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> |
ecb8ac8b | 20 | #include <linux/sched/mm.h> |
17fca131 | 21 | #include <linux/mm_inline.h> |
9a10064f | 22 | #include <linux/string.h> |
ecb8ac8b | 23 | #include <linux/uio.h> |
f8af4da3 | 24 | #include <linux/ksm.h> |
3f31d075 | 25 | #include <linux/fs.h> |
9ab4233d | 26 | #include <linux/file.h> |
1998cc04 | 27 | #include <linux/blkdev.h> |
66114cad | 28 | #include <linux/backing-dev.h> |
a520110e | 29 | #include <linux/pagewalk.h> |
1998cc04 SL |
30 | #include <linux/swap.h> |
31 | #include <linux/swapops.h> | |
3a4f8a0b | 32 | #include <linux/shmem_fs.h> |
854e9ed0 MK |
33 | #include <linux/mmu_notifier.h> |
34 | ||
35 | #include <asm/tlb.h> | |
1da177e4 | 36 | |
23519073 | 37 | #include "internal.h" |
014bb1de | 38 | #include "swap.h" |
23519073 | 39 | |
d616d512 MK |
40 | struct madvise_walk_private { |
41 | struct mmu_gather *tlb; | |
42 | bool pageout; | |
43 | }; | |
44 | ||
0a27a14a NP |
45 | /* |
46 | * Any behaviour which results in changes to the vma->vm_flags needs to | |
c1e8d7c6 | 47 | * take mmap_lock for writing. Others, which simply traverse vmas, need |
0a27a14a NP |
48 | * to only take it for reading. |
49 | */ | |
50 | static int madvise_need_mmap_write(int behavior) | |
51 | { | |
52 | switch (behavior) { | |
53 | case MADV_REMOVE: | |
54 | case MADV_WILLNEED: | |
55 | case MADV_DONTNEED: | |
9457056a | 56 | case MADV_DONTNEED_LOCKED: |
9c276cc6 | 57 | case MADV_COLD: |
1a4e58cc | 58 | case MADV_PAGEOUT: |
854e9ed0 | 59 | case MADV_FREE: |
4ca9b385 DH |
60 | case MADV_POPULATE_READ: |
61 | case MADV_POPULATE_WRITE: | |
7d8faaf1 | 62 | case MADV_COLLAPSE: |
0a27a14a NP |
63 | return 0; |
64 | default: | |
65 | /* be safe, default to 1. list exceptions explicitly */ | |
66 | return 1; | |
67 | } | |
68 | } | |
69 | ||
9a10064f | 70 | #ifdef CONFIG_ANON_VMA_NAME |
5c26f6ac | 71 | struct anon_vma_name *anon_vma_name_alloc(const char *name) |
78db3412 SB |
72 | { |
73 | struct anon_vma_name *anon_name; | |
74 | size_t count; | |
75 | ||
76 | /* Add 1 for NUL terminator at the end of the anon_name->name */ | |
77 | count = strlen(name) + 1; | |
78 | anon_name = kmalloc(struct_size(anon_name, name, count), GFP_KERNEL); | |
79 | if (anon_name) { | |
80 | kref_init(&anon_name->kref); | |
81 | memcpy(anon_name->name, name, count); | |
82 | } | |
83 | ||
84 | return anon_name; | |
85 | } | |
86 | ||
5c26f6ac | 87 | void anon_vma_name_free(struct kref *kref) |
78db3412 SB |
88 | { |
89 | struct anon_vma_name *anon_name = | |
90 | container_of(kref, struct anon_vma_name, kref); | |
91 | kfree(anon_name); | |
92 | } | |
93 | ||
5c26f6ac | 94 | struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma) |
9a10064f | 95 | { |
9a10064f CC |
96 | mmap_assert_locked(vma->vm_mm); |
97 | ||
5c26f6ac | 98 | return vma->anon_name; |
9a10064f CC |
99 | } |
100 | ||
101 | /* mmap_lock should be write-locked */ | |
5c26f6ac SB |
102 | static int replace_anon_vma_name(struct vm_area_struct *vma, |
103 | struct anon_vma_name *anon_name) | |
9a10064f | 104 | { |
5c26f6ac | 105 | struct anon_vma_name *orig_name = anon_vma_name(vma); |
78db3412 | 106 | |
5c26f6ac SB |
107 | if (!anon_name) { |
108 | vma->anon_name = NULL; | |
109 | anon_vma_name_put(orig_name); | |
9a10064f CC |
110 | return 0; |
111 | } | |
112 | ||
5c26f6ac SB |
113 | if (anon_vma_name_eq(orig_name, anon_name)) |
114 | return 0; | |
9a10064f | 115 | |
96403e11 | 116 | vma->anon_name = anon_vma_name_reuse(anon_name); |
