Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/mm/madvise.c | |
4 | * | |
5 | * Copyright (C) 1999 Linus Torvalds | |
6 | * Copyright (C) 2002 Christoph Hellwig | |
7 | */ | |
8 | ||
9 | #include <linux/mman.h> | |
10 | #include <linux/pagemap.h> | |
11 | #include <linux/syscalls.h> | |
05b74384 | 12 | #include <linux/mempolicy.h> |
afcf938e | 13 | #include <linux/page-isolation.h> |
9c276cc6 | 14 | #include <linux/page_idle.h> |
05ce7724 | 15 | #include <linux/userfaultfd_k.h> |
1da177e4 | 16 | #include <linux/hugetlb.h> |
3f31d075 | 17 | #include <linux/falloc.h> |
692fe624 | 18 | #include <linux/fadvise.h> |
e8edc6e0 | 19 | #include <linux/sched.h> |
f8af4da3 | 20 | #include <linux/ksm.h> |
3f31d075 | 21 | #include <linux/fs.h> |
9ab4233d | 22 | #include <linux/file.h> |
1998cc04 | 23 | #include <linux/blkdev.h> |
66114cad | 24 | #include <linux/backing-dev.h> |
a520110e | 25 | #include <linux/pagewalk.h> |
1998cc04 SL |
26 | #include <linux/swap.h> |
27 | #include <linux/swapops.h> | |
3a4f8a0b | 28 | #include <linux/shmem_fs.h> |
854e9ed0 MK |
29 | #include <linux/mmu_notifier.h> |
30 | ||
31 | #include <asm/tlb.h> | |
1da177e4 | 32 | |
23519073 KS |
33 | #include "internal.h" |
34 | ||
0a27a14a NP |
35 | /* |
36 | * Any behaviour which results in changes to the vma->vm_flags needs to | |
37 | * take mmap_sem for writing. Others, which simply traverse vmas, need | |
38 | * to only take it for reading. | |
39 | */ | |
40 | static int madvise_need_mmap_write(int behavior) | |
41 | { | |
42 | switch (behavior) { | |
43 | case MADV_REMOVE: | |
44 | case MADV_WILLNEED: | |
45 | case MADV_DONTNEED: | |
9c276cc6 | 46 | case MADV_COLD: |
1a4e58cc | 47 | case MADV_PAGEOUT: |
854e9ed0 | 48 | case MADV_FREE: |
0a27a14a NP |
49 | return 0; |
50 | default: | |
51 | /* be safe, default to 1. list exceptions explicitly */ | |
52 | return 1; | |
53 | } | |
54 | } | |
55 | ||
1da177e4 LT |
56 | /* |
57 | * We can potentially split a vm area into separate | |
58 | * areas, each area with its own behavior. | |
59 | */ | |
ec9bed9d | 60 | static long madvise_behavior(struct vm_area_struct *vma, |
05b74384 PM |
61 | struct vm_area_struct **prev, |
62 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 63 | { |
ec9bed9d | 64 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 65 | int error = 0; |
05b74384 | 66 | pgoff_t pgoff; |
3866ea90 | 67 | unsigned long new_flags = vma->vm_flags; |
e798c6e8 PM |
68 | |
69 | switch (behavior) { | |
f8225661 MT |
70 | case MADV_NORMAL: |
71 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
72 | break; | |
e798c6e8 | 73 | case MADV_SEQUENTIAL: |
f8225661 | 74 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; |
e798c6e8 PM |
75 | break; |
76 | case MADV_RANDOM: | |
f8225661 | 77 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; |
e798c6e8 | 78 | break; |
f8225661 MT |
79 | case MADV_DONTFORK: |
80 | new_flags |= VM_DONTCOPY; | |
81 | break; | |
82 | case MADV_DOFORK: | |
3866ea90 HD |
83 | if (vma->vm_flags & VM_IO) { |
84 | error = -EINVAL; | |
85 | goto out; | |
86 | } | |
f8225661 | 87 | new_flags &= ~VM_DONTCOPY; |
e798c6e8 | 88 | break; |
d2cd9ede RR |
89 | case MADV_WIPEONFORK: |
90 | /* MADV_WIPEONFORK is only supported on anonymous memory. */ | |
91 | if (vma->vm_file || vma->vm_flags & VM_SHARED) { | |
92 | error = -EINVAL; | |
93 | goto out; | |
94 | } | |
95 | new_flags |= VM_WIPEONFORK; | |
96 | break; | |
97 | case MADV_KEEPONFORK: | |
98 | new_flags &= ~VM_WIPEONFORK; | |
99 | break; | |
accb61fe | 100 | case MADV_DONTDUMP: |
0103bd16 | 101 | new_flags |= VM_DONTDUMP; |
accb61fe JB |
102 | break; |
103 | case MADV_DODUMP: | |
d41aa525 | 104 | if (!is_vm_hugetlb_page(vma) && new_flags & VM_SPECIAL) { |
0103bd16 KK |
105 | error = -EINVAL; |
106 | goto out; | |
107 | } | |
108 | new_flags &= ~VM_DONTDUMP; | |
accb61fe | 109 | break; |
f8af4da3 HD |
110 | case MADV_MERGEABLE: |
111 | case MADV_UNMERGEABLE: | |
112 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
f3bc0dba MR |
113 | if (error) |
114 | goto out_convert_errno; | |
f8af4da3 | 115 | break; |
0af4e98b | 116 | case MADV_HUGEPAGE: |
a664b2d8 | 117 | case MADV_NOHUGEPAGE: |
60ab3244 | 118 | error = hugepage_madvise(vma, &new_flags, behavior); |
f3bc0dba MR |
119 | if (error) |
120 | goto out_convert_errno; | |
0af4e98b | 121 | break; |
e798c6e8 PM |
122 | } |
123 | ||
05b74384 PM |
124 | if (new_flags == vma->vm_flags) { |
125 | *prev = vma; | |
836d5ffd | 126 | goto out; |
05b74384 PM |
127 | } |
128 | ||
129 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
130 | *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma, | |
19a809af AA |
131 | vma->vm_file, pgoff, vma_policy(vma), |
132 | vma->vm_userfaultfd_ctx); | |
05b74384 PM |
133 | if (*prev) { |
134 | vma = *prev; | |
135 | goto success; | |
136 | } | |
137 | ||
138 | *prev = vma; | |
1da177e4 LT |
139 | |
140 | if (start != vma->vm_start) { | |
def5efe0 DR |
141 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
142 | error = -ENOMEM; | |
1da177e4 | 143 | goto out; |
def5efe0 DR |
144 | } |
145 | error = __split_vma(mm, vma, start, 1); | |
f3bc0dba MR |
146 | if (error) |
147 | goto out_convert_errno; | |
1da177e4 LT |
148 | } |
149 | ||
150 | if (end != vma->vm_end) { | |
def5efe0 DR |
151 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
152 | error = -ENOMEM; | |
1da177e4 | 153 | goto out; |
def5efe0 DR |
154 | } |
155 | error = __split_vma(mm, vma, end, 0); | |
f3bc0dba MR |
156 | if (error) |
157 | goto out_convert_errno; | |
1da177e4 LT |
158 | } |
159 | ||
836d5ffd | 160 | success: |
