]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
05b74384 | 11 | #include <linux/mempolicy.h> |
afcf938e | 12 | #include <linux/page-isolation.h> |
05ce7724 | 13 | #include <linux/userfaultfd_k.h> |
1da177e4 | 14 | #include <linux/hugetlb.h> |
3f31d075 | 15 | #include <linux/falloc.h> |
e8edc6e0 | 16 | #include <linux/sched.h> |
f8af4da3 | 17 | #include <linux/ksm.h> |
3f31d075 | 18 | #include <linux/fs.h> |
9ab4233d | 19 | #include <linux/file.h> |
1998cc04 | 20 | #include <linux/blkdev.h> |
66114cad | 21 | #include <linux/backing-dev.h> |
1998cc04 SL |
22 | #include <linux/swap.h> |
23 | #include <linux/swapops.h> | |
854e9ed0 MK |
24 | #include <linux/mmu_notifier.h> |
25 | ||
26 | #include <asm/tlb.h> | |
1da177e4 | 27 | |
23519073 KS |
28 | #include "internal.h" |
29 | ||
0a27a14a NP |
30 | /* |
31 | * Any behaviour which results in changes to the vma->vm_flags needs to | |
32 | * take mmap_sem for writing. Others, which simply traverse vmas, need | |
33 | * to only take it for reading. | |
34 | */ | |
35 | static int madvise_need_mmap_write(int behavior) | |
36 | { | |
37 | switch (behavior) { | |
38 | case MADV_REMOVE: | |
39 | case MADV_WILLNEED: | |
40 | case MADV_DONTNEED: | |
854e9ed0 | 41 | case MADV_FREE: |
0a27a14a NP |
42 | return 0; |
43 | default: | |
44 | /* be safe, default to 1. list exceptions explicitly */ | |
45 | return 1; | |
46 | } | |
47 | } | |
48 | ||
1da177e4 LT |
49 | /* |
50 | * We can potentially split a vm area into separate | |
51 | * areas, each area with its own behavior. | |
52 | */ | |
ec9bed9d | 53 | static long madvise_behavior(struct vm_area_struct *vma, |
05b74384 PM |
54 | struct vm_area_struct **prev, |
55 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 56 | { |
ec9bed9d | 57 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 58 | int error = 0; |
05b74384 | 59 | pgoff_t pgoff; |
3866ea90 | 60 | unsigned long new_flags = vma->vm_flags; |
e798c6e8 PM |
61 | |
62 | switch (behavior) { | |
f8225661 MT |
63 | case MADV_NORMAL: |
64 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
65 | break; | |
e798c6e8 | 66 | case MADV_SEQUENTIAL: |
f8225661 | 67 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; |
e798c6e8 PM |
68 | break; |
69 | case MADV_RANDOM: | |
f8225661 | 70 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; |
e798c6e8 | 71 | break; |
f8225661 MT |
72 | case MADV_DONTFORK: |
73 | new_flags |= VM_DONTCOPY; | |
74 | break; | |
75 | case MADV_DOFORK: | |
3866ea90 HD |
76 | if (vma->vm_flags & VM_IO) { |
77 | error = -EINVAL; | |
78 | goto out; | |
79 | } | |
f8225661 | 80 | new_flags &= ~VM_DONTCOPY; |
e798c6e8 | 81 | break; |
accb61fe | 82 | case MADV_DONTDUMP: |
0103bd16 | 83 | new_flags |= VM_DONTDUMP; |
accb61fe JB |
84 | break; |
85 | case MADV_DODUMP: | |
0103bd16 KK |
86 | if (new_flags & VM_SPECIAL) { |
87 | error = -EINVAL; | |
88 | goto out; | |
89 | } | |
90 | new_flags &= ~VM_DONTDUMP; | |
accb61fe | 91 | break; |
f8af4da3 HD |
92 | case MADV_MERGEABLE: |
93 | case MADV_UNMERGEABLE: | |
94 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
def5efe0 DR |
95 | if (error) { |
96 | /* | |
97 | * madvise() returns EAGAIN if kernel resources, such as | |
98 | * slab, are temporarily unavailable. | |
99 | */ | |
100 | if (error == -ENOMEM) | |
101 | error = -EAGAIN; | |
f8af4da3 | 102 | goto out; |
def5efe0 | 103 | } |
f8af4da3 | 104 | break; |
0af4e98b | 105 | case MADV_HUGEPAGE: |
a664b2d8 | 106 | case MADV_NOHUGEPAGE: |
60ab3244 | 107 | error = hugepage_madvise(vma, &new_flags, behavior); |
def5efe0 DR |
108 | if (error) { |
109 | /* | |
110 | * madvise() returns EAGAIN if kernel resources, such as | |
