]>
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> |
05ce7724 | 14 | #include <linux/userfaultfd_k.h> |
1da177e4 | 15 | #include <linux/hugetlb.h> |
3f31d075 | 16 | #include <linux/falloc.h> |
fee5077e | 17 | #include <linux/fadvise.h> |
e8edc6e0 | 18 | #include <linux/sched.h> |
f8af4da3 | 19 | #include <linux/ksm.h> |
3f31d075 | 20 | #include <linux/fs.h> |
9ab4233d | 21 | #include <linux/file.h> |
1998cc04 | 22 | #include <linux/blkdev.h> |
66114cad | 23 | #include <linux/backing-dev.h> |
1998cc04 SL |
24 | #include <linux/swap.h> |
25 | #include <linux/swapops.h> | |
3a4f8a0b | 26 | #include <linux/shmem_fs.h> |
854e9ed0 MK |
27 | #include <linux/mmu_notifier.h> |
28 | ||
29 | #include <asm/tlb.h> | |
1da177e4 | 30 | |
23519073 KS |
31 | #include "internal.h" |
32 | ||
0a27a14a NP |
33 | /* |
34 | * Any behaviour which results in changes to the vma->vm_flags needs to | |
35 | * take mmap_sem for writing. Others, which simply traverse vmas, need | |
36 | * to only take it for reading. | |
37 | */ | |
38 | static int madvise_need_mmap_write(int behavior) | |
39 | { | |
40 | switch (behavior) { | |
41 | case MADV_REMOVE: | |
42 | case MADV_WILLNEED: | |
43 | case MADV_DONTNEED: | |
854e9ed0 | 44 | case MADV_FREE: |
0a27a14a NP |
45 | return 0; |
46 | default: | |
47 | /* be safe, default to 1. list exceptions explicitly */ | |
48 | return 1; | |
49 | } | |
50 | } | |
51 | ||
1da177e4 LT |
52 | /* |
53 | * We can potentially split a vm area into separate | |
54 | * areas, each area with its own behavior. | |
55 | */ | |
ec9bed9d | 56 | static long madvise_behavior(struct vm_area_struct *vma, |
05b74384 PM |
57 | struct vm_area_struct **prev, |
58 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 59 | { |
ec9bed9d | 60 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 61 | int error = 0; |
05b74384 | 62 | pgoff_t pgoff; |
3866ea90 | 63 | unsigned long new_flags = vma->vm_flags; |
e798c6e8 PM |
64 | |
65 | switch (behavior) { | |
f8225661 MT |
66 | case MADV_NORMAL: |
67 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
68 | break; | |
e798c6e8 | 69 | case MADV_SEQUENTIAL: |
f8225661 | 70 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; |
e798c6e8 PM |
71 | break; |
72 | case MADV_RANDOM: | |
f8225661 | 73 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; |
e798c6e8 | 74 | break; |
f8225661 MT |
75 | case MADV_DONTFORK: |
76 | new_flags |= VM_DONTCOPY; | |
77 | break; | |
78 | case MADV_DOFORK: | |
3866ea90 HD |
79 | if (vma->vm_flags & VM_IO) { |
80 | error = -EINVAL; | |
81 | goto out; | |
82 | } | |
f8225661 | 83 | new_flags &= ~VM_DONTCOPY; |
e798c6e8 | 84 | break; |
d2cd9ede RR |
85 | case MADV_WIPEONFORK: |
86 | /* MADV_WIPEONFORK is only supported on anonymous memory. */ | |
87 | if (vma->vm_file || vma->vm_flags & VM_SHARED) { | |
88 | error = -EINVAL; | |
89 | goto out; | |
90 | } | |
91 | new_flags |= VM_WIPEONFORK; | |
92 | break; | |
93 | case MADV_KEEPONFORK: | |
94 | new_flags &= ~VM_WIPEONFORK; | |
95 | break; | |
accb61fe | 96 | case MADV_DONTDUMP: |
0103bd16 | 97 | new_flags |= VM_DONTDUMP; |
accb61fe JB |
98 | break; |
99 | case MADV_DODUMP: | |
d41aa525 | 100 | if (!is_vm_hugetlb_page(vma) && new_flags & VM_SPECIAL) { |
0103bd16 KK |
101 | error = -EINVAL; |
102 | goto out; | |
103 | } | |
104 | new_flags &= ~VM_DONTDUMP; | |
accb61fe | 105 | break; |
f8af4da3 HD |
106 | case MADV_MERGEABLE: |
107 | case MADV_UNMERGEABLE: | |
108 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
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; | |
f8af4da3 | 116 | goto out; |
def5efe0 | 117 | } |
f8af4da3 | 118 | break; |
0af4e98b | 119 | case MADV_HUGEPAGE: |
a664b2d8 | 120 | case MADV_NOHUGEPAGE: |
60ab3244 | 121 | error = hugepage_madvise(vma, &new_flags, behavior); |
def5efe0 DR |
122 | if (error) { |
123 | /* | |
124 | * madvise() returns EAGAIN if kernel resources, such as | |
125 | * slab, are temporarily unavailable. | |
126 | */ | |
127 | if (error == -ENOMEM) | |
128 | error = -EAGAIN; | |
0af4e98b | 129 | goto out; |
def5efe0 | 130 | } |
0af4e98b | 131 | break; |
e798c6e8 PM |
132 | } |
133 | ||
05b74384 PM |
