blob: 088a5b22f3975e9b551df4e5a92cf88354713571 [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:
Konstantin Khlebnikov0103bd12012-10-08 16:28:59 -070084 if (new_flags & VM_SPECIAL) {
85 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),
Colin Cross3e4578f2015-10-27 16:42:08 -0700112 vma->vm_userfaultfd_ctx, vma_get_anon_name(vma));
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
Shaohua Li1998cc02013-02-22 16:32:31 -0800231#ifdef CONFIG_SWAP
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100232 if (!file) {
Shaohua Li1998cc02013-02-22 16:32:31 -0800233 *prev = vma;
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)) {
239 *prev = vma;
240 force_shm_swapin_readahead(vma, start, end,
241 file->f_mapping);
242 return 0;
243 }
244#else
Suzuki1bef4002005-10-11 08:29:06 -0700245 if (!file)
246 return -EBADF;
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100247#endif
Suzuki1bef4002005-10-11 08:29:06 -0700248
Matthew Wilcoxe748dcd2015-02-16 15:59:12 -0800249 if (IS_DAX(file_inode(file))) {
Carsten Ottefe77ba62005-06-23 22:05:29 -0700250 /* no bad return value, but ignore advice */
251 return 0;
252 }
253
Prasanna Meda05b74382005-06-21 17:14:37 -0700254 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
256 if (end > vma->vm_end)
257 end = vma->vm_end;
258 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
259
Wu Fengguangf7e839d2009-06-16 15:31:20 -0700260 force_page_cache_readahead(file->f_mapping, file, start, end - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 return 0;
262}
263
Minchan Kim854e9ed2016-01-15 16:54:53 -0800264static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
265 unsigned long end, struct mm_walk *walk)
266
267{
268 struct mmu_gather *tlb = walk->private;
269 struct mm_struct *mm = tlb->mm;
270 struct vm_area_struct *vma = walk->vma;
271 spinlock_t *ptl;
272 pte_t *orig_pte, *pte, ptent;
273 struct page *page;
Minchan Kim64b42bc2016-01-15 16:55:06 -0800274 int nr_swap = 0;
Minchan Kimb8d3c4c2016-01-15 16:55:42 -0800275 unsigned long next;
Minchan Kim854e9ed2016-01-15 16:54:53 -0800276
Minchan Kimb8d3c4c2016-01-15 16:55:42 -0800277 next = pmd_addr_end(addr, end);
278 if (pmd_trans_huge(*pmd))
279 if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next))
280 goto next;
281
Minchan Kim854e9ed2016-01-15 16:54:53 -0800282 if (pmd_trans_unstable(pmd))
283 return 0;
284
285 orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Mel Gorman5a1eef72017-08-02 13:31:52 -0700286 flush_tlb_batched_pending(mm);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800287 arch_enter_lazy_mmu_mode();
288 for (; addr != end; pte++, addr += PAGE_SIZE) {
289 ptent = *pte;
290
Minchan Kim64b42bc2016-01-15 16:55:06 -0800291 if (pte_none(ptent))
Minchan Kim854e9ed2016-01-15 16:54:53 -0800292 continue;
Minchan Kim64b42bc2016-01-15 16:55:06 -0800293 /*
294 * If the pte has swp_entry, just clear page table to
295 * prevent swap-in which is more expensive rather than
296 * (page allocation + zeroing).
297 */
298 if (!pte_present(ptent)) {
299 swp_entry_t entry;
300
301 entry = pte_to_swp_entry(ptent);
302 if (non_swap_entry(entry))
303 continue;
304 nr_swap--;
305 free_swap_and_cache(entry);
306 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
307 continue;
308 }
Minchan Kim854e9ed2016-01-15 16:54:53 -0800309
310 page = vm_normal_page(vma, addr, ptent);
311 if (!page)
312 continue;
313
314 /*
315 * If pmd isn't transhuge but the page is THP and
316 * is owned by only this process, split it and
317 * deactivate all pages.
