIP_NO_HOLES
);
2331 for (scan
= entry
; scan
!= &map
->header
&& scan
->start
< addr
+ size
; scan
= scan
->next
) {
2332 KKASSERT(entry
->wired_count
== 0);
2333 entry
->wired_count
= 1;
2335 vm_map_unclip_range(map
, entry
, addr
, addr
+ size
, countp
, MAP_CLIP_NO_HOLES
);
2341 * Push any dirty cached pages in the address range to their pager.
2342 * If syncio is TRUE, dirty pages are written synchronously.
2343 * If invalidate is TRUE, any cached pages are freed as well.
2345 * Returns an error if any part of the specified range is not mapped.
2348 vm_map_clean(vm_map_t map
, vm_offset_t start
, vm_offset_t end
, boolean_t syncio
,
2349 boolean_t invalidate
)
2351 vm_map_entry_t current
;
2352 vm_map_entry_t entry
;
2355 vm_ooffset_t offset
;
2357 vm_map_lock_read(map
);
2358 VM_MAP_RANGE_CHECK(map
, start
, end
);
2359 if (!vm_map_lookup_entry(map
, start
, &entry
)) {
2360 vm_map_unlock_read(map
);
2361 return (KERN_INVALID_ADDRESS
);
2364 * Make a first pass to check for holes.
2366 for (current
= entry
; current
->start
< end
; current
= current
->next
) {
2367 if (current
->maptype
== VM_MAPTYPE_SUBMAP
) {
2368 vm_map_unlock_read(map
);
2369 return (KERN_INVALID_ARGUMENT
);
2371 if (end
> current
->end
&&
2372 (current
->next
== &map
->header
||
2373 current
->end
!= current
->next
->start
)) {
2374 vm_map_unlock_read(map
);
2375 return (KERN_INVALID_ADDRESS
);
2380 pmap_remove(vm_map_pmap(map
), start
, end
);
2382 * Make a second pass, cleaning/uncaching pages from the indicated
2385 for (current
= entry
; current
->start
< end
; current
= current
->next
) {
2386 offset
= current
->offset
+ (start
- current
->start
);
2387 size
= (end
<= current
->end
? end
: current
->end
) - start
;
2388 if (current
->maptype
== VM_MAPTYPE_SUBMAP
) {
2390 vm_map_entry_t tentry
;
2393 smap
= current
->object
.sub_map
;
2394 vm_map_lock_read(smap
);
2395 vm_map_lookup_entry(smap
, offset
, &tentry
);
2396 tsize
= tentry
->end
- offset
;
2399 object
= tentry
->object
.vm_object
;
2400 offset
= tentry
->offset
+ (offset
- tentry
->start
);
2401 vm_map_unlock_read(smap
);
2403 object
= current
->object
.vm_object
;
2406 * Note that there is absolutely no sense in writing out
2407 * anonymous objects, so we track down the vnode object
2409 * We invalidate (remove) all pages from the address space
2410 * anyway, for semantic correctness.
2412 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2413 * may start out with a NULL object.
2415 while (object
&& object
->backing_object
) {
2416 offset
+= object
->backing_object_offset
;
2417 object
= object
->backing_object
;
2418 if (object
->size
< OFF_TO_IDX( offset
+ size
))
2419 size
= IDX_TO_OFF(object
->size
) - offset
;
2421 if (object
&& (object
->type
== OBJT_VNODE
) &&
2422 (current
->protection
& VM_PROT_WRITE
)) {
2424 * Flush pages if writing is allowed, invalidate them
2425 * if invalidation requested. Pages undergoing I/O
2426 * will be ignored by vm_object_page_remove().
2428 * We cannot lock the vnode and then wait for paging
2429 * to complete without deadlocking against vm_fault.
2430 * Instead we simply call vm_object_page_remove() and
2431 * allow it to block internally on a page-by-page
2432 * basis when it encounters pages undergoing async
2437 vm_object_reference(object
);
2438 vn_lock(object
->handle
, LK_EXCLUSIVE
| LK_RETRY
);
2439 flags
= (syncio
|| invalidate
) ? OBJPC_SYNC
: 0;
2440 flags
|= invalidate
? OBJPC_INVAL
: 0;
2443 * When operating on a virtual page table just
2444 * flush the whole object. XXX we probably ought
2447 switch(current
->maptype
) {
2448 case VM_MAPTYPE_NORMAL
:
2449 vm_object_page_clean(object
,
2451 OFF_TO_IDX(offset
+ size
+ PAGE_MASK
),
2454 case VM_MAPTYPE_VPAGETABLE
:
2455 vm_object_page_clean(object
, 0, 0, flags
);
2458 vn_unlock(((struct vnode
*)object
->handle
));
2459 vm_object_deallocate(object
);
2461 if (object
&& invalidate
&&
2462 ((object
->type
== OBJT_VNODE
) ||
2463 (object
->type
== OBJT_DEVICE
))) {
2465 (object
->type
== OBJT_DEVICE
) ? FALSE
: TRUE
;
2466 vm_object_reference(object
);
2467 switch(current
->maptype
) {
2468 case VM_MAPTYPE_NORMAL
:
2469 vm_object_page_remove(object
,
2471 OFF_TO_IDX(offset
+ size
+ PAGE_MASK
),
2474 case VM_MAPTYPE_VPAGETABLE
:
2475 vm_object_page_remove(object
, 0, 0, clean_only
);
2478 vm_object_deallocate(object
);
2483 vm_map_unlock_read(map
);
2484 return (KERN_SUCCESS
);
2488 * vm_map_entry_unwire: [ internal use only ]
2490 * Make the region specified by this entry pageable.