134 | if (new_flags == vma->vm_flags) { |
135 | *prev = vma; | |
836d5ffd | 136 | goto out; |
05b74384 PM |
137 | } |
138 | ||
139 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
140 | *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma, | |
19a809af AA |
141 | vma->vm_file, pgoff, vma_policy(vma), |
142 | vma->vm_userfaultfd_ctx); | |
05b74384 PM |
143 | if (*prev) { |
144 | vma = *prev; | |
145 | goto success; | |
146 | } | |
147 | ||
148 | *prev = vma; | |
1da177e4 LT |
149 | |
150 | if (start != vma->vm_start) { | |
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, start, 1); | |
156 | if (error) { | |
157 | /* | |
158 | * madvise() returns EAGAIN if kernel resources, such as | |
159 | * slab, are temporarily unavailable. | |
160 | */ | |
161 | if (error == -ENOMEM) | |
162 | error = -EAGAIN; | |
163 | goto out; | |
164 | } | |
1da177e4 LT |
165 | } |
166 | ||
167 | if (end != vma->vm_end) { | |
def5efe0 DR |
168 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
169 | error = -ENOMEM; | |
1da177e4 | 170 | goto out; |
def5efe0 DR |
171 | } |
172 | error = __split_vma(mm, vma, end, 0); | |
173 | if (error) { | |
174 | /* | |
175 | * madvise() returns EAGAIN if kernel resources, such as | |
176 | * slab, are temporarily unavailable. | |
177 | */ | |
178 | if (error == -ENOMEM) | |
179 | error = -EAGAIN; | |
180 | goto out; | |
181 | } | |
1da177e4 LT |
182 | } |
183 | ||
836d5ffd | 184 | success: |
1da177e4 LT |
185 | /* |
186 | * vm_flags is protected by the mmap_sem held in write mode. | |
187 | */ | |
e798c6e8 | 188 | vma->vm_flags = new_flags; |
1da177e4 | 189 | out: |
1da177e4 LT |
190 | return error; |
191 | } | |
192 | ||
1998cc04 SL |
193 | #ifdef CONFIG_SWAP |
194 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
195 | unsigned long end, struct mm_walk *walk) | |
196 | { | |
197 | pte_t *orig_pte; | |
198 | struct vm_area_struct *vma = walk->private; | |
199 | unsigned long index; | |
200 | ||
201 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
202 | return 0; | |
203 | ||
204 | for (index = start; index != end; index += PAGE_SIZE) { | |
205 | pte_t pte; | |
206 | swp_entry_t entry; | |
207 | struct page *page; | |
208 | spinlock_t *ptl; | |
209 | ||
210 | orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); | |
211 | pte = *(orig_pte + ((index - start) / PAGE_SIZE)); | |
212 | pte_unmap_unlock(orig_pte, ptl); | |
213 | ||
0661a336 | 214 | if (pte_present(pte) || pte_none(pte)) |
1998cc04 SL |
215 | continue; |
216 | entry = pte_to_swp_entry(pte); | |
217 | if (unlikely(non_swap_entry(entry))) | |
218 | continue; | |
219 | ||
220 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, | |
23955622 | 221 | vma, index, false); |
1998cc04 | 222 | if (page) |
09cbfeaf | 223 | put_page(page); |
1998cc04 SL |
224 | } |
225 | ||
226 | return 0; | |
227 | } | |
228 | ||
229 | static void force_swapin_readahead(struct vm_area_struct *vma, | |
230 | unsigned long start, unsigned long end) | |
231 | { | |
232 | struct mm_walk walk = { | |
233 | .mm = vma->vm_mm, | |
234 | .pmd_entry = swapin_walk_pmd_entry, | |
235 | .private = vma, | |
236 | }; | |
237 | ||
238 | walk_page_range(start, end, &walk); | |
239 | ||
240 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
241 | } | |
242 | ||
243 | static void force_shm_swapin_readahead(struct vm_area_struct *vma, | |
244 | unsigned long start, unsigned long end, | |
245 | struct address_space *mapping) | |
246 | { | |
247 | pgoff_t index; | |
248 | struct page *page; | |
249 | swp_entry_t swap; | |
250 | ||
251 | for (; start < end; start += PAGE_SIZE) { | |
252 | index = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
253 | ||
55231e5c | 254 | page = find_get_entry(mapping, index); |
3159f943 | 255 | if (!xa_is_value(page)) { |
1998cc04 | 256 | if (page) |
09cbfeaf | 257 | put_page(page); |
1998cc04 SL |
258 | continue; |
259 | } | |
260 | swap = radix_to_swp_entry(page); | |
261 | page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE, | |
23955622 | 262 | NULL, 0, false); |
1998cc04 | 263 | if (page) |
09cbfeaf | 264 | put_page(page); |
1998cc04 SL |
265 | } |