1da177e4 LT |
161 | /* |
162 | * vm_flags is protected by the mmap_sem held in write mode. | |
163 | */ | |
e798c6e8 | 164 | vma->vm_flags = new_flags; |
f3bc0dba MR |
165 | |
166 | out_convert_errno: | |
167 | /* | |
168 | * madvise() returns EAGAIN if kernel resources, such as | |
169 | * slab, are temporarily unavailable. | |
170 | */ | |
171 | if (error == -ENOMEM) | |
172 | error = -EAGAIN; | |
1da177e4 | 173 | out: |
1da177e4 LT |
174 | return error; |
175 | } | |
176 | ||
1998cc04 SL |
177 | #ifdef CONFIG_SWAP |
178 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
179 | unsigned long end, struct mm_walk *walk) | |
180 | { | |
181 | pte_t *orig_pte; | |
182 | struct vm_area_struct *vma = walk->private; | |
183 | unsigned long index; | |
184 | ||
185 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
186 | return 0; | |
187 | ||
188 | for (index = start; index != end; index += PAGE_SIZE) { | |
189 | pte_t pte; | |
190 | swp_entry_t entry; | |
191 | struct page *page; | |
192 | spinlock_t *ptl; | |
193 | ||
194 | orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); | |
195 | pte = *(orig_pte + ((index - start) / PAGE_SIZE)); | |
196 | pte_unmap_unlock(orig_pte, ptl); | |
197 | ||
0661a336 | 198 | if (pte_present(pte) || pte_none(pte)) |
1998cc04 SL |
199 | continue; |
200 | entry = pte_to_swp_entry(pte); | |
201 | if (unlikely(non_swap_entry(entry))) | |
202 | continue; | |
203 | ||
204 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, | |
23955622 | 205 | vma, index, false); |
1998cc04 | 206 | if (page) |
09cbfeaf | 207 | put_page(page); |
1998cc04 SL |
208 | } |
209 | ||
210 | return 0; | |
211 | } | |
212 | ||
7b86ac33 CH |
213 | static const struct mm_walk_ops swapin_walk_ops = { |
214 | .pmd_entry = swapin_walk_pmd_entry, | |
215 | }; | |
1998cc04 SL |
216 | |
217 | static void force_shm_swapin_readahead(struct vm_area_struct *vma, | |
218 | unsigned long start, unsigned long end, | |
219 | struct address_space *mapping) | |
220 | { | |
221 | pgoff_t index; | |
222 | struct page *page; | |
223 | swp_entry_t swap; | |
224 | ||
225 | for (; start < end; start += PAGE_SIZE) { | |
226 | index = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
227 | ||
55231e5c | 228 | page = find_get_entry(mapping, index); |
3159f943 | 229 | if (!xa_is_value(page)) { |
1998cc04 | 230 | if (page) |
09cbfeaf | 231 | put_page(page); |
1998cc04 SL |
232 | continue; |
233 | } | |
234 | swap = radix_to_swp_entry(page); | |
235 | page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE, | |
23955622 | 236 | NULL, 0, false); |
1998cc04 | 237 | if (page) |
09cbfeaf | 238 | put_page(page); |
1998cc04 SL |
239 | } |
240 | ||
241 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
242 | } | |
243 | #endif /* CONFIG_SWAP */ | |
244 | ||
1da177e4 LT |
245 | /* |
246 | * Schedule all required I/O operations. Do not wait for completion. | |
247 | */ | |
ec9bed9d VC |
248 | static long madvise_willneed(struct vm_area_struct *vma, |
249 | struct vm_area_struct **prev, | |
1da177e4 LT |
250 | unsigned long start, unsigned long end) |
251 | { | |
252 | struct file *file = vma->vm_file; | |
692fe624 | 253 | loff_t offset; |
1da177e4 | 254 | |
6ea8d958 | 255 | *prev = vma; |
1998cc04 | 256 | #ifdef CONFIG_SWAP |
97b713ba | 257 | if (!file) { |
7b86ac33 CH |
258 | walk_page_range(vma->vm_mm, start, end, &swapin_walk_ops, vma); |
259 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
1998cc04 SL |
260 | return 0; |
261 | } | |
1998cc04 | 262 | |
97b713ba | 263 | if (shmem_mapping(file->f_mapping)) { |
97b713ba CH |
264 | force_shm_swapin_readahead(vma, start, end, |
265 | file->f_mapping); | |
266 | return 0; | |
267 | } | |
268 | #else | |
1bef4003 S |
269 | if (!file) |
270 | return -EBADF; | |
97b713ba | 271 | #endif |
1bef4003 | 272 | |
e748dcd0 | 273 | if (IS_DAX(file_inode(file))) { |
fe77ba6f CO |
274 | /* no bad return value, but ignore advice */ |
275 | return 0; | |
276 | } | |
277 | ||
692fe624 JK |
278 | /* |
279 | * Filesystem's fadvise may need to take various locks. We need to | |
280 | * explicitly grab a reference because the vma (and hence the | |
281 | * vma's reference to the file) can go away as soon as we drop | |
282 | * mmap_sem. | |
283 | */ | |
284 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ | |
285 | get_file(file); | |
286 | up_read(¤t->mm->mmap_sem); | |
287 | offset = (loff_t)(start - vma->vm_start) | |
288 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
289 | vfs_fadvise(file, offset, end - start, POSIX_FADV_WILLNEED); | |
290 | fput(file); | |
291 | down_read(¤t->mm->mmap_sem); | |
1da177e4 LT |
292 | return 0; |
293 | } | |
294 | ||
9c276cc6 MK |
295 | static int madvise_cold_pte_range(pmd_t *pmd, unsigned long addr, |
296 | unsigned long end, struct mm_walk *walk) | |
297 | { | |
298 | struct mmu_gather *tlb = walk->private; | |
299 | struct mm_struct *mm = tlb->mm; | |
300 | struct vm_area_struct *vma = walk->vma; | |
301 | pte_t *orig_pte, *pte, ptent; | |
302 | spinlock_t *ptl; | |
303 | struct page *page; | |
304 | ||
305 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
306 | if (pmd_trans_huge(*pmd)) { | |
307 | pmd_t orig_pmd; | |
308 | unsigned long next = pmd_addr_end(addr, end); | |
309 | ||
310 | tlb_change_page_size(tlb, HPAGE_PMD_SIZE); | |
311 | ptl = pmd_trans_huge_lock(pmd, vma); | |
312 | if (!ptl) | |
313 | return 0; | |
314 | ||
315 | orig_pmd = *pmd; | |
316 | if (is_huge_zero_pmd(orig_pmd)) | |
317 | goto huge_unlock; | |
318 | ||
319 | if (unlikely(!pmd_present(orig_pmd))) { | |