2492 * The map in question should be locked.
2493 * [This is the reason for this routine's existence.]
2496 vm_map_entry_unwire(vm_map_t map
, vm_map_entry_t entry
)
2498 entry
->eflags
&= ~MAP_ENTRY_USER_WIRED
;
2499 entry
->wired_count
= 0;
2500 vm_fault_unwire(map
, entry
);
2504 * vm_map_entry_delete: [ internal use only ]
2506 * Deallocate the given entry from the target map.
2509 vm_map_entry_delete(vm_map_t map
, vm_map_entry_t entry
, int *countp
)
2511 vm_map_entry_unlink(map
, entry
);
2512 map
->size
-= entry
->end
- entry
->start
;
2514 switch(entry
->maptype
) {
2515 case VM_MAPTYPE_NORMAL
:
2516 case VM_MAPTYPE_VPAGETABLE
:
2517 vm_object_deallocate(entry
->object
.vm_object
);
2523 vm_map_entry_dispose(map
, entry
, countp
);
2527 * vm_map_delete: [ internal use only ]
2529 * Deallocates the given address range from the target
2533 vm_map_delete(vm_map_t map
, vm_offset_t start
, vm_offset_t end
, int *countp
)
2536 vm_map_entry_t entry
;
2537 vm_map_entry_t first_entry
;
2541 * Find the start of the region, and clip it. Set entry to point
2542 * at the first record containing the requested address or, if no
2543 * such record exists, the next record with a greater address. The
2544 * loop will run from this point until a record beyond the termination
2545 * address is encountered.
2547 * map->hint must be adjusted to not point to anything we delete,
2548 * so set it to the entry prior to the one being deleted.
2550 * GGG see other GGG comment.
2552 if (vm_map_lookup_entry(map
, start
, &first_entry
)) {
2553 entry
= first_entry
;
2554 vm_map_clip_start(map
, entry
, start
, countp
);
2555 map
->hint
= entry
->prev
; /* possible problem XXX */
2557 map
->hint
= first_entry
; /* possible problem XXX */
2558 entry
= first_entry
->next
;
2562 * If a hole opens up prior to the current first_free then
2563 * adjust first_free. As with map->hint, map->first_free
2564 * cannot be left set to anything we might delete.
2566 if (entry
== &map
->header
) {
2567 map
->first_free
= &map
->header
;
2568 } else if (map
->first_free
->start
>= start
) {
2569 map
->first_free
= entry
->prev
;
2573 * Step through all entries in this region
2576 while ((entry
!= &map
->header
) && (entry
->start
< end
)) {
2577 vm_map_entry_t next
;
2579 vm_pindex_t offidxstart
, offidxend
, count
;
2582 * If we hit an in-transition entry we have to sleep and
2583 * retry. It's easier (and not really slower) to just retry
2584 * since this case occurs so rarely and the hint is already
2585 * pointing at the right place. We have to reset the
2586 * start offset so as not to accidently delete an entry
2587 * another process just created in vacated space.
2589 if (entry
->eflags
& MAP_ENTRY_IN_TRANSITION
) {
2590 entry
->eflags
|= MAP_ENTRY_NEEDS_WAKEUP
;
2591 start
= entry
->start
;
2592 ++mycpu
->gd_cnt
.v_intrans_coll
;
2593 ++mycpu
->gd_cnt
.v_intrans_wait
;
2594 vm_map_transition_wait(map
);
2597 vm_map_clip_end(map
, entry
, end
, countp
);
2603 offidxstart
= OFF_TO_IDX(entry
->offset
);
2604 count
= OFF_TO_IDX(e
- s
);
2605 object
= entry
->object
.vm_object
;
2608 * Unwire before removing addresses from the pmap; otherwise,
2609 * unwiring will put the entries back in the pmap.
2611 if (entry
->wired_count
!= 0)
2612 vm_map_entry_unwire(map
, entry
);
2614 offidxend
= offidxstart
+ count
;
2616 if (object
== &kernel_object
) {
2617 vm_object_page_remove(object
, offidxstart
, offidxend
, FALSE
);
2619 pmap_remove(map
->pmap
, s
, e
);
2620 if (object
!= NULL
&&
2621 object
->ref_count
!= 1 &&
2622 (object
->flags
& (OBJ_NOSPLIT
|OBJ_ONEMAPPING
)) == OBJ_ONEMAPPING
&&
2623 (object
->type
== OBJT_DEFAULT
|| object
->type
== OBJT_SWAP
)) {
2624 vm_object_collapse(object
);
2625 vm_object_page_remove(object
, offidxstart
, offidxend
, FALSE
);
2626 if (object
->type
== OBJT_SWAP
) {
2627 swap_pager_freespace(object
, offidxstart
, count
);
2629 if (offidxend
>= object
->size
&&
2630 offidxstart
< object
->size
) {
2631 object
->size
= offidxstart
;
2637 * Delete the entry (which may delete the object) only after
2638 * removing all pmap entries pointing to its pages.
