blob: 4a01c4bd786c7f839c24b91127c2ce0b37ac4217 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/madvise.c
3 *
4 * Copyright (C) 1999 Linus Torvalds
5 * Copyright (C) 2002 Christoph Hellwig
6 */
7
8#include <linux/mman.h>
9#include <linux/pagemap.h>
10#include <linux/syscalls.h>
Prasanna Meda05b74382005-06-21 17:14:37 -070011#include <linux/mempolicy.h>
Andi Kleenafcf9382009-12-16 12:20:00 +010012#include <linux/page-isolation.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/hugetlb.h>
Hugh Dickins3f31d072012-05-29 15:06:40 -070014#include <linux/falloc.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040015#include <linux/sched.h>
Hugh Dickinsf8af4da2009-09-21 17:01:57 -070016#include <linux/ksm.h>
Hugh Dickins3f31d072012-05-29 15:06:40 -070017#include <linux/fs.h>
Andy Lutomirski9ab42332012-07-05 16:00:11 -070018#include <linux/file.h>
Shaohua Li1998cc02013-02-22 16:32:31 -080019#include <linux/blkdev.h>
Tejun Heo66114ca2015-05-22 17:13:32 -040020#include <linux/backing-dev.h>
Shaohua Li1998cc02013-02-22 16:32:31 -080021#include <linux/swap.h>
22#include <linux/swapops.h>
Minchan Kim854e9ed2016-01-15 16:54:53 -080023#include <linux/mmu_notifier.h>
Mel Gorman5a1eef72017-08-02 13:31:52 -070024#include "internal.h"
Minchan Kim854e9ed2016-01-15 16:54:53 -080025
26#include <asm/tlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28/*
Nick Piggin0a27a142007-05-06 14:49:53 -070029 * Any behaviour which results in changes to the vma->vm_flags needs to
30 * take mmap_sem for writing. Others, which simply traverse vmas, need
31 * to only take it for reading.
32 */
33static int madvise_need_mmap_write(int behavior)
34{
35 switch (behavior) {
36 case MADV_REMOVE:
37 case MADV_WILLNEED:
38 case MADV_DONTNEED:
Minchan Kim854e9ed2016-01-15 16:54:53 -080039 case MADV_FREE:
Nick Piggin0a27a142007-05-06 14:49:53 -070040 return 0;
41 default:
42 /* be safe, default to 1. list exceptions explicitly */
43 return 1;
44 }
45}
46
47/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 * We can potentially split a vm area into separate
49 * areas, each area with its own behavior.
50 */
Vladimir Cernovec9bed92013-09-11 14:20:15 -070051static long madvise_behavior(struct vm_area_struct *vma,
Prasanna Meda05b74382005-06-21 17:14:37 -070052 struct vm_area_struct **prev,
53 unsigned long start, unsigned long end, int behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -070054{
Vladimir Cernovec9bed92013-09-11 14:20:15 -070055 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 int error = 0;
Prasanna Meda05b74382005-06-21 17:14:37 -070057 pgoff_t pgoff;
Hugh Dickins3866ea92009-09-21 17:01:52 -070058 unsigned long new_flags = vma->vm_flags;
Prasanna Medae798c6e2005-06-21 17:14:36 -070059
60 switch (behavior) {
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080061 case MADV_NORMAL:
62 new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ;
63 break;
Prasanna Medae798c6e2005-06-21 17:14:36 -070064 case MADV_SEQUENTIAL:
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080065 new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ;
Prasanna Medae798c6e2005-06-21 17:14:36 -070066 break;
67 case MADV_RANDOM:
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080068 new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ;
Prasanna Medae798c6e2005-06-21 17:14:36 -070069 break;
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080070 case MADV_DONTFORK:
71 new_flags |= VM_DONTCOPY;
72 break;
73 case MADV_DOFORK:
Hugh Dickins3866ea92009-09-21 17:01:52 -070074 if (vma->vm_flags & VM_IO) {
75 error = -EINVAL;
76 goto out;
77 }
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080078 new_flags &= ~VM_DONTCOPY;
Prasanna Medae798c6e2005-06-21 17:14:36 -070079 break;
Jason Baronaccb61f2012-03-23 15:02:51 -070080 case MADV_DONTDUMP:
Konstantin Khlebnikov0103bd12012-10-08 16:28:59 -070081 new_flags |= VM_DONTDUMP;
Jason Baronaccb61f2012-03-23 15:02:51 -070082 break;
83 case MADV_DODUMP:
Daniel Black00a28d942018-10-05 15:52:19 -070084 if (!is_vm_hugetlb_page(vma) && new_flags & VM_SPECIAL) {
Konstantin Khlebnikov0103bd12012-10-08 16:28:59 -070085 error = -EINVAL;
86 goto out;
87 }
88 new_flags &= ~VM_DONTDUMP;
Jason Baronaccb61f2012-03-23 15:02:51 -070089 break;
Hugh Dickinsf8af4da2009-09-21 17:01:57 -070090 case MADV_MERGEABLE:
91 case MADV_UNMERGEABLE:
92 error = ksm_madvise(vma, start, end, behavior, &new_flags);
93 if (error)
94 goto out;
95 break;
Andrea Arcangeli0af4e982011-01-13 15:46:55 -080096 case MADV_HUGEPAGE:
Andrea Arcangelia664b2d2011-01-13 15:47:17 -080097 case MADV_NOHUGEPAGE:
Andrea Arcangeli60ab3242011-01-13 15:47:18 -080098 error = hugepage_madvise(vma, &new_flags, behavior);
Andrea Arcangeli0af4e982011-01-13 15:46:55 -080099 if (error)
100 goto out;
101 break;
Prasanna Medae798c6e2005-06-21 17:14:36 -0700102 }
103
Prasanna Meda05b74382005-06-21 17:14:37 -0700104 if (new_flags == vma->vm_flags) {
105 *prev = vma;
Hugh Dickins836d5ff2005-09-03 15:54:53 -0700106 goto out;
Prasanna Meda05b74382005-06-21 17:14:37 -0700107 }
108
109 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
110 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700111 vma->vm_file, pgoff, vma_policy(vma),
112 vma->vm_userfaultfd_ctx);
Prasanna Meda05b74382005-06-21 17:14:37 -0700113 if (*prev) {
114 vma = *prev;
115 goto success;
116 }
117
118 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
120 if (start != vma->vm_start) {
121 error = split_vma(mm, vma, start, 1);
122 if (error)
123 goto out;
124 }
125
126 if (end != vma->vm_end) {
127 error = split_vma(mm, vma, end, 0);
128 if (error)
129 goto out;
130 }
131
Hugh Dickins836d5ff2005-09-03 15:54:53 -0700132success:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 /*
134 * vm_flags is protected by the mmap_sem held in write mode.
135 */
Prasanna Medae798c6e2005-06-21 17:14:36 -0700136 vma->vm_flags = new_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138out:
139 if (error == -ENOMEM)
140 error = -EAGAIN;
141 return error;
142}
143
Shaohua Li1998cc02013-02-22 16:32:31 -0800144#ifdef CONFIG_SWAP
145static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start,
146 unsigned long end, struct mm_walk *walk)
147{
148 pte_t *orig_pte;
149 struct vm_area_struct *vma = walk->private;
150 unsigned long index;
151
152 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
153 return 0;
154
155 for (index = start; index != end; index += PAGE_SIZE) {
156 pte_t pte;
157 swp_entry_t entry;
158 struct page *page;
159 spinlock_t *ptl;
160
161 orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
162 pte = *(orig_pte + ((index - start) / PAGE_SIZE));
163 pte_unmap_unlock(orig_pte, ptl);
164
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800165 if (pte_present(pte) || pte_none(pte))
Shaohua Li1998cc02013-02-22 16:32:31 -0800166 continue;
167 entry = pte_to_swp_entry(pte);
168 if (unlikely(non_swap_entry(entry)))
169 continue;
170
171 page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE,
172 vma, index);
173 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300174 put_page(page);
Shaohua Li1998cc02013-02-22 16:32:31 -0800175 }
176
177 return 0;
178}
179
180static void force_swapin_readahead(struct vm_area_struct *vma,
181 unsigned long start, unsigned long end)
182{
183 struct mm_walk walk = {
184 .mm = vma->vm_mm,
185 .pmd_entry = swapin_walk_pmd_entry,
186 .private = vma,
187 };
188
189 walk_page_range(start, end, &walk);
190
191 lru_add_drain(); /* Push any new pages onto the LRU now */
192}
193
194static void force_shm_swapin_readahead(struct vm_area_struct *vma,
195 unsigned long start, unsigned long end,
196 struct address_space *mapping)
197{
198 pgoff_t index;
199 struct page *page;
200 swp_entry_t swap;
201
202 for (; start < end; start += PAGE_SIZE) {
203 index = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
204
Johannes Weiner55231e52014-05-22 11:54:17 -0700205 page = find_get_entry(mapping, index);
Shaohua Li1998cc02013-02-22 16:32:31 -0800206 if (!radix_tree_exceptional_entry(page)) {
207 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300208 put_page(page);
Shaohua Li1998cc02013-02-22 16:32:31 -0800209 continue;
210 }
211 swap = radix_to_swp_entry(page);
212 page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE,
213 NULL, 0);
214 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300215 put_page(page);
Shaohua Li1998cc02013-02-22 16:32:31 -0800216 }
217
218 lru_add_drain(); /* Push any new pages onto the LRU now */
219}
220#endif /* CONFIG_SWAP */
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222/*
223 * Schedule all required I/O operations. Do not wait for completion.
