]>
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; |
d2cd9ede RR |
83 | case MADV_WIPEONFORK: |
84 | /* MADV_WIPEONFORK is only supported on anonymous memory. */ | |
85 | if (vma->vm_file || vma->vm_flags & VM_SHARED) { | |
86 | error = -EINVAL; | |
87 | goto out; | |
88 | } | |
89 | new_flags |= VM_WIPEONFORK; | |
90 | break; | |
91 | case MADV_KEEPONFORK: | |
92 | new_flags &= ~VM_WIPEONFORK; | |
93 | break; | |
accb61fe | 94 | case MADV_DONTDUMP: |
0103bd16 | 95 | new_flags |= VM_DONTDUMP; |
accb61fe JB |
96 | break; |
97 | case MADV_DODUMP: | |
0103bd16 KK |
98 | if (new_flags & VM_SPECIAL) { |
99 | error = -EINVAL; | |
100 | goto out; | |
101 | } | |
102 | new_flags &= ~VM_DONTDUMP; | |
accb61fe | 103 | break; |
f8af4da3 HD |
104 | case MADV_MERGEABLE: |
105 | case MADV_UNMERGEABLE: | |
106 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
def5efe0 DR |
107 | if (error) { |
108 | /* | |
109 | * madvise() returns EAGAIN if kernel resources, such as | |
110 | * slab, are temporarily unavailable. | |
111 | */ | |
112 | if (error == -ENOMEM) | |
113 | error = -EAGAIN; | |
f8af4da3 | 114 | goto out; |
def5efe0 | 115 | } |
f8af4da3 | 116 | break; |
0af4e98b | 117 | case MADV_HUGEPAGE: |
a664b2d8 | 118 | case MADV_NOHUGEPAGE: |
60ab3244 | 119 | error = hugepage_madvise(vma, &new_flags, behavior); |
def5efe0 DR |
120 | if (error) { |
121 | /* | |
122 | * madvise() returns EAGAIN if kernel resources, such as | |
123 | * slab, are temporarily unavailable. | |
124 | */ | |
125 | if (error == -ENOMEM) | |
126 | error = -EAGAIN; | |
0af4e98b | 127 | goto out; |
def5efe0 | 128 | } |
0af4e98b | 129 | break; |
e798c6e8 PM |
130 | } |
131 | ||
05b74384 PM |
132 | if (new_flags == vma->vm_flags) { |
133 | *prev = vma; | |
836d5ffd | 134 | goto out; |
05b74384 PM |
135 | } |
136 | ||
137 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
138 | *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma, | |
19a809af AA |
139 | vma->vm_file, pgoff, vma_policy(vma), |
140 | vma->vm_userfaultfd_ctx); | |
05b74384 PM |
141 | if (*prev) { |
142 | vma = *prev; | |
143 | goto success; | |
144 | } | |
145 | ||
146 | *prev = vma; | |
1da177e4 LT |
147 | |
148 | if (start != vma->vm_start) { | |
def5efe0 DR |
149 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
150 | error = -ENOMEM; | |
1da177e4 | 151 | goto out; |
def5efe0 DR |
152 | } |
153 | error = __split_vma(mm, vma, start, 1); | |
154 | if (error) { | |
155 | /* | |
156 | * madvise() returns EAGAIN if kernel resources, such as | |
157 | * slab, are temporarily unavailable. | |
158 | */ | |
159 | if (error == -ENOMEM) | |
160 | error = -EAGAIN; | |
161 | goto out; | |
162 | } | |
1da177e4 LT |
163 | } |
164 | ||
165 | if (end != vma->vm_end) { | |
def5efe0 DR |
166 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
167 | error = -ENOMEM; | |
1da177e4 | 168 | goto out; |
def5efe0 DR |
169 | } |
170 | error = __split_vma(mm, vma, end, 0); | |
171 | if (error) { | |
172 | /* | |
173 | * madvise() returns EAGAIN if kernel resources, such as | |
174 | * slab, are temporarily unavailable. | |
175 | */ | |
176 | if (error == -ENOMEM) | |
177 | error = -EAGAIN; | |
178 | goto out; | |
179 | } | |
1da177e4 LT |
180 | } |
181 | ||
836d5ffd | 182 | success: |
1da177e4 LT |
183 | /* |
184 | * vm_flags is protected by the mmap_sem held in write mode. | |
185 | */ | |
e798c6e8 | 186 | vma->vm_flags = new_flags; |
1da177e4 | 187 | out: |
1da177e4 LT |
188 | return error; |
189 | } | |
190 | ||
1998cc04 SL |
191 | #ifdef CONFIG_SWAP |
192 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
193 | unsigned long end, struct mm_walk *walk) | |
194 | { | |
195 | pte_t *orig_pte; | |
