4 * Copyright (C) 1999 Linus Torvalds
5 * Copyright (C) 2002 Christoph Hellwig
8 #include <linux/mman.h>
9 #include <linux/pagemap.h>
10 #include <linux/syscalls.h>
11 #include <linux/mempolicy.h>
12 #include <linux/hugetlb.h>
13 #include <linux/sched.h>
16 * Any behaviour which results in changes to the vma->vm_flags needs to
17 * take mmap_sem for writing. Others, which simply traverse vmas, need
18 * to only take it for reading.
20 static int madvise_need_mmap_write(int behavior
)
28 /* be safe, default to 1. list exceptions explicitly */
34 * We can potentially split a vm area into separate
35 * areas, each area with its own behavior.
37 static long madvise_behavior(struct vm_area_struct
* vma
,
38 struct vm_area_struct
**prev
,
39 unsigned long start
, unsigned long end
, int behavior
)
41 struct mm_struct
* mm
= vma
->vm_mm
;
44 unsigned long new_flags
= vma
->vm_flags
;
48 new_flags
= new_flags
& ~VM_RAND_READ
& ~VM_SEQ_READ
;
51 new_flags
= (new_flags
& ~VM_RAND_READ
) | VM_SEQ_READ
;
54 new_flags
= (new_flags
& ~VM_SEQ_READ
) | VM_RAND_READ
;
57 new_flags
|= VM_DONTCOPY
;
60 if (vma
->vm_flags
& VM_IO
) {
64 new_flags
&= ~VM_DONTCOPY
;
68 if (new_flags
== vma
->vm_flags
) {
73 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
74 *prev
= vma_merge(mm
, *prev
, start
, end
, new_flags
, vma
->anon_vma
,
75 vma
->vm_file
, pgoff
, vma_policy(vma
));
83 if (start
!= vma
->vm_start
) {
84 error
= split_vma(mm
, vma
, start
, 1);
89 if (end
!= vma
->vm_end
) {
90 error
= split_vma(mm
, vma
, end
, 0);
97 * vm_flags is protected by the mmap_sem held in write mode.
99 vma
->vm_flags
= new_flags
;
102 if (error
== -ENOMEM
)
108 * Schedule all required I/O operations. Do not wait for completion.
110 static long madvise_willneed(struct vm_area_struct
* vma
,
111 struct vm_area_struct
** prev
,
112 unsigned long start
, unsigned long end
)
114 struct file
*file
= vma
->vm_file
;
119 if (file
->f_mapping
->a_ops
->get_xip_mem
) {
120 /* no bad return value, but ignore advice */
125 start
= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
126 if (end
> vma
->vm_end
)
128 end
= ((end
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
130 force_page_cache_readahead(file
->f_mapping
, file
, start
, end
- start
);
135 * Application no longer needs these pages. If the pages are dirty,
136 * it's OK to just throw them away. The app will be more careful about
137 * data it wants to keep. Be sure to free swap resources too. The
138 * zap_page_range call sets things up for shrink_active_list to actually free
139 * these pages later if no one else has touched them in the meantime,
140 * although we could add these pages to a global reuse list for
141 * shrink_active_list to pick up before reclaiming other pages.
143 * NB: This interface discards data rather than pushes it out to swap,
144 * as some implementations do. This has performance implications for
145 * applications like large transactional databases which want to discard
146 * pages in anonymous maps after committing to backing store the data
147 * that was kept in them. There is no reason to write this data out to
148 * the swap area if the application is discarding it.
150 * An interface that causes the system to free clean pages and flush
151 * dirty pages is already available as msync(MS_INVALIDATE).
153 static long madvise_dontneed(struct vm_area_struct
* vma
,
154 struct vm_area_struct
** prev
,
155 unsigned long start
, unsigned long end
)
158 if (vma
->vm_flags
& (VM_LOCKED
|VM_HUGETLB
|VM_PFNMAP
))
161 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
)) {
162 struct zap_details details
= {
163 .nonlinear_vma
= vma
,
164 .last_index
= ULONG_MAX
,
166 zap_page_range(vma
, start
, end
- start
, &details
);
168 zap_page_range(vma
, start
, end
- start
, NULL
);
173 * Application wants to free up the pages and associated backing store.
174 * This is effectively punching a hole into the middle of a file.
176 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
177 * Other filesystems return -ENOSYS.
