6 #include <linux/slab.h>
8 #include <linux/mman.h>
9 #include <linux/pagemap.h>
10 #include <linux/swap.h>
11 #include <linux/swapctl.h>
12 #include <linux/smp_lock.h>
13 #include <linux/init.h>
14 #include <linux/file.h>
16 #include <asm/uaccess.h>
17 #include <asm/pgalloc.h>
19 /* description of effects of mapping type and prot in current implementation.
20 * this is due to the limited x86 page protection hardware. The expected
21 * behavior is in parens:
24 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
25 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
26 * w: (no) no w: (no) no w: (yes) yes w: (no) no
27 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
29 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
30 * w: (no) no w: (no) no w: (copy) copy w: (no) no
31 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
34 pgprot_t protection_map
[16] = {
35 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
36 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
39 /* SLAB cache for vm_area_struct's. */
40 kmem_cache_t
*vm_area_cachep
;
42 int sysctl_overcommit_memory
;
44 /* Check that a process has enough memory to allocate a
45 * new virtual mapping.
47 int vm_enough_memory(long pages
)
49 /* Stupid algorithm to decide if we have enough memory: while
50 * simple, it hopefully works in most obvious cases.. Easy to
51 * fool it, but this should catch most mistakes.
53 /* 23/11/98 NJC: Somewhat less stupid version of algorithm,
54 * which tries to do "TheRightThing". Instead of using half of
55 * (buffers+cache), use the minimum values. Allow an extra 2%
56 * of num_physpages for safety margin.
61 /* Sometimes we want to use more memory than we have. */
62 if (sysctl_overcommit_memory
)
65 free
= atomic_read(&buffermem_pages
);
66 free
+= atomic_read(&page_cache_size
);
67 free
+= nr_free_pages();
68 free
+= nr_swap_pages
;
72 /* Remove one vm structure from the inode's i_mmap ring. */
73 static inline void remove_shared_vm_struct(struct vm_area_struct
*vma
)
75 struct file
* file
= vma
->vm_file
;
78 if (vma
->vm_flags
& VM_DENYWRITE
)
79 atomic_inc(&file
->f_dentry
->d_inode
->i_writecount
);
80 spin_lock(&file
->f_dentry
->d_inode
->i_shared_lock
);
81 if(vma
->vm_next_share
)
82 vma
->vm_next_share
->vm_pprev_share
= vma
->vm_pprev_share
;
83 *vma
->vm_pprev_share
= vma
->vm_next_share
;
84 spin_unlock(&file
->f_dentry
->d_inode
->i_shared_lock
);
89 * sys_brk() for the most part doesn't need the global kernel
90 * lock, except when an application is doing something nasty
91 * like trying to un-brk an area that has already been mapped
92 * to a regular file. in this case, the unmapping will need
93 * to invoke file system routines that need the global lock.
95 asmlinkage
unsigned long sys_brk(unsigned long brk
)
97 unsigned long rlim
, retval
;
98 unsigned long newbrk
, oldbrk
;
99 struct mm_struct
*mm
= current
->mm
;
103 if (brk
< mm
->end_code
)
105 newbrk
= PAGE_ALIGN(brk
);
106 oldbrk
= PAGE_ALIGN(mm
->brk
);
107 if (oldbrk
== newbrk
)
110 /* Always allow shrinking brk. */
111 if (brk
<= mm
->brk
) {
112 if (!do_munmap(newbrk
, oldbrk
-newbrk
))
117 /* Check against rlimit and stack.. */
118 rlim
= current
->rlim
[RLIMIT_DATA
].rlim_cur
;
119 if (rlim
< RLIM_INFINITY
&& brk
- mm
->end_code
> rlim
)
122 /* Check against existing mmap mappings. */
123 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
126 /* Check if we have enough memory.. */
127 if (!vm_enough_memory((newbrk
-oldbrk
) >> PAGE_SHIFT
))
130 /* Ok, looks good - let it rip. */
131 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
141 /* Combine the mmap "prot" and "flags" argument into one "vm_flags" used
142 * internally. Essentially, translate the "PROT_xxx" and "MAP_xxx" bits
145 static inline unsigned long vm_flags(unsigned long prot
, unsigned long flags
)
147 #define _trans(x,bit1,bit2) \