266 | ||
267 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
268 | } | |
269 | #endif /* CONFIG_SWAP */ | |
270 | ||
1da177e4 LT |
271 | /* |
272 | * Schedule all required I/O operations. Do not wait for completion. | |
273 | */ | |
ec9bed9d VC |
274 | static long madvise_willneed(struct vm_area_struct *vma, |
275 | struct vm_area_struct **prev, | |
1da177e4 LT |
276 | unsigned long start, unsigned long end) |
277 | { | |
278 | struct file *file = vma->vm_file; | |
fee5077e | 279 | loff_t offset; |
1da177e4 | 280 | |
6ea8d958 | 281 | *prev = vma; |
1998cc04 | 282 | #ifdef CONFIG_SWAP |
97b713ba | 283 | if (!file) { |
97b713ba | 284 | force_swapin_readahead(vma, start, end); |
1998cc04 SL |
285 | return 0; |
286 | } | |
1998cc04 | 287 | |
97b713ba | 288 | if (shmem_mapping(file->f_mapping)) { |
97b713ba CH |
289 | force_shm_swapin_readahead(vma, start, end, |
290 | file->f_mapping); | |
291 | return 0; | |
292 | } | |
293 | #else | |
1bef4003 S |
294 | if (!file) |
295 | return -EBADF; | |
97b713ba | 296 | #endif |
1bef4003 | 297 | |
e748dcd0 | 298 | if (IS_DAX(file_inode(file))) { |
fe77ba6f CO |
299 | /* no bad return value, but ignore advice */ |
300 | return 0; | |
301 | } | |
302 | ||
fee5077e JK |
303 | /* |
304 | * Filesystem's fadvise may need to take various locks. We need to | |
305 | * explicitly grab a reference because the vma (and hence the | |
306 | * vma's reference to the file) can go away as soon as we drop | |
307 | * mmap_sem. | |
308 | */ | |
309 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ | |
310 | get_file(file); | |
311 | up_read(¤t->mm->mmap_sem); | |
312 | offset = (loff_t)(start - vma->vm_start) | |
313 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
314 | vfs_fadvise(file, offset, end - start, POSIX_FADV_WILLNEED); | |
315 | fput(file); | |
316 | down_read(¤t->mm->mmap_sem); | |
1da177e4 LT |
317 | return 0; |
318 | } | |
319 | ||
854e9ed0 MK |
320 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
321 | unsigned long end, struct mm_walk *walk) | |
322 | ||
323 | { | |
324 | struct mmu_gather *tlb = walk->private; | |
325 | struct mm_struct *mm = tlb->mm; | |
326 | struct vm_area_struct *vma = walk->vma; | |
327 | spinlock_t *ptl; | |
328 | pte_t *orig_pte, *pte, ptent; | |
329 | struct page *page; | |
64b42bc1 | 330 | int nr_swap = 0; |
b8d3c4c3 MK |
331 | unsigned long next; |
332 | ||
333 | next = pmd_addr_end(addr, end); | |
334 | if (pmd_trans_huge(*pmd)) | |
335 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
336 | goto next; | |
854e9ed0 | 337 | |
854e9ed0 MK |
338 | if (pmd_trans_unstable(pmd)) |
339 | return 0; | |
340 | ||
ed6a7935 | 341 | tlb_change_page_size(tlb, PAGE_SIZE); |
854e9ed0 | 342 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
3ea27719 | 343 | flush_tlb_batched_pending(mm); |
854e9ed0 MK |
344 | arch_enter_lazy_mmu_mode(); |
345 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
346 | ptent = *pte; | |
347 | ||
64b42bc1 | 348 | if (pte_none(ptent)) |
854e9ed0 | 349 | continue; |
64b42bc1 MK |
350 | /* |
351 | * If the pte has swp_entry, just clear page table to | |
352 | * prevent swap-in which is more expensive rather than | |
353 | * (page allocation + zeroing). | |
354 | */ | |
355 | if (!pte_present(ptent)) { | |
356 | swp_entry_t entry; | |
357 | ||
358 | entry = pte_to_swp_entry(ptent); | |
359 | if (non_swap_entry(entry)) | |
360 | continue; | |
361 | nr_swap--; | |
362 | free_swap_and_cache(entry); | |
363 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
364 | continue; | |
365 | } | |
854e9ed0 | 366 | |
25b2995a | 367 | page = vm_normal_page(vma, addr, ptent); |
854e9ed0 MK |
368 | if (!page) |
369 | continue; | |
370 | ||
371 | /* | |
372 | * If pmd isn't transhuge but the page is THP and | |
373 | * is owned by only this process, split it and | |
374 | * deactivate all pages. | |
375 | */ | |
376 | if (PageTransCompound(page)) { | |
377 | if (page_mapcount(page) != 1) | |
378 | goto out; | |
379 | get_page(page); | |
380 | if (!trylock_page(page)) { | |
381 | put_page(page); | |
382 | goto out; | |
383 | } | |
384 | pte_unmap_unlock(orig_pte, ptl); | |
