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/backing-dev.h>
21 #include <linux/swap.h>
22 #include <linux/swapops.h>
23 #include <linux/mmu_notifier.h>
28 * Any behaviour which results in changes to the vma->vm_flags needs to
29 * take mmap_sem for writing. Others, which simply traverse vmas, need
30 * to only take it for reading.
32 static int madvise_need_mmap_write(int behavior
)
41 /* be safe, default to 1. list exceptions explicitly */
47 * We can potentially split a vm area into separate
48 * areas, each area with its own behavior.
50 static long madvise_behavior(struct vm_area_struct
*vma
,
51 struct vm_area_struct
**prev
,
52 unsigned long start
, unsigned long end
, int behavior
)
54 struct mm_struct
*mm
= vma
->vm_mm
;
57 unsigned long new_flags
= vma
->vm_flags
;
61 new_flags
= new_flags
& ~VM_RAND_READ
& ~VM_SEQ_READ
;
64 new_flags
= (new_flags
& ~VM_RAND_READ
) | VM_SEQ_READ
;
67 new_flags
= (new_flags
& ~VM_SEQ_READ
) | VM_RAND_READ
;
70 new_flags
|= VM_DONTCOPY
;
73 if (vma
->vm_flags
& VM_IO
) {
77 new_flags
&= ~VM_DONTCOPY
;
80 new_flags
|= VM_DONTDUMP
;
83 if (new_flags
& VM_SPECIAL
) {
87 new_flags
&= ~VM_DONTDUMP
;
90 case MADV_UNMERGEABLE
:
91 error
= ksm_madvise(vma
, start
, end
, behavior
, &new_flags
);
97 error
= hugepage_madvise(vma
, &new_flags
, behavior
);
103 if (new_flags
== vma
->vm_flags
) {
108 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
109 *prev
= vma_merge(mm
, *prev
, start
, end
, new_flags
, vma
->anon_vma
,
110 vma
->vm_file
, pgoff
, vma_policy(vma
),
111 vma
->vm_userfaultfd_ctx
);
119 if (start
!= vma
->vm_start
) {
120 error
= split_vma(mm
, vma
, start
, 1);
125 if (end
!= vma
->vm_end
) {
126 error
= split_vma(mm
, vma
, end
, 0);
133 * vm_flags is protected by the mmap_sem held in write mode.
135 vma
->vm_flags
= new_flags
;
138 if (error
== -ENOMEM
)
144 static int swapin_walk_pmd_entry(pmd_t
*pmd
, unsigned long start
,
145 unsigned long end
, struct mm_walk
*walk
)
148 struct vm_area_struct
*vma
= walk
->private;
151 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
154 for (index
= start
; index
!= end
; index
+= PAGE_SIZE
) {
160 orig_pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, start
, &ptl
);
161 pte
= *(orig_pte
+ ((index
- start
) / PAGE_SIZE
));
162 pte_unmap_unlock(orig_pte
, ptl
);
164 if (pte_present(pte
) || pte_none(pte
))
166 entry
= pte_to_swp_entry(pte
);
167 if (unlikely(non_swap_entry(entry
)))
170 page
= read_swap_cache_async(entry
, GFP_HIGHUSER_MOVABLE
,
173 page_cache_release(page
);
179 static void force_swapin_readahead(struct vm_area_struct
*vma
,
180 unsigned long start
, unsigned long end
)
182 struct mm_walk walk
= {
184 .pmd_entry
= swapin_walk_pmd_entry
,
188 walk_page_range(start
, end
, &walk
);
190 lru_add_drain(); /* Push any new pages onto the LRU now */
193 static void force_shm_swapin_readahead(struct vm_area_struct
*vma
,
194 unsigned long start
, unsigned long end
,
195 struct address_space
*mapping
)
201 for (; start
< end
; start
+= PAGE_SIZE
) {
202 index
= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
204 page
= find_get_entry(mapping
, index
);
205 if (!radix_tree_exceptional_entry(page
)) {
207 page_cache_release(page
);
210 swap
= radix_to_swp_entry(page
);
211 page
= read_swap_cache_async(swap
, GFP_HIGHUSER_MOVABLE
,
214 page_cache_release(page
);
217 lru_add_drain(); /* Push any new pages onto the LRU now */
219 #endif /* CONFIG_SWAP */
222 * Schedule all required I/O operations. Do not wait for completion.