224 */
Vladimir Cernovec9bed92013-09-11 14:20:15 -0700225static long madvise_willneed(struct vm_area_struct *vma,
226 struct vm_area_struct **prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 unsigned long start, unsigned long end)
228{
229 struct file *file = vma->vm_file;
230
chenjieba32d7d2017-11-29 16:10:54 -0800231 *prev = vma;
Shaohua Li1998cc02013-02-22 16:32:31 -0800232#ifdef CONFIG_SWAP
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100233 if (!file) {
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100234 force_swapin_readahead(vma, start, end);
Shaohua Li1998cc02013-02-22 16:32:31 -0800235 return 0;
236 }
Shaohua Li1998cc02013-02-22 16:32:31 -0800237
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100238 if (shmem_mapping(file->f_mapping)) {
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100239 force_shm_swapin_readahead(vma, start, end,
240 file->f_mapping);
241 return 0;
242 }
243#else
Suzuki1bef4002005-10-11 08:29:06 -0700244 if (!file)
245 return -EBADF;
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100246#endif
Suzuki1bef4002005-10-11 08:29:06 -0700247
Matthew Wilcoxe748dcd2015-02-16 15:59:12 -0800248 if (IS_DAX(file_inode(file))) {
Carsten Ottefe77ba62005-06-23 22:05:29 -0700249 /* no bad return value, but ignore advice */
250 return 0;
251 }
252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
254 if (end > vma->vm_end)
255 end = vma->vm_end;
256 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
257
Wu Fengguangf7e839d2009-06-16 15:31:20 -0700258 force_page_cache_readahead(file->f_mapping, file, start, end - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 return 0;
260}
261
Minchan Kim854e9ed2016-01-15 16:54:53 -0800262static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
263 unsigned long end, struct mm_walk *walk)
264
265{
266 struct mmu_gather *tlb = walk->private;
267 struct mm_struct *mm = tlb->mm;
268 struct vm_area_struct *vma = walk->vma;
269 spinlock_t *ptl;
270 pte_t *orig_pte, *pte, ptent;
271 struct page *page;
Minchan Kim64b42bc2016-01-15 16:55:06 -0800272 int nr_swap = 0;
Minchan Kimb8d3c4c2016-01-15 16:55:42 -0800273 unsigned long next;
Minchan Kim854e9ed2016-01-15 16:54:53 -0800274
Minchan Kimb8d3c4c2016-01-15 16:55:42 -0800275 next = pmd_addr_end(addr, end);
276 if (pmd_trans_huge(*pmd))
277 if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next))
278 goto next;
279
Minchan Kim854e9ed2016-01-15 16:54:53 -0800280 if (pmd_trans_unstable(pmd))
281 return 0;
282
283 orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Mel Gorman5a1eef72017-08-02 13:31:52 -0700284 flush_tlb_batched_pending(mm);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800285 arch_enter_lazy_mmu_mode();
286 for (; addr != end; pte++, addr += PAGE_SIZE) {
287 ptent = *pte;
288
Minchan Kim64b42bc2016-01-15 16:55:06 -0800289 if (pte_none(ptent))
Minchan Kim854e9ed2016-01-15 16:54:53 -0800290 continue;
Minchan Kim64b42bc2016-01-15 16:55:06 -0800291 /*
292 * If the pte has swp_entry, just clear page table to
293 * prevent swap-in which is more expensive rather than
294 * (page allocation + zeroing).