196 | struct vm_area_struct *vma = walk->private; | |
197 | unsigned long index; | |
198 | ||
199 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
200 | return 0; | |
201 | ||
202 | for (index = start; index != end; index += PAGE_SIZE) { | |
203 | pte_t pte; | |
204 | swp_entry_t entry; | |
205 | struct page *page; | |
206 | spinlock_t *ptl; | |
207 | ||
208 | orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); | |
209 | pte = *(orig_pte + ((index - start) / PAGE_SIZE)); | |
210 | pte_unmap_unlock(orig_pte, ptl); | |
211 | ||
0661a336 | 212 | if (pte_present(pte) || pte_none(pte)) |
1998cc04 SL |
213 | continue; |
214 | entry = pte_to_swp_entry(pte); | |
215 | if (unlikely(non_swap_entry(entry))) | |
216 | continue; | |
217 | ||
218 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, | |
23955622 | 219 | vma, index, false); |
1998cc04 | 220 | if (page) |
09cbfeaf | 221 | put_page(page); |
1998cc04 SL |
222 | } |
223 | ||
224 | return 0; | |
225 | } | |
226 | ||
227 | static void force_swapin_readahead(struct vm_area_struct *vma, | |
228 | unsigned long start, unsigned long end) | |
229 | { | |
230 | struct mm_walk walk = { | |
231 | .mm = vma->vm_mm, | |
232 | .pmd_entry = swapin_walk_pmd_entry, | |
233 | .private = vma, | |
234 | }; | |
235 | ||
236 | walk_page_range(start, end, &walk); | |
237 | ||
238 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
239 | } | |
240 | ||
241 | static void force_shm_swapin_readahead(struct vm_area_struct *vma, | |
242 | unsigned long start, unsigned long end, | |
243 | struct address_space *mapping) | |
244 | { | |
245 | pgoff_t index; | |
246 | struct page *page; | |
247 | swp_entry_t swap; | |
248 | ||
249 | for (; start < end; start += PAGE_SIZE) { | |
250 | index = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
251 | ||
55231e5c | 252 | page = find_get_entry(mapping, index); |
1998cc04 SL |
253 | if (!radix_tree_exceptional_entry(page)) { |
254 | if (page) | |
09cbfeaf | 255 | put_page(page); |
1998cc04 SL |
256 | continue; |
257 | } | |
258 | swap = radix_to_swp_entry(page); | |
259 | page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE, | |
23955622 | 260 | NULL, 0, false); |
1998cc04 | 261 | if (page) |
09cbfeaf | 262 | put_page(page); |
1998cc04 SL |
263 | } |
264 | ||
265 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
266 | } | |
267 | #endif /* CONFIG_SWAP */ | |
268 | ||
1da177e4 LT |
269 | /* |
270 | * Schedule all required I/O operations. Do not wait for completion. | |
271 | */ | |
ec9bed9d VC |
272 | static long madvise_willneed(struct vm_area_struct *vma, |
273 | struct vm_area_struct **prev, | |
1da177e4 LT |
274 | unsigned long start, unsigned long end) |
275 | { | |
276 | struct file *file = vma->vm_file; | |
277 | ||
1998cc04 | 278 | #ifdef CONFIG_SWAP |
97b713ba | 279 | if (!file) { |
1998cc04 | 280 | *prev = vma; |
97b713ba | 281 | force_swapin_readahead(vma, start, end); |
1998cc04 SL |
282 | return 0; |
283 | } | |
1998cc04 | 284 | |
97b713ba CH |
285 | if (shmem_mapping(file->f_mapping)) { |
286 | *prev = vma; | |
287 | force_shm_swapin_readahead(vma, start, end, | |
288 | file->f_mapping); | |
289 | return 0; | |
290 | } | |
291 | #else | |
1bef4003 S |
292 | if (!file) |
293 | return -EBADF; | |
97b713ba | 294 | #endif |
1bef4003 | 295 | |
e748dcd0 | 296 | if (IS_DAX(file_inode(file))) { |
fe77ba6f CO |
297 | /* no bad return value, but ignore advice */ |
298 | return 0; | |
299 | } | |
300 | ||
05b74384 | 301 | *prev = vma; |
1da177e4 LT |
302 | start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
303 | if (end > vma->vm_end) | |
304 | end = vma->vm_end; | |
305 | end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