385 | if (split_huge_page(page)) { | |
386 | unlock_page(page); | |
387 | put_page(page); | |
388 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
389 | goto out; | |
390 | } | |
854e9ed0 | 391 | unlock_page(page); |
263630e8 | 392 | put_page(page); |
854e9ed0 MK |
393 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
394 | pte--; | |
395 | addr -= PAGE_SIZE; | |
396 | continue; | |
397 | } | |
398 | ||
399 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
400 | ||
401 | if (PageSwapCache(page) || PageDirty(page)) { | |
402 | if (!trylock_page(page)) | |
403 | continue; | |
404 | /* | |
405 | * If page is shared with others, we couldn't clear | |
406 | * PG_dirty of the page. | |
407 | */ | |
408 | if (page_mapcount(page) != 1) { | |
409 | unlock_page(page); | |
410 | continue; | |
411 | } | |
412 | ||
413 | if (PageSwapCache(page) && !try_to_free_swap(page)) { | |
414 | unlock_page(page); | |
415 | continue; | |
416 | } | |
417 | ||
418 | ClearPageDirty(page); | |
419 | unlock_page(page); | |
420 | } | |
421 | ||
422 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
423 | /* | |
424 | * Some of architecture(ex, PPC) don't update TLB | |
425 | * with set_pte_at and tlb_remove_tlb_entry so for | |
426 | * the portability, remap the pte with old|clean | |
427 | * after pte clearing. | |
428 | */ | |
429 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
430 | tlb->fullmm); | |
431 | ||
432 | ptent = pte_mkold(ptent); | |
433 | ptent = pte_mkclean(ptent); | |
434 | set_pte_at(mm, addr, pte, ptent); | |
435 | tlb_remove_tlb_entry(tlb, pte, addr); | |
436 | } | |
802a3a92 | 437 | mark_page_lazyfree(page); |
854e9ed0 MK |
438 | } |
439 | out: | |
64b42bc1 MK |
440 | if (nr_swap) { |
441 | if (current->mm == mm) | |
442 | sync_mm_rss(mm); | |
443 | ||
444 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); | |
445 | } | |
854e9ed0 MK |
446 | arch_leave_lazy_mmu_mode(); |
447 | pte_unmap_unlock(orig_pte, ptl); | |
448 | cond_resched(); | |
b8d3c4c3 | 449 | next: |
854e9ed0 MK |
450 | return 0; |
451 | } | |
452 | ||
453 | static void madvise_free_page_range(struct mmu_gather *tlb, | |
454 | struct vm_area_struct *vma, | |
455 | unsigned long addr, unsigned long end) | |
456 | { | |
457 | struct mm_walk free_walk = { | |
458 | .pmd_entry = madvise_free_pte_range, | |
459 | .mm = vma->vm_mm, | |
460 | .private = tlb, | |
461 | }; | |
462 | ||
463 | tlb_start_vma(tlb, vma); | |
464 | walk_page_range(addr, end, &free_walk); | |
465 | tlb_end_vma(tlb, vma); | |
466 | } | |
467 | ||
468 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
469 | unsigned long start_addr, unsigned long end_addr) | |
470 | { | |
854e9ed0 | 471 | struct mm_struct *mm = vma->vm_mm; |
ac46d4f3 | 472 | struct mmu_notifier_range range; |
854e9ed0 MK |
473 | struct mmu_gather tlb; |
474 | ||
854e9ed0 MK |
475 | /* MADV_FREE works for only anon vma at the moment */ |
476 | if (!vma_is_anonymous(vma)) | |
477 | return -EINVAL; | |
478 | ||
ac46d4f3 JG |
479 | range.start = max(vma->vm_start, start_addr); |
480 | if (range.start >= vma->vm_end) | |
854e9ed0 | 481 | return -EINVAL; |
ac46d4f3 JG |
482 | range.end = min(vma->vm_end, end_addr); |
483 | if (range.end <= vma->vm_start) | |
854e9ed0 | 484 | return -EINVAL; |
7269f999 | 485 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, |
6f4f13e8 | 486 | range.start, range.end); |
854e9ed0 MK |
487 | |
488 | lru_add_drain(); | |
ac46d4f3 | 489 | tlb_gather_mmu(&tlb, mm, range.start, range.end); |
854e9ed0 MK |
490 | update_hiwater_rss(mm); |
491 | ||
ac46d4f3 JG |
492 | mmu_notifier_invalidate_range_start(&range); |
493 | madvise_free_page_range(&tlb, vma, range.start, range.end); | |
494 | mmu_notifier_invalidate_range_end(&range); | |
495 | tlb_finish_mmu(&tlb, range.start, range.end); | |
854e9ed0 MK |
496 | |
497 | return 0; | |
498 | } | |
499 | ||
1da177e4 LT |
500 | /* |
501 | * Application no longer needs these pages. If the pages are dirty, | |
502 | * it's OK to just throw them away. The app will be more careful about | |
503 | * data it wants to keep. Be sure to free swap resources too. The | |