320 | VM_BUG_ON(thp_migration_supported() && | |
321 | !is_pmd_migration_entry(orig_pmd)); | |
322 | goto huge_unlock; | |
323 | } | |
324 | ||
325 | page = pmd_page(orig_pmd); | |
326 | if (next - addr != HPAGE_PMD_SIZE) { | |
327 | int err; | |
328 | ||
329 | if (page_mapcount(page) != 1) | |
330 | goto huge_unlock; | |
331 | ||
332 | get_page(page); | |
333 | spin_unlock(ptl); | |
334 | lock_page(page); | |
335 | err = split_huge_page(page); | |
336 | unlock_page(page); | |
337 | put_page(page); | |
338 | if (!err) | |
339 | goto regular_page; | |
340 | return 0; | |
341 | } | |
342 | ||
343 | if (pmd_young(orig_pmd)) { | |
344 | pmdp_invalidate(vma, addr, pmd); | |
345 | orig_pmd = pmd_mkold(orig_pmd); | |
346 | ||
347 | set_pmd_at(mm, addr, pmd, orig_pmd); | |
348 | tlb_remove_pmd_tlb_entry(tlb, pmd, addr); | |
349 | } | |
350 | ||
351 | test_and_clear_page_young(page); | |
352 | deactivate_page(page); | |
353 | huge_unlock: | |
354 | spin_unlock(ptl); | |
355 | return 0; | |
356 | } | |
357 | ||
358 | if (pmd_trans_unstable(pmd)) | |
359 | return 0; | |
360 | regular_page: | |
361 | #endif | |
362 | tlb_change_page_size(tlb, PAGE_SIZE); | |
363 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
364 | flush_tlb_batched_pending(mm); | |
365 | arch_enter_lazy_mmu_mode(); | |
366 | for (; addr < end; pte++, addr += PAGE_SIZE) { | |
367 | ptent = *pte; | |
368 | ||
369 | if (pte_none(ptent)) | |
370 | continue; | |
371 | ||
372 | if (!pte_present(ptent)) | |
373 | continue; | |
374 | ||
375 | page = vm_normal_page(vma, addr, ptent); | |
376 | if (!page) | |
377 | continue; | |
378 | ||
379 | /* | |
380 | * Creating a THP page is expensive so split it only if we | |
381 | * are sure it's worth. Split it if we are only owner. | |
382 | */ | |
383 | if (PageTransCompound(page)) { | |
384 | if (page_mapcount(page) != 1) | |
385 | break; | |
386 | get_page(page); | |
387 | if (!trylock_page(page)) { | |
388 | put_page(page); | |
389 | break; | |
390 | } | |
391 | pte_unmap_unlock(orig_pte, ptl); | |
392 | if (split_huge_page(page)) { | |
393 | unlock_page(page); | |
394 | put_page(page); | |
395 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
396 | break; | |
397 | } | |
398 | unlock_page(page); | |
399 | put_page(page); | |
400 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
401 | pte--; | |
402 | addr -= PAGE_SIZE; | |
403 | continue; | |
404 | } | |
405 | ||
406 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
407 | ||
408 | if (pte_young(ptent)) { | |
409 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
410 | tlb->fullmm); | |
411 | ptent = pte_mkold(ptent); | |
412 | set_pte_at(mm, addr, pte, ptent); | |
413 | tlb_remove_tlb_entry(tlb, pte, addr); | |
414 | } | |
415 | ||
416 | /* | |
417 | * We are deactivating a page for accelerating reclaiming. | |
418 | * VM couldn't reclaim the page unless we clear PG_young. | |
419 | * As a side effect, it makes confuse idle-page tracking | |
420 | * because they will miss recent referenced history. | |
421 | */ | |
422 | test_and_clear_page_young(page); | |
423 | deactivate_page(page); | |
424 | } | |
425 | ||
426 | arch_leave_lazy_mmu_mode(); | |
427 | pte_unmap_unlock(orig_pte, ptl); | |
428 | cond_resched(); | |
429 | ||
430 | return 0; | |
431 | } | |
432 | ||
433 | static const struct mm_walk_ops cold_walk_ops = { | |
434 | .pmd_entry = madvise_cold_pte_range, | |
435 | }; | |
436 | ||
437 | static void madvise_cold_page_range(struct mmu_gather *tlb, | |
438 | struct vm_area_struct *vma, | |
439 | unsigned long addr, unsigned long end) | |
440 | { | |
441 | tlb_start_vma(tlb, vma); | |
442 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, NULL); | |
443 | tlb_end_vma(tlb, vma); | |
444 | } | |
445 | ||
446 | static long madvise_cold(struct vm_area_struct *vma, | |
447 | struct vm_area_struct **prev, | |
448 | unsigned long start_addr, unsigned long end_addr) | |
449 | { | |
450 | struct mm_struct *mm = vma->vm_mm; | |
451 | struct mmu_gather tlb; | |
452 | ||
453 | *prev = vma; | |
454 | if (!can_madv_lru_vma(vma)) | |
455 | return -EINVAL; | |
456 | ||
457 | lru_add_drain(); | |
458 | tlb_gather_mmu(&tlb, mm, start_addr, end_addr); | |
459 | madvise_cold_page_range(&tlb, vma, start_addr, end_addr); | |
460 | tlb_finish_mmu(&tlb, start_addr, end_addr); | |
461 | ||
462 | return 0; | |
463 | } | |
464 | ||
1a4e58cc MK |
465 | static int madvise_pageout_pte_range(pmd_t *pmd, unsigned long addr, |
466 | unsigned long end, struct mm_walk *walk) | |
467 | { | |
468 | struct mmu_gather *tlb = walk->private; | |
469 | struct mm_struct *mm = tlb->mm; | |
470 | struct vm_area_struct *vma = walk->vma; | |
471 | pte_t *orig_pte, *pte, ptent; | |
472 | spinlock_t *ptl; | |
473 | LIST_HEAD(page_list); | |
474 | struct page *page; | |
475 | ||
476 | if (fatal_signal_pending(current)) | |
477 | return -EINTR; | |
478 | ||
479 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
480 | if (pmd_trans_huge(*pmd)) { | |
481 | pmd_t orig_pmd; | |
482 | unsigned long next = pmd_addr_end(addr, end); | |
483 | ||
484 | tlb_change_page_size(tlb, HPAGE_PMD_SIZE); | |
485 | ptl = pmd_trans_huge_lock(pmd, vma); | |
486 | if (!ptl) | |
487 | return 0; | |
488 | ||
489 | orig_pmd = *pmd; | |
490 | if (is_huge_zero_pmd(orig_pmd)) | |
491 | goto huge_unlock; | |
492 | ||
493 | if (unlikely(!pmd_present(orig_pmd))) { | |
494 | VM_BUG_ON(thp_migration_supported() && | |
495 | !is_pmd_migration_entry(orig_pmd)); | |
496 | goto huge_unlock; | |
497 | } | |
498 | ||
499 | page = pmd_page(orig_pmd); | |
500 | if (next - addr != HPAGE_PMD_SIZE) { | |
