]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/madvise.c | |
3 | * | |
4 | * Copyright (C) 1999 Linus Torvalds | |
5 | * Copyright (C) 2002 Christoph Hellwig | |
6 | */ | |
7 | ||
8 | #include <linux/mman.h> | |
9 | #include <linux/pagemap.h> | |
10 | #include <linux/syscalls.h> | |
05b74384 | 11 | #include <linux/mempolicy.h> |
afcf938e | 12 | #include <linux/page-isolation.h> |
05ce7724 | 13 | #include <linux/userfaultfd_k.h> |
1da177e4 | 14 | #include <linux/hugetlb.h> |
3f31d075 | 15 | #include <linux/falloc.h> |
e8edc6e0 | 16 | #include <linux/sched.h> |
f8af4da3 | 17 | #include <linux/ksm.h> |
3f31d075 | 18 | #include <linux/fs.h> |
9ab4233d | 19 | #include <linux/file.h> |
1998cc04 | 20 | #include <linux/blkdev.h> |
66114cad | 21 | #include <linux/backing-dev.h> |
1998cc04 SL |
22 | #include <linux/swap.h> |
23 | #include <linux/swapops.h> | |
3a4f8a0b | 24 | #include <linux/shmem_fs.h> |
854e9ed0 MK |
25 | #include <linux/mmu_notifier.h> |
26 | ||
27 | #include <asm/tlb.h> | |
1da177e4 | 28 | |
23519073 KS |
29 | #include "internal.h" |
30 | ||
0a27a14a NP |
31 | /* |
32 | * Any behaviour which results in changes to the vma->vm_flags needs to | |
33 | * take mmap_sem for writing. Others, which simply traverse vmas, need | |
34 | * to only take it for reading. | |
35 | */ | |
36 | static int madvise_need_mmap_write(int behavior) | |
37 | { | |
38 | switch (behavior) { | |
39 | case MADV_REMOVE: | |
40 | case MADV_WILLNEED: | |
41 | case MADV_DONTNEED: | |
854e9ed0 | 42 | case MADV_FREE: |
0a27a14a NP |
43 | return 0; |
44 | default: | |
45 | /* be safe, default to 1. list exceptions explicitly */ | |
46 | return 1; | |
47 | } | |
48 | } | |
49 | ||
1da177e4 LT |
50 | /* |
51 | * We can potentially split a vm area into separate | |
52 | * areas, each area with its own behavior. | |
53 | */ | |
ec9bed9d | 54 | static long madvise_behavior(struct vm_area_struct *vma, |
05b74384 PM |
55 | struct vm_area_struct **prev, |
56 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 57 | { |
ec9bed9d | 58 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 59 | int error = 0; |
05b74384 | 60 | pgoff_t pgoff; |
3866ea90 | 61 | unsigned long new_flags = vma->vm_flags; |
e798c6e8 PM |
62 | |
63 | switch (behavior) { | |
f8225661 MT |
64 | case MADV_NORMAL: |
65 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
66 | break; | |
e798c6e8 | 67 | case MADV_SEQUENTIAL: |
f8225661 | 68 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; |
e798c6e8 PM |
69 | break; |
70 | case MADV_RANDOM: | |
f8225661 | 71 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; |
e798c6e8 | 72 | break; |
f8225661 MT |
73 | case MADV_DONTFORK: |
74 | new_flags |= VM_DONTCOPY; | |
75 | break; | |
76 | case MADV_DOFORK: | |
3866ea90 HD |
77 | if (vma->vm_flags & VM_IO) { |
78 | error = -EINVAL; | |
79 | goto out; | |
80 | } | |
f8225661 | 81 | new_flags &= ~VM_DONTCOPY; |
e798c6e8 | 82 | break; |
accb61fe | 83 | case MADV_DONTDUMP: |
0103bd16 | 84 | new_flags |= VM_DONTDUMP; |
accb61fe JB |
85 | break; |
86 | case MADV_DODUMP: | |
0103bd16 KK |
87 | if (new_flags & VM_SPECIAL) { |
88 | error = -EINVAL; | |
89 | goto out; | |
90 | } | |
91 | new_flags &= ~VM_DONTDUMP; | |
accb61fe | 92 | break; |
f8af4da3 HD |
93 | case MADV_MERGEABLE: |
94 | case MADV_UNMERGEABLE: | |
95 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
def5efe0 DR |
96 | if (error) { |
97 | /* | |
98 | * madvise() returns EAGAIN if kernel resources, such as | |
99 | * slab, are temporarily unavailable. | |
100 | */ | |