111 | * slab, are temporarily unavailable. | |
112 | */ | |
113 | if (error == -ENOMEM) | |
114 | error = -EAGAIN; | |
0af4e98b | 115 | goto out; |
def5efe0 | 116 | } |
0af4e98b | 117 | break; |
e798c6e8 PM |
118 | } |
119 | ||
05b74384 PM |
120 | if (new_flags == vma->vm_flags) { |
121 | *prev = vma; | |
836d5ffd | 122 | goto out; |
05b74384 PM |
123 | } |
124 | ||
125 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
126 | *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma, | |
19a809af AA |
127 | vma->vm_file, pgoff, vma_policy(vma), |
128 | vma->vm_userfaultfd_ctx); | |
05b74384 PM |
129 | if (*prev) { |
130 | vma = *prev; | |
131 | goto success; | |
132 | } | |
133 | ||
134 | *prev = vma; | |
1da177e4 LT |
135 | |
136 | if (start != vma->vm_start) { | |
def5efe0 DR |
137 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
138 | error = -ENOMEM; | |
1da177e4 | 139 | goto out; |
def5efe0 DR |
140 | } |
141 | error = __split_vma(mm, vma, start, 1); | |
142 | if (error) { | |
143 | /* | |
144 | * madvise() returns EAGAIN if kernel resources, such as | |
145 | * slab, are temporarily unavailable. | |
146 | */ | |
147 | if (error == -ENOMEM) | |
148 | error = -EAGAIN; | |
149 | goto out; | |
150 | } | |
1da177e4 LT |
151 | } |
152 | ||
153 | if (end != vma->vm_end) { | |
def5efe0 DR |
154 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
155 | error = -ENOMEM; | |
1da177e4 | 156 | goto out; |
def5efe0 DR |
157 | } |
158 | error = __split_vma(mm, vma, end, 0); | |
159 | if (error) { | |
160 | /* | |
161 | * madvise() returns EAGAIN if kernel resources, such as | |
162 | * slab, are temporarily unavailable. | |
163 | */ | |
164 | if (error == -ENOMEM) | |
165 | error = -EAGAIN; | |
166 | goto out; | |
167 | } | |
1da177e4 LT |
168 | } |
169 | ||
836d5ffd | 170 | success: |
1da177e4 LT |
171 | /* |
172 | * vm_flags is protected by the mmap_sem held in write mode. | |
173 | */ | |
e798c6e8 | 174 | vma->vm_flags = new_flags; |
1da177e4 | 175 | out: |
1da177e4 LT |
176 | return error; |
177 | } | |
178 | ||
1998cc04 SL |
179 | #ifdef CONFIG_SWAP |
180 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
181 | unsigned long end, struct mm_walk *walk) | |
182 | { | |
183 | pte_t *orig_pte; | |
184 | struct vm_area_struct *vma = walk->private; | |
185 | unsigned long index; | |
186 | ||
187 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
188 | return 0; | |
189 | ||
190 | for (index = start; index != end; index += PAGE_SIZE) { | |
191 | pte_t pte; | |
192 | swp_entry_t entry; | |
193 | struct page *page; | |
194 | spinlock_t *ptl; | |
195 | ||
196 | orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); | |
197 | pte = *(orig_pte + ((index - start) / PAGE_SIZE)); | |
198 | pte_unmap_unlock(orig_pte, ptl); | |
199 | ||
0661a336 | 200 | if (pte_present(pte) || pte_none(pte)) |
1998cc04 SL |
201 | continue; |
202 | entry = pte_to_swp_entry(pte); | |
203 | if (unlikely(non_swap_entry(entry))) | |
204 | continue; | |
205 | ||
206 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, | |
207 | vma, index); | |
208 | if (page) | |
09cbfeaf | 209 | put_page(page); |
1998cc04 SL |
210 | } |
211 | ||
212 | return 0; | |
213 | } | |
214 | ||
215 | static void force_swapin_readahead(struct vm_area_struct *vma, | |
216 | unsigned long start, unsigned long end) | |
217 | { | |
218 | struct mm_walk walk = { | |
219 | .mm = vma->vm_mm, | |
220 | .pmd_entry = swapin_walk_pmd_entry, | |
221 | .private = vma, | |
222 | }; | |
223 | ||
224 | walk_page_range(start, end, &walk); | |
225 | ||
226 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
227 | } | |
228 | ||