501 | int err; | |
502 | ||
503 | if (page_mapcount(page) != 1) | |
504 | goto huge_unlock; | |
505 | get_page(page); | |
506 | spin_unlock(ptl); | |
507 | lock_page(page); | |
508 | err = split_huge_page(page); | |
509 | unlock_page(page); | |
510 | put_page(page); | |
511 | if (!err) | |
512 | goto regular_page; | |
513 | return 0; | |
514 | } | |
515 | ||
516 | if (pmd_young(orig_pmd)) { | |
517 | pmdp_invalidate(vma, addr, pmd); | |
518 | orig_pmd = pmd_mkold(orig_pmd); | |
519 | ||
520 | set_pmd_at(mm, addr, pmd, orig_pmd); | |
521 | tlb_remove_tlb_entry(tlb, pmd, addr); | |
522 | } | |
523 | ||
524 | ClearPageReferenced(page); | |
525 | test_and_clear_page_young(page); | |
526 | ||
527 | if (!isolate_lru_page(page)) | |
528 | list_add(&page->lru, &page_list); | |
529 | huge_unlock: | |
530 | spin_unlock(ptl); | |
531 | reclaim_pages(&page_list); | |
532 | return 0; | |
533 | } | |
534 | ||
535 | if (pmd_trans_unstable(pmd)) | |
536 | return 0; | |
537 | regular_page: | |
538 | #endif | |
539 | tlb_change_page_size(tlb, PAGE_SIZE); | |
540 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
541 | flush_tlb_batched_pending(mm); | |
542 | arch_enter_lazy_mmu_mode(); | |
543 | for (; addr < end; pte++, addr += PAGE_SIZE) { | |
544 | ptent = *pte; | |
545 | if (!pte_present(ptent)) | |
546 | continue; | |
547 | ||
548 | page = vm_normal_page(vma, addr, ptent); | |
549 | if (!page) | |
550 | continue; | |
551 | ||
552 | /* | |
553 | * creating a THP page is expensive so split it only if we | |
554 | * are sure it's worth. Split it if we are only owner. | |
555 | */ | |
556 | if (PageTransCompound(page)) { | |
557 | if (page_mapcount(page) != 1) | |
558 | break; | |
559 | get_page(page); | |
560 | if (!trylock_page(page)) { | |
561 | put_page(page); | |
562 | break; | |
563 | } | |
564 | pte_unmap_unlock(orig_pte, ptl); | |
565 | if (split_huge_page(page)) { | |
566 | unlock_page(page); | |
567 | put_page(page); | |
568 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
569 | break; | |
570 | } | |
571 | unlock_page(page); | |
572 | put_page(page); | |
573 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
574 | pte--; | |
575 | addr -= PAGE_SIZE; | |
576 | continue; | |
577 | } | |
578 | ||
579 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
580 | ||
581 | if (pte_young(ptent)) { | |
582 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
583 | tlb->fullmm); | |
584 | ptent = pte_mkold(ptent); | |
585 | set_pte_at(mm, addr, pte, ptent); | |
586 | tlb_remove_tlb_entry(tlb, pte, addr); | |
587 | } | |
588 | ClearPageReferenced(page); | |
589 | test_and_clear_page_young(page); | |
590 | ||
591 | if (!isolate_lru_page(page)) | |
592 | list_add(&page->lru, &page_list); | |
593 | } | |
594 | ||
595 | arch_leave_lazy_mmu_mode(); | |
596 | pte_unmap_unlock(orig_pte, ptl); | |
597 | reclaim_pages(&page_list); | |
598 | cond_resched(); | |
599 | ||
600 | return 0; | |
601 | } | |
602 | ||
603 | static void madvise_pageout_page_range(struct mmu_gather *tlb, | |
604 | struct vm_area_struct *vma, | |
605 | unsigned long addr, unsigned long end) | |
606 | { | |
607 | tlb_start_vma(tlb, vma); | |
608 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, NULL); | |
609 | tlb_end_vma(tlb, vma); | |
610 | } | |
611 | ||
612 | static inline bool can_do_pageout(struct vm_area_struct *vma) | |
613 | { | |
614 | if (vma_is_anonymous(vma)) | |
615 | return true; | |
616 | if (!vma->vm_file) | |
617 | return false; | |
618 | /* | |
619 | * paging out pagecache only for non-anonymous mappings that correspond | |
620 | * to the files the calling process could (if tried) open for writing; | |
621 | * otherwise we'd be including shared non-exclusive mappings, which | |
622 | * opens a side channel. | |
623 | */ | |
624 | return inode_owner_or_capable(file_inode(vma->vm_file)) || | |
625 | inode_permission(file_inode(vma->vm_file), MAY_WRITE) == 0; | |
626 | } | |
627 | ||
628 | static long madvise_pageout(struct vm_area_struct *vma, | |
629 | struct vm_area_struct **prev, | |
630 | unsigned long start_addr, unsigned long end_addr) | |
631 | { | |
632 | struct mm_struct *mm = vma->vm_mm; | |
633 | struct mmu_gather tlb; | |
634 | ||
635 | *prev = vma; | |
636 | if (!can_madv_lru_vma(vma)) | |
637 | return -EINVAL; | |
638 | ||
639 | if (!can_do_pageout(vma)) | |
640 | return 0; | |
641 | ||
642 | lru_add_drain(); | |
643 | tlb_gather_mmu(&tlb, mm, start_addr, end_addr); | |
644 | madvise_pageout_page_range(&tlb, vma, start_addr, end_addr); | |
645 | tlb_finish_mmu(&tlb, start_addr, end_addr); | |
646 | ||
647 | return 0; | |
648 | } | |
649 | ||
854e9ed0 MK |
650 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
651 | unsigned long end, struct mm_walk *walk) | |
652 | ||
653 | { | |
654 | struct mmu_gather *tlb = walk->private; | |
655 | struct mm_struct *mm = tlb->mm; | |
656 | struct vm_area_struct *vma = walk->vma; | |
657 | spinlock_t *ptl; | |
658 | pte_t *orig_pte, *pte, ptent; | |
659 | struct page *page; | |
64b42bc1 | 660 | int nr_swap = 0; |
b8d3c4c3 MK |
661 | unsigned long next; |
662 | ||
663 | next = pmd_addr_end(addr, end); | |
664 | if (pmd_trans_huge(*pmd)) | |
665 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
666 | goto next; | |
854e9ed0 | 667 | |
854e9ed0 MK |
668 | if (pmd_trans_unstable(pmd)) |
669 | return 0; | |
670 | ||
ed6a7935 | 671 | tlb_change_page_size(tlb, PAGE_SIZE); |
854e9ed0 | 672 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
3ea27719 | 673 | flush_tlb_batched_pending(mm); |
854e9ed0 MK |
674 | arch_enter_lazy_mmu_mode(); |