148 ((bit1==bit2)?(x&bit1):(x&bit1)?bit2:0)
150 unsigned long prot_bits
, flag_bits
;
152 _trans(prot
, PROT_READ
, VM_READ
) |
153 _trans(prot
, PROT_WRITE
, VM_WRITE
) |
154 _trans(prot
, PROT_EXEC
, VM_EXEC
);
156 _trans(flags
, MAP_GROWSDOWN
, VM_GROWSDOWN
) |
157 _trans(flags
, MAP_DENYWRITE
, VM_DENYWRITE
) |
158 _trans(flags
, MAP_EXECUTABLE
, VM_EXECUTABLE
);
159 return prot_bits
| flag_bits
;
163 unsigned long do_mmap_pgoff(struct file
* file
, unsigned long addr
, unsigned long len
,
164 unsigned long prot
, unsigned long flags
, unsigned long pgoff
)
166 struct mm_struct
* mm
= current
->mm
;
167 struct vm_area_struct
* vma
;
170 if (file
&& (!file
->f_op
|| !file
->f_op
->mmap
))
173 if ((len
= PAGE_ALIGN(len
)) == 0)
176 if (len
> TASK_SIZE
|| addr
> TASK_SIZE
-len
)
179 /* offset overflow? */
180 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
183 /* Too many mappings? */
184 if (mm
->map_count
> MAX_MAP_COUNT
)
187 /* mlock MCL_FUTURE? */
188 if (mm
->def_flags
& VM_LOCKED
) {
189 unsigned long locked
= mm
->locked_vm
<< PAGE_SHIFT
;
191 if (locked
> current
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
)
195 /* Do simple checking here so the lower-level routines won't have
196 * to. we assume access permissions have been handled by the open
197 * of the memory object, so we don't do any here.
200 switch (flags
& MAP_TYPE
) {
202 if ((prot
& PROT_WRITE
) && !(file
->f_mode
& 2))
205 /* Make sure we don't allow writing to an append-only file.. */
206 if (IS_APPEND(file
->f_dentry
->d_inode
) && (file
->f_mode
& 2))
209 /* make sure there are no mandatory locks on the file. */
210 if (locks_verify_locked(file
->f_dentry
->d_inode
))
215 if (!(file
->f_mode
& 1))
222 } else if ((flags
& MAP_TYPE
) != MAP_PRIVATE
)
225 /* Obtain the address to map to. we verify (or select) it and ensure
226 * that it represents a valid section of the address space.
228 if (flags
& MAP_FIXED
) {
229 if (addr
& ~PAGE_MASK
)
232 addr
= get_unmapped_area(addr
, len
);
237 /* Determine the object being mapped and call the appropriate
238 * specific mapper. the address has already been validated, but
239 * not unmapped, but the maps are removed from the list.
241 vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
246 vma
->vm_start
= addr
;
247 vma
->vm_end
= addr
+ len
;
248 vma
->vm_flags
= vm_flags(prot
,flags
) | mm
->def_flags
;
251 if (file
->f_mode
& 1)
252 vma
->vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
253 if (flags
& MAP_SHARED
) {
254 vma
->vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
256 /* This looks strange, but when we don't have the file open
257 * for writing, we can demote the shared mapping to a simpler
258 * private mapping. That also takes care of a security hole
259 * with ptrace() writing to a shared mapping without write
262 * We leave the VM_MAYSHARE bit on, just to get correct output
263 * from /proc/xxx/maps..
265 if (!(file
->f_mode
& 2))
266 vma
->vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
269 vma
->vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
270 vma
->vm_page_prot
= protection_map
[vma
->vm_flags
& 0x0f];
272 vma
->vm_pgoff
= pgoff
;
274 vma
->vm_private_data
= NULL
;
278 if (do_munmap(addr
, len
))
281 /* Check against address space limit. */
282 if ((mm
->total_vm
<< PAGE_SHIFT
) + len
283 > current
->rlim
[RLIMIT_AS
].rlim_cur
)
286 /* Private writable mapping? Check memory availability.. */
287 if ((vma
->vm_flags
& (VM_SHARED
| VM_WRITE
)) == VM_WRITE
&&
288 !(flags
& MAP_NORESERVE
) &&
289 !vm_enough_memory(len
>> PAGE_SHIFT
))
293 int correct_wcount
= 0;
294 if (vma
->vm_flags
& VM_DENYWRITE
) {
295 if (atomic_read(&file
->f_dentry
->d_inode
->i_writecount
) > 0) {
299 /* f_op->mmap might possibly sleep
300 * (generic_file_mmap doesn't, but other code
301 * might). In any case, this takes care of any
302 * race that this might cause.
