]>
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> | |
3a4f8a0b | 24 | #include <linux/shmem_fs.h> |
854e9ed0 MK |
25 | #include <linux/mmu_notifier.h> |
26 | ||
27 | #include <asm/tlb.h> | |
1da177e4 | 28 | |
23519073 KS |
29 | #include "internal.h" |
30 | ||
0a27a14a NP |
31 | /* |
32 | * Any behaviour which results in changes to the vma->vm_flags needs to | |
33 | * take mmap_sem for writing. Others, which simply traverse vmas, need | |
34 | * to only take it for reading. | |
35 | */ | |
36 | static int madvise_need_mmap_write(int behavior) | |
37 | { | |
38 | switch (behavior) { | |
39 | case MADV_REMOVE: | |
40 | case MADV_WILLNEED: | |
41 | case MADV_DONTNEED: | |
854e9ed0 | 42 | case MADV_FREE: |
0a27a14a NP |
43 | return 0; |
44 | default: | |
45 | /* be safe, default to 1. list exceptions explicitly */ | |
46 | return 1; | |
47 | } | |
48 | } | |
49 | ||
1da177e4 LT |
50 | /* |
51 | * We can potentially split a vm area into separate | |
52 | * areas, each area with its own behavior. | |
53 | */ | |
ec9bed9d | 54 | static long madvise_behavior(struct vm_area_struct *vma, |
05b74384 PM |
55 | struct vm_area_struct **prev, |
56 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 57 | { |
ec9bed9d | 58 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 59 | int error = 0; |
05b74384 | 60 | pgoff_t pgoff; |
3866ea90 | 61 | unsigned long new_flags = vma->vm_flags; |
e798c6e8 PM |
62 | |
63 | switch (behavior) { | |
f8225661 MT |
64 | case MADV_NORMAL: |
65 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
66 | break; | |
e798c6e8 | 67 | case MADV_SEQUENTIAL: |
f8225661 | 68 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; |
e798c6e8 PM |
69 | break; |
70 | case MADV_RANDOM: | |
f8225661 | 71 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; |
e798c6e8 | 72 | break; |
f8225661 MT |
73 | case MADV_DONTFORK: |
74 | new_flags |= VM_DONTCOPY; | |
75 | break; | |
76 | case MADV_DOFORK: | |
3866ea90 HD |
77 | if (vma->vm_flags & VM_IO) { |
78 | error = -EINVAL; | |
79 | goto out; | |
80 | } | |
f8225661 | 81 | new_flags &= ~VM_DONTCOPY; |
e798c6e8 | 82 | break; |
accb61fe | 83 | case MADV_DONTDUMP: |
0103bd16 | 84 | new_flags |= VM_DONTDUMP; |
accb61fe JB |
85 | break; |
86 | case MADV_DODUMP: | |
0103bd16 KK |
87 | if (new_flags & VM_SPECIAL) { |
88 | error = -EINVAL; | |
89 | goto out; | |
90 | } | |
91 | new_flags &= ~VM_DONTDUMP; | |
accb61fe | 92 | break; |
f8af4da3 HD |
93 | case MADV_MERGEABLE: |
94 | case MADV_UNMERGEABLE: | |
95 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
def5efe0 DR |
96 | if (error) { |
97 | /* | |
98 | * madvise() returns EAGAIN if kernel resources, such as | |
99 | * slab, are temporarily unavailable. | |
100 | */ | |
101 | if (error == -ENOMEM) | |
102 | error = -EAGAIN; | |
f8af4da3 | 103 | goto out; |
def5efe0 | 104 | } |
f8af4da3 | 105 | break; |
0af4e98b | 106 | case MADV_HUGEPAGE: |
a664b2d8 | 107 | case MADV_NOHUGEPAGE: |
60ab3244 | 108 | error = hugepage_madvise(vma, &new_flags, behavior); |
def5efe0 DR |
109 | if (error) { |
110 | /* | |
111 | * madvise() returns EAGAIN if kernel resources, such as | |
112 | * slab, are temporarily unavailable. | |
113 | */ | |
114 | if (error == -ENOMEM) | |
115 | error = -EAGAIN; | |
0af4e98b | 116 | goto out; |
def5efe0 | 117 | } |
0af4e98b | 118 | break; |
e798c6e8 PM |
119 | } |
120 | ||
05b74384 PM |
121 | if (new_flags == vma->vm_flags) { |
122 | *prev = vma; | |
836d5ffd | 123 | goto out; |
05b74384 PM |
124 | } |
125 | ||