101 | if (error == -ENOMEM) | |
102 | error = -EAGAIN; | |
f8af4da3 | 103 | goto out; |
def5efe0 | 104 | } |
f8af4da3 | 105 | break; |
0af4e98b | 106 | case MADV_HUGEPAGE: |
a664b2d8 | 107 | case MADV_NOHUGEPAGE: |
60ab3244 | 108 | error = hugepage_madvise(vma, &new_flags, behavior); |
def5efe0 DR |
109 | if (error) { |
110 | /* | |
111 | * madvise() returns EAGAIN if kernel resources, such as | |
112 | * slab, are temporarily unavailable. | |
113 | */ | |
114 | if (error == -ENOMEM) | |
115 | error = -EAGAIN; | |
0af4e98b | 116 | goto out; |
def5efe0 | 117 | } |
0af4e98b | 118 | break; |
e798c6e8 PM |
119 | } |
120 | ||
05b74384 PM |
121 | if (new_flags == vma->vm_flags) { |
122 | *prev = vma; | |
836d5ffd | 123 | goto out; |
05b74384 PM |
124 | } |
125 | ||
126 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
127 | *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma, | |
19a809af AA |
128 | vma->vm_file, pgoff, vma_policy(vma), |
129 | vma->vm_userfaultfd_ctx); | |
05b74384 PM |
130 | if (*prev) { |
131 | vma = *prev; | |
132 | goto success; | |
133 | } | |
134 | ||
135 | *prev = vma; | |
1da177e4 LT |
136 | |
137 | if (start != vma->vm_start) { | |
def5efe0 DR |
138 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
139 | error = -ENOMEM; | |
1da177e4 | 140 | goto out; |
def5efe0 DR |
141 | } |
142 | error = __split_vma(mm, vma, start, 1); | |
143 | if (error) { | |
144 | /* | |
145 | * madvise() returns EAGAIN if kernel resources, such as | |
146 | * slab, are temporarily unavailable. | |
147 | */ | |
148 | if (error == -ENOMEM) | |
149 | error = -EAGAIN; | |
150 | goto out; | |
151 | } | |
1da177e4 LT |
152 | } |
153 | ||
154 | if (end != vma->vm_end) { | |
def5efe0 DR |
155 | if (unlikely(mm->map_count >= sysctl_max_map_count)) { |
156 | error = -ENOMEM; | |
1da177e4 | 157 | goto out; |
def5efe0 DR |
158 | } |
159 | error = __split_vma(mm, vma, end, 0); | |
160 | if (error) { | |
161 | /* | |
162 | * madvise() returns EAGAIN if kernel resources, such as | |
163 | * slab, are temporarily unavailable. | |
164 | */ | |
165 | if (error == -ENOMEM) | |
166 | error = -EAGAIN; | |
167 | goto out; | |
168 | } | |
1da177e4 LT |
169 | } |
170 | ||
836d5ffd | 171 | success: |
1da177e4 LT |
172 | /* |
173 | * vm_flags is protected by the mmap_sem held in write mode. | |
174 | */ | |
e798c6e8 | 175 | vma->vm_flags = new_flags; |
1da177e4 | 176 | out: |
1da177e4 LT |
177 | return error; |
178 | } | |
179 | ||
1998cc04 SL |
180 | #ifdef CONFIG_SWAP |
181 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
182 | unsigned long end, struct mm_walk *walk) | |
183 | { | |
184 | pte_t *orig_pte; | |
185 | struct vm_area_struct *vma = walk->private; | |
186 | unsigned long index; | |
187 | ||
188 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
189 | return 0; | |
190 | ||
191 | for (index = start; index != end; index += PAGE_SIZE) { | |
192 | pte_t pte; | |
193 | swp_entry_t entry; | |
194 | struct page *page; | |
195 | spinlock_t *ptl; | |
196 | ||
197 | orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); | |
198 | pte = *(orig_pte + ((index - start) / PAGE_SIZE)); | |
199 | pte_unmap_unlock(orig_pte, ptl); | |
200 | ||
0661a336 | 201 | if (pte_present(pte) || pte_none(pte)) |
1998cc04 SL |
202 | continue; |
203 | entry = pte_to_swp_entry(pte); | |
204 | if (unlikely(non_swap_entry(entry))) | |
205 | continue; | |
206 | ||
207 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, | |
208 | vma, index); | |
209 | if (page) | |
09cbfeaf | 210 | put_page(page); |
1998cc04 SL |