675 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
676 | ptent = *pte; | |
677 | ||
64b42bc1 | 678 | if (pte_none(ptent)) |
854e9ed0 | 679 | continue; |
64b42bc1 MK |
680 | /* |
681 | * If the pte has swp_entry, just clear page table to | |
682 | * prevent swap-in which is more expensive rather than | |
683 | * (page allocation + zeroing). | |
684 | */ | |
685 | if (!pte_present(ptent)) { | |
686 | swp_entry_t entry; | |
687 | ||
688 | entry = pte_to_swp_entry(ptent); | |
689 | if (non_swap_entry(entry)) | |
690 | continue; | |
691 | nr_swap--; | |
692 | free_swap_and_cache(entry); | |
693 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
694 | continue; | |
695 | } | |
854e9ed0 | 696 | |
25b2995a | 697 | page = vm_normal_page(vma, addr, ptent); |
854e9ed0 MK |
698 | if (!page) |
699 | continue; | |
700 | ||
701 | /* | |
702 | * If pmd isn't transhuge but the page is THP and | |
703 | * is owned by only this process, split it and | |
704 | * deactivate all pages. | |
705 | */ | |
706 | if (PageTransCompound(page)) { | |
707 | if (page_mapcount(page) != 1) | |
708 | goto out; | |
709 | get_page(page); | |
710 | if (!trylock_page(page)) { | |
711 | put_page(page); | |
712 | goto out; | |
713 | } | |
714 | pte_unmap_unlock(orig_pte, ptl); | |
715 | if (split_huge_page(page)) { | |
716 | unlock_page(page); | |
717 | put_page(page); | |
718 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
719 | goto out; | |
720 | } | |
854e9ed0 | 721 | unlock_page(page); |
263630e8 | 722 | put_page(page); |
854e9ed0 MK |
723 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
724 | pte--; | |
725 | addr -= PAGE_SIZE; | |
726 | continue; | |
727 | } | |
728 | ||
729 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
730 | ||
731 | if (PageSwapCache(page) || PageDirty(page)) { | |
732 | if (!trylock_page(page)) | |
733 | continue; | |
734 | /* | |
735 | * If page is shared with others, we couldn't clear | |
736 | * PG_dirty of the page. | |
737 | */ | |
738 | if (page_mapcount(page) != 1) { | |
739 | unlock_page(page); | |
740 | continue; | |
741 | } | |
742 | ||
743 | if (PageSwapCache(page) && !try_to_free_swap(page)) { | |
744 | unlock_page(page); | |
745 | continue; | |
746 | } | |
747 | ||
748 | ClearPageDirty(page); | |
749 | unlock_page(page); | |
750 | } | |
751 | ||
752 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
753 | /* | |
754 | * Some of architecture(ex, PPC) don't update TLB | |
755 | * with set_pte_at and tlb_remove_tlb_entry so for | |
756 | * the portability, remap the pte with old|clean | |
757 | * after pte clearing. | |
758 | */ | |
759 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
760 | tlb->fullmm); | |
761 | ||
762 | ptent = pte_mkold(ptent); | |
763 | ptent = pte_mkclean(ptent); | |
764 | set_pte_at(mm, addr, pte, ptent); | |
765 | tlb_remove_tlb_entry(tlb, pte, addr); | |
766 | } | |
802a3a92 | 767 | mark_page_lazyfree(page); |
854e9ed0 MK |
768 | } |
769 | out: | |
64b42bc1 MK |
770 | if (nr_swap) { |
771 | if (current->mm == mm) | |
772 | sync_mm_rss(mm); | |
773 | ||
774 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); | |
775 | } | |
854e9ed0 MK |
776 | arch_leave_lazy_mmu_mode(); |
777 | pte_unmap_unlock(orig_pte, ptl); | |
778 | cond_resched(); | |
b8d3c4c3 | 779 | next: |
854e9ed0 MK |
780 | return 0; |
781 | } | |
782 | ||
7b86ac33 CH |
783 | static const struct mm_walk_ops madvise_free_walk_ops = { |
784 | .pmd_entry = madvise_free_pte_range, | |
785 | }; | |
854e9ed0 MK |
786 | |
787 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
788 | unsigned long start_addr, unsigned long end_addr) | |
789 | { | |
854e9ed0 | 790 | struct mm_struct *mm = vma->vm_mm; |
ac46d4f3 | 791 | struct mmu_notifier_range range; |
854e9ed0 MK |
792 | struct mmu_gather tlb; |
793 | ||
854e9ed0 MK |
794 | /* MADV_FREE works for only anon vma at the moment */ |
795 | if (!vma_is_anonymous(vma)) | |
796 | return -EINVAL; | |
797 | ||
ac46d4f3 JG |
798 | range.start = max(vma->vm_start, start_addr); |
799 | if (range.start >= vma->vm_end) | |
854e9ed0 | 800 | return -EINVAL; |
ac46d4f3 JG |
801 | range.end = min(vma->vm_end, end_addr); |
802 | if (range.end <= vma->vm_start) | |
854e9ed0 | 803 | return -EINVAL; |
7269f999 | 804 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, |
6f4f13e8 | 805 | range.start, range.end); |
854e9ed0 MK |
806 | |
807 | lru_add_drain(); | |
ac46d4f3 | 808 | tlb_gather_mmu(&tlb, mm, range.start, range.end); |
854e9ed0 MK |
809 | update_hiwater_rss(mm); |
810 | ||
ac46d4f3 | 811 | mmu_notifier_invalidate_range_start(&range); |
7b86ac33 CH |
812 | tlb_start_vma(&tlb, vma); |
813 | walk_page_range(vma->vm_mm, range.start, range.end, | |
814 | &madvise_free_walk_ops, &tlb); | |
815 | tlb_end_vma(&tlb, vma); | |
ac46d4f3 JG |
816 | mmu_notifier_invalidate_range_end(&range); |
817 | tlb_finish_mmu(&tlb, range.start, range.end); | |
854e9ed0 MK |
818 | |
819 | return 0; | |
820 | } | |
821 | ||
1da177e4 LT |
822 | /* |
823 | * Application no longer needs these pages. If the pages are dirty, | |
824 | * it's OK to just throw them away. The app will be more careful about | |
825 | * data it wants to keep. Be sure to free swap resources too. The | |
7e6cbea3 | 826 | * zap_page_range call sets things up for shrink_active_list to actually free |
1da177e4 LT |
827 | * these pages later if no one else has touched them in the meantime, |
828 | * although we could add these pages to a global reuse list for | |
7e6cbea3 | 829 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
830 | * |
831 | * NB: This interface discards data rather than pushes it out to swap, | |