304 atomic_dec(&file
->f_dentry
->d_inode
->i_writecount
);
307 error
= file
->f_op
->mmap(file
, vma
);
308 /* Fix up the count if necessary, then check for an error */
310 atomic_inc(&file
->f_dentry
->d_inode
->i_writecount
);
312 goto unmap_and_free_vma
;
318 * merge_segments may merge our vma, so we can't refer to it
319 * after the call. Save the values we need now ...
321 flags
= vma
->vm_flags
;
322 addr
= vma
->vm_start
; /* can addr have changed?? */
323 vmlist_modify_lock(mm
);
324 insert_vm_struct(mm
, vma
);
325 merge_segments(mm
, vma
->vm_start
, vma
->vm_end
);
326 vmlist_modify_unlock(mm
);
328 mm
->total_vm
+= len
>> PAGE_SHIFT
;
329 if (flags
& VM_LOCKED
) {
330 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
331 make_pages_present(addr
, addr
+ len
);
336 /* Undo any partial mapping done by a device driver. */
337 flush_cache_range(mm
, vma
->vm_start
, vma
->vm_end
);
338 zap_page_range(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
339 flush_tlb_range(mm
, vma
->vm_start
, vma
->vm_end
);
341 kmem_cache_free(vm_area_cachep
, vma
);
345 /* Get an address range which is currently unmapped.
346 * For mmap() without MAP_FIXED and shmat() with addr=0.
347 * Return value 0 means ENOMEM.
349 unsigned long get_unmapped_area(unsigned long addr
, unsigned long len
)
351 struct vm_area_struct
* vmm
;
356 addr
= TASK_UNMAPPED_BASE
;
357 addr
= PAGE_ALIGN(addr
);
359 for (vmm
= find_vma(current
->mm
, addr
); ; vmm
= vmm
->vm_next
) {
360 /* At this point: (!vmm || addr < vmm->vm_end). */
361 if (TASK_SIZE
- len
< addr
)
363 if (!vmm
|| addr
+ len
<= vmm
->vm_start
)
369 #define vm_avl_empty (struct vm_area_struct *) NULL
371 #include "mmap_avl.c"
373 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
374 struct vm_area_struct
* find_vma(struct mm_struct
* mm
, unsigned long addr
)
376 struct vm_area_struct
*vma
= NULL
;
379 /* Check the cache first. */
380 /* (Cache hit rate is typically around 35%.) */
381 vma
= mm
->mmap_cache
;
382 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
384 /* Go through the linear list. */
386 while (vma
&& vma
->vm_end
<= addr
)
389 /* Then go through the AVL tree quickly. */
390 struct vm_area_struct
* tree
= mm
->mmap_avl
;
393 if (tree
== vm_avl_empty
)
395 if (tree
->vm_end
> addr
) {
397 if (tree
->vm_start
<= addr
)
399 tree
= tree
->vm_avl_left
;
401 tree
= tree
->vm_avl_right
;
405 mm
->mmap_cache
= vma
;
411 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
412 struct vm_area_struct
* find_vma_prev(struct mm_struct
* mm
, unsigned long addr
,
413 struct vm_area_struct
**pprev
)
417 /* Go through the linear list. */
418 struct vm_area_struct
* prev
= NULL
;
419 struct vm_area_struct
* vma
= mm
->mmap
;
420 while (vma
&& vma
->vm_end
<= addr
) {
427 /* Go through the AVL tree quickly. */
428 struct vm_area_struct
* vma
= NULL
;
429 struct vm_area_struct
* last_turn_right
= NULL
;
430 struct vm_area_struct
* prev
= NULL
;
431 struct vm_area_struct
* tree
= mm
->mmap_avl
;
433 if (tree
== vm_avl_empty
)
435 if (tree
->vm_end
> addr
) {
437 prev
= last_turn_right
;
438 if (tree
->vm_start
<= addr
)
440 tree
= tree
->vm_avl_left
;
442 last_turn_right
= tree
;
443 tree
= tree
->vm_avl_right
;
447 if (vma
->vm_avl_left
!= vm_avl_empty
) {
448 prev
= vma
->vm_avl_left
;
449 while (prev
->vm_avl_right
!= vm_avl_empty
)
450 prev
= prev
->vm_avl_right
;
452 if ((prev
? prev
->vm_next
: mm
->mmap
) != vma
)
453 printk("find_vma_prev: tree inconsistent with list\n");
463 struct vm_area_struct
* find_extend_vma(struct task_struct
* tsk
, unsigned long addr
)
465 struct vm_area_struct
* vma
;
469 vma
= find_vma(tsk
->mm
,addr
);
472 if (vma
->vm_start
<= addr
)
474 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
476 start
= vma
->vm_start
;
477 if (expand_stack(vma
, addr
))
479 if (vma
->vm_flags
& VM_LOCKED
) {
480 make_pages_present(addr
, start
);
485 /* Normal function to fix up a mapping
486 * This function is the default for when an area has no specific
487 * function. This may be used as part of a more specific routine.