211 | } |
212 | ||
213 | return 0; | |
214 | } | |
215 | ||
216 | static void force_swapin_readahead(struct vm_area_struct *vma, | |
217 | unsigned long start, unsigned long end) | |
218 | { | |
219 | struct mm_walk walk = { | |
220 | .mm = vma->vm_mm, | |
221 | .pmd_entry = swapin_walk_pmd_entry, | |
222 | .private = vma, | |
223 | }; | |
224 | ||
225 | walk_page_range(start, end, &walk); | |
226 | ||
227 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
228 | } | |
229 | ||
230 | static void force_shm_swapin_readahead(struct vm_area_struct *vma, | |
231 | unsigned long start, unsigned long end, | |
232 | struct address_space *mapping) | |
233 | { | |
234 | pgoff_t index; | |
235 | struct page *page; | |
236 | swp_entry_t swap; | |
237 | ||
238 | for (; start < end; start += PAGE_SIZE) { | |
239 | index = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
240 | ||
55231e5c | 241 | page = find_get_entry(mapping, index); |
1998cc04 SL |
242 | if (!radix_tree_exceptional_entry(page)) { |
243 | if (page) | |
09cbfeaf | 244 | put_page(page); |
1998cc04 SL |
245 | continue; |
246 | } | |
247 | swap = radix_to_swp_entry(page); | |
248 | page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE, | |
249 | NULL, 0); | |
250 | if (page) | |
09cbfeaf | 251 | put_page(page); |
1998cc04 SL |
252 | } |
253 | ||
254 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
255 | } | |
256 | #endif /* CONFIG_SWAP */ | |
257 | ||
1da177e4 LT |
258 | /* |
259 | * Schedule all required I/O operations. Do not wait for completion. | |
260 | */ | |
ec9bed9d VC |
261 | static long madvise_willneed(struct vm_area_struct *vma, |
262 | struct vm_area_struct **prev, | |
1da177e4 LT |
263 | unsigned long start, unsigned long end) |
264 | { | |
265 | struct file *file = vma->vm_file; | |
266 | ||
1998cc04 | 267 | #ifdef CONFIG_SWAP |
97b713ba | 268 | if (!file) { |
1998cc04 | 269 | *prev = vma; |
97b713ba | 270 | force_swapin_readahead(vma, start, end); |
1998cc04 SL |
271 | return 0; |
272 | } | |
1998cc04 | 273 | |
97b713ba CH |
274 | if (shmem_mapping(file->f_mapping)) { |
275 | *prev = vma; | |
276 | force_shm_swapin_readahead(vma, start, end, | |
277 | file->f_mapping); | |
278 | return 0; | |
279 | } | |
280 | #else | |
1bef4003 S |
281 | if (!file) |
282 | return -EBADF; | |
97b713ba | 283 | #endif |
1bef4003 | 284 | |
e748dcd0 | 285 | if (IS_DAX(file_inode(file))) { |
fe77ba6f CO |
286 | /* no bad return value, but ignore advice */ |
287 | return 0; | |
288 | } | |
289 | ||
05b74384 | 290 | *prev = vma; |
1da177e4 LT |
291 | start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
292 | if (end > vma->vm_end) | |
293 | end = vma->vm_end; | |
294 | end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
295 | ||
f7e839dd | 296 | force_page_cache_readahead(file->f_mapping, file, start, end - start); |
1da177e4 LT |
297 | return 0; |
298 | } | |
299 | ||
854e9ed0 MK |
300 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
301 | unsigned long end, struct mm_walk *walk) | |
302 | ||
303 | { | |
304 | struct mmu_gather *tlb = walk->private; | |
305 | struct mm_struct *mm = tlb->mm; | |
306 | struct vm_area_struct *vma = walk->vma; | |
307 | spinlock_t *ptl; | |
308 | pte_t *orig_pte, *pte, ptent; | |
309 | struct page *page; | |
64b42bc1 | 310 | int nr_swap = 0; |
b8d3c4c3 MK |
311 | unsigned long next; |
312 | ||
313 | next = pmd_addr_end(addr, end); | |
314 | if (pmd_trans_huge(*pmd)) | |
315 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
316 | goto next; | |
854e9ed0 | 317 | |
854e9ed0 MK |