126 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
127 | *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma, | |
19a809af AA |
128 | vma->vm_file, pgoff, vma_policy(vma), |
129 | vma->vm_userfaultfd_ctx); | |
05b74384 PM |
130 | if (*prev) { |
131 | vma = *prev; | |
132 | goto success; | |
133 | } | |
134 | ||
135 | *prev = vma; | |
1da177e4 LT |
136 | |
137 | if (start != vma->vm_start) { | |
def5efe0 DR |
138 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
139 | error = -ENOMEM; | |
1da177e4 | 140 | goto out; |
def5efe0 DR |
141 | } |
142 | error = __split_vma(mm, vma, start, 1); | |
143 | if (error) { | |
144 | /* | |
145 | * madvise() returns EAGAIN if kernel resources, such as | |
146 | * slab, are temporarily unavailable. | |
147 | */ | |
148 | if (error == -ENOMEM) | |
149 | error = -EAGAIN; | |
150 | goto out; | |
151 | } | |
1da177e4 LT |
152 | } |
153 | ||
154 | if (end != vma->vm_end) { | |
def5efe0 DR |
155 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
156 | error = -ENOMEM; | |
1da177e4 | 157 | goto out; |
def5efe0 DR |
158 | } |
159 | error = __split_vma(mm, vma, end, 0); | |
160 | if (error) { | |
161 | /* | |
162 | * madvise() returns EAGAIN if kernel resources, such as | |
163 | * slab, are temporarily unavailable. | |
164 | */ | |
165 | if (error == -ENOMEM) | |
166 | error = -EAGAIN; | |
167 | goto out; | |
168 | } | |
1da177e4 LT |
169 | } |
170 | ||
836d5ffd | 171 | success: |
1da177e4 LT |
172 | /* |
173 | * vm_flags is protected by the mmap_sem held in write mode. | |
174 | */ | |
e798c6e8 | 175 | vma->vm_flags = new_flags; |
1da177e4 | 176 | out: |
1da177e4 LT |
177 | return error; |
178 | } | |
179 | ||
1998cc04 SL |
180 | #ifdef CONFIG_SWAP |
181 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
182 | unsigned long end, struct mm_walk *walk) | |
183 | { | |
184 | pte_t *orig_pte; | |
185 | struct vm_area_struct *vma = walk->private; | |
186 | unsigned long index; | |
187 | ||
188 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
189 | return 0; | |
190 | ||
191 | for (index = start; index != end; index += PAGE_SIZE) { | |
192 | pte_t pte; | |
193 | swp_entry_t entry; | |
194 | struct page *page; | |
195 | spinlock_t *ptl; | |
196 | ||
197 | orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); | |
198 | pte = *(orig_pte + ((index - start) / PAGE_SIZE)); | |
199 | pte_unmap_unlock(orig_pte, ptl); | |
200 | ||
0661a336 | 201 | if (pte_present(pte) || pte_none(pte)) |
1998cc04 SL |
202 | continue; |
203 | entry = pte_to_swp_entry(pte); | |
204 | if (unlikely(non_swap_entry(entry))) | |
205 | continue; | |
206 | ||
207 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, | |
23955622 | 208 | vma, index, false); |
1998cc04 | 209 | if (page) |
09cbfeaf | 210 | put_page(page); |
1998cc04 SL |
211 | } |
212 | ||
213 | return 0; | |
214 | } | |
215 | ||
216 | static void force_swapin_readahead(struct vm_area_struct *vma, | |
217 | unsigned long start, unsigned long end) | |
218 | { | |
219 | struct mm_walk walk = { | |
220 | .mm = vma->vm_mm, | |
221 | .pmd_entry = swapin_walk_pmd_entry, | |
222 | .private = vma, | |
223 | }; | |
224 | ||
225 | walk_page_range(start, end, &walk); | |
226 | ||
227 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
228 | } | |
229 | ||
230 | static void force_shm_swapin_readahead(struct vm_area_struct *vma, | |
231 | unsigned long start, unsigned long end, | |
232 | struct address_space *mapping) | |
233 | { | |
234 | pgoff_t index; | |
