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>
19 #include <linux/blkdev.h>
20 #include <linux/swap.h>
21 #include <linux/swapops.h>
24 * Any behaviour which results in changes to the vma->vm_flags needs to
25 * take mmap_sem for writing. Others, which simply traverse vmas, need
26 * to only take it for reading.
28 static int madvise_need_mmap_write(int behavior
)
36 /* be safe, default to 1. list exceptions explicitly */
42 * We can potentially split a vm area into separate
43 * areas, each area with its own behavior.
45 static long madvise_behavior(struct vm_area_struct
*vma
,
46 struct vm_area_struct
**prev
,
47 unsigned long start
, unsigned long end
, int behavior
)
49 struct mm_struct
*mm
= vma
->vm_mm
;
52 unsigned long new_flags
= vma
->vm_flags
;
56 new_flags
= new_flags
& ~VM_RAND_READ
& ~VM_SEQ_READ
;
59 new_flags
= (new_flags
& ~VM_RAND_READ
) | VM_SEQ_READ
;
62 new_flags
= (new_flags
& ~VM_SEQ_READ
) | VM_RAND_READ
;
65 new_flags
|= VM_DONTCOPY
;
68 if (vma
->vm_flags
& VM_IO
) {
72 new_flags
&= ~VM_DONTCOPY
;
75 new_flags
|= VM_DONTDUMP
;
78 if (new_flags
& VM_SPECIAL
) {
82 new_flags
&= ~VM_DONTDUMP
;
85 case MADV_UNMERGEABLE
:
86 error
= ksm_madvise(vma
, start
, end
, behavior
, &new_flags
);
92 error
= hugepage_madvise(vma
, &new_flags
, behavior
);
98 if (new_flags
== vma
->vm_flags
) {
103 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
104 *prev
= vma_merge(mm
, *prev
, start
, end
, new_flags
, vma
->anon_vma
,
105 vma
->vm_file
, pgoff
, vma_policy(vma
));
113 if (start
!= vma
->vm_start
) {
114 error
= split_vma(mm
, vma
, start
, 1);
119 if (end
!= vma
->vm_end
) {
120 error
= split_vma(mm
, vma
, end
, 0);
127 * vm_flags is protected by the mmap_sem held in write mode.
129 vma
->vm_flags
= new_flags
;
132 if (error
== -ENOMEM
)
138 static int swapin_walk_pmd_entry(pmd_t
*pmd
, unsigned long start
,
139 unsigned long end
, struct mm_walk
*walk
)
142 struct vm_area_struct
*vma
= walk
->private;
145 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
148 for (index
= start
; index
!= end
; index
+= PAGE_SIZE
) {
154 orig_pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, start
, &ptl
);
155 pte
= *(orig_pte
+ ((index
- start
) / PAGE_SIZE
));
156 pte_unmap_unlock(orig_pte
, ptl
);
158 if (pte_present(pte
) || pte_none(pte
) || pte_file(pte
))
160 entry
= pte_to_swp_entry(pte
);
161 if (unlikely(non_swap_entry(entry
)))
164 page
= read_swap_cache_async(entry
, GFP_HIGHUSER_MOVABLE
,
167 page_cache_release(page
);
173 static void force_swapin_readahead(struct vm_area_struct
*vma
,
174 unsigned long start
, unsigned long end
)
176 struct mm_walk walk
= {
178 .pmd_entry
= swapin_walk_pmd_entry
,
182 walk_page_range(start
, end
, &walk
);
184 lru_add_drain(); /* Push any new pages onto the LRU now */
187 static void force_shm_swapin_readahead(struct vm_area_struct
*vma
,
188 unsigned long start
, unsigned long end
,
189 struct address_space
*mapping
)
195 for (; start
< end
; start
+= PAGE_SIZE
) {
196 index
= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
198 page
= find_get_entry(mapping
, index
);
199 if (!radix_tree_exceptional_entry(page
)) {
201 page_cache_release(page
);
204 swap
= radix_to_swp_entry(page
);
205 page
= read_swap_cache_async(swap
, GFP_HIGHUSER_MOVABLE
,
208 page_cache_release(page
);
211 lru_add_drain(); /* Push any new pages onto the LRU now */
213 #endif /* CONFIG_SWAP */
216 * Schedule all required I/O operations. Do not wait for completion.