2639 * (Otherwise, its page frames may be reallocated, and any
2640 * modify bits will be set in the wrong object!)
2642 vm_map_entry_delete(map
, entry
, countp
);
2645 return (KERN_SUCCESS
);
2651 * Remove the given address range from the target map.
2652 * This is the exported form of vm_map_delete.
2655 vm_map_remove(vm_map_t map
, vm_offset_t start
, vm_offset_t end
)
2660 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
2662 VM_MAP_RANGE_CHECK(map
, start
, end
);
2663 result
= vm_map_delete(map
, start
, end
, &count
);
2665 vm_map_entry_release(count
);
2671 * vm_map_check_protection:
2673 * Assert that the target map allows the specified
2674 * privilege on the entire address region given.
2675 * The entire region must be allocated.
2678 vm_map_check_protection(vm_map_t map
, vm_offset_t start
, vm_offset_t end
,
2679 vm_prot_t protection
)
2681 vm_map_entry_t entry
;
2682 vm_map_entry_t tmp_entry
;
2684 if (!vm_map_lookup_entry(map
, start
, &tmp_entry
)) {
2689 while (start
< end
) {
2690 if (entry
== &map
->header
) {
2697 if (start
< entry
->start
) {
2701 * Check protection associated with entry.
2704 if ((entry
->protection
& protection
) != protection
) {
2707 /* go to next entry */
2710 entry
= entry
->next
;
2716 * Split the pages in a map entry into a new object. This affords
2717 * easier removal of unused pages, and keeps object inheritance from
2718 * being a negative impact on memory usage.
2721 vm_map_split(vm_map_entry_t entry
)
2724 vm_object_t orig_object
, new_object
, source
;
2726 vm_pindex_t offidxstart
, offidxend
, idx
;
2728 vm_ooffset_t offset
;
2730 orig_object
= entry
->object
.vm_object
;
2731 if (orig_object
->type
!= OBJT_DEFAULT
&& orig_object
->type
!= OBJT_SWAP
)
2733 if (orig_object
->ref_count
<= 1)
2736 offset
= entry
->offset
;
2740 offidxstart
= OFF_TO_IDX(offset
);
2741 offidxend
= offidxstart
+ OFF_TO_IDX(e
- s
);
2742 size
= offidxend
- offidxstart
;
2744 new_object
= vm_pager_allocate(orig_object
->type
, NULL
,
2745 IDX_TO_OFF(size
), VM_PROT_ALL
, 0);
2746 if (new_object
== NULL
)
2749 source
= orig_object
->backing_object
;
2750 if (source
!= NULL
) {
2751 vm_object_reference(source
); /* Referenced by new_object */
2752 LIST_INSERT_HEAD(&source
->shadow_head
,
2753 new_object
, shadow_list
);
2754 vm_object_clear_flag(source
, OBJ_ONEMAPPING
);
2755 new_object
->backing_object_offset
=
2756 orig_object
->backing_object_offset
+ IDX_TO_OFF(offidxstart
);
2757 new_object
->backing_object
= source
;
2758 source
->shadow_count
++;
2759 source
->generation
++;
2762 for (idx
= 0; idx
< size
; idx
++) {
2766 * A critical section is required to avoid a race between
2767 * the lookup and an interrupt/unbusy/free and our busy
2772 m
= vm_page_lookup(orig_object
, offidxstart
+ idx
);
2779 * We must wait for pending I/O to complete before we can
2782 * We do not have to VM_PROT_NONE the page as mappings should
2783 * not be changed by this operation.
2785 if (vm_page_sleep_busy(m
, TRUE
, "spltwt"))
2788 vm_page_rename(m
, new_object
, idx
);
2789 /* page automatically made dirty by rename and cache handled */
2794 if (orig_object
->type
== OBJT_SWAP
) {
2795 vm_object_pip_add(orig_object
, 1);
2797 * copy orig_object pages into new_object
2798 * and destroy unneeded pages in
2801 swap_pager_copy(orig_object
, new_object
, offidxstart
, 0);
2802 vm_object_pip_wakeup(orig_object
);
2806 * Wakeup the pages we played with. No spl protection is needed
2807 * for a simple wakeup.
2809 for (idx
= 0; idx
< size
; idx
++) {
2810 m
= vm_page_lookup(new_object
, idx
);
2815 entry
->object
.vm_object
= new_object
;
2816 entry
->offset
= 0LL;
2817 vm_object_deallocate(orig_object
);
2821 * vm_map_copy_entry:
2823 * Copies the contents of the source entry to the destination
2824 * entry. The entries *must* be aligned properly.
