4 * (C) Copyright 1996 Linus Torvalds
6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
7 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
11 #include <linux/hugetlb.h>
12 #include <linux/slab.h>
13 #include <linux/shm.h>
14 #include <linux/mman.h>
15 #include <linux/swap.h>
16 #include <linux/capability.h>
18 #include <linux/highmem.h>
19 #include <linux/security.h>
20 #include <linux/syscalls.h>
21 #include <linux/mmu_notifier.h>
23 #include <asm/uaccess.h>
24 #include <asm/cacheflush.h>
25 #include <asm/tlbflush.h>
29 static pmd_t
*get_old_pmd(struct mm_struct
*mm
, unsigned long addr
)
35 pgd
= pgd_offset(mm
, addr
);
36 if (pgd_none_or_clear_bad(pgd
))
39 pud
= pud_offset(pgd
, addr
);
40 if (pud_none_or_clear_bad(pud
))
43 pmd
= pmd_offset(pud
, addr
);
44 if (pmd_none_or_clear_bad(pmd
))
50 static pmd_t
*alloc_new_pmd(struct mm_struct
*mm
, unsigned long addr
)
56 pgd
= pgd_offset(mm
, addr
);
57 pud
= pud_alloc(mm
, pgd
, addr
);
61 pmd
= pmd_alloc(mm
, pud
, addr
);
65 if (!pmd_present(*pmd
) && __pte_alloc(mm
, pmd
, addr
))
71 static void move_ptes(struct vm_area_struct
*vma
, pmd_t
*old_pmd
,
72 unsigned long old_addr
, unsigned long old_end
,
73 struct vm_area_struct
*new_vma
, pmd_t
*new_pmd
,
74 unsigned long new_addr
)
76 struct address_space
*mapping
= NULL
;
77 struct mm_struct
*mm
= vma
->vm_mm
;
78 pte_t
*old_pte
, *new_pte
, pte
;
79 spinlock_t
*old_ptl
, *new_ptl
;
80 unsigned long old_start
;
83 mmu_notifier_invalidate_range_start(vma
->vm_mm
,
87 * Subtle point from Rajesh Venkatasubramanian: before
88 * moving file-based ptes, we must lock vmtruncate out,
89 * since it might clean the dst vma before the src vma,
90 * and we propagate stale pages into the dst afterward.
92 mapping
= vma
->vm_file
->f_mapping
;
93 spin_lock(&mapping
->i_mmap_lock
);
94 if (new_vma
->vm_truncate_count
&&
95 new_vma
->vm_truncate_count
!= vma
->vm_truncate_count
)
96 new_vma
->vm_truncate_count
= 0;
100 * We don't have to worry about the ordering of src and dst
101 * pte locks because exclusive mmap_sem prevents deadlock.
103 old_pte
= pte_offset_map_lock(mm
, old_pmd
, old_addr
, &old_ptl
);
104 new_pte
= pte_offset_map_nested(new_pmd
, new_addr
);
105 new_ptl
= pte_lockptr(mm
, new_pmd
);
106 if (new_ptl
!= old_ptl
)
107 spin_lock_nested(new_ptl
, SINGLE_DEPTH_NESTING
);
108 arch_enter_lazy_mmu_mode();
110 for (; old_addr
< old_end
; old_pte
++, old_addr
+= PAGE_SIZE
,
111 new_pte
++, new_addr
+= PAGE_SIZE
) {
112 if (pte_none(*old_pte
))
114 pte
= ptep_clear_flush(vma
, old_addr
, old_pte
);
115 pte
= move_pte(pte
, new_vma
->vm_page_prot
, old_addr
, new_addr
);
116 set_pte_at(mm
, new_addr
, new_pte
, pte
);
119 arch_leave_lazy_mmu_mode();
120 if (new_ptl
!= old_ptl
)
121 spin_unlock(new_ptl
);
122 pte_unmap_nested(new_pte
- 1);
123 pte_unmap_unlock(old_pte
- 1, old_ptl
);
125 spin_unlock(&mapping
->i_mmap_lock
);
126 mmu_notifier_invalidate_range_end(vma
->vm_mm
, old_start
, old_end
);
129 #define LATENCY_LIMIT (64 * PAGE_SIZE)
131 unsigned long move_page_tables(struct vm_area_struct
*vma
,
132 unsigned long old_addr
, struct vm_area_struct
*new_vma
,
133 unsigned long new_addr
, unsigned long len
)
135 unsigned long extent
, next
, old_end
;
136 pmd_t
*old_pmd
, *new_pmd
;
138 old_end
= old_addr
+ len
;
139 flush_cache_range(vma
, old_addr
, old_end
);
141 for (; old_addr
< old_end
; old_addr
+= extent
, new_addr
+= extent
) {
143 next
= (old_addr
+ PMD_SIZE
) & PMD_MASK
;
144 if (next
- 1 > old_end
)
146 extent
= next
- old_addr
;
147 old_pmd
= get_old_pmd(vma
->vm_mm
, old_addr
);
150 new_pmd
