]>
git.proxmox.com Git - mirror_ubuntu-kernels.git/blob - mm/madvise.c
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/page-isolation.h>
13 #include <linux/hugetlb.h>
14 #include <linux/sched.h>
15 #include <linux/ksm.h>
18 * Any behaviour which results in changes to the vma->vm_flags needs to
19 * take mmap_sem for writing. Others, which simply traverse vmas, need
20 * to only take it for reading.
22 static int madvise_need_mmap_write(int behavior
)
30 /* be safe, default to 1. list exceptions explicitly */
36 * We can potentially split a vm area into separate
37 * areas, each area with its own behavior.
39 static long madvise_behavior(struct vm_area_struct
* vma
,
40 struct vm_area_struct
**prev
,
41 unsigned long start
, unsigned long end
, int behavior
)
43 struct mm_struct
* mm
= vma
->vm_mm
;
46 unsigned long new_flags
= vma
->vm_flags
;
50 new_flags
= new_flags
& ~VM_RAND_READ
& ~VM_SEQ_READ
;
53 new_flags
= (new_flags
& ~VM_RAND_READ
) | VM_SEQ_READ
;
56 new_flags
= (new_flags
& ~VM_SEQ_READ
) | VM_RAND_READ
;
59 new_flags
|= VM_DONTCOPY
;
62 if (vma
->vm_flags
& VM_IO
) {
66 new_flags
&= ~VM_DONTCOPY
;
69 case MADV_UNMERGEABLE
:
70 error
= ksm_madvise(vma
, start
, end
, behavior
, &new_flags
);
76 error
= hugepage_madvise(vma
, &new_flags
, behavior
);
82 if (new_flags
== vma
->vm_flags
) {
87 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
88 *prev
= vma_merge(mm
, *prev
, start
, end
, new_flags
, vma
->anon_vma
,
89 vma
->vm_file
, pgoff
, vma_policy(vma
));
97 if (start
!= vma
->vm_start
) {
98 error
= split_vma(mm
, vma
, start
, 1);
103 if (end
!= vma
->vm_end
) {
104 error
= split_vma(mm
, vma
, end
, 0);
111 * vm_flags is protected by the mmap_sem held in write mode.
113 vma
->vm_flags
= new_flags
;
116 if (error
== -ENOMEM
)
122 * Schedule all required I/O operations. Do not wait for completion.
124 static long madvise_willneed(struct vm_area_struct
* vma
,
125 struct vm_area_struct
** prev
,
126 unsigned long start
, unsigned long end
)
128 struct file
*file
= vma
->vm_file
;
133 if (file
->f_mapping
->a_ops
->get_xip_mem
) {
134 /* no bad return value, but ignore advice */
139 start
= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
140 if (end
> vma
->vm_end
)
142 end
= ((end
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
144 force_page_cache_readahead(file
->f_mapping
, file
, start
, end
- start
);
149 * Application no longer needs these pages. If the pages are dirty,
150 * it's OK to just throw them away. The app will be more careful about
151 * data it wants to keep. Be sure to free swap resources too. The
152 * zap_page_range call sets things up for shrink_active_list to actually free
153 * these pages later if no one else has touched them in the meantime,
154 * although we could add these pages to a global reuse list for
155 * shrink_active_list to pick up before reclaiming other pages.
157 * NB: This interface discards data rather than pushes it out to swap,
158 * as some implementations do. This has performance implications for
159 * applications like large transactional databases which want to discard
160 * pages in anonymous maps after committing to backing store the data
161 * that was kept in them. There is no reason to write this data out to
162 * the swap area if the application is discarding it.
164 * An interface that causes the system to free clean pages and flush
165 * dirty pages is already available as msync(MS_INVALIDATE).
167 static long madvise_dontneed(struct vm_area_struct
* vma
,
168 struct vm_area_struct
** prev
,
169 unsigned long start
, unsigned long end
)
172 if (vma
->vm_flags
& (VM_LOCKED
|VM_HUGETLB
|VM_PFNMAP
))
175 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
)) {
176 struct zap_details details
= {
177 .nonlinear_vma
= vma
,
178 .last_index
= ULONG_MAX
,
180 zap_page_range(vma
, start
, end
- start
, &details
);
182 zap_page_range(vma
, start
, end
- start
, NULL
);
187 * Application wants to free up the pages and associated backing store.
188 * This is effectively punching a hole into the middle of a file.
190 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
191 * Other filesystems return -ENOSYS.
193 static long madvise_remove(struct vm_area_struct
*vma
,
194 struct vm_area_struct
**prev
,
195 unsigned long start
, unsigned long end
)
197 struct address_space
*mapping
;
198 loff_t offset
, endoff
;
201 *prev
= NULL
; /* tell sys_madvise we drop mmap_sem */
203 if (vma
->vm_flags
& (VM_LOCKED
|VM_NONLINEAR
|VM_HUGETLB
))
206 if (!vma
->vm_file
|| !vma
->vm_file
->f_mapping
207 || !vma
->vm_file
->f_mapping
->host
) {
211 if ((vma
->vm_flags
& (VM_SHARED
|VM_WRITE
)) != (VM_SHARED
|VM_WRITE
))
214 mapping
= vma
->vm_file
->f_mapping
;
216 offset
= (loff_t
)(start
- vma
->vm_start
)
217 + ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
218 endoff
= (loff_t
)(end
- vma
->vm_start
- 1)
219 + ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
221 /* vmtruncate_range needs to take i_mutex */
222 up_read(¤t
->mm
->mmap_sem
);
223 error
= vmtruncate_range(mapping
->host
, offset
, endoff
);
224 down_read(¤t
->mm
->mmap_sem
);
228 #ifdef CONFIG_MEMORY_FAILURE
230 * Error injection support for memory error handling.