218 static long madvise_willneed(struct vm_area_struct
*vma
,
219 struct vm_area_struct
**prev
,
220 unsigned long start
, unsigned long end
)
222 struct file
*file
= vma
->vm_file
;
225 if (!file
|| mapping_cap_swap_backed(file
->f_mapping
)) {
228 force_swapin_readahead(vma
, start
, end
);
230 force_shm_swapin_readahead(vma
, start
, end
,
239 if (file
->f_mapping
->a_ops
->get_xip_mem
) {
240 /* no bad return value, but ignore advice */
245 start
= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
246 if (end
> vma
->vm_end
)
248 end
= ((end
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
250 force_page_cache_readahead(file
->f_mapping
, file
, start
, end
- start
);
255 * Application no longer needs these pages. If the pages are dirty,
256 * it's OK to just throw them away. The app will be more careful about
257 * data it wants to keep. Be sure to free swap resources too. The
258 * zap_page_range call sets things up for shrink_active_list to actually free
259 * these pages later if no one else has touched them in the meantime,
260 * although we could add these pages to a global reuse list for
261 * shrink_active_list to pick up before reclaiming other pages.
263 * NB: This interface discards data rather than pushes it out to swap,
264 * as some implementations do. This has performance implications for
265 * applications like large transactional databases which want to discard
266 * pages in anonymous maps after committing to backing store the data
267 * that was kept in them. There is no reason to write this data out to
268 * the swap area if the application is discarding it.
270 * An interface that causes the system to free clean pages and flush
271 * dirty pages is already available as msync(MS_INVALIDATE).
273 static long madvise_dontneed(struct vm_area_struct
*vma
,
274 struct vm_area_struct
**prev
,
275 unsigned long start
, unsigned long end
)
278 if (vma
->vm_flags
& (VM_LOCKED
|VM_HUGETLB
|VM_PFNMAP
))
281 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
)) {
282 struct zap_details details
= {
283 .nonlinear_vma
= vma
,
284 .last_index
= ULONG_MAX
,
286 zap_page_range(vma
, start
, end
- start
, &details
);
288 zap_page_range(vma
, start
, end
- start
, NULL
);
293 * Application wants to free up the pages and associated backing store.
294 * This is effectively punching a hole into the middle of a file.
296 static long madvise_remove(struct vm_area_struct
*vma
,
297 struct vm_area_struct
**prev
,
298 unsigned long start
, unsigned long end
)
304 *prev
= NULL
; /* tell sys_madvise we drop mmap_sem */
306 if (vma
->vm_flags
& (VM_LOCKED
|VM_NONLINEAR
|VM_HUGETLB
))
311 if (!f
|| !f
->f_mapping
|| !f
->f_mapping
->host
) {
315 if ((vma
->vm_flags
& (VM_SHARED
|VM_WRITE
)) != (VM_SHARED
|VM_WRITE
))
318 offset
= (loff_t
)(start
- vma
->vm_start
)
319 + ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
322 * Filesystem's fallocate may need to take i_mutex. We need to
323 * explicitly grab a reference because the vma (and hence the
324 * vma's reference to the file) can go away as soon as we drop
328 up_read(¤t
->mm
->mmap_sem
);
329 error
= do_fallocate(f
,
330 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
331 offset
, end
- start
);
333 down_read(¤t
->mm
->mmap_sem
);
337 #ifdef CONFIG_MEMORY_FAILURE
339 * Error injection support for memory error handling.
341 static int madvise_hwpoison(int bhv
, unsigned long start
, unsigned long end
)
344 if (!capable(CAP_SYS_ADMIN
))
346 for (; start
< end
; start
+= PAGE_SIZE
<<
347 compound_order(compound_head(p
))) {
350 ret
= get_user_pages_fast(start
, 1, 0, &p
);
354 if (PageHWPoison(p
)) {
358 if (bhv
== MADV_SOFT_OFFLINE
) {
359 pr_info("Soft offlining page %#lx at %#lx\n",
360 page_to_pfn(p
), start
);
361 ret
= soft_offline_page(p
, MF_COUNT_INCREASED
);
366 pr_info("Injecting memory failure for page %#lx at %#lx\n",
367 page_to_pfn(p
), start
);
368 /* Ignore return value for now */
369 memory_failure(page_to_pfn(p
), 0, MF_COUNT_INCREASED
);
376 madvise_vma(struct vm_area_struct
*vma
, struct vm_area_struct
**prev
,
377 unsigned long start
, unsigned long end
, int behavior
)
381 return madvise_remove(vma
, prev
, start
, end
);
383 return madvise_willneed(vma
, prev
, start
, end
);
385 return madvise_dontneed(vma
, prev
, start
, end
);
387 return madvise_behavior(vma
, prev
, start
, end
, behavior
);
392 madvise_behavior_valid(int behavior
)
398 case MADV_SEQUENTIAL
:
405 case MADV_UNMERGEABLE
:
407 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
409 case MADV_NOHUGEPAGE
:
421 * The madvise(2) system call.