224 static long madvise_willneed(struct vm_area_struct
*vma
,
225 struct vm_area_struct
**prev
,
226 unsigned long start
, unsigned long end
)
228 struct file
*file
= vma
->vm_file
;
233 force_swapin_readahead(vma
, start
, end
);
237 if (shmem_mapping(file
->f_mapping
)) {
239 force_shm_swapin_readahead(vma
, start
, end
,
248 if (IS_DAX(file_inode(file
))) {
249 /* no bad return value, but ignore advice */
254 start
= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
255 if (end
> vma
->vm_end
)
257 end
= ((end
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
259 force_page_cache_readahead(file
->f_mapping
, file
, start
, end
- start
);
263 static int madvise_free_pte_range(pmd_t
*pmd
, unsigned long addr
,
264 unsigned long end
, struct mm_walk
*walk
)
267 struct mmu_gather
*tlb
= walk
->private;
268 struct mm_struct
*mm
= tlb
->mm
;
269 struct vm_area_struct
*vma
= walk
->vma
;
271 pte_t
*orig_pte
, *pte
, ptent
;
274 split_huge_pmd(vma
, pmd
, addr
);
275 if (pmd_trans_unstable(pmd
))
278 orig_pte
= pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
279 arch_enter_lazy_mmu_mode();
280 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
) {
283 if (!pte_present(ptent
))
286 page
= vm_normal_page(vma
, addr
, ptent
);
291 * If pmd isn't transhuge but the page is THP and
292 * is owned by only this process, split it and
293 * deactivate all pages.
295 if (PageTransCompound(page
)) {
296 if (page_mapcount(page
) != 1)
299 if (!trylock_page(page
)) {
303 pte_unmap_unlock(orig_pte
, ptl
);
304 if (split_huge_page(page
)) {
307 pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
312 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
318 VM_BUG_ON_PAGE(PageTransCompound(page
), page
);
320 if (PageSwapCache(page
) || PageDirty(page
)) {
321 if (!trylock_page(page
))
324 * If page is shared with others, we couldn't clear
325 * PG_dirty of the page.
327 if (page_mapcount(page
) != 1) {
332 if (PageSwapCache(page
) && !try_to_free_swap(page
)) {
337 ClearPageDirty(page
);
341 if (pte_young(ptent
) || pte_dirty(ptent
)) {
343 * Some of architecture(ex, PPC) don't update TLB
344 * with set_pte_at and tlb_remove_tlb_entry so for
345 * the portability, remap the pte with old|clean
346 * after pte clearing.
348 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
351 ptent
= pte_mkold(ptent
);
352 ptent
= pte_mkclean(ptent
);
353 set_pte_at(mm
, addr
, pte
, ptent
);
354 tlb_remove_tlb_entry(tlb
, pte
, addr
);
358 arch_leave_lazy_mmu_mode();
359 pte_unmap_unlock(orig_pte
, ptl
);
364 static void madvise_free_page_range(struct mmu_gather
*tlb
,
365 struct vm_area_struct
*vma
,
366 unsigned long addr
, unsigned long end
)
368 struct mm_walk free_walk
= {
369 .pmd_entry
= madvise_free_pte_range
,
374 tlb_start_vma(tlb
, vma
);
375 walk_page_range(addr
, end
, &free_walk
);
376 tlb_end_vma(tlb
, vma
);
379 static int madvise_free_single_vma(struct vm_area_struct
*vma
,
380 unsigned long start_addr
, unsigned long end_addr
)
382 unsigned long start
, end
;
383 struct mm_struct
*mm
= vma
->vm_mm
;
384 struct mmu_gather tlb
;
386 if (vma
->vm_flags
& (VM_LOCKED
|VM_HUGETLB
|VM_PFNMAP
))
389 /* MADV_FREE works for only anon vma at the moment */
390 if (!vma_is_anonymous(vma
))
393 start
= max(vma
->vm_start
, start_addr
);
394 if (start
>= vma
->vm_end
)
396 end
= min(vma
->vm_end
, end_addr
);
397 if (end
<= vma
->vm_start
)
401 tlb_gather_mmu(&tlb
, mm
, start
, end
);
402 update_hiwater_rss(mm
);
404 mmu_notifier_invalidate_range_start(mm
, start
, end
);
405 madvise_free_page_range(&tlb
, vma
, start
, end
);
406 mmu_notifier_invalidate_range_end(mm
, start
, end
);
407 tlb_finish_mmu(&tlb
, start
, end
);
412 static long madvise_free(struct vm_area_struct
*vma
,
413 struct vm_area_struct
**prev
,
414 unsigned long start
, unsigned long end
)
417 return madvise_free_single_vma(vma
, start
, end
);
421 * Application no longer needs these pages. If the pages are dirty,
422 * it's OK to just throw them away. The app will be more careful about
423 * data it wants to keep. Be sure to free swap resources too. The
424 * zap_page_range call sets things up for shrink_active_list to actually free
425 * these pages later if no one else has touched them in the meantime,
426 * although we could add these pages to a global reuse list for
427 * shrink_active_list to pick up before reclaiming other pages.