295 */
296 if (!pte_present(ptent)) {
297 swp_entry_t entry;
298
299 entry = pte_to_swp_entry(ptent);
300 if (non_swap_entry(entry))
301 continue;
302 nr_swap--;
303 free_swap_and_cache(entry);
304 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
305 continue;
306 }
Minchan Kim854e9ed2016-01-15 16:54:53 -0800307
308 page = vm_normal_page(vma, addr, ptent);
309 if (!page)
310 continue;
311
312 /*
313 * If pmd isn't transhuge but the page is THP and
314 * is owned by only this process, split it and
315 * deactivate all pages.
316 */
317 if (PageTransCompound(page)) {
318 if (page_mapcount(page) != 1)
319 goto out;
320 get_page(page);
321 if (!trylock_page(page)) {
322 put_page(page);
323 goto out;
324 }
325 pte_unmap_unlock(orig_pte, ptl);
326 if (split_huge_page(page)) {
327 unlock_page(page);
328 put_page(page);
329 pte_offset_map_lock(mm, pmd, addr, &ptl);
330 goto out;
331 }
Minchan Kim854e9ed2016-01-15 16:54:53 -0800332 unlock_page(page);
Eric Biggers0f49b052017-08-25 15:55:39 -0700333 put_page(page);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800334 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
335 pte--;
336 addr -= PAGE_SIZE;
337 continue;
338 }
339
340 VM_BUG_ON_PAGE(PageTransCompound(page), page);
341
342 if (PageSwapCache(page) || PageDirty(page)) {
343 if (!trylock_page(page))
344 continue;
345 /*
346 * If page is shared with others, we couldn't clear
347 * PG_dirty of the page.
348 */
349 if (page_mapcount(page) != 1) {
350 unlock_page(page);
351 continue;
352 }
353
354 if (PageSwapCache(page) && !try_to_free_swap(page)) {
355 unlock_page(page);
356 continue;
357 }
358
359 ClearPageDirty(page);
360 unlock_page(page);
361 }
362
363 if (pte_young(ptent) || pte_dirty(ptent)) {
364 /*
365 * Some of architecture(ex, PPC) don't update TLB
366 * with set_pte_at and tlb_remove_tlb_entry so for
367 * the portability, remap the pte with old|clean
368 * after pte clearing.
369 */
370 ptent = ptep_get_and_clear_full(mm, addr, pte,
371 tlb->fullmm);
372
373 ptent = pte_mkold(ptent);
374 ptent = pte_mkclean(ptent);
375 set_pte_at(mm, addr, pte, ptent);
Minchan Kim10853a02016-01-15 16:55:11 -0800376 if (PageActive(page))
377 deactivate_page(page);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800378 tlb_remove_tlb_entry(tlb, pte, addr);
379 }
380 }
381out:
Minchan Kim64b42bc2016-01-15 16:55:06 -0800382 if (nr_swap) {
383 if (current->mm == mm)
384 sync_mm_rss(mm);
385
386 add_mm_counter(mm, MM_SWAPENTS, nr_swap);
387 }
Minchan Kim854e9ed2016-01-15 16:54:53 -0800388 arch_leave_lazy_mmu_mode();
389 pte_unmap_unlock(orig_pte, ptl);
390 cond_resched();
Minchan Kimb8d3c4c2016-01-15 16:55:42 -0800391next:
Minchan Kim854e9ed2016-01-15 16:54:53 -0800392 return 0;
393}
394
395static void madvise_free_page_range(struct mmu_gather *tlb,
396 struct vm_area_struct *vma,
397 unsigned long addr, unsigned long end)
398{
399 struct mm_walk free_walk = {
400 .pmd_entry = madvise_free_pte_range,
401 .mm = vma->vm_mm,
402 .private = tlb,
403 };
404
405 tlb_start_vma(tlb, vma);
406 walk_page_range(addr, end, &free_walk);
407 tlb_end_vma(tlb, vma);
408}
409
410static int madvise_free_single_vma(struct vm_area_struct *vma,
411 unsigned long start_addr, unsigned long end_addr)
412{
413 unsigned long start, end;
414 struct mm_struct *mm = vma->vm_mm;
415 struct mmu_gather tlb;
416
417 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
418 return -EINVAL;
419
420 /* MADV_FREE works for only anon vma at the moment */
421 if (!vma_is_anonymous(vma))
422 return -EINVAL;
423
424 start = max(vma->vm_start, start_addr);
425 if (start >= vma->vm_end)
426 return -EINVAL;
427 end = min(vma->vm_end, end_addr);
428 if (end <= vma->vm_start)
429 return -EINVAL;
430
431 lru_add_drain();
432 tlb_gather_mmu(&tlb, mm, start, end);
433 update_hiwater_rss(mm);
434
435 mmu_notifier_invalidate_range_start(mm, start, end);
436 madvise_free_page_range(&tlb, vma, start, end);
437 mmu_notifier_invalidate_range_end(mm, start, end);
438 tlb_finish_mmu(&tlb, start, end);
439
440 return 0;
441}
442