306 | ||
f7e839dd | 307 | force_page_cache_readahead(file->f_mapping, file, start, end - start); |
1da177e4 LT |
308 | return 0; |
309 | } | |
310 | ||
854e9ed0 MK |
311 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
312 | unsigned long end, struct mm_walk *walk) | |
313 | ||
314 | { | |
315 | struct mmu_gather *tlb = walk->private; | |
316 | struct mm_struct *mm = tlb->mm; | |
317 | struct vm_area_struct *vma = walk->vma; | |
318 | spinlock_t *ptl; | |
319 | pte_t *orig_pte, *pte, ptent; | |
320 | struct page *page; | |
64b42bc1 | 321 | int nr_swap = 0; |
b8d3c4c3 MK |
322 | unsigned long next; |
323 | ||
324 | next = pmd_addr_end(addr, end); | |
325 | if (pmd_trans_huge(*pmd)) | |
326 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
327 | goto next; | |
854e9ed0 | 328 | |
854e9ed0 MK |
329 | if (pmd_trans_unstable(pmd)) |
330 | return 0; | |
331 | ||
07e32661 | 332 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
854e9ed0 | 333 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
3ea27719 | 334 | flush_tlb_batched_pending(mm); |
854e9ed0 MK |
335 | arch_enter_lazy_mmu_mode(); |
336 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
337 | ptent = *pte; | |
338 | ||
64b42bc1 | 339 | if (pte_none(ptent)) |
854e9ed0 | 340 | continue; |
64b42bc1 MK |
341 | /* |
342 | * If the pte has swp_entry, just clear page table to | |
343 | * prevent swap-in which is more expensive rather than | |
344 | * (page allocation + zeroing). | |
345 | */ | |
346 | if (!pte_present(ptent)) { | |
347 | swp_entry_t entry; | |
348 | ||
349 | entry = pte_to_swp_entry(ptent); | |
350 | if (non_swap_entry(entry)) | |
351 | continue; | |
352 | nr_swap--; | |
353 | free_swap_and_cache(entry); | |
354 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
355 | continue; | |
356 | } | |
854e9ed0 | 357 | |
df6ad698 | 358 | page = _vm_normal_page(vma, addr, ptent, true); |
854e9ed0 MK |
359 | if (!page) |
360 | continue; | |
361 | ||
362 | /* | |
363 | * If pmd isn't transhuge but the page is THP and | |
364 | * is owned by only this process, split it and | |
365 | * deactivate all pages. | |
366 | */ | |
367 | if (PageTransCompound(page)) { | |
368 | if (page_mapcount(page) != 1) | |
369 | goto out; | |
370 | get_page(page); | |
371 | if (!trylock_page(page)) { | |
372 | put_page(page); | |
373 | goto out; | |
374 | } | |
375 | pte_unmap_unlock(orig_pte, ptl); | |
376 | if (split_huge_page(page)) { | |
377 | unlock_page(page); | |
378 | put_page(page); | |
379 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
380 | goto out; | |
381 | } | |
854e9ed0 | 382 | unlock_page(page); |
263630e8 | 383 | put_page(page); |
854e9ed0 MK |
384 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
385 | pte--; | |
386 | addr -= PAGE_SIZE; | |
387 | continue; | |
388 | } | |
389 | ||
390 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
391 | ||
392 | if (PageSwapCache(page) || PageDirty(page)) { | |
393 | if (!trylock_page(page)) | |
394 | continue; | |
395 | /* | |
396 | * If page is shared with others, we couldn't clear | |
397 | * PG_dirty of the page. | |
398 | */ | |
399 | if (page_mapcount(page) != 1) { | |
400 | unlock_page(page); | |
401 | continue; | |
402 | } | |
403 | ||
404 | if (PageSwapCache(page) && !try_to_free_swap(page)) { | |
405 | unlock_page(page); | |
406 | continue; | |
407 | } | |
408 | ||
409 | ClearPageDirty(page); | |
410 | unlock_page(page); | |
411 | } | |
412 | ||
413 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
414 | /* | |
415 | * Some of architecture(ex, PPC) don't update TLB | |
416 | * with set_pte_at and tlb_remove_tlb_entry so for | |