832 | * as some implementations do. This has performance implications for | |
833 | * applications like large transactional databases which want to discard | |
834 | * pages in anonymous maps after committing to backing store the data | |
835 | * that was kept in them. There is no reason to write this data out to | |
836 | * the swap area if the application is discarding it. | |
837 | * | |
838 | * An interface that causes the system to free clean pages and flush | |
839 | * dirty pages is already available as msync(MS_INVALIDATE). | |
840 | */ | |
230ca982 MR |
841 | static long madvise_dontneed_single_vma(struct vm_area_struct *vma, |
842 | unsigned long start, unsigned long end) | |
843 | { | |
844 | zap_page_range(vma, start, end - start); | |
845 | return 0; | |
846 | } | |
847 | ||
848 | static long madvise_dontneed_free(struct vm_area_struct *vma, | |
849 | struct vm_area_struct **prev, | |
850 | unsigned long start, unsigned long end, | |
851 | int behavior) | |
1da177e4 | 852 | { |
05b74384 | 853 | *prev = vma; |
9c276cc6 | 854 | if (!can_madv_lru_vma(vma)) |
1da177e4 LT |
855 | return -EINVAL; |
856 | ||
70ccb92f AA |
857 | if (!userfaultfd_remove(vma, start, end)) { |
858 | *prev = NULL; /* mmap_sem has been dropped, prev is stale */ | |
859 | ||
860 | down_read(¤t->mm->mmap_sem); | |
861 | vma = find_vma(current->mm, start); | |
862 | if (!vma) | |
863 | return -ENOMEM; | |
864 | if (start < vma->vm_start) { | |
865 | /* | |
866 | * This "vma" under revalidation is the one | |
867 | * with the lowest vma->vm_start where start | |
868 | * is also < vma->vm_end. If start < | |
869 | * vma->vm_start it means an hole materialized | |
870 | * in the user address space within the | |
230ca982 MR |
871 | * virtual range passed to MADV_DONTNEED |
872 | * or MADV_FREE. | |
70ccb92f AA |
873 | */ |
874 | return -ENOMEM; | |
875 | } | |
9c276cc6 | 876 | if (!can_madv_lru_vma(vma)) |
70ccb92f AA |
877 | return -EINVAL; |
878 | if (end > vma->vm_end) { | |
879 | /* | |
880 | * Don't fail if end > vma->vm_end. If the old | |
881 | * vma was splitted while the mmap_sem was | |
882 | * released the effect of the concurrent | |
230ca982 | 883 | * operation may not cause madvise() to |
70ccb92f AA |
884 | * have an undefined result. There may be an |
885 | * adjacent next vma that we'll walk | |
886 | * next. userfaultfd_remove() will generate an | |
887 | * UFFD_EVENT_REMOVE repetition on the | |
888 | * end-vma->vm_end range, but the manager can | |
889 | * handle a repetition fine. | |
890 | */ | |
891 | end = vma->vm_end; | |
892 | } | |
893 | VM_WARN_ON(start >= end); | |
894 | } | |
230ca982 MR |
895 | |
896 | if (behavior == MADV_DONTNEED) | |
897 | return madvise_dontneed_single_vma(vma, start, end); | |
898 | else if (behavior == MADV_FREE) | |
899 | return madvise_free_single_vma(vma, start, end); | |
900 | else | |
901 | return -EINVAL; | |
1da177e4 LT |
902 | } |
903 | ||
f6b3ec23 BP |
904 | /* |
905 | * Application wants to free up the pages and associated backing store. | |
906 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
907 | */ |
908 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 909 | struct vm_area_struct **prev, |
f6b3ec23 BP |
910 | unsigned long start, unsigned long end) |
911 | { | |
3f31d075 | 912 | loff_t offset; |
90ed52eb | 913 | int error; |
9ab4233d | 914 | struct file *f; |
f6b3ec23 | 915 | |
90ed52eb | 916 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ |
00e9fa2d | 917 | |
72079ba0 | 918 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
919 | return -EINVAL; |
920 | ||
9ab4233d AL |
921 | f = vma->vm_file; |
922 | ||
923 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
924 | return -EINVAL; |
925 | } | |
926 | ||
69cf0fac HD |
927 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
928 | return -EACCES; | |
929 | ||
f6b3ec23 BP |
930 | offset = (loff_t)(start - vma->vm_start) |
931 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 932 | |
9ab4233d AL |
933 | /* |
934 | * Filesystem's fallocate may need to take i_mutex. We need to | |
935 | * explicitly grab a reference because the vma (and hence the | |
936 | * vma's reference to the file) can go away as soon as we drop | |
937 | * mmap_sem. | |
938 | */ | |
939 | get_file(f); | |
70ccb92f AA |
940 | if (userfaultfd_remove(vma, start, end)) { |
941 | /* mmap_sem was not released by userfaultfd_remove() */ | |
942 | up_read(¤t->mm->mmap_sem); | |
943 | } | |
72c72bdf | 944 | error = vfs_fallocate(f, |
3f31d075 HD |
945 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
946 | offset, end - start); | |
9ab4233d | 947 | fput(f); |
0a27a14a | 948 | down_read(¤t->mm->mmap_sem); |
90ed52eb | 949 | return error; |
f6b3ec23 BP |
950 | } |
951 | ||
9893e49d AK |
952 | #ifdef CONFIG_MEMORY_FAILURE |
953 | /* | |
954 | * Error injection support for memory error handling. | |
955 | */ | |
97167a76 AK |
956 | static int madvise_inject_error(int behavior, |
957 | unsigned long start, unsigned long end) | |
9893e49d | 958 | { |
97167a76 | 959 | struct page *page; |
c461ad6a | 960 | struct zone *zone; |
19bfbe22 | 961 | unsigned int order; |
97167a76 | 962 | |
9893e49d AK |
963 | if (!capable(CAP_SYS_ADMIN)) |
964 | return -EPERM; | |
97167a76 | 965 | |
19bfbe22 AM |
966 | |
967 | for (; start < end; start += PAGE_SIZE << order) { | |
23e7b5c2 | 968 | unsigned long pfn; |
325c4ef5 AM |
969 | int ret; |
970 | ||
97167a76 | 971 | ret = get_user_pages_fast(start, 1, 0, &page); |
9893e49d AK |
972 | if (ret != 1) |
973 | return ret; | |