488 * This function works out what part of an area is affected and
489 * adjusts the mapping information. Since the actual page
490 * manipulation is done in do_mmap(), none need be done here,
491 * though it would probably be more appropriate.
493 * By the time this function is called, the area struct has been
494 * removed from the process mapping list, so it needs to be
495 * reinserted if necessary.
497 * The 4 main cases are:
498 * Unmapping the whole area
499 * Unmapping from the start of the segment to a point in it
500 * Unmapping from an intermediate point to the end
501 * Unmapping between to intermediate points, making a hole.
503 * Case 4 involves the creation of 2 new areas, for each side of
504 * the hole. If possible, we reuse the existing area rather than
505 * allocate a new one, and the return indicates whether the old
508 static struct vm_area_struct
* unmap_fixup(struct vm_area_struct
*area
,
509 unsigned long addr
, size_t len
, struct vm_area_struct
*extra
)
511 struct vm_area_struct
*mpnt
;
512 unsigned long end
= addr
+ len
;
514 area
->vm_mm
->total_vm
-= len
>> PAGE_SHIFT
;
515 if (area
->vm_flags
& VM_LOCKED
)
516 area
->vm_mm
->locked_vm
-= len
>> PAGE_SHIFT
;
518 /* Unmapping the whole area. */
519 if (addr
== area
->vm_start
&& end
== area
->vm_end
) {
520 if (area
->vm_ops
&& area
->vm_ops
->close
)
521 area
->vm_ops
->close(area
);
524 kmem_cache_free(vm_area_cachep
, area
);
528 /* Work out to one of the ends. */
529 if (end
== area
->vm_end
) {
531 vmlist_modify_lock(current
->mm
);
532 } else if (addr
== area
->vm_start
) {
533 area
->vm_pgoff
+= (end
- area
->vm_start
) >> PAGE_SHIFT
;
534 area
->vm_start
= end
;
535 vmlist_modify_lock(current
->mm
);
537 /* Unmapping a hole: area->vm_start < addr <= end < area->vm_end */
538 /* Add end mapping -- leave beginning for below */
542 mpnt
->vm_mm
= area
->vm_mm
;
543 mpnt
->vm_start
= end
;
544 mpnt
->vm_end
= area
->vm_end
;
545 mpnt
->vm_page_prot
= area
->vm_page_prot
;
546 mpnt
->vm_flags
= area
->vm_flags
;
547 mpnt
->vm_ops
= area
->vm_ops
;
548 mpnt
->vm_pgoff
= area
->vm_pgoff
;
549 area
->vm_pgoff
+= (end
- area
->vm_start
) >> PAGE_SHIFT
;
550 mpnt
->vm_file
= area
->vm_file
;
551 mpnt
->vm_private_data
= area
->vm_private_data
;
553 get_file(mpnt
->vm_file
);
554 if (mpnt
->vm_ops
&& mpnt
->vm_ops
->open
)
555 mpnt
->vm_ops
->open(mpnt
);
556 area
->vm_end
= addr
; /* Truncate area */
557 vmlist_modify_lock(current
->mm
);
558 insert_vm_struct(current
->mm
, mpnt
);
561 insert_vm_struct(current
->mm
, area
);
562 vmlist_modify_unlock(current
->mm
);
567 * Try to free as many page directory entries as we can,
568 * without having to work very hard at actually scanning
569 * the page tables themselves.
