]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.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/falloc.h>
15 #include <linux/sched.h>
16 #include <linux/ksm.h>
18 #include <linux/file.h>
21 * Any behaviour which results in changes to the vma->vm_flags needs to
22 * take mmap_sem for writing. Others, which simply traverse vmas, need
23 * to only take it for reading.
25 static int madvise_need_mmap_write(int behavior
)
33 /* be safe, default to 1. list exceptions explicitly */
39 * We can potentially split a vm area into separate
40 * areas, each area with its own behavior.
42 static long madvise_behavior(struct vm_area_struct
* vma
,
43 struct vm_area_struct
**prev
,
44 unsigned long start
, unsigned long end
, int behavior
)
46 struct mm_struct
* mm
= vma
->vm_mm
;
49 unsigned long new_flags
= vma
->vm_flags
;
53 new_flags
= new_flags
& ~VM_RAND_READ
& ~VM_SEQ_READ
;
56 new_flags
= (new_flags
& ~VM_RAND_READ
) | VM_SEQ_READ
;
59 new_flags
= (new_flags
& ~VM_SEQ_READ
) | VM_RAND_READ
;
62 new_flags
|= VM_DONTCOPY
;
65 if (vma
->vm_flags
& VM_IO
) {
69 new_flags
&= ~VM_DONTCOPY
;
72 new_flags
|= VM_DONTDUMP
;
75 if (new_flags
& VM_SPECIAL
) {
79 new_flags
&= ~VM_DONTDUMP
;
82 case MADV_UNMERGEABLE
:
83 error
= ksm_madvise(vma
, start
, end
, behavior
, &new_flags
);
89 error
= hugepage_madvise(vma
, &new_flags
, behavior
);
95 if (new_flags
== vma
->vm_flags
) {
100 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
101 *prev
= vma_merge(mm
, *prev
, start
, end
, new_flags
, vma
->anon_vma
,
102 vma
->vm_file
, pgoff
, vma_policy(vma
));
110 if (start
!= vma
->vm_start
) {
111 error
= split_vma(mm
, vma
, start
, 1);
116 if (end
!= vma
->vm_end
) {
117 error
= split_vma(mm
, vma
, end
, 0);
124 * vm_flags is protected by the mmap_sem held in write mode.
126 vma
->vm_flags
= new_flags
;
129 if (error
== -ENOMEM
)
135 * Schedule all required I/O operations. Do not wait for completion.
137 static long madvise_willneed(struct vm_area_struct
* vma
,
138 struct vm_area_struct
** prev
,
139 unsigned long start
, unsigned long end
)
141 struct file
*file
= vma
->vm_file
;
146 if (file
->f_mapping
->a_ops
->get_xip_mem
) {
147 /* no bad return value, but ignore advice */
152 start
= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
153 if (end
> vma
->vm_end
)
155 end
= ((end
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
157 force_page_cache_readahead(file
->f_mapping
, file
, start
, end
- start
);
162 * Application no longer needs these pages. If the pages are dirty,
163 * it's OK to just throw them away. The app will be more careful about
164 * data it wants to keep. Be sure to free swap resources too. The
165 * zap_page_range call sets things up for shrink_active_list to actually free
166 * these pages later if no one else has touched them in the meantime,
167 * although we could add these pages to a global reuse list for
168 * shrink_active_list to pick up before reclaiming other pages.
170 * NB: This interface discards data rather than pushes it out to swap,
171 * as some implementations do. This has performance implications for
172 * applications like large transactional databases which want to discard
173 * pages in anonymous maps after committing to backing store the data
174 * that was kept in them. There is no reason to write this data out to
175 * the swap area if the application is discarding it.
177 * An interface that causes the system to free clean pages and flush
178 * dirty pages is already available as msync(MS_INVALIDATE).
180 static long madvise_dontneed(struct vm_area_struct
* vma
,
181 struct vm_area_struct
** prev
,
182 unsigned long start
, unsigned long end
)
185 if (vma
->vm_flags
& (VM_LOCKED
|VM_HUGETLB
|VM_PFNMAP
))
188 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
)) {
189 struct zap_details details
= {
190 .nonlinear_vma
= vma
,
191 .last_index
= ULONG_MAX
,
193 zap_page_range(vma
, start
, end
- start
, &details
);
195 zap_page_range(vma
, start
, end
- start
, NULL
);
200 * Application wants to free up the pages and associated backing store.
201 * This is effectively punching a hole into the middle of a file.
203 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
204 * Other filesystems return -ENOSYS.
206 static long madvise_remove(struct vm_area_struct
*vma
,
207 struct vm_area_struct
**prev
,
208 unsigned long start
, unsigned long end
)
214 *prev
= NULL
; /* tell sys_madvise we drop mmap_sem */
216 if (vma
->vm_flags
& (VM_LOCKED
|VM_NONLINEAR
|VM_HUGETLB
))
221 if (!f
|| !f
->f_mapping
|| !f
->f_mapping
->host
) {
225 if ((vma
->vm_flags
& (VM_SHARED
|VM_WRITE
)) != (VM_SHARED
|VM_WRITE
))
228 offset
= (loff_t
)(start
- vma
->vm_start
)
229 + ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
232 * Filesystem's fallocate may need to take i_mutex. We need to
233 * explicitly grab a reference because the vma (and hence the
234 * vma's reference to the file) can go away as soon as we drop
238 up_read(¤t
->mm
->mmap_sem
);
239 error
= do_fallocate(f
,
240 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
241 offset
, end
- start
);
243 down_read(¤t
->mm
->mmap_sem
);
247 #ifdef CONFIG_MEMORY_FAILURE
249 * Error injection support for memory error handling.