5c26f6ac | 117 | anon_vma_name_put(orig_name); |
9a10064f CC |
118 | |
119 | return 0; | |
120 | } | |
121 | #else /* CONFIG_ANON_VMA_NAME */ | |
5c26f6ac SB |
122 | static int replace_anon_vma_name(struct vm_area_struct *vma, |
123 | struct anon_vma_name *anon_name) | |
9a10064f | 124 | { |
5c26f6ac | 125 | if (anon_name) |
9a10064f CC |
126 | return -EINVAL; |
127 | ||
128 | return 0; | |
129 | } | |
130 | #endif /* CONFIG_ANON_VMA_NAME */ | |
1da177e4 | 131 | /* |
ac1e9acc | 132 | * Update the vm_flags on region of a vma, splitting it or merging it as |
8651a137 | 133 | * necessary. Must be called with mmap_lock held for writing; |
942341dc SB |
134 | * Caller should ensure anon_name stability by raising its refcount even when |
135 | * anon_name belongs to a valid vma because this function might free that vma. | |
1da177e4 | 136 | */ |
ac1e9acc CC |
137 | static int madvise_update_vma(struct vm_area_struct *vma, |
138 | struct vm_area_struct **prev, unsigned long start, | |
9a10064f | 139 | unsigned long end, unsigned long new_flags, |
5c26f6ac | 140 | struct anon_vma_name *anon_name) |
1da177e4 | 141 | { |
ec9bed9d | 142 | struct mm_struct *mm = vma->vm_mm; |
ac1e9acc | 143 | int error; |
05b74384 | 144 | pgoff_t pgoff; |
85ab779e | 145 | VMA_ITERATOR(vmi, mm, start); |
e798c6e8 | 146 | |
5c26f6ac | 147 | if (new_flags == vma->vm_flags && anon_vma_name_eq(anon_vma_name(vma), anon_name)) { |
05b74384 | 148 | *prev = vma; |
ac1e9acc | 149 | return 0; |
05b74384 PM |
150 | } |
151 | ||
152 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
9760ebff | 153 | *prev = vma_merge(&vmi, mm, *prev, start, end, new_flags, |
178e22ac | 154 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
5c26f6ac | 155 | vma->vm_userfaultfd_ctx, anon_name); |
05b74384 PM |
156 | if (*prev) { |
157 | vma = *prev; | |
158 | goto success; | |
159 | } | |
160 | ||
161 | *prev = vma; | |
1da177e4 LT |
162 | |
163 | if (start != vma->vm_start) { | |
85ab779e | 164 | error = split_vma(&vmi, vma, start, 1); |
f3bc0dba | 165 | if (error) |
ac1e9acc | 166 | return error; |
1da177e4 LT |
167 | } |
168 | ||
169 | if (end != vma->vm_end) { | |
85ab779e | 170 | error = split_vma(&vmi, vma, end, 0); |
f3bc0dba | 171 | if (error) |
ac1e9acc | 172 | return error; |
1da177e4 LT |
173 | } |
174 | ||
836d5ffd | 175 | success: |
1da177e4 | 176 | /* |
c1e8d7c6 | 177 | * vm_flags is protected by the mmap_lock held in write mode. |
1da177e4 | 178 | */ |
1c71222e | 179 | vm_flags_reset(vma, new_flags); |
d09e8ca6 | 180 | if (!vma->vm_file || vma_is_anon_shmem(vma)) { |
5c26f6ac | 181 | error = replace_anon_vma_name(vma, anon_name); |
9a10064f CC |
182 | if (error) |
183 | return error; | |
184 | } | |
f3bc0dba | 185 | |
ac1e9acc | 186 | return 0; |
1da177e4 LT |
187 | } |
188 | ||
1998cc04 SL |
189 | #ifdef CONFIG_SWAP |
190 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
191 | unsigned long end, struct mm_walk *walk) | |
192 | { | |
1998cc04 SL |
193 | struct vm_area_struct *vma = walk->private; |
194 | unsigned long index; | |
5169b844 | 195 | struct swap_iocb *splug = NULL; |
1998cc04 SL |
196 | |
197 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
198 | return 0; | |
199 | ||
200 | for (index = start; index != end; index += PAGE_SIZE) { | |
201 | pte_t pte; | |
202 | swp_entry_t entry; | |
203 | struct page *page; | |
204 | spinlock_t *ptl; | |
f7cc67ae | 205 | pte_t *ptep; |
1998cc04 | 206 | |
f7cc67ae ML |
207 | ptep = pte_offset_map_lock(vma->vm_mm, pmd, index, &ptl); |