571 * Right now we try to free page tables if we have a nice
572 * PGDIR-aligned area that got free'd up. We could be more
573 * granular if we want to, but this is fast and simple,
574 * and covers the bad cases.
576 * "prev", if it exists, points to a vma before the one
577 * we just free'd - but there's no telling how much before.
579 static void free_pgtables(struct mm_struct
* mm
, struct vm_area_struct
*prev
,
580 unsigned long start
, unsigned long end
)
582 unsigned long first
= start
& PGDIR_MASK
;
583 unsigned long last
= (end
+ PGDIR_SIZE
- 1) & PGDIR_MASK
;
589 if (prev
->vm_end
> start
) {
590 if (last
> prev
->vm_start
)
591 last
= prev
->vm_start
;
596 struct vm_area_struct
*next
= prev
->vm_next
;
599 if (next
->vm_start
< start
) {
603 if (last
> next
->vm_start
)
604 last
= next
->vm_start
;
606 if (prev
->vm_end
> first
)
607 first
= prev
->vm_end
+ PGDIR_SIZE
- 1;
611 first
= first
>> PGDIR_SHIFT
;
612 last
= last
>> PGDIR_SHIFT
;
614 clear_page_tables(mm
, first
, last
-first
);
617 /* Munmap is split into 2 main parts -- this part which finds
618 * what needs doing, and the areas themselves, which do the
619 * work. This now handles partial unmappings.
620 * Jeremy Fitzhardine <jeremy@sw.oz.au>
622 int do_munmap(unsigned long addr
, size_t len
)
624 struct mm_struct
* mm
;
625 struct vm_area_struct
*mpnt
, *prev
, **npp
, *free
, *extra
;
627 if ((addr
& ~PAGE_MASK
) || addr
> TASK_SIZE
|| len
> TASK_SIZE
-addr
)
630 if ((len
= PAGE_ALIGN(len
)) == 0)
633 /* Check if this memory area is ok - put it on the temporary
634 * list if so.. The checks here are pretty simple --
635 * every area affected in some way (by any overlap) is put
636 * on the list. If nothing is put on, nothing is affected.
639 mpnt
= find_vma_prev(mm
, addr
, &prev
);
642 /* we have addr < mpnt->vm_end */
644 if (mpnt
->vm_start
>= addr
+len
)
647 /* If we'll make "hole", check the vm areas limit */
648 if ((mpnt
->vm_start
< addr
&& mpnt
->vm_end
> addr
+len
)
649 && mm
->map_count
>= MAX_MAP_COUNT
)
653 * We may need one additional vma to fix up the mappings ...
654 * and this is the last chance for an easy error exit.
656 extra
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
660 npp
= (prev
? &prev
->vm_next
: &mm
->mmap
);
662 vmlist_modify_lock(mm
);
663 for ( ; mpnt
&& mpnt
->vm_start
< addr
+len
; mpnt
= *npp
) {
664 *npp
= mpnt
->vm_next
;
665 mpnt
->vm_next
= free
;
668 avl_remove(mpnt
, &mm
->mmap_avl
);
670 mm
->mmap_cache
= NULL
; /* Kill the cache. */
671 vmlist_modify_unlock(mm
);
673 /* Ok - we have the memory areas we should free on the 'free' list,
674 * so release them, and unmap the page range..
675 * If the one of the segments is only being partially unmapped,
676 * it will put new vm_area_struct(s) into the address space.
678 while ((mpnt
= free
) != NULL
) {
679 unsigned long st
, end
, size
;
681 free
= free
->vm_next
;
683 st
= addr
< mpnt
->vm_start
? mpnt
->vm_start
: addr
;
685 end
= end
> mpnt
->vm_end
? mpnt
->vm_end
: end
;
689 if (mpnt
->vm_ops
&& mpnt
->vm_ops
->unmap
)
690 mpnt
->vm_ops
->unmap(mpnt
, st
, size
);
693 remove_shared_vm_struct(mpnt
);
696 flush_cache_range(mm
, st
, end
);
697 zap_page_range(mm
, st
, size
);
698 flush_tlb_range(mm
, st
, end
);
701 * Fix the mapping, and free the old area if it wasn't reused.