429 * NB: This interface discards data rather than pushes it out to swap,
430 * as some implementations do. This has performance implications for
431 * applications like large transactional databases which want to discard
432 * pages in anonymous maps after committing to backing store the data
433 * that was kept in them. There is no reason to write this data out to
434 * the swap area if the application is discarding it.
436 * An interface that causes the system to free clean pages and flush
437 * dirty pages is already available as msync(MS_INVALIDATE).
439 static long madvise_dontneed(struct vm_area_struct
*vma
,
440 struct vm_area_struct
**prev
,
441 unsigned long start
, unsigned long end
)
444 if (vma
->vm_flags
& (VM_LOCKED
|VM_HUGETLB
|VM_PFNMAP
))
447 zap_page_range(vma
, start
, end
- start
, NULL
);
452 * Application wants to free up the pages and associated backing store.
453 * This is effectively punching a hole into the middle of a file.
455 static long madvise_remove(struct vm_area_struct
*vma
,
456 struct vm_area_struct
**prev
,
457 unsigned long start
, unsigned long end
)
463 *prev
= NULL
; /* tell sys_madvise we drop mmap_sem */
465 if (vma
->vm_flags
& VM_LOCKED
)
470 if (!f
|| !f
->f_mapping
|| !f
->f_mapping
->host
) {
474 if ((vma
->vm_flags
& (VM_SHARED
|VM_WRITE
)) != (VM_SHARED
|VM_WRITE
))
477 offset
= (loff_t
)(start
- vma
->vm_start
)
478 + ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
481 * Filesystem's fallocate may need to take i_mutex. We need to
482 * explicitly grab a reference because the vma (and hence the
483 * vma's reference to the file) can go away as soon as we drop
487 up_read(¤t
->mm
->mmap_sem
);
488 error
= vfs_fallocate(f
,
489 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
490 offset
, end
- start
);
492 down_read(¤t
->mm
->mmap_sem
);
496 #ifdef CONFIG_MEMORY_FAILURE
498 * Error injection support for memory error handling.
500 static int madvise_hwpoison(int bhv
, unsigned long start
, unsigned long end
)
503 if (!capable(CAP_SYS_ADMIN
))
505 for (; start
< end
; start
+= PAGE_SIZE
<<
506 compound_order(compound_head(p
))) {
509 ret
= get_user_pages_fast(start
, 1, 0, &p
);
513 if (PageHWPoison(p
)) {
517 if (bhv
== MADV_SOFT_OFFLINE
) {
518 pr_info("Soft offlining page %#lx at %#lx\n",
519 page_to_pfn(p
), start
);
520 ret
= soft_offline_page(p
, MF_COUNT_INCREASED
);
525 pr_info("Injecting memory failure for page %#lx at %#lx\n",
526 page_to_pfn(p
), start
);
527 /* Ignore return value for now */
528 memory_failure(page_to_pfn(p
), 0, MF_COUNT_INCREASED
);
535 madvise_vma(struct vm_area_struct
*vma
, struct vm_area_struct
**prev
,
536 unsigned long start
, unsigned long end
, int behavior
)
540 return madvise_remove(vma
, prev
, start
, end
);
542 return madvise_willneed(vma
, prev
, start
, end
);
545 * XXX: In this implementation, MADV_FREE works like
546 * MADV_DONTNEED on swapless system or full swap.