235 | struct page *page; | |
236 | swp_entry_t swap; | |
237 | ||
238 | for (; start < end; start += PAGE_SIZE) { | |
239 | index = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
240 | ||
55231e5c | 241 | page = find_get_entry(mapping, index); |
1998cc04 SL |
242 | if (!radix_tree_exceptional_entry(page)) { |
243 | if (page) | |
09cbfeaf | 244 | put_page(page); |
1998cc04 SL |
245 | continue; |
246 | } | |
247 | swap = radix_to_swp_entry(page); | |
248 | page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE, | |
23955622 | 249 | NULL, 0, false); |
1998cc04 | 250 | if (page) |
09cbfeaf | 251 | put_page(page); |
1998cc04 SL |
252 | } |
253 | ||
254 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
255 | } | |
256 | #endif /* CONFIG_SWAP */ | |
257 | ||
1da177e4 LT |
258 | /* |
259 | * Schedule all required I/O operations. Do not wait for completion. | |
260 | */ | |
ec9bed9d VC |
261 | static long madvise_willneed(struct vm_area_struct *vma, |
262 | struct vm_area_struct **prev, | |
1da177e4 LT |
263 | unsigned long start, unsigned long end) |
264 | { | |
265 | struct file *file = vma->vm_file; | |
266 | ||
dff4aaf7 | 267 | *prev = vma; |
1998cc04 | 268 | #ifdef CONFIG_SWAP |
97b713ba | 269 | if (!file) { |
97b713ba | 270 | force_swapin_readahead(vma, start, end); |
1998cc04 SL |
271 | return 0; |
272 | } | |
1998cc04 | 273 | |
97b713ba | 274 | if (shmem_mapping(file->f_mapping)) { |
97b713ba CH |
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 | ||
1da177e4 LT |
289 | start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
290 | if (end > vma->vm_end) | |
291 | end = vma->vm_end; | |
292 | end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
293 | ||
f7e839dd | 294 | force_page_cache_readahead(file->f_mapping, file, start, end - start); |
1da177e4 LT |
295 | return 0; |
296 | } | |
297 | ||
854e9ed0 MK |
298 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
299 | unsigned long end, struct mm_walk *walk) | |
300 | ||
301 | { | |
302 | struct mmu_gather *tlb = walk->private; | |
303 | struct mm_struct *mm = tlb->mm; | |
304 | struct vm_area_struct *vma = walk->vma; | |
305 | spinlock_t *ptl; | |
306 | pte_t *orig_pte, *pte, ptent; | |
307 | struct page *page; | |
64b42bc1 | 308 | int nr_swap = 0; |
b8d3c4c3 MK |
309 | unsigned long next; |
310 | ||
311 | next = pmd_addr_end(addr, end); | |
312 | if (pmd_trans_huge(*pmd)) | |
313 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
314 | goto next; | |
854e9ed0 | 315 | |
854e9ed0 MK |
316 | if (pmd_trans_unstable(pmd)) |
317 | return 0; | |
318 | ||
07e32661 | 319 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
854e9ed0 | 320 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
3ea27719 | 321 | flush_tlb_batched_pending(mm); |
854e9ed0 MK |
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 | } | |
854e9ed0 | 369 | unlock_page(page); |
263630e8 | 370 | put_page(page); |
854e9ed0 MK |
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); | |
413 | tlb_remove_tlb_entry(tlb, pte, addr); | |
414 | } | |
802a3a92 | 415 | mark_page_lazyfree(page); |
854e9ed0 MK |
416 | } |
417 | out: | |
64b42bc1 MK |
418 | if (nr_swap) { |
419 | if (current->mm == mm) | |
420 | sync_mm_rss(mm); | |
421 | ||
422 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); | |
423 | } | |
854e9ed0 MK |
424 | arch_leave_lazy_mmu_mode(); |
425 | pte_unmap_unlock(orig_pte, ptl); | |
426 | cond_resched(); | |
b8d3c4c3 | 427 | next: |
854e9ed0 MK |
428 | return 0; |
429 | } | |