2827 vm_map_copy_entry(vm_map_t src_map
, vm_map_t dst_map
,
2828 vm_map_entry_t src_entry
, vm_map_entry_t dst_entry
)
2830 vm_object_t src_object
;
2832 if (dst_entry
->maptype
== VM_MAPTYPE_SUBMAP
)
2834 if (src_entry
->maptype
== VM_MAPTYPE_SUBMAP
)
2837 if (src_entry
->wired_count
== 0) {
2839 * If the source entry is marked needs_copy, it is already
2842 if ((src_entry
->eflags
& MAP_ENTRY_NEEDS_COPY
) == 0) {
2843 pmap_protect(src_map
->pmap
,
2846 src_entry
->protection
& ~VM_PROT_WRITE
);
2850 * Make a copy of the object.
2852 if ((src_object
= src_entry
->object
.vm_object
) != NULL
) {
2853 if ((src_object
->handle
== NULL
) &&
2854 (src_object
->type
== OBJT_DEFAULT
||
2855 src_object
->type
== OBJT_SWAP
)) {
2856 vm_object_collapse(src_object
);
2857 if ((src_object
->flags
& (OBJ_NOSPLIT
|OBJ_ONEMAPPING
)) == OBJ_ONEMAPPING
) {
2858 vm_map_split(src_entry
);
2859 src_object
= src_entry
->object
.vm_object
;
2863 vm_object_reference(src_object
);
2864 vm_object_clear_flag(src_object
, OBJ_ONEMAPPING
);
2865 dst_entry
->object
.vm_object
= src_object
;
2866 src_entry
->eflags
|= (MAP_ENTRY_COW
|MAP_ENTRY_NEEDS_COPY
);
2867 dst_entry
->eflags
|= (MAP_ENTRY_COW
|MAP_ENTRY_NEEDS_COPY
);
2868 dst_entry
->offset
= src_entry
->offset
;
2870 dst_entry
->object
.vm_object
= NULL
;
2871 dst_entry
->offset
= 0;
2874 pmap_copy(dst_map
->pmap
, src_map
->pmap
, dst_entry
->start
,
2875 dst_entry
->end
- dst_entry
->start
, src_entry
->start
);
2878 * Of course, wired down pages can't be set copy-on-write.
2879 * Cause wired pages to be copied into the new map by
2880 * simulating faults (the new pages are pageable)
2882 vm_fault_copy_entry(dst_map
, src_map
, dst_entry
, src_entry
);
2888 * Create a new process vmspace structure and vm_map
2889 * based on those of an existing process. The new map
2890 * is based on the old map, according to the inheritance
2891 * values on the regions in that map.
2893 * The source map must not be locked.
2896 vmspace_fork(struct vmspace
*vm1
)
2898 struct vmspace
*vm2
;
2899 vm_map_t old_map
= &vm1
->vm_map
;
2901 vm_map_entry_t old_entry
;
2902 vm_map_entry_t new_entry
;
2906 vm_map_lock(old_map
);
2907 old_map
->infork
= 1;
2910 * XXX Note: upcalls are not copied.
2912 vm2
= vmspace_alloc(old_map
->min_offset
, old_map
->max_offset
);
2913 bcopy(&vm1
->vm_startcopy
, &vm2
->vm_startcopy
,
2914 (caddr_t
)&vm1
->vm_endcopy
- (caddr_t
)&vm1
->vm_startcopy
);
2915 new_map
= &vm2
->vm_map
; /* XXX */
2916 new_map
->timestamp
= 1;
2919 old_entry
= old_map
->header
.next
;
2920 while (old_entry
!= &old_map
->header
) {
2922 old_entry
= old_entry
->next
;
2925 count
= vm_map_entry_reserve(count
+ MAP_RESERVE_COUNT
);
2927 old_entry
= old_map
->header
.next
;
2928 while (old_entry
!= &old_map
->header
) {
2929 if (old_entry
->maptype
== VM_MAPTYPE_SUBMAP
)
2930 panic("vm_map_fork: encountered a submap");
2932 switch (old_entry
->inheritance
) {
2933 case VM_INHERIT_NONE
:
2936 case VM_INHERIT_SHARE
:
2938 * Clone the entry, creating the shared object if
2941 object
= old_entry
->object
.vm_object
;
2942 if (object
== NULL
) {
2943 vm_map_entry_allocate_object(old_entry
);
2944 object
= old_entry
->object
.vm_object
;
2948 * Add the reference before calling vm_map_entry_shadow
2949 * to insure that a shadow object is created.
2951 vm_object_reference(object
);
2952 if (old_entry
->eflags
& MAP_ENTRY_NEEDS_COPY
) {
2953 vm_map_entry_shadow(old_entry
);
2954 /* Transfer the second reference too. */
2955 vm_object_reference(
2956 old_entry
->object
.vm_object
);
2957 vm_object_deallocate(object
);
2958 object
= old_entry
->object
.vm_object
;
2960 vm_object_clear_flag(object
, OBJ_ONEMAPPING
);
2963 * Clone the entry, referencing the shared object.
2965 new_entry
= vm_map_entry_create(new_map
, &count
);
2966 *new_entry
= *old_entry
;
2967 new_entry
->eflags
&= ~MAP_ENTRY_USER_WIRED
;
2968 new_entry
->wired_count
= 0;
2971 * Insert the entry into the new map -- we know we're
2972 * inserting at the end of the new map.