179 static long madvise_remove(struct vm_area_struct
*vma
,
180 struct vm_area_struct
**prev
,
181 unsigned long start
, unsigned long end
)
183 struct address_space
*mapping
;
184 loff_t offset
, endoff
;
187 *prev
= NULL
; /* tell sys_madvise we drop mmap_sem */
189 if (vma
->vm_flags
& (VM_LOCKED
|VM_NONLINEAR
|VM_HUGETLB
))
192 if (!vma
->vm_file
|| !vma
->vm_file
->f_mapping
193 || !vma
->vm_file
->f_mapping
->host
) {
197 if ((vma
->vm_flags
& (VM_SHARED
|VM_WRITE
)) != (VM_SHARED
|VM_WRITE
))
200 mapping
= vma
->vm_file
->f_mapping
;
202 offset
= (loff_t
)(start
- vma
->vm_start
)
203 + ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
204 endoff
= (loff_t
)(end
- vma
->vm_start
- 1)
205 + ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
207 /* vmtruncate_range needs to take i_mutex and i_alloc_sem */
208 up_read(¤t
->mm
->mmap_sem
);
209 error
= vmtruncate_range(mapping
->host
, offset
, endoff
);
210 down_read(¤t
->mm
->mmap_sem
);
215 madvise_vma(struct vm_area_struct
*vma
, struct vm_area_struct
**prev
,
216 unsigned long start
, unsigned long end
, int behavior
)
220 return madvise_remove(vma
, prev
, start
, end
);
222 return madvise_willneed(vma
, prev
, start
, end
);
224 return madvise_dontneed(vma
, prev
, start
, end
);
226 return madvise_behavior(vma
, prev
, start
, end
, behavior
);
231 madvise_behavior_valid(int behavior
)
237 case MADV_SEQUENTIAL
:
250 * The madvise(2) system call.
252 * Applications can use madvise() to advise the kernel how it should
253 * handle paging I/O in this VM area. The idea is to help the kernel
254 * use appropriate read-ahead and caching techniques. The information
255 * provided is advisory only, and can be safely disregarded by the
256 * kernel without affecting the correct operation of the application.
259 * MADV_NORMAL - the default behavior is to read clusters. This
260 * results in some read-ahead and read-behind.
261 * MADV_RANDOM - the system should read the minimum amount of data
262 * on any access, since it is unlikely that the appli-
263 * cation will need more than what it asks for.
264 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
265 * once, so they can be aggressively read ahead, and
266 * can be freed soon after they are accessed.
267 * MADV_WILLNEED - the application is notifying the system to read
269 * MADV_DONTNEED - the application is finished with the given range,
270 * so the kernel can free resources associated with it.
271 * MADV_REMOVE - the application wants to free up the given range of
272 * pages and associated backing store.
273 * MADV_DONTFORK - omit this area from child's address space when forking:
274 * typically, to avoid COWing pages pinned by get_user_pages().
275 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
279 * -EINVAL - start + len < 0, start is not page-aligned,
280 * "behavior" is not a valid value, or application
281 * is attempting to release locked or shared pages.
282 * -ENOMEM - addresses in the specified range are not currently
283 * mapped, or are outside the AS of the process.
284 * -EIO - an I/O error occurred while paging in data.
285 * -EBADF - map exists, but area maps something that isn't a file.
286 * -EAGAIN - a kernel resource was temporarily unavailable.
288 SYSCALL_DEFINE3(madvise
, unsigned long, start
, size_t, len_in
, int, behavior
)
290 unsigned long end
, tmp
;
291 struct vm_area_struct
* vma
, *prev
;
292 int unmapped_error
= 0;
297 if (!madvise_behavior_valid(behavior
))
300 write
= madvise_need_mmap_write(behavior
);
302 down_write(¤t
->mm
->mmap_sem
);
304 down_read(¤t
->mm
->mmap_sem
);
306 if (start
& ~PAGE_MASK
)
308 len
= (len_in
+ ~PAGE_MASK
) & PAGE_MASK
;
310 /* Check to see whether len was rounded up from small -ve to zero */
323 * If the interval [start,end) covers some unmapped address
324 * ranges, just ignore them, but return -ENOMEM at the end.
325 * - different from the way of handling in mlock etc.
327 vma
= find_vma_prev(current
->mm
, start
, &prev
);
328 if (vma
&& start
> vma
->vm_start
)
332 /* Still start < end. */
337 /* Here start < (end|vma->vm_end). */
338 if (start
< vma
->vm_start
) {
339 unmapped_error
= -ENOMEM
;
340 start
= vma
->vm_start
;
345 /* Here vma->vm_start <= start < (end|vma->vm_end) */
350 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
351 error
= madvise_vma(vma
, &prev
, start
, tmp
, behavior
);
355 if (prev
&& start
< prev
->vm_end
)
356 start
= prev
->vm_end
;
357 error
= unmapped_error
;
362 else /* madvise_remove dropped mmap_sem */
363 vma
= find_vma(current
->mm
, start
);
367 up_write(¤t
->mm
->mmap_sem
);
369 up_read(¤t
->mm
->mmap_sem
);