229 | static void force_shm_swapin_readahead(struct vm_area_struct *vma, | |
230 | unsigned long start, unsigned long end, | |
231 | struct address_space *mapping) | |
232 | { | |
233 | pgoff_t index; | |
234 | struct page *page; | |
235 | swp_entry_t swap; | |
236 | ||
237 | for (; start < end; start += PAGE_SIZE) { | |
238 | index = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
239 | ||
55231e5c | 240 | page = find_get_entry(mapping, index); |
1998cc04 SL |
241 | if (!radix_tree_exceptional_entry(page)) { |
242 | if (page) | |
09cbfeaf | 243 | put_page(page); |
1998cc04 SL |
244 | continue; |
245 | } | |
246 | swap = radix_to_swp_entry(page); | |
247 | page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE, | |
248 | NULL, 0); | |
249 | if (page) | |
09cbfeaf | 250 | put_page(page); |
1998cc04 SL |
251 | } |
252 | ||
253 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
254 | } | |
255 | #endif /* CONFIG_SWAP */ | |
256 | ||
1da177e4 LT |
257 | /* |
258 | * Schedule all required I/O operations. Do not wait for completion. | |
259 | */ | |
ec9bed9d VC |
260 | static long madvise_willneed(struct vm_area_struct *vma, |
261 | struct vm_area_struct **prev, | |
1da177e4 LT |
262 | unsigned long start, unsigned long end) |
263 | { | |
264 | struct file *file = vma->vm_file; | |
265 | ||
1998cc04 | 266 | #ifdef CONFIG_SWAP |
97b713ba | 267 | if (!file) { |
1998cc04 | 268 | *prev = vma; |
97b713ba | 269 | force_swapin_readahead(vma, start, end); |
1998cc04 SL |
270 | return 0; |
271 | } | |
1998cc04 | 272 | |
97b713ba CH |
273 | if (shmem_mapping(file->f_mapping)) { |
274 | *prev = vma; | |
275 | force_shm_swapin_readahead(vma, start, end, | |
276 | file->f_mapping); | |
277 | return 0; | |
278 | } | |
279 | #else | |
1bef4003 S |
280 | if (!file) |
281 | return -EBADF; | |
97b713ba | 282 | #endif |
1bef4003 | 283 | |
e748dcd0 | 284 | if (IS_DAX(file_inode(file))) { |
fe77ba6f CO |
285 | /* no bad return value, but ignore advice */ |
286 | return 0; | |
287 | } | |
288 | ||
05b74384 | 289 | *prev = vma; |
1da177e4 LT |
290 | start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
291 | if (end > vma->vm_end) | |
292 | end = vma->vm_end; | |
293 | end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
294 | ||
f7e839dd | 295 | force_page_cache_readahead(file->f_mapping, file, start, end - start); |
1da177e4 LT |
296 | return 0; |
297 | } | |
298 | ||
854e9ed0 MK |
299 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
300 | unsigned long end, struct mm_walk *walk) | |
301 | ||
302 | { | |
303 | struct mmu_gather *tlb = walk->private; | |
304 | struct mm_struct *mm = tlb->mm; | |
305 | struct vm_area_struct *vma = walk->vma; | |
306 | spinlock_t *ptl; | |
307 | pte_t *orig_pte, *pte, ptent; | |
308 | struct page *page; | |
64b42bc1 | 309 | int nr_swap = 0; |
b8d3c4c3 MK |
310 | unsigned long next; |
311 | ||
312 | next = pmd_addr_end(addr, end); | |
313 | if (pmd_trans_huge(*pmd)) | |
314 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
315 | goto next; | |
854e9ed0 | 316 | |
854e9ed0 MK |
317 | if (pmd_trans_unstable(pmd)) |
318 | return 0; | |
319 | ||
07e32661 | 320 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
854e9ed0 MK |
321 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
322 | arch_enter_lazy_mmu_mode(); | |
323 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
324 | ptent = *pte; | |
325 | ||
64b42bc1 | 326 | if (pte_none(ptent)) |
854e9ed0 | 327 | continue; |
64b42bc1 MK |
328 | /* |
329 | * If the pte has swp_entry, just clear page table to | |
330 | * prevent swap-in which is more expensive rather than | |
331 | * (page allocation + zeroing). | |
332 | */ | |