251 static int madvise_hwpoison(int bhv
, unsigned long start
, unsigned long end
)
255 if (!capable(CAP_SYS_ADMIN
))
257 for (; start
< end
; start
+= PAGE_SIZE
) {
259 int ret
= get_user_pages_fast(start
, 1, 0, &p
);
262 if (bhv
== MADV_SOFT_OFFLINE
) {
263 printk(KERN_INFO
"Soft offlining page %lx at %lx\n",
264 page_to_pfn(p
), start
);
265 ret
= soft_offline_page(p
, MF_COUNT_INCREASED
);
270 printk(KERN_INFO
"Injecting memory failure for page %lx at %lx\n",
271 page_to_pfn(p
), start
);
272 /* Ignore return value for now */
273 memory_failure(page_to_pfn(p
), 0, MF_COUNT_INCREASED
);
280 madvise_vma(struct vm_area_struct
*vma
, struct vm_area_struct
**prev
,
281 unsigned long start
, unsigned long end
, int behavior
)
285 return madvise_remove(vma
, prev
, start
, end
);
287 return madvise_willneed(vma
, prev
, start
, end
);
289 return madvise_dontneed(vma
, prev
, start
, end
);
291 return madvise_behavior(vma
, prev
, start
, end
, behavior
);
296 madvise_behavior_valid(int behavior
)
302 case MADV_SEQUENTIAL
:
309 case MADV_UNMERGEABLE
:
311 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
313 case MADV_NOHUGEPAGE
:
325 * The madvise(2) system call.
327 * Applications can use madvise() to advise the kernel how it should
328 * handle paging I/O in this VM area. The idea is to help the kernel
329 * use appropriate read-ahead and caching techniques. The information
330 * provided is advisory only, and can be safely disregarded by the
331 * kernel without affecting the correct operation of the application.
334 * MADV_NORMAL - the default behavior is to read clusters. This
335 * results in some read-ahead and read-behind.
336 * MADV_RANDOM - the system should read the minimum amount of data
337 * on any access, since it is unlikely that the appli-
338 * cation will need more than what it asks for.
339 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
340 * once, so they can be aggressively read ahead, and
341 * can be freed soon after they are accessed.
342 * MADV_WILLNEED - the application is notifying the system to read
344 * MADV_DONTNEED - the application is finished with the given range,
345 * so the kernel can free resources associated with it.
346 * MADV_REMOVE - the application wants to free up the given range of
347 * pages and associated backing store.
348 * MADV_DONTFORK - omit this area from child's address space when forking:
349 * typically, to avoid COWing pages pinned by get_user_pages().
350 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
351 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
352 * this area with pages of identical content from other such areas.
353 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
357 * -EINVAL - start + len < 0, start is not page-aligned,
358 * "behavior" is not a valid value, or application
359 * is attempting to release locked or shared pages.
360 * -ENOMEM - addresses in the specified range are not currently
361 * mapped, or are outside the AS of the process.
362 * -EIO - an I/O error occurred while paging in data.
363 * -EBADF - map exists, but area maps something that isn't a file.
364 * -EAGAIN - a kernel resource was temporarily unavailable.
366 SYSCALL_DEFINE3(madvise
, unsigned long, start
, size_t, len_in
, int, behavior
)
368 unsigned long end
, tmp
;
369 struct vm_area_struct
* vma
, *prev
;
370 int unmapped_error
= 0;
375 #ifdef CONFIG_MEMORY_FAILURE
376 if (behavior
== MADV_HWPOISON
|| behavior
== MADV_SOFT_OFFLINE
)
377 return madvise_hwpoison(behavior
, start
, start
+len_in
);
379 if (!madvise_behavior_valid(behavior
))
382 write
= madvise_need_mmap_write(behavior
);
384 down_write(¤t
->mm
->mmap_sem
);
386 down_read(¤t
->mm
->mmap_sem
);
388 if (start
& ~PAGE_MASK
)
390 len
= (len_in
+ ~PAGE_MASK
) & PAGE_MASK
;
392 /* Check to see whether len was rounded up from small -ve to zero */
405 * If the interval [start,end) covers some unmapped address
406 * ranges, just ignore them, but return -ENOMEM at the end.
407 * - different from the way of handling in mlock etc.
409 vma
= find_vma_prev(current
->mm
, start
, &prev
);
410 if (vma
&& start
> vma
->vm_start
)
414 /* Still start < end. */
419 /* Here start < (end|vma->vm_end). */
420 if (start
< vma
->vm_start
) {
421 unmapped_error
= -ENOMEM
;
422 start
= vma
->vm_start
;
427 /* Here vma->vm_start <= start < (end|vma->vm_end) */
432 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
433 error
= madvise_vma(vma
, &prev
, start
, tmp
, behavior
);
437 if (prev
&& start
< prev
->vm_end
)
438 start
= prev
->vm_end
;
439 error
= unmapped_error
;
444 else /* madvise_remove dropped mmap_sem */
445 vma
= find_vma(current
->mm
, start
);
449 up_write(¤t
->mm
->mmap_sem
);
451 up_read(¤t
->mm
->mmap_sem
);