430 | ||
431 | static void madvise_free_page_range(struct mmu_gather *tlb, | |
432 | struct vm_area_struct *vma, | |
433 | unsigned long addr, unsigned long end) | |
434 | { | |
435 | struct mm_walk free_walk = { | |
436 | .pmd_entry = madvise_free_pte_range, | |
437 | .mm = vma->vm_mm, | |
438 | .private = tlb, | |
439 | }; | |
440 | ||
441 | tlb_start_vma(tlb, vma); | |
442 | walk_page_range(addr, end, &free_walk); | |
443 | tlb_end_vma(tlb, vma); | |
444 | } | |
445 | ||
446 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
447 | unsigned long start_addr, unsigned long end_addr) | |
448 | { | |
449 | unsigned long start, end; | |
450 | struct mm_struct *mm = vma->vm_mm; | |
451 | struct mmu_gather tlb; | |
452 | ||
854e9ed0 MK |
453 | /* MADV_FREE works for only anon vma at the moment */ |
454 | if (!vma_is_anonymous(vma)) | |
455 | return -EINVAL; | |
456 | ||
457 | start = max(vma->vm_start, start_addr); | |
458 | if (start >= vma->vm_end) | |
459 | return -EINVAL; | |
460 | end = min(vma->vm_end, end_addr); | |
461 | if (end <= vma->vm_start) | |
462 | return -EINVAL; | |
463 | ||
464 | lru_add_drain(); | |
465 | tlb_gather_mmu(&tlb, mm, start, end); | |
466 | update_hiwater_rss(mm); | |
467 | ||
468 | mmu_notifier_invalidate_range_start(mm, start, end); | |
469 | madvise_free_page_range(&tlb, vma, start, end); | |
470 | mmu_notifier_invalidate_range_end(mm, start, end); | |
471 | tlb_finish_mmu(&tlb, start, end); | |
472 | ||
473 | return 0; | |
474 | } | |
475 | ||
1da177e4 LT |
476 | /* |
477 | * Application no longer needs these pages. If the pages are dirty, | |
478 | * it's OK to just throw them away. The app will be more careful about | |
479 | * data it wants to keep. Be sure to free swap resources too. The | |
7e6cbea3 | 480 | * zap_page_range call sets things up for shrink_active_list to actually free |
1da177e4 LT |
481 | * these pages later if no one else has touched them in the meantime, |
482 | * although we could add these pages to a global reuse list for | |
7e6cbea3 | 483 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
484 | * |
485 | * NB: This interface discards data rather than pushes it out to swap, | |
486 | * as some implementations do. This has performance implications for | |
487 | * applications like large transactional databases which want to discard | |
488 | * pages in anonymous maps after committing to backing store the data | |
489 | * that was kept in them. There is no reason to write this data out to | |
490 | * the swap area if the application is discarding it. | |
491 | * | |
492 | * An interface that causes the system to free clean pages and flush | |
493 | * dirty pages is already available as msync(MS_INVALIDATE). | |
494 | */ | |
230ca982 MR |
495 | static long madvise_dontneed_single_vma(struct vm_area_struct *vma, |
496 | unsigned long start, unsigned long end) | |
497 | { | |
498 | zap_page_range(vma, start, end - start); | |
499 | return 0; | |
500 | } | |
501 | ||
502 | static long madvise_dontneed_free(struct vm_area_struct *vma, | |
503 | struct vm_area_struct **prev, | |
504 | unsigned long start, unsigned long end, | |
505 | int behavior) | |
1da177e4 | 506 | { |
05b74384 | 507 | *prev = vma; |
23519073 | 508 | if (!can_madv_dontneed_vma(vma)) |
1da177e4 LT |
509 | return -EINVAL; |
510 | ||
70ccb92f AA |
511 | if (!userfaultfd_remove(vma, start, end)) { |
512 | *prev = NULL; /* mmap_sem has been dropped, prev is stale */ | |
513 | ||
514 | down_read(¤t->mm->mmap_sem); | |