443static long madvise_free(struct vm_area_struct *vma,
444 struct vm_area_struct **prev,
445 unsigned long start, unsigned long end)
446{
447 *prev = vma;
448 return madvise_free_single_vma(vma, start, end);
449}
450
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451/*
452 * Application no longer needs these pages. If the pages are dirty,
453 * it's OK to just throw them away. The app will be more careful about
454 * data it wants to keep. Be sure to free swap resources too. The
Fernando Luis Vazquez Cao7e6cbea2008-07-29 22:33:39 -0700455 * zap_page_range call sets things up for shrink_active_list to actually free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 * these pages later if no one else has touched them in the meantime,
457 * although we could add these pages to a global reuse list for
Fernando Luis Vazquez Cao7e6cbea2008-07-29 22:33:39 -0700458 * shrink_active_list to pick up before reclaiming other pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 *
460 * NB: This interface discards data rather than pushes it out to swap,
461 * as some implementations do. This has performance implications for
462 * applications like large transactional databases which want to discard
463 * pages in anonymous maps after committing to backing store the data
464 * that was kept in them. There is no reason to write this data out to
465 * the swap area if the application is discarding it.
466 *
467 * An interface that causes the system to free clean pages and flush
468 * dirty pages is already available as msync(MS_INVALIDATE).
469 */
Vladimir Cernovec9bed92013-09-11 14:20:15 -0700470static long madvise_dontneed(struct vm_area_struct *vma,
471 struct vm_area_struct **prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 unsigned long start, unsigned long end)
473{
Prasanna Meda05b74382005-06-21 17:14:37 -0700474 *prev = vma;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800475 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 return -EINVAL;
477
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -0800478 zap_page_range(vma, start, end - start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 return 0;
480}
481
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800482/*
483 * Application wants to free up the pages and associated backing store.
484 * This is effectively punching a hole into the middle of a file.
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800485 */
486static long madvise_remove(struct vm_area_struct *vma,
Nick Piggin00e9fa22007-03-16 13:38:10 -0800487 struct vm_area_struct **prev,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800488 unsigned long start, unsigned long end)
489{
Hugh Dickins3f31d072012-05-29 15:06:40 -0700490 loff_t offset;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700491 int error;
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700492 struct file *f;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800493
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700494 *prev = NULL; /* tell sys_madvise we drop mmap_sem */
Nick Piggin00e9fa22007-03-16 13:38:10 -0800495
Mike Kravetz72079ba2015-09-08 15:01:57 -0700496 if (vma->vm_flags & VM_LOCKED)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800497 return -EINVAL;
498
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700499 f = vma->vm_file;
500
501 if (!f || !f->f_mapping || !f->f_mapping->host) {
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800502 return -EINVAL;
503 }
504
Hugh Dickins69cf0fa2006-04-17 22:46:32 +0100505 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE))
506 return -EACCES;
507
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800508 offset = (loff_t)(start - vma->vm_start)
509 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700510
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700511 /*
512 * Filesystem's fallocate may need to take i_mutex. We need to
513 * explicitly grab a reference because the vma (and hence the
514 * vma's reference to the file) can go away as soon as we drop
515 * mmap_sem.