318 | if (pmd_trans_unstable(pmd)) |
319 | return 0; | |
320 | ||
07e32661 | 321 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
854e9ed0 MK |
322 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
323 | arch_enter_lazy_mmu_mode(); | |
324 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
325 | ptent = *pte; | |
326 | ||
64b42bc1 | 327 | if (pte_none(ptent)) |
854e9ed0 | 328 | continue; |
64b42bc1 MK |
329 | /* |
330 | * If the pte has swp_entry, just clear page table to | |
331 | * prevent swap-in which is more expensive rather than | |
332 | * (page allocation + zeroing). | |
333 | */ | |
334 | if (!pte_present(ptent)) { | |
335 | swp_entry_t entry; | |
336 | ||
337 | entry = pte_to_swp_entry(ptent); | |
338 | if (non_swap_entry(entry)) | |
339 | continue; | |
340 | nr_swap--; | |
341 | free_swap_and_cache(entry); | |
342 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
343 | continue; | |
344 | } | |
854e9ed0 MK |
345 | |
346 | page = vm_normal_page(vma, addr, ptent); | |
347 | if (!page) | |
348 | continue; | |
349 | ||
350 | /* | |
351 | * If pmd isn't transhuge but the page is THP and | |
352 | * is owned by only this process, split it and | |
353 | * deactivate all pages. | |
354 | */ | |
355 | if (PageTransCompound(page)) { | |
356 | if (page_mapcount(page) != 1) | |
357 | goto out; | |
358 | get_page(page); | |
359 | if (!trylock_page(page)) { | |
360 | put_page(page); | |
361 | goto out; | |
362 | } | |
363 | pte_unmap_unlock(orig_pte, ptl); | |
364 | if (split_huge_page(page)) { | |
365 | unlock_page(page); | |
366 | put_page(page); | |
367 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
368 | goto out; | |
369 | } | |
370 | put_page(page); | |
371 | unlock_page(page); | |
372 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
373 | pte--; | |
374 | addr -= PAGE_SIZE; | |
375 | continue; | |
376 | } | |
377 | ||
378 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
379 | ||
380 | if (PageSwapCache(page) || PageDirty(page)) { | |
381 | if (!trylock_page(page)) | |
382 | continue; | |
383 | /* | |
384 | * If page is shared with others, we couldn't clear | |
385 | * PG_dirty of the page. | |
386 | */ | |
387 | if (page_mapcount(page) != 1) { | |
388 | unlock_page(page); | |
389 | continue; | |
390 | } | |
391 | ||
392 | if (PageSwapCache(page) && !try_to_free_swap(page)) { | |
393 | unlock_page(page); | |
394 | continue; | |
395 | } | |
396 | ||
397 | ClearPageDirty(page); | |
398 | unlock_page(page); | |
399 | } | |
400 | ||
401 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
402 | /* | |
403 | * Some of architecture(ex, PPC) don't update TLB | |
404 | * with set_pte_at and tlb_remove_tlb_entry so for | |
405 | * the portability, remap the pte with old|clean | |
406 | * after pte clearing. | |
407 | */ | |
408 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
409 | tlb->fullmm); | |
410 | ||
411 | ptent = pte_mkold(ptent); | |
412 | ptent = pte_mkclean(ptent); | |
413 | set_pte_at(mm, addr, pte, ptent); | |
10853a03 MK |
414 | if (PageActive(page)) |
415 | deactivate_page(page); | |
854e9ed0 MK |
416 | tlb_remove_tlb_entry(tlb, pte, addr); |
417 | } | |
418 | } | |
419 | out: | |
64b42bc1 MK |
420 | if (nr_swap) { |
421 | if (current->mm == mm) | |
422 | sync_mm_rss(mm); | |
423 | ||
424 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); | |
425 | } | |
854e9ed0 MK |
426 | arch_leave_lazy_mmu_mode(); |
427 | pte_unmap_unlock(orig_pte, ptl); | |
428 | cond_resched(); | |
b8d3c4c3 | 429 | next: |
854e9ed0 MK |
430 | return 0; |