515 | vma = find_vma(current->mm, start); | |
516 | if (!vma) | |
517 | return -ENOMEM; | |
518 | if (start < vma->vm_start) { | |
519 | /* | |
520 | * This "vma" under revalidation is the one | |
521 | * with the lowest vma->vm_start where start | |
522 | * is also < vma->vm_end. If start < | |
523 | * vma->vm_start it means an hole materialized | |
524 | * in the user address space within the | |
230ca982 MR |
525 | * virtual range passed to MADV_DONTNEED |
526 | * or MADV_FREE. | |
70ccb92f AA |
527 | */ |
528 | return -ENOMEM; | |
529 | } | |
530 | if (!can_madv_dontneed_vma(vma)) | |
531 | return -EINVAL; | |
532 | if (end > vma->vm_end) { | |
533 | /* | |
534 | * Don't fail if end > vma->vm_end. If the old | |
535 | * vma was splitted while the mmap_sem was | |
536 | * released the effect of the concurrent | |
230ca982 | 537 | * operation may not cause madvise() to |
70ccb92f AA |
538 | * have an undefined result. There may be an |
539 | * adjacent next vma that we'll walk | |
540 | * next. userfaultfd_remove() will generate an | |
541 | * UFFD_EVENT_REMOVE repetition on the | |
542 | * end-vma->vm_end range, but the manager can | |
543 | * handle a repetition fine. | |
544 | */ | |
545 | end = vma->vm_end; | |
546 | } | |
547 | VM_WARN_ON(start >= end); | |
548 | } | |
230ca982 MR |
549 | |
550 | if (behavior == MADV_DONTNEED) | |
551 | return madvise_dontneed_single_vma(vma, start, end); | |
552 | else if (behavior == MADV_FREE) | |
553 | return madvise_free_single_vma(vma, start, end); | |
554 | else | |
555 | return -EINVAL; | |
1da177e4 LT |
556 | } |
557 | ||
f6b3ec23 BP |
558 | /* |
559 | * Application wants to free up the pages and associated backing store. | |
560 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
561 | */ |
562 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 563 | struct vm_area_struct **prev, |
f6b3ec23 BP |
564 | unsigned long start, unsigned long end) |
565 | { | |
3f31d075 | 566 | loff_t offset; |
90ed52eb | 567 | int error; |
9ab4233d | 568 | struct file *f; |
f6b3ec23 | 569 | |
90ed52eb | 570 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ |
00e9fa2d | 571 | |
72079ba0 | 572 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
573 | return -EINVAL; |
574 | ||
9ab4233d AL |
575 | f = vma->vm_file; |
576 | ||
577 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
578 | return -EINVAL; |
579 | } | |
580 | ||
69cf0fac HD |
581 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
582 | return -EACCES; | |
583 | ||
f6b3ec23 BP |
584 | offset = (loff_t)(start - vma->vm_start) |
585 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 586 | |
9ab4233d AL |
587 | /* |
588 | * Filesystem's fallocate may need to take i_mutex. We need to | |
589 | * explicitly grab a reference because the vma (and hence the | |
590 | * vma's reference to the file) can go away as soon as we drop | |
591 | * mmap_sem. | |
592 | */ | |
593 | get_file(f); | |
70ccb92f AA |
594 | if (userfaultfd_remove(vma, start, end)) { |
595 | /* mmap_sem was not released by userfaultfd_remove() */ | |
596 | up_read(¤t->mm->mmap_sem); | |
597 | } | |
72c72bdf | 598 | error = vfs_fallocate(f, |
3f31d075 HD |
599 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
600 | offset, end - start); | |
9ab4233d | 601 | fput(f); |
0a27a14a | 602 | down_read(¤t->mm->mmap_sem); |
90ed52eb | 603 | return error; |
f6b3ec23 BP |
604 | } |
605 | ||
9893e49d AK |
606 | #ifdef CONFIG_MEMORY_FAILURE |
607 | /* | |