7e6cbea3 | 504 | * zap_page_range call sets things up for shrink_active_list to actually free |
1da177e4 LT |
505 | * these pages later if no one else has touched them in the meantime, |
506 | * although we could add these pages to a global reuse list for | |
7e6cbea3 | 507 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
508 | * |
509 | * NB: This interface discards data rather than pushes it out to swap, | |
510 | * as some implementations do. This has performance implications for | |
511 | * applications like large transactional databases which want to discard | |
512 | * pages in anonymous maps after committing to backing store the data | |
513 | * that was kept in them. There is no reason to write this data out to | |
514 | * the swap area if the application is discarding it. | |
515 | * | |
516 | * An interface that causes the system to free clean pages and flush | |
517 | * dirty pages is already available as msync(MS_INVALIDATE). | |
518 | */ | |
230ca982 MR |
519 | static long madvise_dontneed_single_vma(struct vm_area_struct *vma, |
520 | unsigned long start, unsigned long end) | |
521 | { | |
522 | zap_page_range(vma, start, end - start); | |
523 | return 0; | |
524 | } | |
525 | ||
526 | static long madvise_dontneed_free(struct vm_area_struct *vma, | |
527 | struct vm_area_struct **prev, | |
528 | unsigned long start, unsigned long end, | |
529 | int behavior) | |
1da177e4 | 530 | { |
05b74384 | 531 | *prev = vma; |
23519073 | 532 | if (!can_madv_dontneed_vma(vma)) |
1da177e4 LT |
533 | return -EINVAL; |
534 | ||
70ccb92f AA |
535 | if (!userfaultfd_remove(vma, start, end)) { |
536 | *prev = NULL; /* mmap_sem has been dropped, prev is stale */ | |
537 | ||
538 | down_read(¤t->mm->mmap_sem); | |
539 | vma = find_vma(current->mm, start); | |
540 | if (!vma) | |
541 | return -ENOMEM; | |
542 | if (start < vma->vm_start) { | |
543 | /* | |
544 | * This "vma" under revalidation is the one | |
545 | * with the lowest vma->vm_start where start | |
546 | * is also < vma->vm_end. If start < | |
547 | * vma->vm_start it means an hole materialized | |
548 | * in the user address space within the | |
230ca982 MR |
549 | * virtual range passed to MADV_DONTNEED |
550 | * or MADV_FREE. | |
70ccb92f AA |
551 | */ |
552 | return -ENOMEM; | |
553 | } | |
554 | if (!can_madv_dontneed_vma(vma)) | |
555 | return -EINVAL; | |
556 | if (end > vma->vm_end) { | |
557 | /* | |
558 | * Don't fail if end > vma->vm_end. If the old | |
559 | * vma was splitted while the mmap_sem was | |
560 | * released the effect of the concurrent | |
230ca982 | 561 | * operation may not cause madvise() to |
70ccb92f AA |
562 | * have an undefined result. There may be an |
563 | * adjacent next vma that we'll walk | |
564 | * next. userfaultfd_remove() will generate an | |
565 | * UFFD_EVENT_REMOVE repetition on the | |
566 | * end-vma->vm_end range, but the manager can | |
567 | * handle a repetition fine. | |
568 | */ | |
569 | end = vma->vm_end; | |
570 | } | |
571 | VM_WARN_ON(start >= end); | |
572 | } | |
230ca982 MR |
573 | |
574 | if (behavior == MADV_DONTNEED) | |
575 | return madvise_dontneed_single_vma(vma, start, end); | |
576 | else if (behavior == MADV_FREE) | |
577 | return madvise_free_single_vma(vma, start, end); | |
578 | else | |
579 | return -EINVAL; | |
1da177e4 LT |
580 | } |
581 | ||
f6b3ec23 BP |
582 | /* |
583 | * Application wants to free up the pages and associated backing store. | |
584 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
585 | */ |
586 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 587 | struct vm_area_struct **prev, |
f6b3ec23 BP |
588 | unsigned long start, unsigned long end) |
589 | { | |
3f31d075 | 590 | loff_t offset; |
90ed52eb | 591 | int error; |
9ab4233d | 592 | struct file *f; |
f6b3ec23 | 593 | |
90ed52eb | 594 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ |
00e9fa2d | 595 | |
72079ba0 | 596 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
597 | return -EINVAL; |
598 | ||
9ab4233d AL |
599 | f = vma->vm_file; |
600 | ||
601 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
602 | return -EINVAL; |
603 | } | |
604 | ||