333 | if (!pte_present(ptent)) { | |
334 | swp_entry_t entry; | |
335 | ||
336 | entry = pte_to_swp_entry(ptent); | |
337 | if (non_swap_entry(entry)) | |
338 | continue; | |
339 | nr_swap--; | |
340 | free_swap_and_cache(entry); | |
341 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
342 | continue; | |
343 | } | |
854e9ed0 MK |
344 | |
345 | page = vm_normal_page(vma, addr, ptent); | |
346 | if (!page) | |
347 | continue; | |
348 | ||
349 | /* | |
350 | * If pmd isn't transhuge but the page is THP and | |
351 | * is owned by only this process, split it and | |
352 | * deactivate all pages. | |
353 | */ | |
354 | if (PageTransCompound(page)) { | |
355 | if (page_mapcount(page) != 1) | |
356 | goto out; | |
357 | get_page(page); | |
358 | if (!trylock_page(page)) { | |
359 | put_page(page); | |
360 | goto out; | |
361 | } | |
362 | pte_unmap_unlock(orig_pte, ptl); | |
363 | if (split_huge_page(page)) { | |
364 | unlock_page(page); | |
365 | put_page(page); | |
366 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
367 | goto out; | |
368 | } | |
369 | put_page(page); | |
370 | unlock_page(page); | |
371 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
372 | pte--; | |
373 | addr -= PAGE_SIZE; | |
374 | continue; | |
375 | } | |
376 | ||
377 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
378 | ||
379 | if (PageSwapCache(page) || PageDirty(page)) { | |
380 | if (!trylock_page(page)) | |
381 | continue; | |
382 | /* | |
383 | * If page is shared with others, we couldn't clear | |
384 | * PG_dirty of the page. | |
385 | */ | |
386 | if (page_mapcount(page) != 1) { | |
387 | unlock_page(page); | |
388 | continue; | |
389 | } | |
390 | ||
391 | if (PageSwapCache(page) && !try_to_free_swap(page)) { | |
392 | unlock_page(page); | |
393 | continue; | |
394 | } | |
395 | ||
396 | ClearPageDirty(page); | |
397 | unlock_page(page); | |
398 | } | |
399 | ||
400 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
401 | /* | |
402 | * Some of architecture(ex, PPC) don't update TLB | |
403 | * with set_pte_at and tlb_remove_tlb_entry so for | |
404 | * the portability, remap the pte with old|clean | |
405 | * after pte clearing. | |
406 | */ | |
407 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
408 | tlb->fullmm); | |
409 | ||
410 | ptent = pte_mkold(ptent); | |
411 | ptent = pte_mkclean(ptent); | |
412 | set_pte_at(mm, addr, pte, ptent); | |
10853a03 MK |
413 | if (PageActive(page)) |
414 | deactivate_page(page); | |
854e9ed0 MK |
415 | tlb_remove_tlb_entry(tlb, pte, addr); |
416 | } | |
417 | } | |
418 | out: | |
64b42bc1 MK |
419 | if (nr_swap) { |
420 | if (current->mm == mm) | |
421 | sync_mm_rss(mm); | |
422 | ||
423 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); | |
424 | } | |
854e9ed0 MK |
425 | arch_leave_lazy_mmu_mode(); |
426 | pte_unmap_unlock(orig_pte, ptl); | |
427 | cond_resched(); | |
b8d3c4c3 | 428 | next: |
854e9ed0 MK |
429 | return 0; |
430 | } | |
431 | ||
432 | static void madvise_free_page_range(struct mmu_gather *tlb, | |
433 | struct vm_area_struct *vma, | |
434 | unsigned long addr, unsigned long end) | |
435 | { | |
436 | struct mm_walk free_walk = { | |
437 | .pmd_entry = madvise_free_pte_range, | |
438 | .mm = vma->vm_mm, | |
439 | .private = tlb, | |
440 | }; | |
441 | ||
442 | tlb_start_vma(tlb, vma); | |
443 | walk_page_range(addr, end, &free_walk); | |
444 | tlb_end_vma(tlb, vma); | |
445 | } | |
446 | ||
447 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
448 | unsigned long start_addr, unsigned long end_addr) | |
449 | { | |
450 | unsigned long start, end; | |
451 | struct mm_struct *mm = vma->vm_mm; | |
452 | struct mmu_gather tlb; | |