431 | } | |
432 | ||
433 | static void madvise_free_page_range(struct mmu_gather *tlb, | |
434 | struct vm_area_struct *vma, | |
435 | unsigned long addr, unsigned long end) | |
436 | { | |
437 | struct mm_walk free_walk = { | |
438 | .pmd_entry = madvise_free_pte_range, | |
439 | .mm = vma->vm_mm, | |
440 | .private = tlb, | |
441 | }; | |
442 | ||
443 | tlb_start_vma(tlb, vma); | |
444 | walk_page_range(addr, end, &free_walk); | |
445 | tlb_end_vma(tlb, vma); | |
446 | } | |
447 | ||
448 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
449 | unsigned long start_addr, unsigned long end_addr) | |
450 | { | |
451 | unsigned long start, end; | |
452 | struct mm_struct *mm = vma->vm_mm; | |
453 | struct mmu_gather tlb; | |
454 | ||
455 | if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP)) | |
456 | return -EINVAL; | |
457 | ||
458 | /* MADV_FREE works for only anon vma at the moment */ | |
459 | if (!vma_is_anonymous(vma)) | |
460 | return -EINVAL; | |
461 | ||
462 | start = max(vma->vm_start, start_addr); | |
463 | if (start >= vma->vm_end) | |
464 | return -EINVAL; | |
465 | end = min(vma->vm_end, end_addr); | |
466 | if (end <= vma->vm_start) | |
467 | return -EINVAL; | |
468 | ||
469 | lru_add_drain(); | |
470 | tlb_gather_mmu(&tlb, mm, start, end); | |
471 | update_hiwater_rss(mm); | |
472 | ||
473 | mmu_notifier_invalidate_range_start(mm, start, end); | |
474 | madvise_free_page_range(&tlb, vma, start, end); | |
475 | mmu_notifier_invalidate_range_end(mm, start, end); | |
476 | tlb_finish_mmu(&tlb, start, end); | |
477 | ||
478 | return 0; | |
479 | } | |
480 | ||
481 | static long madvise_free(struct vm_area_struct *vma, | |
482 | struct vm_area_struct **prev, | |
483 | unsigned long start, unsigned long end) | |
484 | { | |
485 | *prev = vma; | |
486 | return madvise_free_single_vma(vma, start, end); | |
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 | */ | |
ec9bed9d VC |
508 | static long madvise_dontneed(struct vm_area_struct *vma, |
509 | struct vm_area_struct **prev, | |
1da177e4 LT |
510 | unsigned long start, unsigned long end) |
511 | { | |
05b74384 | 512 | *prev = vma; |
23519073 | 513 | if (!can_madv_dontneed_vma(vma)) |
1da177e4 LT |
514 | return -EINVAL; |
515 | ||
d811914d | 516 | userfaultfd_remove(vma, prev, start, end); |
ecf1385d | 517 | zap_page_range(vma, start, end - start); |
1da177e4 LT |
518 | return 0; |
519 | } | |
520 | ||
f6b3ec23 BP |
521 | /* |
522 | * Application wants to free up the pages and associated backing store. | |
523 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
524 | */ |
525 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 526 | struct vm_area_struct **prev, |
f6b3ec23 BP |
527 | unsigned long start, unsigned long end) |
528 | { | |
3f31d075 | 529 | loff_t offset; |
90ed52eb | 530 | int error; |
9ab4233d | 531 | struct file *f; |
f6b3ec23 | 532 | |
90ed52eb | 533 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ |
00e9fa2d | 534 | |
72079ba0 | 535 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
536 | return -EINVAL; |
537 | ||
9ab4233d AL |
538 | f = vma->vm_file; |
539 | ||
540 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
541 | return -EINVAL; |
542 | } | |
543 | ||
69cf0fac HD |
544 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
545 | return -EACCES; | |
546 | ||
f6b3ec23 BP |
547 | offset = (loff_t)(start - vma->vm_start) |
548 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 549 | |
9ab4233d AL |
550 | /* |
551 | * Filesystem's fallocate may need to take i_mutex. We need to | |