608 | * Error injection support for memory error handling. | |
609 | */ | |
97167a76 AK |
610 | static int madvise_inject_error(int behavior, |
611 | unsigned long start, unsigned long end) | |
9893e49d | 612 | { |
97167a76 | 613 | struct page *page; |
c461ad6a | 614 | struct zone *zone; |
922b7863 | 615 | unsigned int order; |
97167a76 | 616 | |
9893e49d AK |
617 | if (!capable(CAP_SYS_ADMIN)) |
618 | return -EPERM; | |
97167a76 | 619 | |
922b7863 AM |
620 | |
621 | for (; start < end; start += PAGE_SIZE << order) { | |
325c4ef5 AM |
622 | int ret; |
623 | ||
97167a76 | 624 | ret = get_user_pages_fast(start, 1, 0, &page); |
9893e49d AK |
625 | if (ret != 1) |
626 | return ret; | |
325c4ef5 | 627 | |
922b7863 AM |
628 | /* |
629 | * When soft offlining hugepages, after migrating the page | |
630 | * we dissolve it, therefore in the second loop "page" will | |
631 | * no longer be a compound page, and order will be 0. | |
632 | */ | |
633 | order = compound_order(compound_head(page)); | |
634 | ||
97167a76 AK |
635 | if (PageHWPoison(page)) { |
636 | put_page(page); | |
29b4eede WL |
637 | continue; |
638 | } | |
97167a76 AK |
639 | |
640 | if (behavior == MADV_SOFT_OFFLINE) { | |
641 | pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n", | |
642 | page_to_pfn(page), start); | |
643 | ||
644 | ret = soft_offline_page(page, MF_COUNT_INCREASED); | |
afcf938e | 645 | if (ret) |
8302423b | 646 | return ret; |
afcf938e AK |
647 | continue; |
648 | } | |
97167a76 AK |
649 | pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n", |
650 | page_to_pfn(page), start); | |
651 | ||
652 | ret = memory_failure(page_to_pfn(page), 0, MF_COUNT_INCREASED); | |
23a003bf NH |
653 | if (ret) |
654 | return ret; | |
9893e49d | 655 | } |
c461ad6a MG |
656 | |
657 | /* Ensure that all poisoned pages are removed from per-cpu lists */ | |
658 | for_each_populated_zone(zone) | |
659 | drain_all_pages(zone); | |
660 | ||
325c4ef5 | 661 | return 0; |
9893e49d AK |
662 | } |
663 | #endif | |
664 | ||
165cd402 | 665 | static long |
666 | madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, | |
667 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 668 | { |
1da177e4 | 669 | switch (behavior) { |
f6b3ec23 | 670 | case MADV_REMOVE: |
3866ea90 | 671 | return madvise_remove(vma, prev, start, end); |
1da177e4 | 672 | case MADV_WILLNEED: |
3866ea90 | 673 | return madvise_willneed(vma, prev, start, end); |
854e9ed0 | 674 | case MADV_FREE: |
1da177e4 | 675 | case MADV_DONTNEED: |
230ca982 | 676 | return madvise_dontneed_free(vma, prev, start, end, behavior); |
1da177e4 | 677 | default: |
3866ea90 | 678 | return madvise_behavior(vma, prev, start, end, behavior); |
1da177e4 | 679 | } |
1da177e4 LT |
680 | } |
681 | ||
1ecef9ed | 682 | static bool |
75927af8 NP |
683 | madvise_behavior_valid(int behavior) |
684 | { | |
685 | switch (behavior) { | |
686 | case MADV_DOFORK: | |
687 | case MADV_DONTFORK: | |
688 | case MADV_NORMAL: | |
689 | case MADV_SEQUENTIAL: | |
690 | case MADV_RANDOM: | |
691 | case MADV_REMOVE: | |
692 | case MADV_WILLNEED: | |
693 | case MADV_DONTNEED: | |
854e9ed0 | 694 | case MADV_FREE: |
f8af4da3 HD |
695 | #ifdef CONFIG_KSM |
696 | case MADV_MERGEABLE: | |
697 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
698 | #endif |
699 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
700 | case MADV_HUGEPAGE: | |
a664b2d8 | 701 | case MADV_NOHUGEPAGE: |
f8af4da3 | 702 | #endif |
accb61fe JB |
703 | case MADV_DONTDUMP: |
704 | case MADV_DODUMP: | |
5e451be7 AK |