208 | pte = *ptep; | |
209 | pte_unmap_unlock(ptep, ptl); | |
1998cc04 | 210 | |
f7cc67ae | 211 | if (!is_swap_pte(pte)) |
1998cc04 SL |
212 | continue; |
213 | entry = pte_to_swp_entry(pte); | |
214 | if (unlikely(non_swap_entry(entry))) | |
215 | continue; | |
216 | ||
217 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, | |
5169b844 | 218 | vma, index, false, &splug); |
1998cc04 | 219 | if (page) |
09cbfeaf | 220 | put_page(page); |
1998cc04 | 221 | } |
5169b844 | 222 | swap_read_unplug(splug); |
de2e5171 | 223 | cond_resched(); |
1998cc04 SL |
224 | |
225 | return 0; | |
226 | } | |
227 | ||
7b86ac33 CH |
228 | static const struct mm_walk_ops swapin_walk_ops = { |
229 | .pmd_entry = swapin_walk_pmd_entry, | |
230 | }; | |
1998cc04 SL |
231 | |
232 | static void force_shm_swapin_readahead(struct vm_area_struct *vma, | |
233 | unsigned long start, unsigned long end, | |
234 | struct address_space *mapping) | |
235 | { | |
e6e88712 | 236 | XA_STATE(xas, &mapping->i_pages, linear_page_index(vma, start)); |
66383800 | 237 | pgoff_t end_index = linear_page_index(vma, end + PAGE_SIZE - 1); |
1998cc04 | 238 | struct page *page; |
5169b844 | 239 | struct swap_iocb *splug = NULL; |
1998cc04 | 240 | |
e6e88712 MWO |
241 | rcu_read_lock(); |
242 | xas_for_each(&xas, page, end_index) { | |
243 | swp_entry_t swap; | |
1998cc04 | 244 | |
e6e88712 | 245 | if (!xa_is_value(page)) |
1998cc04 | 246 | continue; |
ba6851b4 ML |
247 | swap = radix_to_swp_entry(page); |
248 | /* There might be swapin error entries in shmem mapping. */ | |
249 | if (non_swap_entry(swap)) | |
250 | continue; | |
e6e88712 MWO |
251 | xas_pause(&xas); |
252 | rcu_read_unlock(); | |
253 | ||
1998cc04 | 254 | page = read_swap_cache_async(swap, GFP_HIGHUSER_MOVABLE, |
5169b844 | 255 | NULL, 0, false, &splug); |
1998cc04 | 256 | if (page) |
09cbfeaf | 257 | put_page(page); |
e6e88712 MWO |
258 | |
259 | rcu_read_lock(); | |
1998cc04 | 260 | } |
e6e88712 | 261 | rcu_read_unlock(); |
5169b844 | 262 | swap_read_unplug(splug); |
1998cc04 SL |
263 | |
264 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
265 | } | |
266 | #endif /* CONFIG_SWAP */ | |
267 | ||
1da177e4 LT |
268 | /* |
269 | * Schedule all required I/O operations. Do not wait for completion. | |
270 | */ | |
ec9bed9d VC |
271 | static long madvise_willneed(struct vm_area_struct *vma, |
272 | struct vm_area_struct **prev, | |
1da177e4 LT |
273 | unsigned long start, unsigned long end) |
274 | { | |
0726b01e | 275 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 276 | struct file *file = vma->vm_file; |
692fe624 | 277 | loff_t offset; |
1da177e4 | 278 | |
6ea8d958 | 279 | *prev = vma; |
1998cc04 | 280 | #ifdef CONFIG_SWAP |
97b713ba | 281 | if (!file) { |
7b86ac33 CH |
282 | walk_page_range(vma->vm_mm, start, end, &swapin_walk_ops, vma); |
283 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
1998cc04 SL |
284 | return 0; |
285 | } | |
1998cc04 | 286 | |
97b713ba | 287 | if (shmem_mapping(file->f_mapping)) { |
97b713ba CH |
288 | force_shm_swapin_readahead(vma, start, end, |
289 | file->f_mapping); | |
290 | return 0; | |
291 | } | |
292 | #else | |
1bef4003 S |
293 | if (!file) |
294 | return -EBADF; | |
97b713ba | 295 | #endif |
1bef4003 | 296 | |
e748dcd0 | 297 | if (IS_DAX(file_inode(file))) { |
fe77ba6f CO |
298 | /* no bad return value, but ignore advice */ |
299 | return 0; | |
300 | } | |
301 | ||
692fe624 JK |
302 | /* |
303 | * Filesystem's fadvise may need to take various locks. We need to | |