2975 vm_map_entry_link(new_map
, new_map
->header
.prev
,
2979 * Update the physical map
2982 pmap_copy(new_map
->pmap
, old_map
->pmap
,
2984 (old_entry
->end
- old_entry
->start
),
2988 case VM_INHERIT_COPY
:
2990 * Clone the entry and link into the map.
2992 new_entry
= vm_map_entry_create(new_map
, &count
);
2993 *new_entry
= *old_entry
;
2994 new_entry
->eflags
&= ~MAP_ENTRY_USER_WIRED
;
2995 new_entry
->wired_count
= 0;
2996 new_entry
->object
.vm_object
= NULL
;
2997 vm_map_entry_link(new_map
, new_map
->header
.prev
,
2999 vm_map_copy_entry(old_map
, new_map
, old_entry
,
3003 old_entry
= old_entry
->next
;
3006 new_map
->size
= old_map
->size
;
3007 old_map
->infork
= 0;
3008 vm_map_unlock(old_map
);
3009 vm_map_entry_release(count
);
3015 vm_map_stack (vm_map_t map
, vm_offset_t addrbos
, vm_size_t max_ssize
,
3016 int flags
, vm_prot_t prot
, vm_prot_t max
, int cow
)
3018 vm_map_entry_t prev_entry
;
3019 vm_map_entry_t new_stack_entry
;
3020 vm_size_t init_ssize
;
3023 vm_offset_t tmpaddr
;
3025 cow
|= MAP_IS_STACK
;
3027 if (max_ssize
< sgrowsiz
)
3028 init_ssize
= max_ssize
;
3030 init_ssize
= sgrowsiz
;
3032 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
3036 * Find space for the mapping
3038 if ((flags
& (MAP_FIXED
| MAP_TRYFIXED
)) == 0) {
3039 if (vm_map_findspace(map
, addrbos
, max_ssize
, 1,
3042 vm_map_entry_release(count
);
3043 return (KERN_NO_SPACE
);
3048 /* If addr is already mapped, no go */
3049 if (vm_map_lookup_entry(map
, addrbos
, &prev_entry
)) {
3051 vm_map_entry_release(count
);
3052 return (KERN_NO_SPACE
);
3056 /* XXX already handled by kern_mmap() */
3057 /* If we would blow our VMEM resource limit, no go */
3058 if (map
->size
+ init_ssize
>
3059 curproc
->p_rlimit
[RLIMIT_VMEM
].rlim_cur
) {
3061 vm_map_entry_release(count
);
3062 return (KERN_NO_SPACE
);
3067 * If we can't accomodate max_ssize in the current mapping,
3068 * no go. However, we need to be aware that subsequent user
3069 * mappings might map into the space we have reserved for
3070 * stack, and currently this space is not protected.
3072 * Hopefully we will at least detect this condition
3073 * when we try to grow the stack.
3075 if ((prev_entry
->next
!= &map
->header
) &&
3076 (prev_entry
->next
->start
< addrbos
+ max_ssize
)) {
3078 vm_map_entry_release(count
);
3079 return (KERN_NO_SPACE
);
3083 * We initially map a stack of only init_ssize. We will
3084 * grow as needed later. Since this is to be a grow
3085 * down stack, we map at the top of the range.
3087 * Note: we would normally expect prot and max to be
3088 * VM_PROT_ALL, and cow to be 0. Possibly we should
3089 * eliminate these as input parameters, and just
3090 * pass these values here in the insert call.
3092 rv
= vm_map_insert(map
, &count
,
3093 NULL
, 0, addrbos
+ max_ssize
- init_ssize
,
3094 addrbos
+ max_ssize
,
3099 /* Now set the avail_ssize amount */
3100 if (rv
== KERN_SUCCESS
) {
3101 if (prev_entry
!= &map
->header
)
3102 vm_map_clip_end(map
, prev_entry
, addrbos
+ max_ssize
- init_ssize
, &count
);
3103 new_stack_entry
= prev_entry
->next
;
3104 if (new_stack_entry
->end
!= addrbos
+ max_ssize
||
3105 new_stack_entry
->start
!= addrbos
+ max_ssize
- init_ssize
)
3106 panic ("Bad entry start/end for new stack entry");
3108 new_stack_entry
->aux
.avail_ssize
= max_ssize
- init_ssize
;
3112 vm_map_entry_release(count
);
3116 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3117 * desired address is already mapped, or if we successfully grow
3118 * the stack. Also returns KERN_SUCCESS if addr is outside the
3119 * stack range (this is strange, but preserves compatibility with
3120 * the grow function in vm_machdep.c).