417 | * the portability, remap the pte with old|clean | |
418 | * after pte clearing. | |
419 | */ | |
420 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
421 | tlb->fullmm); | |
422 | ||
423 | ptent = pte_mkold(ptent); | |
424 | ptent = pte_mkclean(ptent); | |
425 | set_pte_at(mm, addr, pte, ptent); | |
426 | tlb_remove_tlb_entry(tlb, pte, addr); | |
427 | } | |
802a3a92 | 428 | mark_page_lazyfree(page); |
854e9ed0 MK |
429 | } |
430 | out: | |
64b42bc1 MK |
431 | if (nr_swap) { |
432 | if (current->mm == mm) | |
433 | sync_mm_rss(mm); | |
434 | ||
435 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); | |
436 | } | |
854e9ed0 MK |
437 | arch_leave_lazy_mmu_mode(); |
438 | pte_unmap_unlock(orig_pte, ptl); | |
439 | cond_resched(); | |
b8d3c4c3 | 440 | next: |
854e9ed0 MK |
441 | return 0; |
442 | } | |
443 | ||
444 | static void madvise_free_page_range(struct mmu_gather *tlb, | |
445 | struct vm_area_struct *vma, | |
446 | unsigned long addr, unsigned long end) | |
447 | { | |
448 | struct mm_walk free_walk = { | |
449 | .pmd_entry = madvise_free_pte_range, | |
450 | .mm = vma->vm_mm, | |
451 | .private = tlb, | |
452 | }; | |
453 | ||
454 | tlb_start_vma(tlb, vma); | |
455 | walk_page_range(addr, end, &free_walk); | |
456 | tlb_end_vma(tlb, vma); | |
457 | } | |
458 | ||
459 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
460 | unsigned long start_addr, unsigned long end_addr) | |
461 | { | |
462 | unsigned long start, end; | |
463 | struct mm_struct *mm = vma->vm_mm; | |
464 | struct mmu_gather tlb; | |
465 | ||
854e9ed0 MK |
466 | /* MADV_FREE works for only anon vma at the moment */ |
467 | if (!vma_is_anonymous(vma)) | |
468 | return -EINVAL; | |
469 | ||
470 | start = max(vma->vm_start, start_addr); | |
471 | if (start >= vma->vm_end) | |
472 | return -EINVAL; | |
473 | end = min(vma->vm_end, end_addr); | |
474 | if (end <= vma->vm_start) | |
475 | return -EINVAL; | |
476 | ||
477 | lru_add_drain(); | |
478 | tlb_gather_mmu(&tlb, mm, start, end); | |
479 | update_hiwater_rss(mm); | |
480 | ||
481 | mmu_notifier_invalidate_range_start(mm, start, end); | |
482 | madvise_free_page_range(&tlb, vma, start, end); | |
483 | mmu_notifier_invalidate_range_end(mm, start, end); | |
484 | tlb_finish_mmu(&tlb, start, end); | |
485 | ||
486 | return 0; | |
487 | } | |
488 | ||
1da177e4 LT |
489 | /* |
490 | * Application no longer needs these pages. If the pages are dirty, | |
491 | * it's OK to just throw them away. The app will be more careful about | |
492 | * data it wants to keep. Be sure to free swap resources too. The | |
7e6cbea3 | 493 | * zap_page_range call sets things up for shrink_active_list to actually free |
1da177e4 LT |
494 | * these pages later if no one else has touched them in the meantime, |
495 | * although we could add these pages to a global reuse list for | |
7e6cbea3 | 496 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
497 | * |
498 | * NB: This interface discards data rather than pushes it out to swap, | |
499 | * as some implementations do. This has performance implications for | |
500 | * applications like large transactional databases which want to discard | |
501 | * pages in anonymous maps after committing to backing store the data | |
502 | * that was kept in them. There is no reason to write this data out to | |
503 | * the swap area if the application is discarding it. | |
504 | * | |
505 | * An interface that causes the system to free clean pages and flush | |
506 | * dirty pages is already available as msync(MS_INVALIDATE). | |
507 | */ | |
230ca982 MR |
508 | static long madvise_dontneed_single_vma(struct vm_area_struct *vma, |
509 | unsigned long start, unsigned long end) | |