552 | * explicitly grab a reference because the vma (and hence the | |
553 | * vma's reference to the file) can go away as soon as we drop | |
554 | * mmap_sem. | |
555 | */ | |
556 | get_file(f); | |
a6bf53eb | 557 | userfaultfd_remove(vma, prev, start, end); |
0a27a14a | 558 | up_read(¤t->mm->mmap_sem); |
72c72bdf | 559 | error = vfs_fallocate(f, |
3f31d075 HD |
560 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
561 | offset, end - start); | |
9ab4233d | 562 | fput(f); |
0a27a14a | 563 | down_read(¤t->mm->mmap_sem); |
90ed52eb | 564 | return error; |
f6b3ec23 BP |
565 | } |
566 | ||
9893e49d AK |
567 | #ifdef CONFIG_MEMORY_FAILURE |
568 | /* | |
569 | * Error injection support for memory error handling. | |
570 | */ | |
afcf938e | 571 | static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end) |
9893e49d | 572 | { |
20cb6cab | 573 | struct page *p; |
9893e49d AK |
574 | if (!capable(CAP_SYS_ADMIN)) |
575 | return -EPERM; | |
20cb6cab WL |
576 | for (; start < end; start += PAGE_SIZE << |
577 | compound_order(compound_head(p))) { | |
325c4ef5 AM |
578 | int ret; |
579 | ||
580 | ret = get_user_pages_fast(start, 1, 0, &p); | |
9893e49d AK |
581 | if (ret != 1) |
582 | return ret; | |
325c4ef5 | 583 | |
29b4eede WL |
584 | if (PageHWPoison(p)) { |
585 | put_page(p); | |
586 | continue; | |
587 | } | |
afcf938e | 588 | if (bhv == MADV_SOFT_OFFLINE) { |
b194b8cd | 589 | pr_info("Soft offlining page %#lx at %#lx\n", |
afcf938e AK |
590 | page_to_pfn(p), start); |
591 | ret = soft_offline_page(p, MF_COUNT_INCREASED); | |
592 | if (ret) | |
8302423b | 593 | return ret; |
afcf938e AK |
594 | continue; |
595 | } | |
b194b8cd | 596 | pr_info("Injecting memory failure for page %#lx at %#lx\n", |
9893e49d | 597 | page_to_pfn(p), start); |
23a003bf NH |
598 | ret = memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED); |
599 | if (ret) | |
600 | return ret; | |
9893e49d | 601 | } |
325c4ef5 | 602 | return 0; |
9893e49d AK |
603 | } |
604 | #endif | |
605 | ||
165cd402 | 606 | static long |
607 | madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, | |
608 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 609 | { |
1da177e4 | 610 | switch (behavior) { |
f6b3ec23 | 611 | case MADV_REMOVE: |
3866ea90 | 612 | return madvise_remove(vma, prev, start, end); |
1da177e4 | 613 | case MADV_WILLNEED: |
3866ea90 | 614 | return madvise_willneed(vma, prev, start, end); |
854e9ed0 MK |
615 | case MADV_FREE: |
616 | /* | |
617 | * XXX: In this implementation, MADV_FREE works like | |
618 | * MADV_DONTNEED on swapless system or full swap. | |
619 | */ | |
620 | if (get_nr_swap_pages() > 0) | |
621 | return madvise_free(vma, prev, start, end); | |
622 | /* passthrough */ | |
1da177e4 | 623 | case MADV_DONTNEED: |
3866ea90 | 624 | return madvise_dontneed(vma, prev, start, end); |
1da177e4 | 625 | default: |
3866ea90 | 626 | return madvise_behavior(vma, prev, start, end, behavior); |
1da177e4 | 627 | } |
1da177e4 LT |
628 | } |
629 | ||
1ecef9ed | 630 | static bool |
75927af8 NP |
631 | madvise_behavior_valid(int behavior) |
632 | { | |
633 | switch (behavior) { | |
634 | case MADV_DOFORK: | |
635 | case MADV_DONTFORK: | |
636 | case MADV_NORMAL: | |
637 | case MADV_SEQUENTIAL: | |
638 | case MADV_RANDOM: | |
639 | case MADV_REMOVE: | |
640 | case MADV_WILLNEED: | |
641 | case MADV_DONTNEED: | |
854e9ed0 | 642 | case MADV_FREE: |
f8af4da3 HD |
643 | #ifdef CONFIG_KSM |
644 | case MADV_MERGEABLE: | |
645 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
646 | #endif |