69cf0fac HD |
605 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
606 | return -EACCES; | |
607 | ||
f6b3ec23 BP |
608 | offset = (loff_t)(start - vma->vm_start) |
609 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 610 | |
9ab4233d AL |
611 | /* |
612 | * Filesystem's fallocate may need to take i_mutex. We need to | |
613 | * explicitly grab a reference because the vma (and hence the | |
614 | * vma's reference to the file) can go away as soon as we drop | |
615 | * mmap_sem. | |
616 | */ | |
617 | get_file(f); | |
70ccb92f AA |
618 | if (userfaultfd_remove(vma, start, end)) { |
619 | /* mmap_sem was not released by userfaultfd_remove() */ | |
620 | up_read(¤t->mm->mmap_sem); | |
621 | } | |
72c72bdf | 622 | error = vfs_fallocate(f, |
3f31d075 HD |
623 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
624 | offset, end - start); | |
9ab4233d | 625 | fput(f); |
0a27a14a | 626 | down_read(¤t->mm->mmap_sem); |
90ed52eb | 627 | return error; |
f6b3ec23 BP |
628 | } |
629 | ||
9893e49d AK |
630 | #ifdef CONFIG_MEMORY_FAILURE |
631 | /* | |
632 | * Error injection support for memory error handling. | |
633 | */ | |
97167a76 AK |
634 | static int madvise_inject_error(int behavior, |
635 | unsigned long start, unsigned long end) | |
9893e49d | 636 | { |
97167a76 | 637 | struct page *page; |
c461ad6a | 638 | struct zone *zone; |
19bfbe22 | 639 | unsigned int order; |
97167a76 | 640 | |
9893e49d AK |
641 | if (!capable(CAP_SYS_ADMIN)) |
642 | return -EPERM; | |
97167a76 | 643 | |
19bfbe22 AM |
644 | |
645 | for (; start < end; start += PAGE_SIZE << order) { | |
23e7b5c2 | 646 | unsigned long pfn; |
325c4ef5 AM |
647 | int ret; |
648 | ||
97167a76 | 649 | ret = get_user_pages_fast(start, 1, 0, &page); |
9893e49d AK |
650 | if (ret != 1) |
651 | return ret; | |
23e7b5c2 | 652 | pfn = page_to_pfn(page); |
325c4ef5 | 653 | |
19bfbe22 AM |
654 | /* |
655 | * When soft offlining hugepages, after migrating the page | |
656 | * we dissolve it, therefore in the second loop "page" will | |
657 | * no longer be a compound page, and order will be 0. | |
658 | */ | |
659 | order = compound_order(compound_head(page)); | |
660 | ||
97167a76 AK |
661 | if (PageHWPoison(page)) { |
662 | put_page(page); | |
29b4eede WL |
663 | continue; |
664 | } | |
97167a76 AK |
665 | |
666 | if (behavior == MADV_SOFT_OFFLINE) { | |
667 | pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n", | |
23e7b5c2 | 668 | pfn, start); |
97167a76 AK |
669 | |
670 | ret = soft_offline_page(page, MF_COUNT_INCREASED); | |
afcf938e | 671 | if (ret) |
8302423b | 672 | return ret; |
afcf938e AK |
673 | continue; |
674 | } | |
23e7b5c2 | 675 | |
97167a76 | 676 | pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n", |
23e7b5c2 | 677 | pfn, start); |
97167a76 | 678 | |
23e7b5c2 DW |
679 | /* |
680 | * Drop the page reference taken by get_user_pages_fast(). In | |
681 | * the absence of MF_COUNT_INCREASED the memory_failure() | |
682 | * routine is responsible for pinning the page to prevent it | |
683 | * from being released back to the page allocator. | |
684 | */ | |
685 | put_page(page); | |
686 | ret = memory_failure(pfn, 0); | |
23a003bf NH |
687 | if (ret) |
688 | return ret; | |
9893e49d | 689 | } |
c461ad6a MG |
690 | |
691 | /* Ensure that all poisoned pages are removed from per-cpu lists */ | |
692 | for_each_populated_zone(zone) | |
693 | drain_all_pages(zone); | |
694 | ||
325c4ef5 | 695 | return 0; |
9893e49d AK |
696 | } |
697 | #endif | |
698 | ||
165cd402 | 699 | static long |
700 | madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, | |
701 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 702 | { |
1da177e4 | 703 | switch (behavior) { |
f6b3ec23 | 704 | case MADV_REMOVE: |
3866ea90 | 705 | return madvise_remove(vma, prev, start, end); |
1da177e4 | 706 | case MADV_WILLNEED: |
3866ea90 | 707 | return madvise_willneed(vma, prev, start, end); |
854e9ed0 | 708 | case MADV_FREE: |
1da177e4 | 709 | case MADV_DONTNEED: |
230ca982 | 710 | return madvise_dontneed_free(vma, prev, start, end, behavior); |
1da177e4 | 711 | default: |
3866ea90 | 712 | return madvise_behavior(vma, prev, start, end, behavior); |