304 | * explicitly grab a reference because the vma (and hence the | |
305 | * vma's reference to the file) can go away as soon as we drop | |
c1e8d7c6 | 306 | * mmap_lock. |
692fe624 | 307 | */ |
c1e8d7c6 | 308 | *prev = NULL; /* tell sys_madvise we drop mmap_lock */ |
692fe624 | 309 | get_file(file); |
692fe624 JK |
310 | offset = (loff_t)(start - vma->vm_start) |
311 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
0726b01e | 312 | mmap_read_unlock(mm); |
692fe624 JK |
313 | vfs_fadvise(file, offset, end - start, POSIX_FADV_WILLNEED); |
314 | fput(file); | |
0726b01e | 315 | mmap_read_lock(mm); |
1da177e4 LT |
316 | return 0; |
317 | } | |
318 | ||
fd3b1bc3 PK |
319 | static inline bool can_do_file_pageout(struct vm_area_struct *vma) |
320 | { | |
321 | if (!vma->vm_file) | |
322 | return false; | |
323 | /* | |
324 | * paging out pagecache only for non-anonymous mappings that correspond | |
325 | * to the files the calling process could (if tried) open for writing; | |
326 | * otherwise we'd be including shared non-exclusive mappings, which | |
327 | * opens a side channel. | |
328 | */ | |
01beba79 | 329 | return inode_owner_or_capable(&nop_mnt_idmap, |
fd3b1bc3 PK |
330 | file_inode(vma->vm_file)) || |
331 | file_permission(vma->vm_file, MAY_WRITE) == 0; | |
332 | } | |
333 | ||
d616d512 MK |
334 | static int madvise_cold_or_pageout_pte_range(pmd_t *pmd, |
335 | unsigned long addr, unsigned long end, | |
336 | struct mm_walk *walk) | |
9c276cc6 | 337 | { |
d616d512 MK |
338 | struct madvise_walk_private *private = walk->private; |
339 | struct mmu_gather *tlb = private->tlb; | |
340 | bool pageout = private->pageout; | |
9c276cc6 MK |
341 | struct mm_struct *mm = tlb->mm; |
342 | struct vm_area_struct *vma = walk->vma; | |
343 | pte_t *orig_pte, *pte, ptent; | |
344 | spinlock_t *ptl; | |
07e8c82b VMO |
345 | struct folio *folio = NULL; |
346 | LIST_HEAD(folio_list); | |
fd3b1bc3 | 347 | bool pageout_anon_only_filter; |
d616d512 MK |
348 | |
349 | if (fatal_signal_pending(current)) | |
350 | return -EINTR; | |
9c276cc6 | 351 | |
fd3b1bc3 PK |
352 | pageout_anon_only_filter = pageout && !vma_is_anonymous(vma) && |
353 | !can_do_file_pageout(vma); | |
354 | ||
9c276cc6 MK |
355 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
356 | if (pmd_trans_huge(*pmd)) { | |
357 | pmd_t orig_pmd; | |
358 | unsigned long next = pmd_addr_end(addr, end); | |
359 | ||
360 | tlb_change_page_size(tlb, HPAGE_PMD_SIZE); | |
361 | ptl = pmd_trans_huge_lock(pmd, vma); | |
362 | if (!ptl) | |
363 | return 0; | |
364 | ||
365 | orig_pmd = *pmd; | |
366 | if (is_huge_zero_pmd(orig_pmd)) | |
367 | goto huge_unlock; | |
368 | ||
369 | if (unlikely(!pmd_present(orig_pmd))) { | |
370 | VM_BUG_ON(thp_migration_supported() && | |
371 | !is_pmd_migration_entry(orig_pmd)); | |
372 | goto huge_unlock; | |
373 | } | |
374 | ||
07e8c82b | 375 | folio = pfn_folio(pmd_pfn(orig_pmd)); |
12e967fd | 376 | |
07e8c82b VMO |
377 | /* Do not interfere with other mappings of this folio */ |
378 | if (folio_mapcount(folio) != 1) | |
12e967fd MH |
379 | goto huge_unlock; |
380 | ||
07e8c82b | 381 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 PK |
382 | goto huge_unlock; |
383 | ||
9c276cc6 MK |
384 | if (next - addr != HPAGE_PMD_SIZE) { |
385 | int err; | |
386 | ||
07e8c82b | 387 | folio_get(folio); |
9c276cc6 | 388 | spin_unlock(ptl); |
07e8c82b VMO |
389 | folio_lock(folio); |
390 | err = split_folio(folio); | |
391 | folio_unlock(folio); | |
392 | folio_put(folio); | |