318 */
319 if (PageTransCompound(page)) {
320 if (page_mapcount(page) != 1)
321 goto out;
322 get_page(page);
323 if (!trylock_page(page)) {
324 put_page(page);
325 goto out;
326 }
327 pte_unmap_unlock(orig_pte, ptl);
328 if (split_huge_page(page)) {
329 unlock_page(page);
330 put_page(page);
331 pte_offset_map_lock(mm, pmd, addr, &ptl);
332 goto out;
333 }
Minchan Kim854e9ed2016-01-15 16:54:53 -0800334 unlock_page(page);
Eric Biggers0f49b052017-08-25 15:55:39 -0700335 put_page(page);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800336 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
337 pte--;
338 addr -= PAGE_SIZE;
339 continue;
340 }
341
342 VM_BUG_ON_PAGE(PageTransCompound(page), page);
343
344 if (PageSwapCache(page) || PageDirty(page)) {
345 if (!trylock_page(page))
346 continue;
347 /*
348 * If page is shared with others, we couldn't clear
349 * PG_dirty of the page.
350 */
351 if (page_mapcount(page) != 1) {
352 unlock_page(page);
353 continue;
354 }
355
356 if (PageSwapCache(page) && !try_to_free_swap(page)) {
357 unlock_page(page);
358 continue;
359 }
360
361 ClearPageDirty(page);
362 unlock_page(page);
363 }
364
365 if (pte_young(ptent) || pte_dirty(ptent)) {
366 /*
367 * Some of architecture(ex, PPC) don't update TLB
368 * with set_pte_at and tlb_remove_tlb_entry so for
369 * the portability, remap the pte with old|clean
370 * after pte clearing.
371 */
372 ptent = ptep_get_and_clear_full(mm, addr, pte,
373 tlb->fullmm);
374
375 ptent = pte_mkold(ptent);
376 ptent = pte_mkclean(ptent);
377 set_pte_at(mm, addr, pte, ptent);
Minchan Kim10853a02016-01-15 16:55:11 -0800378 if (PageActive(page))
379 deactivate_page(page);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800380 tlb_remove_tlb_entry(tlb, pte, addr);
381 }
382 }
383out:
Minchan Kim64b42bc2016-01-15 16:55:06 -0800384 if (nr_swap) {
385 if (current->mm == mm)
386 sync_mm_rss(mm);
387
388 add_mm_counter(mm, MM_SWAPENTS, nr_swap);
389 }
Minchan Kim854e9ed2016-01-15 16:54:53 -0800390 arch_leave_lazy_mmu_mode();
391 pte_unmap_unlock(orig_pte, ptl);
392 cond_resched();
Minchan Kimb8d3c4c2016-01-15 16:55:42 -0800393next:
Minchan Kim854e9ed2016-01-15 16:54:53 -0800394 return 0;
395}
396
397static void madvise_free_page_range(struct mmu_gather *tlb,
398 struct vm_area_struct *vma,
399 unsigned long addr, unsigned long end)
400{
401 struct mm_walk free_walk = {
402 .pmd_entry = madvise_free_pte_range,
403 .mm = vma->vm_mm,
404 .private = tlb,
405 };
406
407 tlb_start_vma(tlb, vma);
408 walk_page_range(addr, end, &free_walk);
409 tlb_end_vma(tlb, vma);
410}
411
412static int madvise_free_single_vma(struct vm_area_struct *vma,
413 unsigned long start_addr, unsigned long end_addr)
414{
415 unsigned long start, end;
416 struct mm_struct *mm = vma->vm_mm;
417 struct mmu_gather tlb;
418
419 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
420 return -EINVAL;
421
422 /* MADV_FREE works for only anon vma at the moment */
423 if (!vma_is_anonymous(vma))
424 return -EINVAL;
425
426 start = max(vma->vm_start, start_addr);
427 if (start >= vma->vm_end)
428 return -EINVAL;
429 end = min(vma->vm_end, end_addr);
430 if (end <= vma->vm_start)
431 return -EINVAL;
432
433 lru_add_drain();
434 tlb_gather_mmu(&tlb, mm, start, end);
435 update_hiwater_rss(mm);
436
437 mmu_notifier_invalidate_range_start(mm, start, end);
438 madvise_free_page_range(&tlb, vma, start, end);
439 mmu_notifier_invalidate_range_end(mm, start, end);
440 tlb_finish_mmu(&tlb, start, end);
441
442 return 0;
443}
444
445static long madvise_free(struct vm_area_struct *vma,
446 struct vm_area_struct **prev,
447 unsigned long start, unsigned long end)
448{
449 *prev = vma;
450 return madvise_free_single_vma(vma, start, end);
451}
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453/*
454 * Application no longer needs these pages. If the pages are dirty,
455 * it's OK to just throw them away. The app will be more careful about
456 * data it wants to keep. Be sure to free swap resources too. The
Fernando Luis Vazquez Cao7e6cbea2008-07-29 22:33:39 -0700457 * zap_page_range call sets things up for shrink_active_list to actually free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 * these pages later if no one else has touched them in the meantime,