516 */
517 get_file(f);
Nick Piggin0a27a142007-05-06 14:49:53 -0700518 up_read(&current->mm->mmap_sem);
Anna Schumaker72c72bd2014-11-07 14:44:25 -0500519 error = vfs_fallocate(f,
Hugh Dickins3f31d072012-05-29 15:06:40 -0700520 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
521 offset, end - start);
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700522 fput(f);
Nick Piggin0a27a142007-05-06 14:49:53 -0700523 down_read(&current->mm->mmap_sem);
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700524 return error;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800525}
526
Andi Kleen9893e492009-09-16 11:50:17 +0200527#ifdef CONFIG_MEMORY_FAILURE
528/*
529 * Error injection support for memory error handling.
530 */
Andi Kleenafcf9382009-12-16 12:20:00 +0100531static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
Andi Kleen9893e492009-09-16 11:50:17 +0200532{
Wanpeng Li20cb6ca2013-09-30 13:45:21 -0700533 struct page *p;
Mel Gorman8cc3acf2017-08-31 16:15:30 -0700534 struct zone *zone;
535
Andi Kleen9893e492009-09-16 11:50:17 +0200536 if (!capable(CAP_SYS_ADMIN))
537 return -EPERM;
Wanpeng Li20cb6ca2013-09-30 13:45:21 -0700538 for (; start < end; start += PAGE_SIZE <<
539 compound_order(compound_head(p))) {
Andrew Morton325c4ef2013-09-11 14:23:03 -0700540 int ret;
541
542 ret = get_user_pages_fast(start, 1, 0, &p);
Andi Kleen9893e492009-09-16 11:50:17 +0200543 if (ret != 1)
544 return ret;
Andrew Morton325c4ef2013-09-11 14:23:03 -0700545
Wanpeng Li29b4eed2013-09-11 14:22:59 -0700546 if (PageHWPoison(p)) {
547 put_page(p);
548 continue;
549 }
Andi Kleenafcf9382009-12-16 12:20:00 +0100550 if (bhv == MADV_SOFT_OFFLINE) {
Wanpeng Lib194b8c2013-09-11 14:22:57 -0700551 pr_info("Soft offlining page %#lx at %#lx\n",
Andi Kleenafcf9382009-12-16 12:20:00 +0100552 page_to_pfn(p), start);
553 ret = soft_offline_page(p, MF_COUNT_INCREASED);
554 if (ret)
Wanpeng Li83024232013-09-11 14:23:02 -0700555 return ret;
Andi Kleenafcf9382009-12-16 12:20:00 +0100556 continue;
557 }
Wanpeng Lib194b8c2013-09-11 14:22:57 -0700558 pr_info("Injecting memory failure for page %#lx at %#lx\n",
Andi Kleen9893e492009-09-16 11:50:17 +0200559 page_to_pfn(p), start);
Naoya Horiguchi23a003b2016-03-15 14:56:36 -0700560 ret = memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED);
561 if (ret)
562 return ret;
Andi Kleen9893e492009-09-16 11:50:17 +0200563 }
Mel Gorman8cc3acf2017-08-31 16:15:30 -0700564
565 /* Ensure that all poisoned pages are removed from per-cpu lists */
566 for_each_populated_zone(zone)
567 drain_all_pages(zone);
568
Andrew Morton325c4ef2013-09-11 14:23:03 -0700569 return 0;
Andi Kleen9893e492009-09-16 11:50:17 +0200570}
571#endif
572
suzuki165cd402005-07-27 11:43:59 -0700573static long
574madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
575 unsigned long start, unsigned long end, int behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 switch (behavior) {
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800578 case MADV_REMOVE:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700579 return madvise_remove(vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 case MADV_WILLNEED:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700581 return madvise_willneed(vma, prev, start, end);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800582 case MADV_FREE:
583 /*
584 * XXX: In this implementation, MADV_FREE works like
585 * MADV_DONTNEED on swapless system or full swap.