705 | #ifdef CONFIG_MEMORY_FAILURE |
706 | case MADV_SOFT_OFFLINE: | |
707 | case MADV_HWPOISON: | |
708 | #endif | |
1ecef9ed | 709 | return true; |
75927af8 NP |
710 | |
711 | default: | |
1ecef9ed | 712 | return false; |
75927af8 NP |
713 | } |
714 | } | |
3866ea90 | 715 | |
1da177e4 LT |
716 | /* |
717 | * The madvise(2) system call. | |
718 | * | |
719 | * Applications can use madvise() to advise the kernel how it should | |
720 | * handle paging I/O in this VM area. The idea is to help the kernel | |
721 | * use appropriate read-ahead and caching techniques. The information | |
722 | * provided is advisory only, and can be safely disregarded by the | |
723 | * kernel without affecting the correct operation of the application. | |
724 | * | |
725 | * behavior values: | |
726 | * MADV_NORMAL - the default behavior is to read clusters. This | |
727 | * results in some read-ahead and read-behind. | |
728 | * MADV_RANDOM - the system should read the minimum amount of data | |
729 | * on any access, since it is unlikely that the appli- | |
730 | * cation will need more than what it asks for. | |
731 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
732 | * once, so they can be aggressively read ahead, and | |
733 | * can be freed soon after they are accessed. | |
734 | * MADV_WILLNEED - the application is notifying the system to read | |
735 | * some pages ahead. | |
736 | * MADV_DONTNEED - the application is finished with the given range, | |
737 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
738 | * MADV_FREE - the application marks pages in the given range as lazy free, |
739 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
740 | * MADV_REMOVE - the application wants to free up the given range of |
741 | * pages and associated backing store. | |
3866ea90 HD |
742 | * MADV_DONTFORK - omit this area from child's address space when forking: |
743 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
744 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
d7206a70 NH |
745 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
746 | * were corrupted by unrecoverable hardware memory failure. | |
747 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
748 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
749 | * this area with pages of identical content from other such areas. | |
750 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
751 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
752 | * huge pages in the future. Existing pages might be coalesced and | |
753 | * new pages might be allocated as THP. | |
754 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
755 | * transparent huge pages so the existing pages will not be | |
756 | * coalesced into THP and new pages will not be allocated as THP. | |
757 | * MADV_DONTDUMP - the application wants to prevent pages in the given range | |
758 | * from being included in its core dump. | |
759 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
1da177e4 LT |
760 | * |
761 | * return values: | |
762 | * zero - success | |
763 | * -EINVAL - start + len < 0, start is not page-aligned, | |
764 | * "behavior" is not a valid value, or application | |
765 | * is attempting to release locked or shared pages. | |
766 | * -ENOMEM - addresses in the specified range are not currently | |
767 | * mapped, or are outside the AS of the process. | |
768 | * -EIO - an I/O error occurred while paging in data. | |
769 | * -EBADF - map exists, but area maps something that isn't a file. | |