3123 vm_map_growstack (struct proc
*p
, vm_offset_t addr
)
3125 vm_map_entry_t prev_entry
;
3126 vm_map_entry_t stack_entry
;
3127 vm_map_entry_t new_stack_entry
;
3128 struct vmspace
*vm
= p
->p_vmspace
;
3129 vm_map_t map
= &vm
->vm_map
;
3132 int rv
= KERN_SUCCESS
;
3134 int use_read_lock
= 1;
3137 count
= vm_map_entry_reserve(MAP_RESERVE_COUNT
);
3140 vm_map_lock_read(map
);
3144 /* If addr is already in the entry range, no need to grow.*/
3145 if (vm_map_lookup_entry(map
, addr
, &prev_entry
))
3148 if ((stack_entry
= prev_entry
->next
) == &map
->header
)
3150 if (prev_entry
== &map
->header
)
3151 end
= stack_entry
->start
- stack_entry
->aux
.avail_ssize
;
3153 end
= prev_entry
->end
;
3156 * This next test mimics the old grow function in vm_machdep.c.
3157 * It really doesn't quite make sense, but we do it anyway
3158 * for compatibility.
3160 * If not growable stack, return success. This signals the
3161 * caller to proceed as he would normally with normal vm.
3163 if (stack_entry
->aux
.avail_ssize
< 1 ||
3164 addr
>= stack_entry
->start
||
3165 addr
< stack_entry
->start
- stack_entry
->aux
.avail_ssize
) {
3169 /* Find the minimum grow amount */
3170 grow_amount
= roundup (stack_entry
->start
- addr
, PAGE_SIZE
);
3171 if (grow_amount
> stack_entry
->aux
.avail_ssize
) {
3177 * If there is no longer enough space between the entries
3178 * nogo, and adjust the available space. Note: this
3179 * should only happen if the user has mapped into the
3180 * stack area after the stack was created, and is
3181 * probably an error.
3183 * This also effectively destroys any guard page the user
3184 * might have intended by limiting the stack size.
3186 if (grow_amount
> stack_entry
->start
- end
) {
3187 if (use_read_lock
&& vm_map_lock_upgrade(map
)) {
3192 stack_entry
->aux
.avail_ssize
= stack_entry
->start
- end
;
3197 is_procstack
= addr
>= (vm_offset_t
)vm
->vm_maxsaddr
;
3199 /* If this is the main process stack, see if we're over the
3202 if (is_procstack
&& (ctob(vm
->vm_ssize
) + grow_amount
>
3203 p
->p_rlimit
[RLIMIT_STACK
].rlim_cur
)) {
3208 /* Round up the grow amount modulo SGROWSIZ */
3209 grow_amount
= roundup (grow_amount
, sgrowsiz
);
3210 if (grow_amount
> stack_entry
->aux
.avail_ssize
) {
3211 grow_amount
= stack_entry
->aux
.avail_ssize
;
3213 if (is_procstack
&& (ctob(vm
->vm_ssize
) + grow_amount
>
3214 p
->p_rlimit
[RLIMIT_STACK
].rlim_cur
)) {
3215 grow_amount
= p
->p_rlimit
[RLIMIT_STACK
].rlim_cur
-
3219 /* If we would blow our VMEM resource limit, no go */
3220 if (map
->size
+ grow_amount
> p
->p_rlimit
[RLIMIT_VMEM
].rlim_cur
) {
3225 if (use_read_lock
&& vm_map_lock_upgrade(map
)) {
3231 /* Get the preliminary new entry start value */
3232 addr
= stack_entry
->start
- grow_amount
;
3234 /* If this puts us into the previous entry, cut back our growth
3235 * to the available space. Also, see the note above.
3238 stack_entry
->aux
.avail_ssize
= stack_entry
->start
- end
;
3242 rv
= vm_map_insert(map
, &count
,
3243 NULL
, 0, addr
, stack_entry
->start
,
3245 VM_PROT_ALL
, VM_PROT_ALL
,
3248 /* Adjust the available stack space by the amount we grew. */
3249 if (rv
== KERN_SUCCESS
) {
3250 if (prev_entry
!= &map
->header
)
3251 vm_map_clip_end(map
, prev_entry
, addr
, &count
);
3252 new_stack_entry
= prev_entry
->next
;
3253 if (new_stack_entry
->end
!= stack_entry
->start
||
3254 new_stack_entry
->start
!= addr
)
3255 panic ("Bad stack grow start/end in new stack entry");
3257 new_stack_entry
->aux
.avail_ssize
=
3258 stack_entry
->aux
.avail_ssize
-
3259 (new_stack_entry
->end
- new_stack_entry
->start
);
3261 vm
->vm_ssize
+= btoc(new_stack_entry
->end
-
3262 new_stack_entry
->start
);
3268 vm_map_unlock_read(map
);
3271 vm_map_entry_release(count
);
3276 * Unshare the specified VM space for exec. If other processes are
3277 * mapped to it, then create a new one. The new vmspace is null.
3280 vmspace_exec(struct proc
*p
, struct vmspace
*vmcopy
)
3282 struct vmspace
*oldvmspace
= p
->p_vmspace
;
3283 struct vmspace
*newvmspace
;
3284 vm_map_t map
= &p
->p_vmspace
->vm_map
;
3287 * If we are execing a resident vmspace we fork it, otherwise
3288 * we create a new vmspace. Note that exitingcnt and upcalls
3289 * are not copied to the new vmspace.