459 * although we could add these pages to a global reuse list for
Fernando Luis Vazquez Cao7e6cbea2008-07-29 22:33:39 -0700460 * shrink_active_list to pick up before reclaiming other pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 *
462 * NB: This interface discards data rather than pushes it out to swap,
463 * as some implementations do. This has performance implications for
464 * applications like large transactional databases which want to discard
465 * pages in anonymous maps after committing to backing store the data
466 * that was kept in them. There is no reason to write this data out to
467 * the swap area if the application is discarding it.
468 *
469 * An interface that causes the system to free clean pages and flush
470 * dirty pages is already available as msync(MS_INVALIDATE).
471 */
Vladimir Cernovec9bed92013-09-11 14:20:15 -0700472static long madvise_dontneed(struct vm_area_struct *vma,
473 struct vm_area_struct **prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 unsigned long start, unsigned long end)
475{
Prasanna Meda05b74382005-06-21 17:14:37 -0700476 *prev = vma;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800477 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 return -EINVAL;
479
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -0800480 zap_page_range(vma, start, end - start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 return 0;
482}
483
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800484/*
485 * Application wants to free up the pages and associated backing store.
486 * This is effectively punching a hole into the middle of a file.
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800487 */
488static long madvise_remove(struct vm_area_struct *vma,
Nick Piggin00e9fa22007-03-16 13:38:10 -0800489 struct vm_area_struct **prev,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800490 unsigned long start, unsigned long end)
491{
Hugh Dickins3f31d072012-05-29 15:06:40 -0700492 loff_t offset;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700493 int error;
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700494 struct file *f;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800495
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700496 *prev = NULL; /* tell sys_madvise we drop mmap_sem */
Nick Piggin00e9fa22007-03-16 13:38:10 -0800497
Mike Kravetz72079ba2015-09-08 15:01:57 -0700498 if (vma->vm_flags & VM_LOCKED)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800499 return -EINVAL;
500
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700501 f = vma->vm_file;
502
503 if (!f || !f->f_mapping || !f->f_mapping->host) {
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800504 return -EINVAL;
505 }
506
Hugh Dickins69cf0fa2006-04-17 22:46:32 +0100507 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE))
508 return -EACCES;
509
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800510 offset = (loff_t)(start - vma->vm_start)
511 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700512
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700513 /*
514 * Filesystem's fallocate may need to take i_mutex. We need to
515 * explicitly grab a reference because the vma (and hence the
516 * vma's reference to the file) can go away as soon as we drop
517 * mmap_sem.
518 */
519 get_file(f);
Nick Piggin0a27a142007-05-06 14:49:53 -0700520 up_read(&current->mm->mmap_sem);
Anna Schumaker72c72bd2014-11-07 14:44:25 -0500521 error = vfs_fallocate(f,
Hugh Dickins3f31d072012-05-29 15:06:40 -0700522 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
523 offset, end - start);
Andy Lutomirski9ab42332012-07-05 16:00:11 -0700524 fput(f);
Nick Piggin0a27a142007-05-06 14:49:53 -0700525 down_read(&current->mm->mmap_sem);
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700526 return error;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800527}
528
Andi Kleen9893e492009-09-16 11:50:17 +0200529#ifdef CONFIG_MEMORY_FAILURE
530/*
531 * Error injection support for memory error handling.