9c276cc6 | 393 | if (!err) |
07e8c82b | 394 | goto regular_folio; |
9c276cc6 MK |
395 | return 0; |
396 | } | |
397 | ||
398 | if (pmd_young(orig_pmd)) { | |
399 | pmdp_invalidate(vma, addr, pmd); | |
400 | orig_pmd = pmd_mkold(orig_pmd); | |
401 | ||
402 | set_pmd_at(mm, addr, pmd, orig_pmd); | |
403 | tlb_remove_pmd_tlb_entry(tlb, pmd, addr); | |
404 | } | |
405 | ||
07e8c82b VMO |
406 | folio_clear_referenced(folio); |
407 | folio_test_clear_young(folio); | |
d616d512 | 408 | if (pageout) { |
be2d5756 | 409 | if (folio_isolate_lru(folio)) { |
07e8c82b VMO |
410 | if (folio_test_unevictable(folio)) |
411 | folio_putback_lru(folio); | |
82072962 | 412 | else |
07e8c82b | 413 | list_add(&folio->lru, &folio_list); |
82072962 | 414 | } |
d616d512 | 415 | } else |
5a9e3474 | 416 | folio_deactivate(folio); |
9c276cc6 MK |
417 | huge_unlock: |
418 | spin_unlock(ptl); | |
d616d512 | 419 | if (pageout) |
07e8c82b | 420 | reclaim_pages(&folio_list); |
9c276cc6 MK |
421 | return 0; |
422 | } | |
423 | ||
07e8c82b | 424 | regular_folio: |
9c276cc6 MK |
425 | if (pmd_trans_unstable(pmd)) |
426 | return 0; | |
9c276cc6 MK |
427 | #endif |
428 | tlb_change_page_size(tlb, PAGE_SIZE); | |
429 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
430 | flush_tlb_batched_pending(mm); | |
431 | arch_enter_lazy_mmu_mode(); | |
432 | for (; addr < end; pte++, addr += PAGE_SIZE) { | |
433 | ptent = *pte; | |
434 | ||
435 | if (pte_none(ptent)) | |
436 | continue; | |
437 | ||
438 | if (!pte_present(ptent)) | |
439 | continue; | |
440 | ||
07e8c82b VMO |
441 | folio = vm_normal_folio(vma, addr, ptent); |
442 | if (!folio || folio_is_zone_device(folio)) | |
9c276cc6 MK |
443 | continue; |
444 | ||
445 | /* | |
446 | * Creating a THP page is expensive so split it only if we | |
447 | * are sure it's worth. Split it if we are only owner. | |
448 | */ | |
07e8c82b VMO |
449 | if (folio_test_large(folio)) { |
450 | if (folio_mapcount(folio) != 1) | |
9c276cc6 | 451 | break; |
07e8c82b | 452 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 | 453 | break; |
07e8c82b VMO |
454 | folio_get(folio); |
455 | if (!folio_trylock(folio)) { | |
456 | folio_put(folio); | |
9c276cc6 MK |
457 | break; |
458 | } | |
459 | pte_unmap_unlock(orig_pte, ptl); | |
07e8c82b VMO |
460 | if (split_folio(folio)) { |
461 | folio_unlock(folio); | |
462 | folio_put(folio); | |
f3b9e8cc | 463 | orig_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
9c276cc6 MK |
464 | break; |
465 | } | |
07e8c82b VMO |
466 | folio_unlock(folio); |
467 | folio_put(folio); | |
f3b9e8cc | 468 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
9c276cc6 MK |
469 | pte--; |
470 | addr -= PAGE_SIZE; | |
471 | continue; | |
472 | } | |
473 | ||
58d426a7 | 474 | /* |
07e8c82b VMO |
475 | * Do not interfere with other mappings of this folio and |
476 | * non-LRU folio. | |
58d426a7 | 477 | */ |
07e8c82b | 478 | if (!folio_test_lru(folio) || folio_mapcount(folio) != 1) |
12e967fd MH |
479 | continue; |
480 | ||
07e8c82b | 481 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 PK |
482 | continue; |
483 | ||
07e8c82b | 484 | VM_BUG_ON_FOLIO(folio_test_large(folio), folio); |
9c276cc6 MK |
485 | |
486 | if (pte_young(ptent)) { | |
487 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
488 | tlb->fullmm); | |
489 | ptent = pte_mkold(ptent); | |
490 | set_pte_at(mm, addr, pte, ptent); | |
491 | tlb_remove_tlb_entry(tlb, pte, addr); | |
492 | } | |
493 | ||
494 | /* | |
07e8c82b VMO |