3292 newvmspace
= vmspace_fork(vmcopy
);
3294 newvmspace
= vmspace_alloc(map
->min_offset
, map
->max_offset
);
3295 bcopy(&oldvmspace
->vm_startcopy
, &newvmspace
->vm_startcopy
,
3296 (caddr_t
)&oldvmspace
->vm_endcopy
-
3297 (caddr_t
)&oldvmspace
->vm_startcopy
);
3301 * Finish initializing the vmspace before assigning it
3302 * to the process. The vmspace will become the current vmspace
3305 pmap_pinit2(vmspace_pmap(newvmspace
));
3306 pmap_replacevm(p
, newvmspace
, 0);
3307 sysref_put(&oldvmspace
->vm_sysref
);
3311 * Unshare the specified VM space for forcing COW. This
3312 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3314 * The exitingcnt test is not strictly necessary but has been
3315 * included for code sanity (to make the code a bit more deterministic).
3319 vmspace_unshare(struct proc
*p
)
3321 struct vmspace
*oldvmspace
= p
->p_vmspace
;
3322 struct vmspace
*newvmspace
;
3324 if (oldvmspace
->vm_sysref
.refcnt
== 1 && oldvmspace
->vm_exitingcnt
== 0)
3326 newvmspace
= vmspace_fork(oldvmspace
);
3327 pmap_pinit2(vmspace_pmap(newvmspace
));
3328 pmap_replacevm(p
, newvmspace
, 0);
3329 sysref_put(&oldvmspace
->vm_sysref
);
3335 * Finds the VM object, offset, and
3336 * protection for a given virtual address in the
3337 * specified map, assuming a page fault of the
3340 * Leaves the map in question locked for read; return
3341 * values are guaranteed until a vm_map_lookup_done
3342 * call is performed. Note that the map argument
3343 * is in/out; the returned map must be used in
3344 * the call to vm_map_lookup_done.
3346 * A handle (out_entry) is returned for use in
3347 * vm_map_lookup_done, to make that fast.
3349 * If a lookup is requested with "write protection"
3350 * specified, the map may be changed to perform virtual
3351 * copying operations, although the data referenced will
3355 vm_map_lookup(vm_map_t
*var_map
, /* IN/OUT */
3357 vm_prot_t fault_typea
,
3358 vm_map_entry_t
*out_entry
, /* OUT */
3359 vm_object_t
*object
, /* OUT */
3360 vm_pindex_t
*pindex
, /* OUT */
3361 vm_prot_t
*out_prot
, /* OUT */
3362 boolean_t
*wired
) /* OUT */
3364 vm_map_entry_t entry
;
3365 vm_map_t map
= *var_map
;
3367 vm_prot_t fault_type
= fault_typea
;
3368 int use_read_lock
= 1;
3369 int rv
= KERN_SUCCESS
;
3373 vm_map_lock_read(map
);
3378 * If the map has an interesting hint, try it before calling full
3379 * blown lookup routine.
3384 if ((entry
== &map
->header
) ||
3385 (vaddr
< entry
->start
) || (vaddr
>= entry
->end
)) {
3386 vm_map_entry_t tmp_entry
;
3389 * Entry was either not a valid hint, or the vaddr was not
3390 * contained in the entry, so do a full lookup.
3392 if (!vm_map_lookup_entry(map
, vaddr
, &tmp_entry
)) {
3393 rv
= KERN_INVALID_ADDRESS
;
3404 if (entry
->maptype
== VM_MAPTYPE_SUBMAP
) {
3405 vm_map_t old_map
= map
;
3407 *var_map
= map
= entry
->object
.sub_map
;
3409 vm_map_unlock_read(old_map
);
3411 vm_map_unlock(old_map
);
3417 * Check whether this task is allowed to have this page.
3418 * Note the special case for MAP_ENTRY_COW
3419 * pages with an override. This is to implement a forced
3420 * COW for debuggers.
3423 if (fault_type
& VM_PROT_OVERRIDE_WRITE
)
3424 prot
= entry
->max_protection
;
3426 prot
= entry
->protection
;
3428 fault_type
&= (VM_PROT_READ
|VM_PROT_WRITE
|VM_PROT_EXECUTE
);
3429 if ((fault_type
& prot
) != fault_type
) {
3430 rv
= KERN_PROTECTION_FAILURE
;
3434 if ((entry
->eflags
& MAP_ENTRY_USER_WIRED
) &&
3435 (entry
->eflags
& MAP_ENTRY_COW
) &&
3436 (fault_type
& VM_PROT_WRITE
) &&
3437 (fault_typea
& VM_PROT_OVERRIDE_WRITE
) == 0) {
3438 rv
= KERN_PROTECTION_FAILURE
;
3443 * If this page is not pageable, we have to get it for all possible
3446 *wired
= (entry
->wired_count
!= 0);
3448 prot
= fault_type
= entry
->protection
;
3451 * Virtual page tables may need to update the accessed (A) bit
3452 * in a page table entry. Upgrade the fault to a write fault for
3453 * that case if the map will support it. If the map does not support
3454 * it the page table entry simply will not be updated.