510 | { | |
511 | zap_page_range(vma, start, end - start); | |
512 | return 0; | |
513 | } | |
514 | ||
515 | static long madvise_dontneed_free(struct vm_area_struct *vma, | |
516 | struct vm_area_struct **prev, | |
517 | unsigned long start, unsigned long end, | |
518 | int behavior) | |
1da177e4 | 519 | { |
05b74384 | 520 | *prev = vma; |
23519073 | 521 | if (!can_madv_dontneed_vma(vma)) |
1da177e4 LT |
522 | return -EINVAL; |
523 | ||
70ccb92f AA |
524 | if (!userfaultfd_remove(vma, start, end)) { |
525 | *prev = NULL; /* mmap_sem has been dropped, prev is stale */ | |
526 | ||
527 | down_read(¤t->mm->mmap_sem); | |
528 | vma = find_vma(current->mm, start); | |
529 | if (!vma) | |
530 | return -ENOMEM; | |
531 | if (start < vma->vm_start) { | |
532 | /* | |
533 | * This "vma" under revalidation is the one | |
534 | * with the lowest vma->vm_start where start | |
535 | * is also < vma->vm_end. If start < | |
536 | * vma->vm_start it means an hole materialized | |
537 | * in the user address space within the | |
230ca982 MR |
538 | * virtual range passed to MADV_DONTNEED |
539 | * or MADV_FREE. | |
70ccb92f AA |
540 | */ |
541 | return -ENOMEM; | |
542 | } | |
543 | if (!can_madv_dontneed_vma(vma)) | |
544 | return -EINVAL; | |
545 | if (end > vma->vm_end) { | |
546 | /* | |
547 | * Don't fail if end > vma->vm_end. If the old | |
548 | * vma was splitted while the mmap_sem was | |
549 | * released the effect of the concurrent | |
230ca982 | 550 | * operation may not cause madvise() to |
70ccb92f AA |
551 | * have an undefined result. There may be an |
552 | * adjacent next vma that we'll walk | |
553 | * next. userfaultfd_remove() will generate an | |
554 | * UFFD_EVENT_REMOVE repetition on the | |
555 | * end-vma->vm_end range, but the manager can | |
556 | * handle a repetition fine. | |
557 | */ | |
558 | end = vma->vm_end; | |
559 | } | |
560 | VM_WARN_ON(start >= end); | |
561 | } | |
230ca982 MR |
562 | |
563 | if (behavior == MADV_DONTNEED) | |
564 | return madvise_dontneed_single_vma(vma, start, end); | |
565 | else if (behavior == MADV_FREE) | |
566 | return madvise_free_single_vma(vma, start, end); | |
567 | else | |
568 | return -EINVAL; | |
1da177e4 LT |
569 | } |
570 | ||
f6b3ec23 BP |
571 | /* |
572 | * Application wants to free up the pages and associated backing store. | |
573 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
574 | */ |
575 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 576 | struct vm_area_struct **prev, |
f6b3ec23 BP |
577 | unsigned long start, unsigned long end) |
578 | { | |
3f31d075 | 579 | loff_t offset; |
90ed52eb | 580 | int error; |
9ab4233d | 581 | struct file *f; |
f6b3ec23 | 582 | |
90ed52eb | 583 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ |
00e9fa2d | 584 | |
72079ba0 | 585 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
586 | return -EINVAL; |
587 | ||
9ab4233d AL |
588 | f = vma->vm_file; |
589 | ||
590 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
591 | return -EINVAL; |
592 | } | |
593 | ||
69cf0fac HD |
594 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
595 | return -EACCES; | |
596 | ||
f6b3ec23 BP |
597 | offset = (loff_t)(start - vma->vm_start) |
598 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 599 | |
9ab4233d AL |
600 | /* |
601 | * Filesystem's fallocate may need to take i_mutex. We need to | |
602 | * explicitly grab a reference because the vma (and hence the | |
603 | * vma's reference to the file) can go away as soon as we drop | |
604 | * mmap_sem. | |
605 | */ | |
606 | get_file(f); | |
70ccb92f AA |
607 | if (userfaultfd_remove(vma, start, end)) { |
608 | /* mmap_sem was not released by userfaultfd_remove() */ | |