23e7b5c2 | 974 | pfn = page_to_pfn(page); |
325c4ef5 | 975 | |
19bfbe22 AM |
976 | /* |
977 | * When soft offlining hugepages, after migrating the page | |
978 | * we dissolve it, therefore in the second loop "page" will | |
979 | * no longer be a compound page, and order will be 0. | |
980 | */ | |
981 | order = compound_order(compound_head(page)); | |
982 | ||
97167a76 AK |
983 | if (PageHWPoison(page)) { |
984 | put_page(page); | |
29b4eede WL |
985 | continue; |
986 | } | |
97167a76 AK |
987 | |
988 | if (behavior == MADV_SOFT_OFFLINE) { | |
989 | pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n", | |
23e7b5c2 | 990 | pfn, start); |
97167a76 AK |
991 | |
992 | ret = soft_offline_page(page, MF_COUNT_INCREASED); | |
afcf938e | 993 | if (ret) |
8302423b | 994 | return ret; |
afcf938e AK |
995 | continue; |
996 | } | |
23e7b5c2 | 997 | |
97167a76 | 998 | pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n", |
23e7b5c2 | 999 | pfn, start); |
97167a76 | 1000 | |
23e7b5c2 DW |
1001 | /* |
1002 | * Drop the page reference taken by get_user_pages_fast(). In | |
1003 | * the absence of MF_COUNT_INCREASED the memory_failure() | |
1004 | * routine is responsible for pinning the page to prevent it | |
1005 | * from being released back to the page allocator. | |
1006 | */ | |
1007 | put_page(page); | |
1008 | ret = memory_failure(pfn, 0); | |
23a003bf NH |
1009 | if (ret) |
1010 | return ret; | |
9893e49d | 1011 | } |
c461ad6a MG |
1012 | |
1013 | /* Ensure that all poisoned pages are removed from per-cpu lists */ | |
1014 | for_each_populated_zone(zone) | |
1015 | drain_all_pages(zone); | |
1016 | ||
325c4ef5 | 1017 | return 0; |
9893e49d AK |
1018 | } |
1019 | #endif | |
1020 | ||
165cd402 | 1021 | static long |
1022 | madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, | |
1023 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 1024 | { |
1da177e4 | 1025 | switch (behavior) { |
f6b3ec23 | 1026 | case MADV_REMOVE: |
3866ea90 | 1027 | return madvise_remove(vma, prev, start, end); |
1da177e4 | 1028 | case MADV_WILLNEED: |
3866ea90 | 1029 | return madvise_willneed(vma, prev, start, end); |
9c276cc6 MK |
1030 | case MADV_COLD: |
1031 | return madvise_cold(vma, prev, start, end); | |
1a4e58cc MK |
1032 | case MADV_PAGEOUT: |
1033 | return madvise_pageout(vma, prev, start, end); | |
854e9ed0 | 1034 | case MADV_FREE: |
1da177e4 | 1035 | case MADV_DONTNEED: |
230ca982 | 1036 | return madvise_dontneed_free(vma, prev, start, end, behavior); |
1da177e4 | 1037 | default: |
3866ea90 | 1038 | return madvise_behavior(vma, prev, start, end, behavior); |
1da177e4 | 1039 | } |
1da177e4 LT |
1040 | } |
1041 | ||
1ecef9ed | 1042 | static bool |
75927af8 NP |
1043 | madvise_behavior_valid(int behavior) |
1044 | { | |
1045 | switch (behavior) { | |
1046 | case MADV_DOFORK: | |
1047 | case MADV_DONTFORK: | |
1048 | case MADV_NORMAL: | |
1049 | case MADV_SEQUENTIAL: | |
1050 | case MADV_RANDOM: | |
1051 | case MADV_REMOVE: | |
1052 | case MADV_WILLNEED: | |
1053 | case MADV_DONTNEED: | |
854e9ed0 | 1054 | case MADV_FREE: |
9c276cc6 | 1055 | case MADV_COLD: |
1a4e58cc | 1056 | case MADV_PAGEOUT: |
f8af4da3 HD |
1057 | #ifdef CONFIG_KSM |
1058 | case MADV_MERGEABLE: | |
1059 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
1060 | #endif |
1061 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1062 | case MADV_HUGEPAGE: | |
a664b2d8 | 1063 | case MADV_NOHUGEPAGE: |
f8af4da3 | 1064 | #endif |
accb61fe JB |
1065 | case MADV_DONTDUMP: |
1066 | case MADV_DODUMP: | |
d2cd9ede RR |
1067 | case MADV_WIPEONFORK: |
1068 | case MADV_KEEPONFORK: | |
5e451be7 AK |
1069 | #ifdef CONFIG_MEMORY_FAILURE |
1070 | case MADV_SOFT_OFFLINE: | |
1071 | case MADV_HWPOISON: | |
1072 | #endif | |
1ecef9ed | 1073 | return true; |
75927af8 NP |
1074 | |
1075 | default: | |
1ecef9ed | 1076 | return false; |
75927af8 NP |
1077 | } |
1078 | } | |
3866ea90 | 1079 | |
1da177e4 LT |
1080 | /* |
1081 | * The madvise(2) system call. | |
1082 | * | |
1083 | * Applications can use madvise() to advise the kernel how it should | |
1084 | * handle paging I/O in this VM area. The idea is to help the kernel | |
1085 | * use appropriate read-ahead and caching techniques. The information | |
1086 | * provided is advisory only, and can be safely disregarded by the | |
1087 | * kernel without affecting the correct operation of the application. | |
1088 | * | |
1089 | * behavior values: | |
1090 | * MADV_NORMAL - the default behavior is to read clusters. This | |
1091 | * results in some read-ahead and read-behind. | |
1092 | * MADV_RANDOM - the system should read the minimum amount of data | |
1093 | * on any access, since it is unlikely that the appli- | |
1094 | * cation will need more than what it asks for. | |
1095 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
1096 | * once, so they can be aggressively read ahead, and | |
1097 | * can be freed soon after they are accessed. | |
1098 | * MADV_WILLNEED - the application is notifying the system to read | |
1099 | * some pages ahead. | |
1100 | * MADV_DONTNEED - the application is finished with the given range, | |
1101 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
1102 | * MADV_FREE - the application marks pages in the given range as lazy free, |
1103 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
1104 | * MADV_REMOVE - the application wants to free up the given range of |
1105 | * pages and associated backing store. | |
3866ea90 HD |
1106 | * MADV_DONTFORK - omit this area from child's address space when forking: |
1107 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