453 | ||
454 | if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP)) | |
455 | return -EINVAL; | |
456 | ||
457 | /* MADV_FREE works for only anon vma at the moment */ | |
458 | if (!vma_is_anonymous(vma)) | |
459 | return -EINVAL; | |
460 | ||
461 | start = max(vma->vm_start, start_addr); | |
462 | if (start >= vma->vm_end) | |
463 | return -EINVAL; | |
464 | end = min(vma->vm_end, end_addr); | |
465 | if (end <= vma->vm_start) | |
466 | return -EINVAL; | |
467 | ||
468 | lru_add_drain(); | |
469 | tlb_gather_mmu(&tlb, mm, start, end); | |
470 | update_hiwater_rss(mm); | |
471 | ||
472 | mmu_notifier_invalidate_range_start(mm, start, end); | |
473 | madvise_free_page_range(&tlb, vma, start, end); | |
474 | mmu_notifier_invalidate_range_end(mm, start, end); | |
475 | tlb_finish_mmu(&tlb, start, end); | |
476 | ||
477 | return 0; | |
478 | } | |
479 | ||
480 | static long madvise_free(struct vm_area_struct *vma, | |
481 | struct vm_area_struct **prev, | |
482 | unsigned long start, unsigned long end) | |
483 | { | |
484 | *prev = vma; | |
485 | return madvise_free_single_vma(vma, start, end); | |
486 | } | |
487 | ||
1da177e4 LT |
488 | /* |
489 | * Application no longer needs these pages. If the pages are dirty, | |
490 | * it's OK to just throw them away. The app will be more careful about | |
491 | * data it wants to keep. Be sure to free swap resources too. The | |
7e6cbea3 | 492 | * zap_page_range call sets things up for shrink_active_list to actually free |
1da177e4 LT |
493 | * these pages later if no one else has touched them in the meantime, |
494 | * although we could add these pages to a global reuse list for | |
7e6cbea3 | 495 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
496 | * |
497 | * NB: This interface discards data rather than pushes it out to swap, | |
498 | * as some implementations do. This has performance implications for | |
499 | * applications like large transactional databases which want to discard | |
500 | * pages in anonymous maps after committing to backing store the data | |
501 | * that was kept in them. There is no reason to write this data out to | |
502 | * the swap area if the application is discarding it. | |
503 | * | |
504 | * An interface that causes the system to free clean pages and flush | |
505 | * dirty pages is already available as msync(MS_INVALIDATE). | |
506 | */ | |
ec9bed9d VC |
507 | static long madvise_dontneed(struct vm_area_struct *vma, |
508 | struct vm_area_struct **prev, | |
1da177e4 LT |
509 | unsigned long start, unsigned long end) |
510 | { | |
05b74384 | 511 | *prev = vma; |
23519073 | 512 | if (!can_madv_dontneed_vma(vma)) |
1da177e4 LT |
513 | return -EINVAL; |
514 | ||
d811914d | 515 | userfaultfd_remove(vma, prev, start, end); |
ecf1385d | 516 | zap_page_range(vma, start, end - start); |
1da177e4 LT |
517 | return 0; |
518 | } | |
519 | ||
f6b3ec23 BP |
520 | /* |
521 | * Application wants to free up the pages and associated backing store. | |
522 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
523 | */ |
524 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 525 | struct vm_area_struct **prev, |
f6b3ec23 BP |
526 | unsigned long start, unsigned long end) |
527 | { | |
3f31d075 | 528 | loff_t offset; |
90ed52eb | 529 | int error; |
9ab4233d | 530 | struct file *f; |
f6b3ec23 | 531 | |
90ed52eb | 532 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ |
00e9fa2d | 533 | |
72079ba0 | 534 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
535 | return -EINVAL; |
536 | ||
9ab4233d AL |
537 | f = vma->vm_file; |
538 | ||
539 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
540 | return -EINVAL; |