232 static int madvise_hwpoison(int bhv
, unsigned long start
, unsigned long end
)
236 if (!capable(CAP_SYS_ADMIN
))
238 for (; start
< end
; start
+= PAGE_SIZE
) {
240 int ret
= get_user_pages_fast(start
, 1, 0, &p
);
243 if (bhv
== MADV_SOFT_OFFLINE
) {
244 printk(KERN_INFO
"Soft offlining page %lx at %lx\n",
245 page_to_pfn(p
), start
);
246 ret
= soft_offline_page(p
, MF_COUNT_INCREASED
);
251 printk(KERN_INFO
"Injecting memory failure for page %lx at %lx\n",
252 page_to_pfn(p
), start
);
253 /* Ignore return value for now */
254 __memory_failure(page_to_pfn(p
), 0, MF_COUNT_INCREASED
);
261 madvise_vma(struct vm_area_struct
*vma
, struct vm_area_struct
**prev
,
262 unsigned long start
, unsigned long end
, int behavior
)
266 return madvise_remove(vma
, prev
, start
, end
);
268 return madvise_willneed(vma
, prev
, start
, end
);
270 return madvise_dontneed(vma
, prev
, start
, end
);
272 return madvise_behavior(vma
, prev
, start
, end
, behavior
);
277 madvise_behavior_valid(int behavior
)
283 case MADV_SEQUENTIAL
:
290 case MADV_UNMERGEABLE
:
292 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
294 case MADV_NOHUGEPAGE
:
304 * The madvise(2) system call.
306 * Applications can use madvise() to advise the kernel how it should
307 * handle paging I/O in this VM area. The idea is to help the kernel
308 * use appropriate read-ahead and caching techniques. The information
309 * provided is advisory only, and can be safely disregarded by the
310 * kernel without affecting the correct operation of the application.
313 * MADV_NORMAL - the default behavior is to read clusters. This
314 * results in some read-ahead and read-behind.
315 * MADV_RANDOM - the system should read the minimum amount of data
316 * on any access, since it is unlikely that the appli-
317 * cation will need more than what it asks for.
318 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
319 * once, so they can be aggressively read ahead, and
320 * can be freed soon after they are accessed.
321 * MADV_WILLNEED - the application is notifying the system to read
323 * MADV_DONTNEED - the application is finished with the given range,
324 * so the kernel can free resources associated with it.
325 * MADV_REMOVE - the application wants to free up the given range of
326 * pages and associated backing store.
327 * MADV_DONTFORK - omit this area from child's address space when forking:
328 * typically, to avoid COWing pages pinned by get_user_pages().
329 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
330 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
331 * this area with pages of identical content from other such areas.
332 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
336 * -EINVAL - start + len < 0, start is not page-aligned,
337 * "behavior" is not a valid value, or application
338 * is attempting to release locked or shared pages.
339 * -ENOMEM - addresses in the specified range are not currently
340 * mapped, or are outside the AS of the process.
341 * -EIO - an I/O error occurred while paging in data.
342 * -EBADF - map exists, but area maps something that isn't a file.
343 * -EAGAIN - a kernel resource was temporarily unavailable.
345 SYSCALL_DEFINE3(madvise
, unsigned long, start
, size_t, len_in
, int, behavior
)
347 unsigned long end
, tmp
;
348 struct vm_area_struct
* vma
, *prev
;
349 int unmapped_error
= 0;
354 #ifdef CONFIG_MEMORY_FAILURE
355 if (behavior
== MADV_HWPOISON
|| behavior
== MADV_SOFT_OFFLINE
)
356 return madvise_hwpoison(behavior
, start
, start
+len_in
);
358 if (!madvise_behavior_valid(behavior
))
361 write
= madvise_need_mmap_write(behavior
);
363 down_write(¤t
->mm
->mmap_sem
);
365 down_read(¤t
->mm
->mmap_sem
);
367 if (start
& ~PAGE_MASK
)
369 len
= (len_in
+ ~PAGE_MASK
) & PAGE_MASK
;
371 /* Check to see whether len was rounded up from small -ve to zero */
384 * If the interval [start,end) covers some unmapped address
385 * ranges, just ignore them, but return -ENOMEM at the end.
386 * - different from the way of handling in mlock etc.
388 vma
= find_vma_prev(current
->mm
, start
, &prev
);
389 if (vma
&& start
> vma
->vm_start
)
393 /* Still start < end. */
398 /* Here start < (end|vma->vm_end). */
399 if (start
< vma
->vm_start
) {
400 unmapped_error
= -ENOMEM
;
401 start
= vma
->vm_start
;
406 /* Here vma->vm_start <= start < (end|vma->vm_end) */
411 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
412 error
= madvise_vma(vma
, &prev
, start
, tmp
, behavior
);
416 if (prev
&& start
< prev
->vm_end
)
417 start
= prev
->vm_end
;
418 error
= unmapped_error
;
423 else /* madvise_remove dropped mmap_sem */
424 vma
= find_vma(current
->mm
, start
);
428 up_write(¤t
->mm
->mmap_sem
);
430 up_read(¤t
->mm
->mmap_sem
);