1108 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
c02c3009 YS |
1109 | * MADV_WIPEONFORK - present the child process with zero-filled memory in this |
1110 | * range after a fork. | |
1111 | * MADV_KEEPONFORK - undo the effect of MADV_WIPEONFORK | |
d7206a70 NH |
1112 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
1113 | * were corrupted by unrecoverable hardware memory failure. | |
1114 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
1115 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
1116 | * this area with pages of identical content from other such areas. | |
1117 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
1118 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
1119 | * huge pages in the future. Existing pages might be coalesced and | |
1120 | * new pages might be allocated as THP. | |
1121 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
1122 | * transparent huge pages so the existing pages will not be | |
1123 | * coalesced into THP and new pages will not be allocated as THP. | |
1124 | * MADV_DONTDUMP - the application wants to prevent pages in the given range | |
1125 | * from being included in its core dump. | |
1126 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
1da177e4 LT |
1127 | * |
1128 | * return values: | |
1129 | * zero - success | |
1130 | * -EINVAL - start + len < 0, start is not page-aligned, | |
1131 | * "behavior" is not a valid value, or application | |
c02c3009 YS |
1132 | * is attempting to release locked or shared pages, |
1133 | * or the specified address range includes file, Huge TLB, | |
1134 | * MAP_SHARED or VMPFNMAP range. | |
1da177e4 LT |
1135 | * -ENOMEM - addresses in the specified range are not currently |
1136 | * mapped, or are outside the AS of the process. | |
1137 | * -EIO - an I/O error occurred while paging in data. | |
1138 | * -EBADF - map exists, but area maps something that isn't a file. | |
1139 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
1140 | */ | |
3480b257 | 1141 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) |
1da177e4 | 1142 | { |
05b74384 | 1143 | unsigned long end, tmp; |
ec9bed9d | 1144 | struct vm_area_struct *vma, *prev; |
1da177e4 LT |
1145 | int unmapped_error = 0; |
1146 | int error = -EINVAL; | |
f7977793 | 1147 | int write; |
1da177e4 | 1148 | size_t len; |
1998cc04 | 1149 | struct blk_plug plug; |
1da177e4 | 1150 | |
057d3389 AK |
1151 | start = untagged_addr(start); |
1152 | ||
75927af8 NP |
1153 | if (!madvise_behavior_valid(behavior)) |
1154 | return error; | |
1155 | ||
1da177e4 | 1156 | if (start & ~PAGE_MASK) |
84d96d89 | 1157 | return error; |
1da177e4 LT |
1158 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; |
1159 | ||
1160 | /* Check to see whether len was rounded up from small -ve to zero */ | |
1161 | if (len_in && !len) | |
84d96d89 | 1162 | return error; |
1da177e4 LT |
1163 | |
1164 | end = start + len; | |
1165 | if (end < start) | |
84d96d89 | 1166 | return error; |
1da177e4 LT |
1167 | |
1168 | error = 0; | |
1169 | if (end == start) | |
84d96d89 RV |
1170 | return error; |
1171 | ||
5e451be7 AK |
1172 | #ifdef CONFIG_MEMORY_FAILURE |
1173 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) | |
1174 | return madvise_inject_error(behavior, start, start + len_in); | |
1175 | #endif | |
1176 | ||
84d96d89 | 1177 | write = madvise_need_mmap_write(behavior); |
dc0ef0df MH |
1178 | if (write) { |
1179 | if (down_write_killable(¤t->mm->mmap_sem)) | |
1180 | return -EINTR; | |
1181 | } else { | |
84d96d89 | 1182 | down_read(¤t->mm->mmap_sem); |
dc0ef0df | 1183 | } |
1da177e4 LT |
1184 | |
1185 | /* | |
1186 | * If the interval [start,end) covers some unmapped address | |
1187 | * ranges, just ignore them, but return -ENOMEM at the end. | |
05b74384 | 1188 | * - different from the way of handling in mlock etc. |
1da177e4 | 1189 | */ |
05b74384 | 1190 | vma = find_vma_prev(current->mm, start, &prev); |
836d5ffd HD |
1191 | if (vma && start > vma->vm_start) |
1192 | prev = vma; | |
1193 | ||
1998cc04 | 1194 | blk_start_plug(&plug); |
1da177e4 LT |
1195 | for (;;) { |
1196 | /* Still start < end. */ | |
1197 | error = -ENOMEM; | |
1198 | if (!vma) | |
84d96d89 | 1199 | goto out; |
1da177e4 | 1200 | |
05b74384 | 1201 | /* Here start < (end|vma->vm_end). */ |
1da177e4 LT |
1202 | if (start < vma->vm_start) { |
1203 | unmapped_error = -ENOMEM; | |
1204 | start = vma->vm_start; | |
05b74384 | 1205 | if (start >= end) |
84d96d89 | 1206 | goto out; |
1da177e4 LT |
1207 | } |
1208 | ||
05b74384 PM |
1209 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ |
1210 | tmp = vma->vm_end; | |
1211 | if (end < tmp) | |
1212 | tmp = end; | |
1da177e4 | 1213 | |
05b74384 PM |
1214 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ |
1215 | error = madvise_vma(vma, &prev, start, tmp, behavior); | |
1da177e4 | 1216 | if (error) |
84d96d89 | 1217 | goto out; |
05b74384 | 1218 | start = tmp; |
90ed52eb | 1219 | if (prev && start < prev->vm_end) |
05b74384 PM |
1220 | start = prev->vm_end; |
1221 | error = unmapped_error; | |
1222 | if (start >= end) | |
84d96d89 | 1223 | goto out; |
90ed52eb HD |
1224 | if (prev) |
1225 | vma = prev->vm_next; | |
1226 | else /* madvise_remove dropped mmap_sem */ | |
1227 | vma = find_vma(current->mm, start); | |
1da177e4 | 1228 | } |
1da177e4 | 1229 | out: |
84d96d89 | 1230 | blk_finish_plug(&plug); |
f7977793 | 1231 | if (write) |
0a27a14a NP |
1232 | up_write(¤t->mm->mmap_sem); |
1233 | else | |
1234 | up_read(¤t->mm->mmap_sem); | |
1235 | ||
1da177e4 LT |
1236 | return error; |
1237 | } |