770 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
771 | */ | |
3480b257 | 772 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) |
1da177e4 | 773 | { |
05b74384 | 774 | unsigned long end, tmp; |
ec9bed9d | 775 | struct vm_area_struct *vma, *prev; |
1da177e4 LT |
776 | int unmapped_error = 0; |
777 | int error = -EINVAL; | |
f7977793 | 778 | int write; |
1da177e4 | 779 | size_t len; |
1998cc04 | 780 | struct blk_plug plug; |
1da177e4 | 781 | |
75927af8 NP |
782 | if (!madvise_behavior_valid(behavior)) |
783 | return error; | |
784 | ||
1da177e4 | 785 | if (start & ~PAGE_MASK) |
84d96d89 | 786 | return error; |
1da177e4 LT |
787 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; |
788 | ||
789 | /* Check to see whether len was rounded up from small -ve to zero */ | |
790 | if (len_in && !len) | |
84d96d89 | 791 | return error; |
1da177e4 LT |
792 | |
793 | end = start + len; | |
794 | if (end < start) | |
84d96d89 | 795 | return error; |
1da177e4 LT |
796 | |
797 | error = 0; | |
798 | if (end == start) | |
84d96d89 RV |
799 | return error; |
800 | ||
5e451be7 AK |
801 | #ifdef CONFIG_MEMORY_FAILURE |
802 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) | |
803 | return madvise_inject_error(behavior, start, start + len_in); | |
804 | #endif | |
805 | ||
84d96d89 | 806 | write = madvise_need_mmap_write(behavior); |
dc0ef0df MH |
807 | if (write) { |
808 | if (down_write_killable(¤t->mm->mmap_sem)) | |
809 | return -EINTR; | |
810 | } else { | |
84d96d89 | 811 | down_read(¤t->mm->mmap_sem); |
dc0ef0df | 812 | } |
1da177e4 LT |
813 | |
814 | /* | |
815 | * If the interval [start,end) covers some unmapped address | |
816 | * ranges, just ignore them, but return -ENOMEM at the end. | |
05b74384 | 817 | * - different from the way of handling in mlock etc. |
1da177e4 | 818 | */ |
05b74384 | 819 | vma = find_vma_prev(current->mm, start, &prev); |
836d5ffd HD |
820 | if (vma && start > vma->vm_start) |
821 | prev = vma; | |
822 | ||
1998cc04 | 823 | blk_start_plug(&plug); |
1da177e4 LT |
824 | for (;;) { |
825 | /* Still start < end. */ | |
826 | error = -ENOMEM; | |
827 | if (!vma) | |
84d96d89 | 828 | goto out; |
1da177e4 | 829 | |
05b74384 | 830 | /* Here start < (end|vma->vm_end). */ |
1da177e4 LT |
831 | if (start < vma->vm_start) { |
832 | unmapped_error = -ENOMEM; | |
833 | start = vma->vm_start; | |
05b74384 | 834 | if (start >= end) |
84d96d89 | 835 | goto out; |
1da177e4 LT |
836 | } |
837 | ||
05b74384 PM |
838 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ |
839 | tmp = vma->vm_end; | |
840 | if (end < tmp) | |
841 | tmp = end; | |
1da177e4 | 842 | |
05b74384 PM |
843 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ |
844 | error = madvise_vma(vma, &prev, start, tmp, behavior); | |
1da177e4 | 845 | if (error) |
84d96d89 | 846 | goto out; |
05b74384 | 847 | start = tmp; |
90ed52eb | 848 | if (prev && start < prev->vm_end) |
05b74384 PM |
849 | start = prev->vm_end; |
850 | error = unmapped_error; | |
851 | if (start >= end) | |
84d96d89 | 852 | goto out; |
90ed52eb HD |
853 | if (prev) |
854 | vma = prev->vm_next; | |
855 | else /* madvise_remove dropped mmap_sem */ | |
856 | vma = find_vma(current->mm, start); | |
1da177e4 | 857 | } |
1da177e4 | 858 | out: |
84d96d89 | 859 | blk_finish_plug(&plug); |
f7977793 | 860 | if (write) |
0a27a14a NP |
861 | up_write(¤t->mm->mmap_sem); |
862 | else | |
863 | up_read(¤t->mm->mmap_sem); | |
864 | ||
1da177e4 LT |
865 | return error; |
866 | } |