532 */
Andi Kleenafcf9382009-12-16 12:20:00 +0100533static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
Andi Kleen9893e492009-09-16 11:50:17 +0200534{
Wanpeng Li20cb6ca2013-09-30 13:45:21 -0700535 struct page *p;
Mel Gorman8cc3acf2017-08-31 16:15:30 -0700536 struct zone *zone;
537
Andi Kleen9893e492009-09-16 11:50:17 +0200538 if (!capable(CAP_SYS_ADMIN))
539 return -EPERM;
Wanpeng Li20cb6ca2013-09-30 13:45:21 -0700540 for (; start < end; start += PAGE_SIZE <<
541 compound_order(compound_head(p))) {
Andrew Morton325c4ef2013-09-11 14:23:03 -0700542 int ret;
543
544 ret = get_user_pages_fast(start, 1, 0, &p);
Andi Kleen9893e492009-09-16 11:50:17 +0200545 if (ret != 1)
546 return ret;
Andrew Morton325c4ef2013-09-11 14:23:03 -0700547
Wanpeng Li29b4eed2013-09-11 14:22:59 -0700548 if (PageHWPoison(p)) {
549 put_page(p);
550 continue;
551 }
Andi Kleenafcf9382009-12-16 12:20:00 +0100552 if (bhv == MADV_SOFT_OFFLINE) {
Wanpeng Lib194b8c2013-09-11 14:22:57 -0700553 pr_info("Soft offlining page %#lx at %#lx\n",
Andi Kleenafcf9382009-12-16 12:20:00 +0100554 page_to_pfn(p), start);
555 ret = soft_offline_page(p, MF_COUNT_INCREASED);
556 if (ret)
Wanpeng Li83024232013-09-11 14:23:02 -0700557 return ret;
Andi Kleenafcf9382009-12-16 12:20:00 +0100558 continue;
559 }
Wanpeng Lib194b8c2013-09-11 14:22:57 -0700560 pr_info("Injecting memory failure for page %#lx at %#lx\n",
Andi Kleen9893e492009-09-16 11:50:17 +0200561 page_to_pfn(p), start);
Naoya Horiguchi23a003b2016-03-15 14:56:36 -0700562 ret = memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED);
563 if (ret)
564 return ret;
Andi Kleen9893e492009-09-16 11:50:17 +0200565 }
Mel Gorman8cc3acf2017-08-31 16:15:30 -0700566
567 /* Ensure that all poisoned pages are removed from per-cpu lists */
568 for_each_populated_zone(zone)
569 drain_all_pages(zone);
570
Andrew Morton325c4ef2013-09-11 14:23:03 -0700571 return 0;
Andi Kleen9893e492009-09-16 11:50:17 +0200572}
573#endif
574
suzuki165cd402005-07-27 11:43:59 -0700575static long
576madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
577 unsigned long start, unsigned long end, int behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 switch (behavior) {
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800580 case MADV_REMOVE:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700581 return madvise_remove(vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 case MADV_WILLNEED:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700583 return madvise_willneed(vma, prev, start, end);
Minchan Kim854e9ed2016-01-15 16:54:53 -0800584 case MADV_FREE:
585 /*
586 * XXX: In this implementation, MADV_FREE works like
587 * MADV_DONTNEED on swapless system or full swap.