586 */
587 if (get_nr_swap_pages() > 0)
588 return madvise_free(vma, prev, start, end);
589 /* passthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 case MADV_DONTNEED:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700591 return madvise_dontneed(vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 default:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700593 return madvise_behavior(vma, prev, start, end, behavior);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
596
Nicholas Krause1ecef9e2015-09-04 15:48:24 -0700597static bool
Nick Piggin75927af2009-06-16 15:32:38 -0700598madvise_behavior_valid(int behavior)
599{
600 switch (behavior) {
601 case MADV_DOFORK:
602 case MADV_DONTFORK:
603 case MADV_NORMAL:
604 case MADV_SEQUENTIAL:
605 case MADV_RANDOM:
606 case MADV_REMOVE:
607 case MADV_WILLNEED:
608 case MADV_DONTNEED:
Minchan Kim854e9ed2016-01-15 16:54:53 -0800609 case MADV_FREE:
Hugh Dickinsf8af4da2009-09-21 17:01:57 -0700610#ifdef CONFIG_KSM
611 case MADV_MERGEABLE:
612 case MADV_UNMERGEABLE:
613#endif
Andrea Arcangeli0af4e982011-01-13 15:46:55 -0800614#ifdef CONFIG_TRANSPARENT_HUGEPAGE
615 case MADV_HUGEPAGE:
Andrea Arcangelia664b2d2011-01-13 15:47:17 -0800616 case MADV_NOHUGEPAGE:
Andrea Arcangeli0af4e982011-01-13 15:46:55 -0800617#endif
Jason Baronaccb61f2012-03-23 15:02:51 -0700618 case MADV_DONTDUMP:
619 case MADV_DODUMP:
Nicholas Krause1ecef9e2015-09-04 15:48:24 -0700620 return true;
Nick Piggin75927af2009-06-16 15:32:38 -0700621
622 default:
Nicholas Krause1ecef9e2015-09-04 15:48:24 -0700623 return false;
Nick Piggin75927af2009-06-16 15:32:38 -0700624 }
625}
Hugh Dickins3866ea92009-09-21 17:01:52 -0700626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627/*
628 * The madvise(2) system call.
629 *
630 * Applications can use madvise() to advise the kernel how it should
631 * handle paging I/O in this VM area. The idea is to help the kernel
632 * use appropriate read-ahead and caching techniques. The information
633 * provided is advisory only, and can be safely disregarded by the
634 * kernel without affecting the correct operation of the application.
635 *
636 * behavior values:
637 * MADV_NORMAL - the default behavior is to read clusters. This
638 * results in some read-ahead and read-behind.
639 * MADV_RANDOM - the system should read the minimum amount of data
640 * on any access, since it is unlikely that the appli-
641 * cation will need more than what it asks for.
642 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
643 * once, so they can be aggressively read ahead, and
644 * can be freed soon after they are accessed.
645 * MADV_WILLNEED - the application is notifying the system to read
646 * some pages ahead.
647 * MADV_DONTNEED - the application is finished with the given range,
648 * so the kernel can free resources associated with it.
Naoya Horiguchid7206a72016-03-15 14:56:58 -0700649 * MADV_FREE - the application marks pages in the given range as lazy free,
650 * where actual purges are postponed until memory pressure happens.
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800651 * MADV_REMOVE - the application wants to free up the given range of
652 * pages and associated backing store.
Hugh Dickins3866ea92009-09-21 17:01:52 -0700653 * MADV_DONTFORK - omit this area from child's address space when forking:
654 * typically, to avoid COWing pages pinned by get_user_pages().
655 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
Naoya Horiguchid7206a72016-03-15 14:56:58 -0700656 * MADV_HWPOISON - trigger memory error handler as if the given memory range
657 * were corrupted by unrecoverable hardware memory failure.
658 * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory.
Hugh Dickinsf8af4da2009-09-21 17:01:57 -0700659 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
660 * this area with pages of identical content from other such areas.
661 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
Naoya Horiguchid7206a72016-03-15 14:56:58 -0700662 * MADV_HUGEPAGE - the application wants to back the given range by transparent
663 * huge pages in the future. Existing pages might be coalesced and
664 * new pages might be allocated as THP.
665 * MADV_NOHUGEPAGE - mark the given range as not worth being backed by
666 * transparent huge pages so the existing pages will not be
667 * coalesced into THP and new pages will not be allocated as THP.
668 * MADV_DONTDUMP - the application wants to prevent pages in the given range
669 * from being included in its core dump.
670 * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 *
672 * return values:
673 * zero - success
674 * -EINVAL - start + len < 0, start is not page-aligned,
675 * "behavior" is not a valid value, or application
676 * is attempting to release locked or shared pages.