703 extra
= unmap_fixup(mpnt
, st
, size
, extra
);
706 /* Release the extra vma struct if it wasn't used */
708 kmem_cache_free(vm_area_cachep
, extra
);
710 free_pgtables(mm
, prev
, addr
, addr
+len
);
715 asmlinkage
long sys_munmap(unsigned long addr
, size_t len
)
719 down(¤t
->mm
->mmap_sem
);
720 ret
= do_munmap(addr
, len
);
721 up(¤t
->mm
->mmap_sem
);
726 * this is really a simplified "do_mmap". it only handles
727 * anonymous maps. eventually we may be able to do some
728 * brk-specific accounting here.
730 unsigned long do_brk(unsigned long addr
, unsigned long len
)
732 struct mm_struct
* mm
= current
->mm
;
733 struct vm_area_struct
* vma
;
734 unsigned long flags
, retval
;
736 len
= PAGE_ALIGN(len
);
743 if (mm
->def_flags
& VM_LOCKED
) {
744 unsigned long locked
= mm
->locked_vm
<< PAGE_SHIFT
;
746 if (locked
> current
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
)
751 * Clear old maps. this also does some error checking for us
753 retval
= do_munmap(addr
, len
);
757 /* Check against address space limits *after* clearing old maps... */
758 if ((mm
->total_vm
<< PAGE_SHIFT
) + len
759 > current
->rlim
[RLIMIT_AS
].rlim_cur
)
762 if (mm
->map_count
> MAX_MAP_COUNT
)
765 if (!vm_enough_memory(len
>> PAGE_SHIFT
))
769 * create a vma struct for an anonymous mapping
771 vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
776 vma
->vm_start
= addr
;
777 vma
->vm_end
= addr
+ len
;
778 vma
->vm_flags
= vm_flags(PROT_READ
|PROT_WRITE
|PROT_EXEC
,
779 MAP_FIXED
|MAP_PRIVATE
) | mm
->def_flags
;
781 vma
->vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
782 vma
->vm_page_prot
= protection_map
[vma
->vm_flags
& 0x0f];
786 vma
->vm_private_data
= NULL
;
789 * merge_segments may merge our vma, so we can't refer to it
790 * after the call. Save the values we need now ...
792 flags
= vma
->vm_flags
;
793 addr
= vma
->vm_start
;
795 vmlist_modify_lock(mm
);
796 insert_vm_struct(mm
, vma
);
797 merge_segments(mm
, vma
->vm_start
, vma
->vm_end
);
798 vmlist_modify_unlock(mm
);
800 mm
->total_vm
+= len
>> PAGE_SHIFT
;
801 if (flags
& VM_LOCKED
) {
802 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
803 make_pages_present(addr
, addr
+ len
);
808 /* Build the AVL tree corresponding to the VMA list. */
809 void build_mmap_avl(struct mm_struct
* mm
)
811 struct vm_area_struct
* vma
;
814 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
)
815 avl_insert(vma
, &mm
->mmap_avl
);
818 /* Release all mmaps. */
819 void exit_mmap(struct mm_struct
* mm
)
821 struct vm_area_struct
* mpnt
;
823 release_segments(mm
);
825 vmlist_modify_lock(mm
);
826 mm
->mmap
= mm
->mmap_avl
= mm
->mmap_cache
= NULL
;
827 vmlist_modify_unlock(mm
);
832 struct vm_area_struct
* next
= mpnt
->vm_next
;
833 unsigned long start
= mpnt
->vm_start
;
834 unsigned long end
= mpnt
->vm_end
;
835 unsigned long size
= end
- start
;
838 if (mpnt
->vm_ops
->unmap
)
839 mpnt
->vm_ops
->unmap(mpnt
, start
, size
);
840 if (mpnt
->vm_ops
->close
)
841 mpnt
->vm_ops
->close(mpnt
);
844 remove_shared_vm_struct(mpnt
);
845 zap_page_range(mm
, start
, size
);
848 kmem_cache_free(vm_area_cachep
, mpnt
);
852 /* This is just debugging */
854 printk("exit_mmap: map count is %d\n", mm
->map_count
);
856 clear_page_tables(mm
, 0, USER_PTRS_PER_PGD
);
859 /* Insert vm structure into process list sorted by address
860 * and into the inode's i_mmap ring.