548 if (get_nr_swap_pages() > 0)
549 return madvise_free(vma
, prev
, start
, end
);
552 return madvise_dontneed(vma
, prev
, start
, end
);
554 return madvise_behavior(vma
, prev
, start
, end
, behavior
);
559 madvise_behavior_valid(int behavior
)
565 case MADV_SEQUENTIAL
:
573 case MADV_UNMERGEABLE
:
575 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
577 case MADV_NOHUGEPAGE
:
589 * The madvise(2) system call.
591 * Applications can use madvise() to advise the kernel how it should
592 * handle paging I/O in this VM area. The idea is to help the kernel
593 * use appropriate read-ahead and caching techniques. The information
594 * provided is advisory only, and can be safely disregarded by the
595 * kernel without affecting the correct operation of the application.
598 * MADV_NORMAL - the default behavior is to read clusters. This
599 * results in some read-ahead and read-behind.
600 * MADV_RANDOM - the system should read the minimum amount of data
601 * on any access, since it is unlikely that the appli-
602 * cation will need more than what it asks for.
603 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
604 * once, so they can be aggressively read ahead, and
605 * can be freed soon after they are accessed.
606 * MADV_WILLNEED - the application is notifying the system to read
608 * MADV_DONTNEED - the application is finished with the given range,
609 * so the kernel can free resources associated with it.
610 * MADV_REMOVE - the application wants to free up the given range of
611 * pages and associated backing store.
612 * MADV_DONTFORK - omit this area from child's address space when forking:
613 * typically, to avoid COWing pages pinned by get_user_pages().
614 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
615 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
616 * this area with pages of identical content from other such areas.
617 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
621 * -EINVAL - start + len < 0, start is not page-aligned,
622 * "behavior" is not a valid value, or application
623 * is attempting to release locked or shared pages.
624 * -ENOMEM - addresses in the specified range are not currently
625 * mapped, or are outside the AS of the process.
626 * -EIO - an I/O error occurred while paging in data.
627 * -EBADF - map exists, but area maps something that isn't a file.
628 * -EAGAIN - a kernel resource was temporarily unavailable.
630 SYSCALL_DEFINE3(madvise
, unsigned long, start
, size_t, len_in
, int, behavior
)
632 unsigned long end
, tmp
;
633 struct vm_area_struct
*vma
, *prev
;
634 int unmapped_error
= 0;
638 struct blk_plug plug
;
640 #ifdef CONFIG_MEMORY_FAILURE
641 if (behavior
== MADV_HWPOISON
|| behavior
== MADV_SOFT_OFFLINE
)
642 return madvise_hwpoison(behavior
, start
, start
+len_in
);
644 if (!madvise_behavior_valid(behavior
))
647 if (start
& ~PAGE_MASK
)
649 len
= (len_in
+ ~PAGE_MASK
) & PAGE_MASK
;
651 /* Check to see whether len was rounded up from small -ve to zero */
663 write
= madvise_need_mmap_write(behavior
);
665 down_write(¤t
->mm
->mmap_sem
);
667 down_read(¤t
->mm
->mmap_sem
);
670 * If the interval [start,end) covers some unmapped address
671 * ranges, just ignore them, but return -ENOMEM at the end.
672 * - different from the way of handling in mlock etc.
674 vma
= find_vma_prev(current
->mm
, start
, &prev
);
675 if (vma
&& start
> vma
->vm_start
)
678 blk_start_plug(&plug
);
680 /* Still start < end. */
685 /* Here start < (end|vma->vm_end). */
686 if (start
< vma
->vm_start
) {
687 unmapped_error
= -ENOMEM
;
688 start
= vma
->vm_start
;
693 /* Here vma->vm_start <= start < (end|vma->vm_end) */
698 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
699 error
= madvise_vma(vma
, &prev
, start
, tmp
, behavior
);
703 if (prev
&& start
< prev
->vm_end
)
704 start
= prev
->vm_end
;
705 error
= unmapped_error
;
710 else /* madvise_remove dropped mmap_sem */
711 vma
= find_vma(current
->mm
, start
);
714 blk_finish_plug(&plug
);
716 up_write(¤t
->mm
->mmap_sem
);
718 up_read(¤t
->mm
->mmap_sem
);
This page took 0.044245 seconds and 6 git commands to generate.