495 | * We are deactivating a folio for accelerating reclaiming. |
496 | * VM couldn't reclaim the folio unless we clear PG_young. | |
9c276cc6 MK |
497 | * As a side effect, it makes confuse idle-page tracking |
498 | * because they will miss recent referenced history. | |
499 | */ | |
07e8c82b VMO |
500 | folio_clear_referenced(folio); |
501 | folio_test_clear_young(folio); | |
d616d512 | 502 | if (pageout) { |
be2d5756 | 503 | if (folio_isolate_lru(folio)) { |
07e8c82b VMO |
504 | if (folio_test_unevictable(folio)) |
505 | folio_putback_lru(folio); | |
82072962 | 506 | else |
07e8c82b | 507 | list_add(&folio->lru, &folio_list); |
82072962 | 508 | } |
d616d512 | 509 | } else |
5a9e3474 | 510 | folio_deactivate(folio); |
9c276cc6 MK |
511 | } |
512 | ||
513 | arch_leave_lazy_mmu_mode(); | |
514 | pte_unmap_unlock(orig_pte, ptl); | |
d616d512 | 515 | if (pageout) |
07e8c82b | 516 | reclaim_pages(&folio_list); |
9c276cc6 MK |
517 | cond_resched(); |
518 | ||
519 | return 0; | |
520 | } | |
521 | ||
522 | static const struct mm_walk_ops cold_walk_ops = { | |
d616d512 | 523 | .pmd_entry = madvise_cold_or_pageout_pte_range, |
9c276cc6 MK |
524 | }; |
525 | ||
526 | static void madvise_cold_page_range(struct mmu_gather *tlb, | |
527 | struct vm_area_struct *vma, | |
528 | unsigned long addr, unsigned long end) | |
529 | { | |
d616d512 MK |
530 | struct madvise_walk_private walk_private = { |
531 | .pageout = false, | |
532 | .tlb = tlb, | |
533 | }; | |
534 | ||
9c276cc6 | 535 | tlb_start_vma(tlb, vma); |
d616d512 | 536 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); |
9c276cc6 MK |
537 | tlb_end_vma(tlb, vma); |
538 | } | |
539 | ||
a213e5cf HD |
540 | static inline bool can_madv_lru_vma(struct vm_area_struct *vma) |
541 | { | |
9457056a | 542 | return !(vma->vm_flags & (VM_LOCKED|VM_PFNMAP|VM_HUGETLB)); |
a213e5cf HD |
543 | } |
544 | ||
9c276cc6 MK |
545 | static long madvise_cold(struct vm_area_struct *vma, |
546 | struct vm_area_struct **prev, | |
547 | unsigned long start_addr, unsigned long end_addr) | |
548 | { | |
549 | struct mm_struct *mm = vma->vm_mm; | |
550 | struct mmu_gather tlb; | |
551 | ||
552 | *prev = vma; | |
553 | if (!can_madv_lru_vma(vma)) | |
554 | return -EINVAL; | |
555 | ||
556 | lru_add_drain(); | |
a72afd87 | 557 | tlb_gather_mmu(&tlb, mm); |
9c276cc6 | 558 | madvise_cold_page_range(&tlb, vma, start_addr, end_addr); |
ae8eba8b | 559 | tlb_finish_mmu(&tlb); |
9c276cc6 MK |
560 | |
561 | return 0; | |
562 | } | |
563 | ||
1a4e58cc MK |
564 | static void madvise_pageout_page_range(struct mmu_gather *tlb, |
565 | struct vm_area_struct *vma, | |
566 | unsigned long addr, unsigned long end) | |
567 | { | |
d616d512 MK |
568 | struct madvise_walk_private walk_private = { |
569 | .pageout = true, | |
570 | .tlb = tlb, | |
571 | }; | |
572 | ||
1a4e58cc | 573 | tlb_start_vma(tlb, vma); |
d616d512 | 574 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); |
1a4e58cc MK |
575 | tlb_end_vma(tlb, vma); |
576 | } | |
577 | ||
1a4e58cc MK |
578 | static long madvise_pageout(struct vm_area_struct *vma, |
579 | struct vm_area_struct **prev, | |
580 | unsigned long start_addr, unsigned long end_addr) | |
581 | { | |
582 | struct mm_struct *mm = vma->vm_mm; | |
583 | struct mmu_gather tlb; | |
584 | ||
585 | *prev = vma; | |
586 | if (!can_madv_lru_vma(vma)) | |
587 | return -EINVAL; | |
588 | ||
fd3b1bc3 PK |
589 | /* |
590 | * If the VMA belongs to a private file mapping, there can be private | |
591 | * dirty pages which can be paged out if even this process is neither | |