862 void insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vmp
)
864 struct vm_area_struct
**pprev
;
869 while (*pprev
&& (*pprev
)->vm_start
<= vmp
->vm_start
)
870 pprev
= &(*pprev
)->vm_next
;
872 struct vm_area_struct
*prev
, *next
;
873 avl_insert_neighbours(vmp
, &mm
->mmap_avl
, &prev
, &next
);
874 pprev
= (prev
? &prev
->vm_next
: &mm
->mmap
);
876 printk("insert_vm_struct: tree inconsistent with list\n");
878 vmp
->vm_next
= *pprev
;
882 if (mm
->map_count
>= AVL_MIN_MAP_COUNT
&& !mm
->mmap_avl
)
887 struct inode
* inode
= file
->f_dentry
->d_inode
;
888 if (vmp
->vm_flags
& VM_DENYWRITE
)
889 atomic_dec(&inode
->i_writecount
);
891 /* insert vmp into inode's share list */
892 spin_lock(&inode
->i_shared_lock
);
893 if((vmp
->vm_next_share
= inode
->i_mmap
) != NULL
)
894 inode
->i_mmap
->vm_pprev_share
= &vmp
->vm_next_share
;
896 vmp
->vm_pprev_share
= &inode
->i_mmap
;
897 spin_unlock(&inode
->i_shared_lock
);
901 /* Merge the list of memory segments if possible.
902 * Redundant vm_area_structs are freed.
903 * This assumes that the list is ordered by address.
904 * We don't need to traverse the entire list, only those segments
905 * which intersect or are adjacent to a given interval.
907 * We must already hold the mm semaphore when we get here..
909 void merge_segments (struct mm_struct
* mm
, unsigned long start_addr
, unsigned long end_addr
)
911 struct vm_area_struct
*prev
, *mpnt
, *next
, *prev1
;
913 mpnt
= find_vma_prev(mm
, start_addr
, &prev1
);
921 mpnt
= mpnt
->vm_next
;
923 mm
->mmap_cache
= NULL
; /* Kill the cache. */
925 /* prev and mpnt cycle through the list, as long as
926 * start_addr < mpnt->vm_end && prev->vm_start < end_addr
928 for ( ; mpnt
&& prev
->vm_start
< end_addr
; prev
= mpnt
, mpnt
= next
) {
929 next
= mpnt
->vm_next
;
931 /* To share, we must have the same file, operations.. */
932 if ((mpnt
->vm_file
!= prev
->vm_file
)||
933 (mpnt
->vm_private_data
!= prev
->vm_private_data
) ||
934 (mpnt
->vm_ops
!= prev
->vm_ops
) ||
935 (mpnt
->vm_flags
!= prev
->vm_flags
) ||
936 (prev
->vm_end
!= mpnt
->vm_start
))
940 * If we have a file or it's a shared memory area
941 * the offsets must be contiguous..
943 if ((mpnt
->vm_file
!= NULL
) || (mpnt
->vm_flags
& VM_SHM
)) {
944 unsigned long off
= prev
->vm_pgoff
;
945 off
+= (prev
->vm_end
- prev
->vm_start
) >> PAGE_SHIFT
;
946 if (off
!= mpnt
->vm_pgoff
)
950 /* merge prev with mpnt and set up pointers so the new
951 * big segment can possibly merge with the next one.
952 * The old unused mpnt is freed.
955 avl_remove(mpnt
, &mm
->mmap_avl
);
956 prev
->vm_end
= mpnt
->vm_end
;
957 prev
->vm_next
= mpnt
->vm_next
;
958 if (mpnt
->vm_ops
&& mpnt
->vm_ops
->close
) {
959 mpnt
->vm_pgoff
+= (mpnt
->vm_end
- mpnt
->vm_start
) >> PAGE_SHIFT
;
960 mpnt
->vm_start
= mpnt
->vm_end
;
961 vmlist_modify_unlock(mm
);
962 mpnt
->vm_ops
->close(mpnt
);
963 vmlist_modify_lock(mm
);
966 remove_shared_vm_struct(mpnt
);
969 kmem_cache_free(vm_area_cachep
, mpnt
);
974 void __init
vma_init(void)
976 vm_area_cachep
= kmem_cache_create("vm_area_struct",
977 sizeof(struct vm_area_struct
),
978 0, SLAB_HWCACHE_ALIGN
,
981 panic("vma_init: Cannot alloc vm_area_struct cache.");
983 mm_cachep
= kmem_cache_create("mm_struct",
984 sizeof(struct mm_struct
),
985 0, SLAB_HWCACHE_ALIGN
,
988 panic("vma_init: Cannot alloc mm_struct cache.");