= alloc_new_pmd(vma
->vm_mm
, new_addr
);
153 next
= (new_addr
+ PMD_SIZE
) & PMD_MASK
;
154 if (extent
> next
- new_addr
)
155 extent
= next
- new_addr
;
156 if (extent
> LATENCY_LIMIT
)
157 extent
= LATENCY_LIMIT
;
158 move_ptes(vma
, old_pmd
, old_addr
, old_addr
+ extent
,
159 new_vma
, new_pmd
, new_addr
);
162 return len
+ old_addr
- old_end
; /* how much done */
165 static unsigned long move_vma(struct vm_area_struct
*vma
,
166 unsigned long old_addr
, unsigned long old_len
,
167 unsigned long new_len
, unsigned long new_addr
)
169 struct mm_struct
*mm
= vma
->vm_mm
;
170 struct vm_area_struct
*new_vma
;
171 unsigned long vm_flags
= vma
->vm_flags
;
172 unsigned long new_pgoff
;
173 unsigned long moved_len
;
174 unsigned long excess
= 0;
175 unsigned long hiwater_vm
;
179 * We'd prefer to avoid failure later on in do_munmap:
180 * which may split one vma into three before unmapping.
182 if (mm
->map_count
>= sysctl_max_map_count
- 3)
185 new_pgoff
= vma
->vm_pgoff
+ ((old_addr
- vma
->vm_start
) >> PAGE_SHIFT
);
186 new_vma
= copy_vma(&vma
, new_addr
, new_len
, new_pgoff
);
190 moved_len
= move_page_tables(vma
, old_addr
, new_vma
, new_addr
, old_len
);
191 if (moved_len
< old_len
) {
193 * On error, move entries back from new area to old,
194 * which will succeed since page tables still there,
195 * and then proceed to unmap new area instead of old.
197 move_page_tables(new_vma
, new_addr
, vma
, old_addr
, moved_len
);
204 /* Conceal VM_ACCOUNT so old reservation is not undone */
205 if (vm_flags
& VM_ACCOUNT
) {
206 vma
->vm_flags
&= ~VM_ACCOUNT
;
207 excess
= vma
->vm_end
- vma
->vm_start
- old_len
;
208 if (old_addr
> vma
->vm_start
&&
209 old_addr
+ old_len
< vma
->vm_end
)
214 * If we failed to move page tables we still do total_vm increment
215 * since do_munmap() will decrement it by old_len == new_len.
217 * Since total_vm is about to be raised artificially high for a
218 * moment, we need to restore high watermark afterwards: if stats
219 * are taken meanwhile, total_vm and hiwater_vm appear too high.
220 * If this were a serious issue, we'd add a flag to do_munmap().
222 hiwater_vm
= mm
->hiwater_vm
;
223 mm
->total_vm
+= new_len
>> PAGE_SHIFT
;
224 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, new_len
>>PAGE_SHIFT
);
226 if (do_munmap(mm
, old_addr
, old_len
) < 0) {
227 /* OOM: unable to split vma, just get accounts right */
228 vm_unacct_memory(excess
>> PAGE_SHIFT
);
231 mm
->hiwater_vm
= hiwater_vm
;
233 /* Restore VM_ACCOUNT if one or two pieces of vma left */
235 vma
->vm_flags
|= VM_ACCOUNT
;
237 vma
->vm_next
->vm_flags
|= VM_ACCOUNT
;
240 if (vm_flags
& VM_LOCKED
) {
241 mm
->locked_vm
+= new_len
>> PAGE_SHIFT
;
242 if (new_len
> old_len
)
243 mlock_vma_pages_range(new_vma
, new_addr
+ old_len
,
251 * Expand (or shrink) an existing mapping, potentially moving it at the
252 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
254 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
255 * This option implies MREMAP_MAYMOVE.
257 unsigned long do_mremap(unsigned long addr
,
258 unsigned long old_len
, unsigned long new_len
,
259 unsigned long flags
, unsigned long new_addr
)
261 struct mm_struct
*mm
= current
->mm
;
262 struct vm_area_struct
*vma
;
263 unsigned long ret
= -EINVAL
;
264 unsigned long charged
= 0;
266 if (flags
& ~(MREMAP_FIXED
| MREMAP_MAYMOVE
))
269 if (addr
& ~PAGE_MASK
)
272 old_len
= PAGE_ALIGN(old_len
);
273 new_len
= PAGE_ALIGN(new_len
);
276 * We allow a zero old-len as a special case
277 * for DOS-emu "duplicate shm area" thing. But
278 * a zero new-len is nonsensical.