592 | * owner nor write capable of the file. We allow private file mappings | |
593 | * further to pageout dirty anon pages. | |
594 | */ | |
595 | if (!vma_is_anonymous(vma) && (!can_do_file_pageout(vma) && | |
596 | (vma->vm_flags & VM_MAYSHARE))) | |
1a4e58cc MK |
597 | return 0; |
598 | ||
599 | lru_add_drain(); | |
a72afd87 | 600 | tlb_gather_mmu(&tlb, mm); |
1a4e58cc | 601 | madvise_pageout_page_range(&tlb, vma, start_addr, end_addr); |
ae8eba8b | 602 | tlb_finish_mmu(&tlb); |
1a4e58cc MK |
603 | |
604 | return 0; | |
605 | } | |
606 | ||
854e9ed0 MK |
607 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
608 | unsigned long end, struct mm_walk *walk) | |
609 | ||
610 | { | |
611 | struct mmu_gather *tlb = walk->private; | |
612 | struct mm_struct *mm = tlb->mm; | |
613 | struct vm_area_struct *vma = walk->vma; | |
614 | spinlock_t *ptl; | |
615 | pte_t *orig_pte, *pte, ptent; | |
98b211d6 | 616 | struct folio *folio; |
64b42bc1 | 617 | int nr_swap = 0; |
b8d3c4c3 MK |
618 | unsigned long next; |
619 | ||
620 | next = pmd_addr_end(addr, end); | |
621 | if (pmd_trans_huge(*pmd)) | |
622 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
623 | goto next; | |
854e9ed0 | 624 | |
854e9ed0 MK |
625 | if (pmd_trans_unstable(pmd)) |
626 | return 0; | |
627 | ||
ed6a7935 | 628 | tlb_change_page_size(tlb, PAGE_SIZE); |
854e9ed0 | 629 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
3ea27719 | 630 | flush_tlb_batched_pending(mm); |
854e9ed0 MK |
631 | arch_enter_lazy_mmu_mode(); |
632 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
633 | ptent = *pte; | |
634 | ||
64b42bc1 | 635 | if (pte_none(ptent)) |
854e9ed0 | 636 | continue; |
64b42bc1 MK |
637 | /* |
638 | * If the pte has swp_entry, just clear page table to | |
639 | * prevent swap-in which is more expensive rather than | |
640 | * (page allocation + zeroing). | |
641 | */ | |
642 | if (!pte_present(ptent)) { | |
643 | swp_entry_t entry; | |
644 | ||
645 | entry = pte_to_swp_entry(ptent); | |
7b49514f ML |
646 | if (!non_swap_entry(entry)) { |
647 | nr_swap--; | |
648 | free_swap_and_cache(entry); | |
649 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
650 | } else if (is_hwpoison_entry(entry) || | |
651 | is_swapin_error_entry(entry)) { | |
652 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
653 | } | |
64b42bc1 MK |
654 | continue; |
655 | } | |
854e9ed0 | 656 | |
4947ed93 KW |
657 | folio = vm_normal_folio(vma, addr, ptent); |
658 | if (!folio || folio_is_zone_device(folio)) | |
854e9ed0 MK |
659 | continue; |
660 | ||
661 | /* | |
98b211d6 | 662 | * If pmd isn't transhuge but the folio is large and |
854e9ed0 MK |
663 | * is owned by only this process, split it and |
664 | * deactivate all pages. | |
665 | */ | |
98b211d6 MWO |
666 | if (folio_test_large(folio)) { |
667 | if (folio_mapcount(folio) != 1) | |
854e9ed0 | 668 | goto out; |
98b211d6 MWO |
669 | folio_get(folio); |
670 | if (!folio_trylock(folio)) { | |
671 | folio_put(folio); | |
854e9ed0 MK |
672 | goto out; |
673 | } | |
674 | pte_unmap_unlock(orig_pte, ptl); | |
98b211d6 MWO |
675 | if (split_folio(folio)) { |
676 | folio_unlock(folio); | |
677 | folio_put(folio); | |
f3b9e8cc | 678 | orig_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
854e9ed0 MK |
679 | goto out; |
680 | } | |
98b211d6 MWO |
681 | folio_unlock(folio); |
682 | folio_put(folio); | |
f3b9e8cc | 683 | orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
854e9ed0 MK |
684 | pte--; |
685 | addr -= PAGE_SIZE; | |