541 | } | |
542 | ||
69cf0fac HD |
543 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
544 | return -EACCES; | |
545 | ||
f6b3ec23 BP |
546 | offset = (loff_t)(start - vma->vm_start) |
547 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 548 | |
9ab4233d AL |
549 | /* |
550 | * Filesystem's fallocate may need to take i_mutex. We need to | |
551 | * explicitly grab a reference because the vma (and hence the | |
552 | * vma's reference to the file) can go away as soon as we drop | |
553 | * mmap_sem. | |
554 | */ | |
555 | get_file(f); | |
a6bf53eb | 556 | userfaultfd_remove(vma, prev, start, end); |
0a27a14a | 557 | up_read(¤t->mm->mmap_sem); |
72c72bdf | 558 | error = vfs_fallocate(f, |
3f31d075 HD |
559 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
560 | offset, end - start); | |
9ab4233d | 561 | fput(f); |
0a27a14a | 562 | down_read(¤t->mm->mmap_sem); |
90ed52eb | 563 | return error; |
f6b3ec23 BP |
564 | } |
565 | ||
9893e49d AK |
566 | #ifdef CONFIG_MEMORY_FAILURE |
567 | /* | |
568 | * Error injection support for memory error handling. | |
569 | */ | |
afcf938e | 570 | static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end) |
9893e49d | 571 | { |
20cb6cab | 572 | struct page *p; |
9893e49d AK |
573 | if (!capable(CAP_SYS_ADMIN)) |
574 | return -EPERM; | |
20cb6cab WL |
575 | for (; start < end; start += PAGE_SIZE << |
576 | compound_order(compound_head(p))) { | |
325c4ef5 AM |
577 | int ret; |
578 | ||
579 | ret = get_user_pages_fast(start, 1, 0, &p); | |
9893e49d AK |
580 | if (ret != 1) |
581 | return ret; | |
325c4ef5 | 582 | |
29b4eede WL |
583 | if (PageHWPoison(p)) { |
584 | put_page(p); | |
585 | continue; | |
586 | } | |
afcf938e | 587 | if (bhv == MADV_SOFT_OFFLINE) { |
b194b8cd | 588 | pr_info("Soft offlining page %#lx at %#lx\n", |
afcf938e AK |
589 | page_to_pfn(p), start); |
590 | ret = soft_offline_page(p, MF_COUNT_INCREASED); | |
591 | if (ret) | |
8302423b | 592 | return ret; |
afcf938e AK |
593 | continue; |
594 | } | |
b194b8cd | 595 | pr_info("Injecting memory failure for page %#lx at %#lx\n", |
9893e49d | 596 | page_to_pfn(p), start); |
23a003bf NH |
597 | ret = memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED); |
598 | if (ret) | |
599 | return ret; | |
9893e49d | 600 | } |
325c4ef5 | 601 | return 0; |
9893e49d AK |
602 | } |
603 | #endif | |
604 | ||
165cd402 | 605 | static long |
606 | madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, | |
607 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 608 | { |
1da177e4 | 609 | switch (behavior) { |
f6b3ec23 | 610 | case MADV_REMOVE: |
3866ea90 | 611 | return madvise_remove(vma, prev, start, end); |
1da177e4 | 612 | case MADV_WILLNEED: |
3866ea90 | 613 | return madvise_willneed(vma, prev, start, end); |
854e9ed0 MK |
614 | case MADV_FREE: |
615 | /* | |
616 | * XXX: In this implementation, MADV_FREE works like | |
617 | * MADV_DONTNEED on swapless system or full swap. | |
618 | */ | |
619 | if (get_nr_swap_pages() > 0) | |
620 | return madvise_free(vma, prev, start, end); | |
621 | /* passthrough */ | |
1da177e4 | 622 | case MADV_DONTNEED: |
3866ea90 | 623 | return madvise_dontneed(vma, prev, start, end); |
1da177e4 | 624 | default: |
3866ea90 | 625 | return madvise_behavior(vma, prev, start, end, behavior); |
1da177e4 | 626 | } |
1da177e4 LT |
627 | } |
628 | ||
1ecef9ed | 629 | static bool |
75927af8 NP |
630 | madvise_behavior_valid(int behavior) |
631 | { | |
632 | switch (behavior) { | |
633 | case MADV_DOFORK: | |
634 | case MADV_DONTFORK: | |
635 | case MADV_NORMAL: | |
636 | case MADV_SEQUENTIAL: | |
637 | case MADV_RANDOM: | |
638 | case MADV_REMOVE: | |
639 | case MADV_WILLNEED: | |