677 * -ENOMEM - addresses in the specified range are not currently
678 * mapped, or are outside the AS of the process.
679 * -EIO - an I/O error occurred while paging in data.
680 * -EBADF - map exists, but area maps something that isn't a file.
681 * -EAGAIN - a kernel resource was temporarily unavailable.
682 */
Heiko Carstens3480b252009-01-14 14:14:16 +0100683SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
Prasanna Meda05b74382005-06-21 17:14:37 -0700685 unsigned long end, tmp;
Vladimir Cernovec9bed92013-09-11 14:20:15 -0700686 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 int unmapped_error = 0;
688 int error = -EINVAL;
Jason Baronf7977792007-07-15 23:38:21 -0700689 int write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 size_t len;
Shaohua Li1998cc02013-02-22 16:32:31 -0800691 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Andi Kleen9893e492009-09-16 11:50:17 +0200693#ifdef CONFIG_MEMORY_FAILURE
Andi Kleenafcf9382009-12-16 12:20:00 +0100694 if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE)
695 return madvise_hwpoison(behavior, start, start+len_in);
Andi Kleen9893e492009-09-16 11:50:17 +0200696#endif
Nick Piggin75927af2009-06-16 15:32:38 -0700697 if (!madvise_behavior_valid(behavior))
698 return error;
699
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700700 if (start & ~PAGE_MASK)
701 return error;
702 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
703
704 /* Check to see whether len was rounded up from small -ve to zero */
705 if (len_in && !len)
706 return error;
707
708 end = start + len;
709 if (end < start)
710 return error;
711
712 error = 0;
713 if (end == start)
714 return error;
715
Jason Baronf7977792007-07-15 23:38:21 -0700716 write = madvise_need_mmap_write(behavior);
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700717 if (write) {
718 if (down_write_killable(&current->mm->mmap_sem))
719 return -EINTR;
720 } else {
Nick Piggin0a27a142007-05-06 14:49:53 -0700721 down_read(&current->mm->mmap_sem);
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700722 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 /*
725 * If the interval [start,end) covers some unmapped address
726 * ranges, just ignore them, but return -ENOMEM at the end.
Prasanna Meda05b74382005-06-21 17:14:37 -0700727 * - different from the way of handling in mlock etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 */
Prasanna Meda05b74382005-06-21 17:14:37 -0700729 vma = find_vma_prev(current->mm, start, &prev);
Hugh Dickins836d5ff2005-09-03 15:54:53 -0700730 if (vma && start > vma->vm_start)
731 prev = vma;
732
Shaohua Li1998cc02013-02-22 16:32:31 -0800733 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 for (;;) {
735 /* Still start < end. */
736 error = -ENOMEM;
737 if (!vma)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700738 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Prasanna Meda05b74382005-06-21 17:14:37 -0700740 /* Here start < (end|vma->vm_end). */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 if (start < vma->vm_start) {
742 unmapped_error = -ENOMEM;
743 start = vma->vm_start;
Prasanna Meda05b74382005-06-21 17:14:37 -0700744 if (start >= end)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700745 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
747
Prasanna Meda05b74382005-06-21 17:14:37 -0700748 /* Here vma->vm_start <= start < (end|vma->vm_end) */
749 tmp = vma->vm_end;
750 if (end < tmp)
751 tmp = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Prasanna Meda05b74382005-06-21 17:14:37 -0700753 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
754 error = madvise_vma(vma, &prev, start, tmp, behavior);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 if (error)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700756 goto out;
Prasanna Meda05b74382005-06-21 17:14:37 -0700757 start = tmp;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700758 if (prev && start < prev->vm_end)
Prasanna Meda05b74382005-06-21 17:14:37 -0700759 start = prev->vm_end;
760 error = unmapped_error;
761 if (start >= end)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700762 goto out;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700763 if (prev)
764 vma = prev->vm_next;
765 else /* madvise_remove dropped mmap_sem */
766 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768out:
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700769 blk_finish_plug(&plug);
Jason Baronf7977792007-07-15 23:38:21 -0700770 if (write)
Nick Piggin0a27a142007-05-06 14:49:53 -0700771 up_write(&current->mm->mmap_sem);
772 else
773 up_read(&current->mm->mmap_sem);
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 return error;
776}