609 | up_read(¤t->mm->mmap_sem); | |
610 | } | |
72c72bdf | 611 | error = vfs_fallocate(f, |
3f31d075 HD |
612 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
613 | offset, end - start); | |
9ab4233d | 614 | fput(f); |
0a27a14a | 615 | down_read(¤t->mm->mmap_sem); |
90ed52eb | 616 | return error; |
f6b3ec23 BP |
617 | } |
618 | ||
9893e49d AK |
619 | #ifdef CONFIG_MEMORY_FAILURE |
620 | /* | |
621 | * Error injection support for memory error handling. | |
622 | */ | |
97167a76 AK |
623 | static int madvise_inject_error(int behavior, |
624 | unsigned long start, unsigned long end) | |
9893e49d | 625 | { |
97167a76 | 626 | struct page *page; |
c461ad6a | 627 | struct zone *zone; |
19bfbe22 | 628 | unsigned int order; |
97167a76 | 629 | |
9893e49d AK |
630 | if (!capable(CAP_SYS_ADMIN)) |
631 | return -EPERM; | |
97167a76 | 632 | |
19bfbe22 AM |
633 | |
634 | for (; start < end; start += PAGE_SIZE << order) { | |
325c4ef5 AM |
635 | int ret; |
636 | ||
97167a76 | 637 | ret = get_user_pages_fast(start, 1, 0, &page); |
9893e49d AK |
638 | if (ret != 1) |
639 | return ret; | |
325c4ef5 | 640 | |
19bfbe22 AM |
641 | /* |
642 | * When soft offlining hugepages, after migrating the page | |
643 | * we dissolve it, therefore in the second loop "page" will | |
644 | * no longer be a compound page, and order will be 0. | |
645 | */ | |
646 | order = compound_order(compound_head(page)); | |
647 | ||
97167a76 AK |
648 | if (PageHWPoison(page)) { |
649 | put_page(page); | |
29b4eede WL |
650 | continue; |
651 | } | |
97167a76 AK |
652 | |
653 | if (behavior == MADV_SOFT_OFFLINE) { | |
654 | pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n", | |
655 | page_to_pfn(page), start); | |
656 | ||
657 | ret = soft_offline_page(page, MF_COUNT_INCREASED); | |
afcf938e | 658 | if (ret) |
8302423b | 659 | return ret; |
afcf938e AK |
660 | continue; |
661 | } | |
97167a76 AK |
662 | pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n", |
663 | page_to_pfn(page), start); | |
664 | ||
665 | ret = memory_failure(page_to_pfn(page), 0, MF_COUNT_INCREASED); | |
23a003bf NH |
666 | if (ret) |
667 | return ret; | |
9893e49d | 668 | } |
c461ad6a MG |
669 | |
670 | /* Ensure that all poisoned pages are removed from per-cpu lists */ | |
671 | for_each_populated_zone(zone) | |
672 | drain_all_pages(zone); | |
673 | ||
325c4ef5 | 674 | return 0; |
9893e49d AK |
675 | } |
676 | #endif | |
677 | ||
165cd402 | 678 | static long |
679 | madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, | |
680 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 681 | { |
1da177e4 | 682 | switch (behavior) { |
f6b3ec23 | 683 | case MADV_REMOVE: |
3866ea90 | 684 | return madvise_remove(vma, prev, start, end); |
1da177e4 | 685 | case MADV_WILLNEED: |
3866ea90 | 686 | return madvise_willneed(vma, prev, start, end); |
854e9ed0 | 687 | case MADV_FREE: |
1da177e4 | 688 | case MADV_DONTNEED: |
230ca982 | 689 | return madvise_dontneed_free(vma, prev, start, end, behavior); |
1da177e4 | 690 | default: |
3866ea90 | 691 | return madvise_behavior(vma, prev, start, end, behavior); |
1da177e4 | 692 | } |
1da177e4 LT |
693 | } |
694 | ||
1ecef9ed | 695 | static bool |
75927af8 NP |
696 | madvise_behavior_valid(int behavior) |
697 | { | |
698 | switch (behavior) { | |
699 | case MADV_DOFORK: | |
700 | case MADV_DONTFORK: | |
701 | case MADV_NORMAL: | |
702 | case MADV_SEQUENTIAL: | |
703 | case MADV_RANDOM: | |
704 | case MADV_REMOVE: | |
705 | case MADV_WILLNEED: | |
706 | case MADV_DONTNEED: | |
854e9ed0 | 707 | case MADV_FREE: |
f8af4da3 HD |
708 | #ifdef CONFIG_KSM |
709 | case MADV_MERGEABLE: | |
710 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