640 | case MADV_DONTNEED: | |
854e9ed0 | 641 | case MADV_FREE: |
f8af4da3 HD |
642 | #ifdef CONFIG_KSM |
643 | case MADV_MERGEABLE: | |
644 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
645 | #endif |
646 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
647 | case MADV_HUGEPAGE: | |
a664b2d8 | 648 | case MADV_NOHUGEPAGE: |
f8af4da3 | 649 | #endif |
accb61fe JB |
650 | case MADV_DONTDUMP: |
651 | case MADV_DODUMP: | |
1ecef9ed | 652 | return true; |
75927af8 NP |
653 | |
654 | default: | |
1ecef9ed | 655 | return false; |
75927af8 NP |
656 | } |
657 | } | |
3866ea90 | 658 | |
1da177e4 LT |
659 | /* |
660 | * The madvise(2) system call. | |
661 | * | |
662 | * Applications can use madvise() to advise the kernel how it should | |
663 | * handle paging I/O in this VM area. The idea is to help the kernel | |
664 | * use appropriate read-ahead and caching techniques. The information | |
665 | * provided is advisory only, and can be safely disregarded by the | |
666 | * kernel without affecting the correct operation of the application. | |
667 | * | |
668 | * behavior values: | |
669 | * MADV_NORMAL - the default behavior is to read clusters. This | |
670 | * results in some read-ahead and read-behind. | |
671 | * MADV_RANDOM - the system should read the minimum amount of data | |
672 | * on any access, since it is unlikely that the appli- | |
673 | * cation will need more than what it asks for. | |
674 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
675 | * once, so they can be aggressively read ahead, and | |
676 | * can be freed soon after they are accessed. | |
677 | * MADV_WILLNEED - the application is notifying the system to read | |
678 | * some pages ahead. | |
679 | * MADV_DONTNEED - the application is finished with the given range, | |
680 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
681 | * MADV_FREE - the application marks pages in the given range as lazy free, |
682 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
683 | * MADV_REMOVE - the application wants to free up the given range of |
684 | * pages and associated backing store. | |
3866ea90 HD |
685 | * MADV_DONTFORK - omit this area from child's address space when forking: |
686 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
687 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
d7206a70 NH |
688 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
689 | * were corrupted by unrecoverable hardware memory failure. | |
690 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
691 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
692 | * this area with pages of identical content from other such areas. | |
693 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
694 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
695 | * huge pages in the future. Existing pages might be coalesced and | |
696 | * new pages might be allocated as THP. | |
697 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
698 | * transparent huge pages so the existing pages will not be | |
699 | * coalesced into THP and new pages will not be allocated as THP. | |
700 | * MADV_DONTDUMP - the application wants to prevent pages in the given range | |
701 | * from being included in its core dump. | |
702 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
1da177e4 LT |
703 | * |
704 | * return values: | |
705 | * zero - success | |
706 | * -EINVAL - start + len < 0, start is not page-aligned, | |
707 | * "behavior" is not a valid value, or application | |
708 | * is attempting to release locked or shared pages. | |
709 | * -ENOMEM - addresses in the specified range are not currently | |