1da177e4 | 713 | } |
1da177e4 LT |
714 | } |
715 | ||
1ecef9ed | 716 | static bool |
75927af8 NP |
717 | madvise_behavior_valid(int behavior) |
718 | { | |
719 | switch (behavior) { | |
720 | case MADV_DOFORK: | |
721 | case MADV_DONTFORK: | |
722 | case MADV_NORMAL: | |
723 | case MADV_SEQUENTIAL: | |
724 | case MADV_RANDOM: | |
725 | case MADV_REMOVE: | |
726 | case MADV_WILLNEED: | |
727 | case MADV_DONTNEED: | |
854e9ed0 | 728 | case MADV_FREE: |
f8af4da3 HD |
729 | #ifdef CONFIG_KSM |
730 | case MADV_MERGEABLE: | |
731 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
732 | #endif |
733 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
734 | case MADV_HUGEPAGE: | |
a664b2d8 | 735 | case MADV_NOHUGEPAGE: |
f8af4da3 | 736 | #endif |
accb61fe JB |
737 | case MADV_DONTDUMP: |
738 | case MADV_DODUMP: | |
d2cd9ede RR |
739 | case MADV_WIPEONFORK: |
740 | case MADV_KEEPONFORK: | |
5e451be7 AK |
741 | #ifdef CONFIG_MEMORY_FAILURE |
742 | case MADV_SOFT_OFFLINE: | |
743 | case MADV_HWPOISON: | |
744 | #endif | |
1ecef9ed | 745 | return true; |
75927af8 NP |
746 | |
747 | default: | |
1ecef9ed | 748 | return false; |
75927af8 NP |
749 | } |
750 | } | |
3866ea90 | 751 | |
1da177e4 LT |
752 | /* |
753 | * The madvise(2) system call. | |
754 | * | |
755 | * Applications can use madvise() to advise the kernel how it should | |
756 | * handle paging I/O in this VM area. The idea is to help the kernel | |
757 | * use appropriate read-ahead and caching techniques. The information | |
758 | * provided is advisory only, and can be safely disregarded by the | |
759 | * kernel without affecting the correct operation of the application. | |
760 | * | |
761 | * behavior values: | |
762 | * MADV_NORMAL - the default behavior is to read clusters. This | |
763 | * results in some read-ahead and read-behind. | |
764 | * MADV_RANDOM - the system should read the minimum amount of data | |
765 | * on any access, since it is unlikely that the appli- | |
766 | * cation will need more than what it asks for. | |
767 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
768 | * once, so they can be aggressively read ahead, and | |
769 | * can be freed soon after they are accessed. | |
770 | * MADV_WILLNEED - the application is notifying the system to read | |
771 | * some pages ahead. | |
772 | * MADV_DONTNEED - the application is finished with the given range, | |
773 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
774 | * MADV_FREE - the application marks pages in the given range as lazy free, |
775 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
776 | * MADV_REMOVE - the application wants to free up the given range of |
777 | * pages and associated backing store. | |
3866ea90 HD |
778 | * MADV_DONTFORK - omit this area from child's address space when forking: |
779 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
780 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
c02c3009 YS |
781 | * MADV_WIPEONFORK - present the child process with zero-filled memory in this |
782 | * range after a fork. | |
783 | * MADV_KEEPONFORK - undo the effect of MADV_WIPEONFORK | |
d7206a70 NH |
784 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
785 | * were corrupted by unrecoverable hardware memory failure. | |
786 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
787 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
788 | * this area with pages of identical content from other such areas. | |
789 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
790 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
791 | * huge pages in the future. Existing pages might be coalesced and | |
792 | * new pages might be allocated as THP. | |
793 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
794 | * transparent huge pages so the existing pages will not be | |
795 | * coalesced into THP and new pages will not be allocated as THP. | |
796 | * MADV_DONTDUMP - the application wants to prevent pages in the given range | |
797 | * from being included in its core dump. | |
798 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
1da177e4 LT |