711 | #endif |
712 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
713 | case MADV_HUGEPAGE: | |
a664b2d8 | 714 | case MADV_NOHUGEPAGE: |
f8af4da3 | 715 | #endif |
accb61fe JB |
716 | case MADV_DONTDUMP: |
717 | case MADV_DODUMP: | |
d2cd9ede RR |
718 | case MADV_WIPEONFORK: |
719 | case MADV_KEEPONFORK: | |
5e451be7 AK |
720 | #ifdef CONFIG_MEMORY_FAILURE |
721 | case MADV_SOFT_OFFLINE: | |
722 | case MADV_HWPOISON: | |
723 | #endif | |
1ecef9ed | 724 | return true; |
75927af8 NP |
725 | |
726 | default: | |
1ecef9ed | 727 | return false; |
75927af8 NP |
728 | } |
729 | } | |
3866ea90 | 730 | |
1da177e4 LT |
731 | /* |
732 | * The madvise(2) system call. | |
733 | * | |
734 | * Applications can use madvise() to advise the kernel how it should | |
735 | * handle paging I/O in this VM area. The idea is to help the kernel | |
736 | * use appropriate read-ahead and caching techniques. The information | |
737 | * provided is advisory only, and can be safely disregarded by the | |
738 | * kernel without affecting the correct operation of the application. | |
739 | * | |
740 | * behavior values: | |
741 | * MADV_NORMAL - the default behavior is to read clusters. This | |
742 | * results in some read-ahead and read-behind. | |
743 | * MADV_RANDOM - the system should read the minimum amount of data | |
744 | * on any access, since it is unlikely that the appli- | |
745 | * cation will need more than what it asks for. | |
746 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
747 | * once, so they can be aggressively read ahead, and | |
748 | * can be freed soon after they are accessed. | |
749 | * MADV_WILLNEED - the application is notifying the system to read | |
750 | * some pages ahead. | |
751 | * MADV_DONTNEED - the application is finished with the given range, | |
752 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
753 | * MADV_FREE - the application marks pages in the given range as lazy free, |
754 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
755 | * MADV_REMOVE - the application wants to free up the given range of |
756 | * pages and associated backing store. | |
3866ea90 HD |
757 | * MADV_DONTFORK - omit this area from child's address space when forking: |
758 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
759 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
d7206a70 NH |
760 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
761 | * were corrupted by unrecoverable hardware memory failure. | |
762 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
763 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
764 | * this area with pages of identical content from other such areas. | |
765 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
766 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
767 | * huge pages in the future. Existing pages might be coalesced and | |
768 | * new pages might be allocated as THP. | |
769 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
770 | * transparent huge pages so the existing pages will not be | |
771 | * coalesced into THP and new pages will not be allocated as THP. | |
772 | * MADV_DONTDUMP - the application wants to prevent pages in the given range | |
773 | * from being included in its core dump. | |
774 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
1da177e4 LT |
775 | * |
776 | * return values: | |
777 | * zero - success | |
778 | * -EINVAL - start + len < 0, start is not page-aligned, | |
779 | * "behavior" is not a valid value, or application | |
780 | * is attempting to release locked or shared pages. | |
781 | * -ENOMEM - addresses in the specified range are not currently | |