710 | * mapped, or are outside the AS of the process. | |
711 | * -EIO - an I/O error occurred while paging in data. | |
712 | * -EBADF - map exists, but area maps something that isn't a file. | |
713 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
714 | */ | |
3480b257 | 715 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) |
1da177e4 | 716 | { |
05b74384 | 717 | unsigned long end, tmp; |
ec9bed9d | 718 | struct vm_area_struct *vma, *prev; |
1da177e4 LT |
719 | int unmapped_error = 0; |
720 | int error = -EINVAL; | |
f7977793 | 721 | int write; |
1da177e4 | 722 | size_t len; |
1998cc04 | 723 | struct blk_plug plug; |
1da177e4 | 724 | |
9893e49d | 725 | #ifdef CONFIG_MEMORY_FAILURE |
afcf938e AK |
726 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) |
727 | return madvise_hwpoison(behavior, start, start+len_in); | |
9893e49d | 728 | #endif |
75927af8 NP |
729 | if (!madvise_behavior_valid(behavior)) |
730 | return error; | |
731 | ||
1da177e4 | 732 | if (start & ~PAGE_MASK) |
84d96d89 | 733 | return error; |
1da177e4 LT |
734 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; |
735 | ||
736 | /* Check to see whether len was rounded up from small -ve to zero */ | |
737 | if (len_in && !len) | |
84d96d89 | 738 | return error; |
1da177e4 LT |
739 | |
740 | end = start + len; | |
741 | if (end < start) | |
84d96d89 | 742 | return error; |
1da177e4 LT |
743 | |
744 | error = 0; | |
745 | if (end == start) | |
84d96d89 RV |
746 | return error; |
747 | ||
748 | write = madvise_need_mmap_write(behavior); | |
dc0ef0df MH |
749 | if (write) { |
750 | if (down_write_killable(¤t->mm->mmap_sem)) | |
751 | return -EINTR; | |
752 | } else { | |
84d96d89 | 753 | down_read(¤t->mm->mmap_sem); |
dc0ef0df | 754 | } |
1da177e4 LT |
755 | |
756 | /* | |
757 | * If the interval [start,end) covers some unmapped address | |
758 | * ranges, just ignore them, but return -ENOMEM at the end. | |
05b74384 | 759 | * - different from the way of handling in mlock etc. |
1da177e4 | 760 | */ |
05b74384 | 761 | vma = find_vma_prev(current->mm, start, &prev); |
836d5ffd HD |
762 | if (vma && start > vma->vm_start) |
763 | prev = vma; | |
764 | ||
1998cc04 | 765 | blk_start_plug(&plug); |
1da177e4 LT |
766 | for (;;) { |
767 | /* Still start < end. */ | |
768 | error = -ENOMEM; | |
769 | if (!vma) | |
84d96d89 | 770 | goto out; |
1da177e4 | 771 | |
05b74384 | 772 | /* Here start < (end|vma->vm_end). */ |
1da177e4 LT |
773 | if (start < vma->vm_start) { |
774 | unmapped_error = -ENOMEM; | |
775 | start = vma->vm_start; | |
05b74384 | 776 | if (start >= end) |
84d96d89 | 777 | goto out; |
1da177e4 LT |
778 | } |
779 | ||
05b74384 PM |
780 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ |
781 | tmp = vma->vm_end; | |
782 | if (end < tmp) | |
783 | tmp = end; | |
1da177e4 | 784 | |
05b74384 PM |
785 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ |
786 | error = madvise_vma(vma, &prev, start, tmp, behavior); | |
1da177e4 | 787 | if (error) |
84d96d89 | 788 | goto out; |
05b74384 | 789 | start = tmp; |
90ed52eb | 790 | if (prev && start < prev->vm_end) |
05b74384 PM |
791 | start = prev->vm_end; |
792 | error = unmapped_error; | |
793 | if (start >= end) | |
84d96d89 | 794 | goto out; |
90ed52eb HD |
795 | if (prev) |
796 | vma = prev->vm_next; | |
797 | else /* madvise_remove dropped mmap_sem */ | |
798 | vma = find_vma(current->mm, start); | |
1da177e4 | 799 | } |
1da177e4 | 800 | out: |
84d96d89 | 801 | blk_finish_plug(&plug); |
f7977793 | 802 | if (write) |
0a27a14a NP |
803 | up_write(¤t->mm->mmap_sem); |
804 | else | |
805 | up_read(¤t->mm->mmap_sem); | |
806 | ||
1da177e4 LT |
807 | return error; |
808 | } |