799 | * |
800 | * return values: | |
801 | * zero - success | |
802 | * -EINVAL - start + len < 0, start is not page-aligned, | |
803 | * "behavior" is not a valid value, or application | |
c02c3009 YS |
804 | * is attempting to release locked or shared pages, |
805 | * or the specified address range includes file, Huge TLB, | |
806 | * MAP_SHARED or VMPFNMAP range. | |
1da177e4 LT |
807 | * -ENOMEM - addresses in the specified range are not currently |
808 | * mapped, or are outside the AS of the process. | |
809 | * -EIO - an I/O error occurred while paging in data. | |
810 | * -EBADF - map exists, but area maps something that isn't a file. | |
811 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
812 | */ | |
3480b257 | 813 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) |
1da177e4 | 814 | { |
05b74384 | 815 | unsigned long end, tmp; |
ec9bed9d | 816 | struct vm_area_struct *vma, *prev; |
1da177e4 LT |
817 | int unmapped_error = 0; |
818 | int error = -EINVAL; | |
f7977793 | 819 | int write; |
1da177e4 | 820 | size_t len; |
1998cc04 | 821 | struct blk_plug plug; |
1da177e4 | 822 | |
75927af8 NP |
823 | if (!madvise_behavior_valid(behavior)) |
824 | return error; | |
825 | ||
1da177e4 | 826 | if (start & ~PAGE_MASK) |
84d96d89 | 827 | return error; |
1da177e4 LT |
828 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; |
829 | ||
830 | /* Check to see whether len was rounded up from small -ve to zero */ | |
831 | if (len_in && !len) | |
84d96d89 | 832 | return error; |
1da177e4 LT |
833 | |
834 | end = start + len; | |
835 | if (end < start) | |
84d96d89 | 836 | return error; |
1da177e4 LT |
837 | |
838 | error = 0; | |
839 | if (end == start) | |
84d96d89 RV |
840 | return error; |
841 | ||
5e451be7 AK |
842 | #ifdef CONFIG_MEMORY_FAILURE |
843 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) | |
844 | return madvise_inject_error(behavior, start, start + len_in); | |
845 | #endif | |
846 | ||
84d96d89 | 847 | write = madvise_need_mmap_write(behavior); |
dc0ef0df MH |
848 | if (write) { |
849 | if (down_write_killable(¤t->mm->mmap_sem)) | |
850 | return -EINTR; | |
851 | } else { | |
84d96d89 | 852 | down_read(¤t->mm->mmap_sem); |
dc0ef0df | 853 | } |
1da177e4 LT |
854 | |
855 | /* | |
856 | * If the interval [start,end) covers some unmapped address | |
857 | * ranges, just ignore them, but return -ENOMEM at the end. | |
05b74384 | 858 | * - different from the way of handling in mlock etc. |
1da177e4 | 859 | */ |
05b74384 | 860 | vma = find_vma_prev(current->mm, start, &prev); |
836d5ffd HD |
861 | if (vma && start > vma->vm_start) |
862 | prev = vma; | |
863 | ||
1998cc04 | 864 | blk_start_plug(&plug); |
1da177e4 LT |
865 | for (;;) { |
866 | /* Still start < end. */ | |
867 | error = -ENOMEM; | |
868 | if (!vma) | |
84d96d89 | 869 | goto out; |
1da177e4 | 870 | |
05b74384 | 871 | /* Here start < (end|vma->vm_end). */ |
1da177e4 LT |
872 | if (start < vma->vm_start) { |
873 | unmapped_error = -ENOMEM; | |
874 | start = vma->vm_start; | |
05b74384 | 875 | if (start >= end) |
84d96d89 | 876 | goto out; |
1da177e4 LT |
877 | } |
878 | ||
05b74384 PM |
879 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ |
880 | tmp = vma->vm_end; | |
881 | if (end < tmp) | |
882 | tmp = end; | |
1da177e4 | 883 | |
05b74384 PM |
884 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ |
885 | error = madvise_vma(vma, &prev, start, tmp, behavior); | |
1da177e4 | 886 | if (error) |
84d96d89 | 887 | goto out; |
05b74384 | 888 | start = tmp; |
90ed52eb | 889 | if (prev && start < prev->vm_end) |
05b74384 PM |
890 | start = prev->vm_end; |
891 | error = unmapped_error; | |
892 | if (start >= end) | |
84d96d89 | 893 | goto out; |
90ed52eb HD |
894 | if (prev) |
895 | vma = prev->vm_next; | |
896 | else /* madvise_remove dropped mmap_sem */ | |
897 | vma = find_vma(current->mm, start); | |
1da177e4 | 898 | } |
1da177e4 | 899 | out: |
84d96d89 | 900 | blk_finish_plug(&plug); |
f7977793 | 901 | if (write) |
0a27a14a NP |
902 | up_write(¤t->mm->mmap_sem); |
903 | else | |
904 | up_read(¤t->mm->mmap_sem); | |
905 | ||
1da177e4 LT |
906 | return error; |
907 | } |