3456 if (entry
->maptype
== VM_MAPTYPE_VPAGETABLE
) {
3457 if (prot
& VM_PROT_WRITE
)
3458 fault_type
|= VM_PROT_WRITE
;
3462 * If the entry was copy-on-write, we either ...
3464 if (entry
->eflags
& MAP_ENTRY_NEEDS_COPY
) {
3466 * If we want to write the page, we may as well handle that
3467 * now since we've got the map locked.
3469 * If we don't need to write the page, we just demote the
3470 * permissions allowed.
3473 if (fault_type
& VM_PROT_WRITE
) {
3475 * Make a new object, and place it in the object
3476 * chain. Note that no new references have appeared
3477 * -- one just moved from the map to the new
3481 if (use_read_lock
&& vm_map_lock_upgrade(map
)) {
3487 vm_map_entry_shadow(entry
);
3490 * We're attempting to read a copy-on-write page --
3491 * don't allow writes.
3494 prot
&= ~VM_PROT_WRITE
;
3499 * Create an object if necessary.
3501 if (entry
->object
.vm_object
== NULL
&&
3503 if (use_read_lock
&& vm_map_lock_upgrade(map
)) {
3508 vm_map_entry_allocate_object(entry
);
3512 * Return the object/offset from this entry. If the entry was
3513 * copy-on-write or empty, it has been fixed up.
3516 *pindex
= OFF_TO_IDX((vaddr
- entry
->start
) + entry
->offset
);
3517 *object
= entry
->object
.vm_object
;
3520 * Return whether this is the only map sharing this data. On
3521 * success we return with a read lock held on the map. On failure
3522 * we return with the map unlocked.
3526 if (rv
== KERN_SUCCESS
) {
3527 if (use_read_lock
== 0)
3528 vm_map_lock_downgrade(map
);
3529 } else if (use_read_lock
) {
3530 vm_map_unlock_read(map
);
3538 * vm_map_lookup_done:
3540 * Releases locks acquired by a vm_map_lookup
3541 * (according to the handle returned by that lookup).
3545 vm_map_lookup_done(vm_map_t map
, vm_map_entry_t entry
, int count
)
3548 * Unlock the main-level map
3550 vm_map_unlock_read(map
);
3552 vm_map_entry_release(count
);
3555 #include "opt_ddb.h"
3557 #include <sys/kernel.h>
3559 #include <ddb/ddb.h>
3562 * vm_map_print: [ debug ]
3564 DB_SHOW_COMMAND(map
, vm_map_print
)
3567 /* XXX convert args. */
3568 vm_map_t map
= (vm_map_t
)addr
;
3569 boolean_t full
= have_addr
;
3571 vm_map_entry_t entry
;
3573 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3575 (void *)map
->pmap
, map
->nentries
, map
->timestamp
);
3578 if (!full
&& db_indent
)
3582 for (entry
= map
->header
.next
; entry
!= &map
->header
;
3583 entry
= entry
->next
) {
3584 db_iprintf("map entry %p: start=%p, end=%p\n",
3585 (void *)entry
, (void *)entry
->start
, (void *)entry
->end
);
3588 static char *inheritance_name
[4] =
3589 {"share", "copy", "none", "donate_copy"};
3591 db_iprintf(" prot=%x/%x/%s",
3593 entry
->max_protection
,
3594 inheritance_name
[(int)(unsigned char)entry
->inheritance
]);
3595 if (entry
->wired_count
!= 0)
3596 db_printf(", wired");
3598 if (entry
->maptype
== VM_MAPTYPE_SUBMAP
) {
3599 /* XXX no %qd in kernel. Truncate entry->offset. */
3600 db_printf(", share=%p, offset=0x%lx\n",
3601 (void *)entry
->object
.sub_map
,
3602 (long)entry
->offset
);
3604 if ((entry
->prev
== &map
->header
) ||
3605 (entry
->prev
->object
.sub_map
!=
3606 entry
->object
.sub_map
)) {
3608 vm_map_print((db_expr_t
)(intptr_t)
3609 entry
->object
.sub_map
,
3614 /* XXX no %qd in kernel. Truncate entry->offset. */
3615 db_printf(", object=%p, offset=0x%lx",
3616 (void *)entry
->object
.vm_object
,
3617 (long)entry
->offset
);
3618 if (entry
->eflags
& MAP_ENTRY_COW
)
3619 db_printf(", copy (%s)",
3620 (entry
->eflags
& MAP_ENTRY_NEEDS_COPY
) ? "needed" : "done");
3624 if ((entry
->prev
== &map
->header
) ||
3625 (entry
->prev
->object
.vm_object
!=
3626 entry
->object
.vm_object
)) {
3628 vm_object_print((db_expr_t
)(intptr_t)
3629 entry
->object
.vm_object
,
3642 DB_SHOW_COMMAND(procvm
, procvm
)
3647 p
= (struct proc
*) addr
;
3652 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3653 (void *)p
, (void *)p
->p_vmspace
, (void *)&p
->p_vmspace
->vm_map
,
3654 (void *)vmspace_pmap(p
->p_vmspace
));
3656 vm_map_print((db_expr_t
)(intptr_t)&p
->p_vmspace
->vm_map
, 1, 0, NULL
);