588 */
589 if (get_nr_swap_pages() > 0)
590 return madvise_free(vma, prev, start, end);
591 /* passthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 case MADV_DONTNEED:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700593 return madvise_dontneed(vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 default:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700595 return madvise_behavior(vma, prev, start, end, behavior);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
598
Nicholas Krause1ecef9e2015-09-04 15:48:24 -0700599static bool
Nick Piggin75927af2009-06-16 15:32:38 -0700600madvise_behavior_valid(int behavior)
601{
602 switch (behavior) {
603 case MADV_DOFORK:
604 case MADV_DONTFORK:
605 case MADV_NORMAL:
606 case MADV_SEQUENTIAL:
607 case MADV_RANDOM:
608 case MADV_REMOVE:
609 case MADV_WILLNEED:
610 case MADV_DONTNEED:
Minchan Kim854e9ed2016-01-15 16:54:53 -0800611 case MADV_FREE:
Hugh Dickinsf8af4da2009-09-21 17:01:57 -0700612#ifdef CONFIG_KSM
613 case MADV_MERGEABLE:
614 case MADV_UNMERGEABLE:
615#endif
Andrea Arcangeli0af4e982011-01-13 15:46:55 -0800616#ifdef CONFIG_TRANSPARENT_HUGEPAGE
617 case MADV_HUGEPAGE:
Andrea Arcangelia664b2d2011-01-13 15:47:17 -0800618 case MADV_NOHUGEPAGE:
Andrea Arcangeli0af4e982011-01-13 15:46:55 -0800619#endif
Jason Baronaccb61f2012-03-23 15:02:51 -0700620 case MADV_DONTDUMP:
621 case MADV_DODUMP:
Nicholas Krause1ecef9e2015-09-04 15:48:24 -0700622 return true;
Nick Piggin75927af2009-06-16 15:32:38 -0700623
624 default:
Nicholas Krause1ecef9e2015-09-04 15:48:24 -0700625 return false;
Nick Piggin75927af2009-06-16 15:32:38 -0700626 }
627}
Hugh Dickins3866ea92009-09-21 17:01:52 -0700628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629/*
630 * The madvise(2) system call.
631 *
632 * Applications can use madvise() to advise the kernel how it should
633 * handle paging I/O in this VM area. The idea is to help the kernel
634 * use appropriate read-ahead and caching techniques. The information
635 * provided is advisory only, and can be safely disregarded by the
636 * kernel without affecting the correct operation of the application.
637 *
638 * behavior values:
639 * MADV_NORMAL - the default behavior is to read clusters. This
640 * results in some read-ahead and read-behind.
641 * MADV_RANDOM - the system should read the minimum amount of data
642 * on any access, since it is unlikely that the appli-
643 * cation will need more than what it asks for.
644 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
645 * once, so they can be aggressively read ahead, and
646 * can be freed soon after they are accessed.
647 * MADV_WILLNEED - the application is notifying the system to read
648 * some pages ahead.
649 * MADV_DONTNEED - the application is finished with the given range,
650 * so the kernel can free resources associated with it.
Naoya Horiguchid7206a72016-03-15 14:56:58 -0700651 * MADV_FREE - the application marks pages in the given range as lazy free,
652 * where actual purges are postponed until memory pressure happens.
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800653 * MADV_REMOVE - the application wants to free up the given range of
654 * pages and associated backing store.
Hugh Dickins3866ea92009-09-21 17:01:52 -0700655 * MADV_DONTFORK - omit this area from child's address space when forking:
656 * typically, to avoid COWing pages pinned by get_user_pages().
657 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
Naoya Horiguchid7206a72016-03-15 14:56:58 -0700658 * MADV_HWPOISON - trigger memory error handler as if the given memory range
659 * were corrupted by unrecoverable hardware memory failure.
660 * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory.
Hugh Dickinsf8af4da2009-09-21 17:01:57 -0700661 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
662 * this area with pages of identical content from other such areas.
663 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
Naoya Horiguchid7206a72016-03-15 14:56:58 -0700664 * MADV_HUGEPAGE - the application wants to back the given range by transparent
665 * huge pages in the future. Existing pages might be coalesced and
666 * new pages might be allocated as THP.
667 * MADV_NOHUGEPAGE - mark the given range as not worth being backed by
668 * transparent huge pages so the existing pages will not be
669 * coalesced into THP and new pages will not be allocated as THP.
670 * MADV_DONTDUMP - the application wants to prevent pages in the given range
671 * from being included in its core dump.
672 * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *
674 * return values:
675 * zero - success
676 * -EINVAL - start + len < 0, start is not page-aligned,
677 * "behavior" is not a valid value, or application
678 * is attempting to release locked or shared pages.
679 * -ENOMEM - addresses in the specified range are not currently
680 * mapped, or are outside the AS of the process.
681 * -EIO - an I/O error occurred while paging in data.
682 * -EBADF - map exists, but area maps something that isn't a file.