782 | * mapped, or are outside the AS of the process. | |
783 | * -EIO - an I/O error occurred while paging in data. | |
784 | * -EBADF - map exists, but area maps something that isn't a file. | |
785 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
786 | */ | |
3480b257 | 787 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) |
1da177e4 | 788 | { |
05b74384 | 789 | unsigned long end, tmp; |
ec9bed9d | 790 | struct vm_area_struct *vma, *prev; |
1da177e4 LT |
791 | int unmapped_error = 0; |
792 | int error = -EINVAL; | |
f7977793 | 793 | int write; |
1da177e4 | 794 | size_t len; |
1998cc04 | 795 | struct blk_plug plug; |
1da177e4 | 796 | |
75927af8 NP |
797 | if (!madvise_behavior_valid(behavior)) |
798 | return error; | |
799 | ||
1da177e4 | 800 | if (start & ~PAGE_MASK) |
84d96d89 | 801 | return error; |
1da177e4 LT |
802 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; |
803 | ||
804 | /* Check to see whether len was rounded up from small -ve to zero */ | |
805 | if (len_in && !len) | |
84d96d89 | 806 | return error; |
1da177e4 LT |
807 | |
808 | end = start + len; | |
809 | if (end < start) | |
84d96d89 | 810 | return error; |
1da177e4 LT |
811 | |
812 | error = 0; | |
813 | if (end == start) | |
84d96d89 RV |
814 | return error; |
815 | ||
5e451be7 AK |
816 | #ifdef CONFIG_MEMORY_FAILURE |
817 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) | |
818 | return madvise_inject_error(behavior, start, start + len_in); | |
819 | #endif | |
820 | ||
84d96d89 | 821 | write = madvise_need_mmap_write(behavior); |
dc0ef0df MH |
822 | if (write) { |
823 | if (down_write_killable(¤t->mm->mmap_sem)) | |
824 | return -EINTR; | |
825 | } else { | |
84d96d89 | 826 | down_read(¤t->mm->mmap_sem); |
dc0ef0df | 827 | } |
1da177e4 LT |
828 | |
829 | /* | |
830 | * If the interval [start,end) covers some unmapped address | |
831 | * ranges, just ignore them, but return -ENOMEM at the end. | |
05b74384 | 832 | * - different from the way of handling in mlock etc. |
1da177e4 | 833 | */ |
05b74384 | 834 | vma = find_vma_prev(current->mm, start, &prev); |
836d5ffd HD |
835 | if (vma && start > vma->vm_start) |
836 | prev = vma; | |
837 | ||
1998cc04 | 838 | blk_start_plug(&plug); |
1da177e4 LT |
839 | for (;;) { |
840 | /* Still start < end. */ | |
841 | error = -ENOMEM; | |
842 | if (!vma) | |
84d96d89 | 843 | goto out; |
1da177e4 | 844 | |
05b74384 | 845 | /* Here start < (end|vma->vm_end). */ |
1da177e4 LT |
846 | if (start < vma->vm_start) { |
847 | unmapped_error = -ENOMEM; | |
848 | start = vma->vm_start; | |
05b74384 | 849 | if (start >= end) |
84d96d89 | 850 | goto out; |
1da177e4 LT |
851 | } |
852 | ||
05b74384 PM |
853 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ |
854 | tmp = vma->vm_end; | |
855 | if (end < tmp) | |
856 | tmp = end; | |
1da177e4 | 857 | |
05b74384 PM |
858 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ |
859 | error = madvise_vma(vma, &prev, start, tmp, behavior); | |
1da177e4 | 860 | if (error) |
84d96d89 | 861 | goto out; |
05b74384 | 862 | start = tmp; |
90ed52eb | 863 | if (prev && start < prev->vm_end) |
05b74384 PM |
864 | start = prev->vm_end; |
865 | error = unmapped_error; | |
866 | if (start >= end) | |
84d96d89 | 867 | goto out; |
90ed52eb HD |
868 | if (prev) |
869 | vma = prev->vm_next; | |
870 | else /* madvise_remove dropped mmap_sem */ | |
871 | vma = find_vma(current->mm, start); | |
1da177e4 | 872 | } |
1da177e4 | 873 | out: |
84d96d89 | 874 | blk_finish_plug(&plug); |
f7977793 | 875 | if (write) |
0a27a14a NP |
876 | up_write(¤t->mm->mmap_sem); |
877 | else | |
878 | up_read(¤t->mm->mmap_sem); | |
879 | ||
1da177e4 LT |
880 | return error; |
881 | } |