283 /* new_addr is only valid if MREMAP_FIXED is specified */
284 if (flags
& MREMAP_FIXED
) {
285 if (new_addr
& ~PAGE_MASK
)
287 if (!(flags
& MREMAP_MAYMOVE
))
290 if (new_len
> TASK_SIZE
|| new_addr
> TASK_SIZE
- new_len
)
293 /* Check if the location we're moving into overlaps the
294 * old location at all, and fail if it does.
296 if ((new_addr
<= addr
) && (new_addr
+new_len
) > addr
)
299 if ((addr
<= new_addr
) && (addr
+old_len
) > new_addr
)
302 ret
= security_file_mmap(NULL
, 0, 0, 0, new_addr
, 1);
306 ret
= do_munmap(mm
, new_addr
, new_len
);
312 * Always allow a shrinking remap: that just unmaps
313 * the unnecessary pages..
314 * do_munmap does all the needed commit accounting
316 if (old_len
>= new_len
) {
317 ret
= do_munmap(mm
, addr
+new_len
, old_len
- new_len
);
318 if (ret
&& old_len
!= new_len
)
321 if (!(flags
& MREMAP_FIXED
) || (new_addr
== addr
))
327 * Ok, we need to grow.. or relocate.
330 vma
= find_vma(mm
, addr
);
331 if (!vma
|| vma
->vm_start
> addr
)
333 if (is_vm_hugetlb_page(vma
)) {
337 /* We can't remap across vm area boundaries */
338 if (old_len
> vma
->vm_end
- addr
)
340 if (vma
->vm_flags
& (VM_DONTEXPAND
| VM_PFNMAP
)) {
341 if (new_len
> old_len
)
344 if (vma
->vm_flags
& VM_LOCKED
) {
345 unsigned long locked
, lock_limit
;
346 locked
= mm
->locked_vm
<< PAGE_SHIFT
;
347 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
348 locked
+= new_len
- old_len
;
350 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
353 if (!may_expand_vm(mm
, (new_len
- old_len
) >> PAGE_SHIFT
)) {
358 if (vma
->vm_flags
& VM_ACCOUNT
) {
359 charged
= (new_len
- old_len
) >> PAGE_SHIFT
;
360 if (security_vm_enough_memory(charged
))
364 /* old_len exactly to the end of the area..
365 * And we're not relocating the area.
367 if (old_len
== vma
->vm_end
- addr
&&
368 !((flags
& MREMAP_FIXED
) && (addr
!= new_addr
)) &&
369 (old_len
!= new_len
|| !(flags
& MREMAP_MAYMOVE
))) {
370 unsigned long max_addr
= TASK_SIZE
;
372 max_addr
= vma
->vm_next
->vm_start
;
373 /* can we just expand the current mapping? */
374 if (max_addr
- addr
>= new_len
) {
375 int pages
= (new_len
- old_len
) >> PAGE_SHIFT
;
377 vma_adjust(vma
, vma
->vm_start
,
378 addr
+ new_len
, vma
->vm_pgoff
, NULL
);
380 mm
->total_vm
+= pages
;
381 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, pages
);
382 if (vma
->vm_flags
& VM_LOCKED
) {
383 mm
->locked_vm
+= pages
;
384 mlock_vma_pages_range(vma
, addr
+ old_len
,
393 * We weren't able to just expand or shrink the area,
394 * we need to create a new one and move it..
397 if (flags
& MREMAP_MAYMOVE
) {
398 if (!(flags
& MREMAP_FIXED
)) {
399 unsigned long map_flags
= 0;
400 if (vma
->vm_flags
& VM_MAYSHARE
)
401 map_flags
|= MAP_SHARED
;
403 new_addr
= get_unmapped_area(vma
->vm_file
, 0, new_len
,
404 vma
->vm_pgoff
, map_flags
);
405 if (new_addr
& ~PAGE_MASK
) {
410 ret
= security_file_mmap(NULL
, 0, 0, 0, new_addr
, 1);
414 ret
= move_vma(vma
, addr
, old_len
, new_len
, new_addr
);
417 if (ret
& ~PAGE_MASK
)
418 vm_unacct_memory(charged
);
423 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
424 unsigned long, new_len
, unsigned long, flags
,
425 unsigned long, new_addr
)
429 down_write(¤t
->mm
->mmap_sem
);
430 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
431 up_write(¤t
->mm
->mmap_sem
);
This page took 0.039983 seconds and 5 git commands to generate.