647 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
648 | case MADV_HUGEPAGE: | |
a664b2d8 | 649 | case MADV_NOHUGEPAGE: |
f8af4da3 | 650 | #endif |
accb61fe JB |
651 | case MADV_DONTDUMP: |
652 | case MADV_DODUMP: | |
1ecef9ed | 653 | return true; |
75927af8 NP |
654 | |
655 | default: | |
1ecef9ed | 656 | return false; |
75927af8 NP |
657 | } |
658 | } | |
3866ea90 | 659 | |
1da177e4 LT |
660 | /* |
661 | * The madvise(2) system call. | |
662 | * | |
663 | * Applications can use madvise() to advise the kernel how it should | |
664 | * handle paging I/O in this VM area. The idea is to help the kernel | |
665 | * use appropriate read-ahead and caching techniques. The information | |
666 | * provided is advisory only, and can be safely disregarded by the | |
667 | * kernel without affecting the correct operation of the application. | |
668 | * | |
669 | * behavior values: | |
670 | * MADV_NORMAL - the default behavior is to read clusters. This | |
671 | * results in some read-ahead and read-behind. | |
672 | * MADV_RANDOM - the system should read the minimum amount of data | |
673 | * on any access, since it is unlikely that the appli- | |
674 | * cation will need more than what it asks for. | |
675 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
676 | * once, so they can be aggressively read ahead, and | |
677 | * can be freed soon after they are accessed. | |
678 | * MADV_WILLNEED - the application is notifying the system to read | |
679 | * some pages ahead. | |
680 | * MADV_DONTNEED - the application is finished with the given range, | |
681 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
682 | * MADV_FREE - the application marks pages in the given range as lazy free, |
683 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
684 | * MADV_REMOVE - the application wants to free up the given range of |
685 | * pages and associated backing store. | |
3866ea90 HD |
686 | * MADV_DONTFORK - omit this area from child's address space when forking: |
687 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
688 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
d7206a70 NH |
689 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
690 | * were corrupted by unrecoverable hardware memory failure. | |
691 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
692 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
693 | * this area with pages of identical content from other such areas. | |
694 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
695 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
696 | * huge pages in the future. Existing pages might be coalesced and | |
697 | * new pages might be allocated as THP. | |
698 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
699 | * transparent huge pages so the existing pages will not be | |
700 | * coalesced into THP and new pages will not be allocated as THP. | |
701 | * MADV_DONTDUMP - the application wants to prevent pages in the given range | |
702 | * from being included in its core dump. | |
703 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
1da177e4 LT |
704 | * |
705 | * return values: | |
706 | * zero - success | |
707 | * -EINVAL - start + len < 0, start is not page-aligned, | |
708 | * "behavior" is not a valid value, or application | |
709 | * is attempting to release locked or shared pages. | |
710 | * -ENOMEM - addresses in the specified range are not currently | |
711 | * mapped, or are outside the AS of the process. | |
712 | * -EIO - an I/O error occurred while paging in data. | |