683 * -EAGAIN - a kernel resource was temporarily unavailable.
684 */
Heiko Carstens3480b252009-01-14 14:14:16 +0100685SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Prasanna Meda05b74382005-06-21 17:14:37 -0700687 unsigned long end, tmp;
Vladimir Cernovec9bed92013-09-11 14:20:15 -0700688 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 int unmapped_error = 0;
690 int error = -EINVAL;
Jason Baronf7977792007-07-15 23:38:21 -0700691 int write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 size_t len;
Shaohua Li1998cc02013-02-22 16:32:31 -0800693 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Andi Kleen9893e492009-09-16 11:50:17 +0200695#ifdef CONFIG_MEMORY_FAILURE
Andi Kleenafcf9382009-12-16 12:20:00 +0100696 if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE)
697 return madvise_hwpoison(behavior, start, start+len_in);
Andi Kleen9893e492009-09-16 11:50:17 +0200698#endif
Nick Piggin75927af2009-06-16 15:32:38 -0700699 if (!madvise_behavior_valid(behavior))
700 return error;
701
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700702 if (start & ~PAGE_MASK)
703 return error;
704 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
705
706 /* Check to see whether len was rounded up from small -ve to zero */
707 if (len_in && !len)
708 return error;
709
710 end = start + len;
711 if (end < start)
712 return error;
713
714 error = 0;
715 if (end == start)
716 return error;
717
Jason Baronf7977792007-07-15 23:38:21 -0700718 write = madvise_need_mmap_write(behavior);
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700719 if (write) {
720 if (down_write_killable(&current->mm->mmap_sem))
721 return -EINTR;
722 } else {
Nick Piggin0a27a142007-05-06 14:49:53 -0700723 down_read(&current->mm->mmap_sem);
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700724 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 /*
727 * If the interval [start,end) covers some unmapped address
728 * ranges, just ignore them, but return -ENOMEM at the end.
Prasanna Meda05b74382005-06-21 17:14:37 -0700729 * - different from the way of handling in mlock etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 */
Prasanna Meda05b74382005-06-21 17:14:37 -0700731 vma = find_vma_prev(current->mm, start, &prev);
Hugh Dickins836d5ff2005-09-03 15:54:53 -0700732 if (vma && start > vma->vm_start)
733 prev = vma;
734
Shaohua Li1998cc02013-02-22 16:32:31 -0800735 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 for (;;) {
737 /* Still start < end. */
738 error = -ENOMEM;
739 if (!vma)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700740 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Prasanna Meda05b74382005-06-21 17:14:37 -0700742 /* Here start < (end|vma->vm_end). */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 if (start < vma->vm_start) {
744 unmapped_error = -ENOMEM;
745 start = vma->vm_start;
Prasanna Meda05b74382005-06-21 17:14:37 -0700746 if (start >= end)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700747 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
749
Prasanna Meda05b74382005-06-21 17:14:37 -0700750 /* Here vma->vm_start <= start < (end|vma->vm_end) */
751 tmp = vma->vm_end;
752 if (end < tmp)
753 tmp = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Prasanna Meda05b74382005-06-21 17:14:37 -0700755 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
756 error = madvise_vma(vma, &prev, start, tmp, behavior);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 if (error)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700758 goto out;
Prasanna Meda05b74382005-06-21 17:14:37 -0700759 start = tmp;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700760 if (prev && start < prev->vm_end)
Prasanna Meda05b74382005-06-21 17:14:37 -0700761 start = prev->vm_end;
762 error = unmapped_error;
763 if (start >= end)
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700764 goto out;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700765 if (prev)
766 vma = prev->vm_next;
767 else /* madvise_remove dropped mmap_sem */
768 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770out:
Rasmus Villemoes84d96d82013-04-29 15:08:23 -0700771 blk_finish_plug(&plug);
Jason Baronf7977792007-07-15 23:38:21 -0700772 if (write)
Nick Piggin0a27a142007-05-06 14:49:53 -0700773 up_write(&current->mm->mmap_sem);
774 else
775 up_read(&current->mm->mmap_sem);
776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 return error;
778}