423 * Applications can use madvise() to advise the kernel how it should
424 * handle paging I/O in this VM area. The idea is to help the kernel
425 * use appropriate read-ahead and caching techniques. The information
426 * provided is advisory only, and can be safely disregarded by the
427 * kernel without affecting the correct operation of the application.
430 * MADV_NORMAL - the default behavior is to read clusters. This
431 * results in some read-ahead and read-behind.
432 * MADV_RANDOM - the system should read the minimum amount of data
433 * on any access, since it is unlikely that the appli-
434 * cation will need more than what it asks for.
435 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
436 * once, so they can be aggressively read ahead, and
437 * can be freed soon after they are accessed.
438 * MADV_WILLNEED - the application is notifying the system to read
440 * MADV_DONTNEED - the application is finished with the given range,
441 * so the kernel can free resources associated with it.
442 * MADV_REMOVE - the application wants to free up the given range of
443 * pages and associated backing store.
444 * MADV_DONTFORK - omit this area from child's address space when forking:
445 * typically, to avoid COWing pages pinned by get_user_pages().
446 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
447 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
448 * this area with pages of identical content from other such areas.
449 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
453 * -EINVAL - start + len < 0, start is not page-aligned,
454 * "behavior" is not a valid value, or application
455 * is attempting to release locked or shared pages.
456 * -ENOMEM - addresses in the specified range are not currently
457 * mapped, or are outside the AS of the process.
458 * -EIO - an I/O error occurred while paging in data.
459 * -EBADF - map exists, but area maps something that isn't a file.
460 * -EAGAIN - a kernel resource was temporarily unavailable.
462 SYSCALL_DEFINE3(madvise
, unsigned long, start
, size_t, len_in
, int, behavior
)
464 unsigned long end
, tmp
;
465 struct vm_area_struct
*vma
, *prev
;
466 int unmapped_error
= 0;
470 struct blk_plug plug
;
472 #ifdef CONFIG_MEMORY_FAILURE
473 if (behavior
== MADV_HWPOISON
|| behavior
== MADV_SOFT_OFFLINE
)
474 return madvise_hwpoison(behavior
, start
, start
+len_in
);
476 if (!madvise_behavior_valid(behavior
))
479 if (start
& ~PAGE_MASK
)
481 len
= (len_in
+ ~PAGE_MASK
) & PAGE_MASK
;
483 /* Check to see whether len was rounded up from small -ve to zero */
495 write
= madvise_need_mmap_write(behavior
);
497 down_write(¤t
->mm
->mmap_sem
);
499 down_read(¤t
->mm
->mmap_sem
);
502 * If the interval [start,end) covers some unmapped address
503 * ranges, just ignore them, but return -ENOMEM at the end.
504 * - different from the way of handling in mlock etc.
506 vma
= find_vma_prev(current
->mm
, start
, &prev
);
507 if (vma
&& start
> vma
->vm_start
)
510 blk_start_plug(&plug
);
512 /* Still start < end. */
517 /* Here start < (end|vma->vm_end). */
518 if (start
< vma
->vm_start
) {
519 unmapped_error
= -ENOMEM
;
520 start
= vma
->vm_start
;
525 /* Here vma->vm_start <= start < (end|vma->vm_end) */
530 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
531 error
= madvise_vma(vma
, &prev
, start
, tmp
, behavior
);
535 if (prev
&& start
< prev
->vm_end
)
536 start
= prev
->vm_end
;
537 error
= unmapped_error
;
542 else /* madvise_remove dropped mmap_sem */
543 vma
= find_vma(current
->mm
, start
);
546 blk_finish_plug(&plug
);
548 up_write(¤t
->mm
->mmap_sem
);
550 up_read(¤t
->mm
->mmap_sem
);
This page took 0.040639 seconds and 5 git commands to generate.