713 | * -EBADF - map exists, but area maps something that isn't a file. | |
714 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
715 | */ | |
3480b257 | 716 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) |
1da177e4 | 717 | { |
05b74384 | 718 | unsigned long end, tmp; |
ec9bed9d | 719 | struct vm_area_struct *vma, *prev; |
1da177e4 LT |
720 | int unmapped_error = 0; |
721 | int error = -EINVAL; | |
f7977793 | 722 | int write; |
1da177e4 | 723 | size_t len; |
1998cc04 | 724 | struct blk_plug plug; |
1da177e4 | 725 | |
9893e49d | 726 | #ifdef CONFIG_MEMORY_FAILURE |
afcf938e AK |
727 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) |
728 | return madvise_hwpoison(behavior, start, start+len_in); | |
9893e49d | 729 | #endif |
75927af8 NP |
730 | if (!madvise_behavior_valid(behavior)) |
731 | return error; | |
732 | ||
1da177e4 | 733 | if (start & ~PAGE_MASK) |
84d96d89 | 734 | return error; |
1da177e4 LT |
735 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; |
736 | ||
737 | /* Check to see whether len was rounded up from small -ve to zero */ | |
738 | if (len_in && !len) | |
84d96d89 | 739 | return error; |
1da177e4 LT |
740 | |
741 | end = start + len; | |
742 | if (end < start) | |
84d96d89 | 743 | return error; |
1da177e4 LT |
744 | |
745 | error = 0; | |
746 | if (end == start) | |
84d96d89 RV |
747 | return error; |
748 | ||
749 | write = madvise_need_mmap_write(behavior); | |
dc0ef0df MH |
750 | if (write) { |
751 | if (down_write_killable(¤t->mm->mmap_sem)) | |
752 | return -EINTR; | |
753 | } else { | |
84d96d89 | 754 | down_read(¤t->mm->mmap_sem); |
dc0ef0df | 755 | } |
1da177e4 LT |
756 | |
757 | /* | |
758 | * If the interval [start,end) covers some unmapped address | |
759 | * ranges, just ignore them, but return -ENOMEM at the end. | |
05b74384 | 760 | * - different from the way of handling in mlock etc. |
1da177e4 | 761 | */ |
05b74384 | 762 | vma = find_vma_prev(current->mm, start, &prev); |
836d5ffd HD |
763 | if (vma && start > vma->vm_start) |
764 | prev = vma; | |
765 | ||
1998cc04 | 766 | blk_start_plug(&plug); |
1da177e4 LT |
767 | for (;;) { |
768 | /* Still start < end. */ | |
769 | error = -ENOMEM; | |
770 | if (!vma) | |
84d96d89 | 771 | goto out; |
1da177e4 | 772 | |
05b74384 | 773 | /* Here start < (end|vma->vm_end). */ |
1da177e4 LT |
774 | if (start < vma->vm_start) { |
775 | unmapped_error = -ENOMEM; | |
776 | start = vma->vm_start; | |
05b74384 | 777 | if (start >= end) |
84d96d89 | 778 | goto out; |
1da177e4 LT |
779 | } |
780 | ||
05b74384 PM |
781 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ |
782 | tmp = vma->vm_end; | |
783 | if (end < tmp) | |
784 | tmp = end; | |
1da177e4 | 785 | |
05b74384 PM |
786 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ |
787 | error = madvise_vma(vma, &prev, start, tmp, behavior); | |
1da177e4 | 788 | if (error) |
84d96d89 | 789 | goto out; |
05b74384 | 790 | start = tmp; |
90ed52eb | 791 | if (prev && start < prev->vm_end) |
05b74384 PM |
792 | start = prev->vm_end; |
793 | error = unmapped_error; | |
794 | if (start >= end) | |
84d96d89 | 795 | goto out; |
90ed52eb HD |
796 | if (prev) |
797 | vma = prev->vm_next; | |
798 | else /* madvise_remove dropped mmap_sem */ | |
799 | vma = find_vma(current->mm, start); | |
1da177e4 | 800 | } |
1da177e4 | 801 | out: |
84d96d89 | 802 | blk_finish_plug(&plug); |
f7977793 | 803 | if (write) |
0a27a14a NP |
804 | up_write(¤t->mm->mmap_sem); |
805 | else | |
806 | up_read(¤t->mm->mmap_sem); | |
807 | ||
1da177e4 LT |
808 | return error; |
809 | } |