686 | continue; | |
687 | } | |
688 | ||
98b211d6 MWO |
689 | if (folio_test_swapcache(folio) || folio_test_dirty(folio)) { |
690 | if (!folio_trylock(folio)) | |
854e9ed0 MK |
691 | continue; |
692 | /* | |
98b211d6 MWO |
693 | * If folio is shared with others, we mustn't clear |
694 | * the folio's dirty flag. | |
854e9ed0 | 695 | */ |
98b211d6 MWO |
696 | if (folio_mapcount(folio) != 1) { |
697 | folio_unlock(folio); | |
854e9ed0 MK |
698 | continue; |
699 | } | |
700 | ||
98b211d6 MWO |
701 | if (folio_test_swapcache(folio) && |
702 | !folio_free_swap(folio)) { | |
703 | folio_unlock(folio); | |
854e9ed0 MK |
704 | continue; |
705 | } | |
706 | ||
98b211d6 MWO |
707 | folio_clear_dirty(folio); |
708 | folio_unlock(folio); | |
854e9ed0 MK |
709 | } |
710 | ||
711 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
712 | /* | |
713 | * Some of architecture(ex, PPC) don't update TLB | |
714 | * with set_pte_at and tlb_remove_tlb_entry so for | |
715 | * the portability, remap the pte with old|clean | |
716 | * after pte clearing. | |
717 | */ | |
718 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
719 | tlb->fullmm); | |
720 | ||
721 | ptent = pte_mkold(ptent); | |
722 | ptent = pte_mkclean(ptent); | |
723 | set_pte_at(mm, addr, pte, ptent); | |
724 | tlb_remove_tlb_entry(tlb, pte, addr); | |
725 | } | |
6a6fe9eb | 726 | folio_mark_lazyfree(folio); |
854e9ed0 MK |
727 | } |
728 | out: | |
64b42bc1 MK |
729 | if (nr_swap) { |
730 | if (current->mm == mm) | |
731 | sync_mm_rss(mm); | |
732 | ||
733 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); | |
734 | } | |
854e9ed0 MK |
735 | arch_leave_lazy_mmu_mode(); |
736 | pte_unmap_unlock(orig_pte, ptl); | |
737 | cond_resched(); | |
b8d3c4c3 | 738 | next: |
854e9ed0 MK |
739 | return 0; |
740 | } | |
741 | ||
7b86ac33 CH |
742 | static const struct mm_walk_ops madvise_free_walk_ops = { |
743 | .pmd_entry = madvise_free_pte_range, | |
744 | }; | |
854e9ed0 MK |
745 | |
746 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
747 | unsigned long start_addr, unsigned long end_addr) | |
748 | { | |
854e9ed0 | 749 | struct mm_struct *mm = vma->vm_mm; |
ac46d4f3 | 750 | struct mmu_notifier_range range; |
854e9ed0 MK |
751 | struct mmu_gather tlb; |
752 | ||
854e9ed0 MK |
753 | /* MADV_FREE works for only anon vma at the moment */ |
754 | if (!vma_is_anonymous(vma)) | |
755 | return -EINVAL; | |
756 | ||
ac46d4f3 JG |
757 | range.start = max(vma->vm_start, start_addr); |
758 | if (range.start >= vma->vm_end) | |
854e9ed0 | 759 | return -EINVAL; |
ac46d4f3 JG |
760 | range.end = min(vma->vm_end, end_addr); |
761 | if (range.end <= vma->vm_start) | |
854e9ed0 | 762 | return -EINVAL; |
7d4a8be0 | 763 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, |
6f4f13e8 | 764 | range.start, range.end); |
854e9ed0 MK |
765 | |
766 | lru_add_drain(); | |
a72afd87 | 767 | tlb_gather_mmu(&tlb, mm); |
854e9ed0 MK |
768 | update_hiwater_rss(mm); |
769 | ||
ac46d4f3 | 770 | mmu_notifier_invalidate_range_start(&range); |
7b86ac33 CH |
771 | tlb_start_vma(&tlb, vma); |
772 | walk_page_range(vma->vm_mm, range.start, range.end, | |
773 | &madvise_free_walk_ops, &tlb); | |
774 | tlb_end_vma(&tlb, vma); | |
ac46d4f3 | 775 | mmu_notifier_invalidate_range_end(&range); |
ae8eba8b | 776 | tlb_finish_mmu(&tlb); |
854e9ed0 MK |
777 | |
778 | return 0; | |
779 | } | |
780 | ||
1da177e4 LT |
781 | /* |
782 | * Application no longer needs these pages. If the pages are dirty, | |
783 | * it's OK to just throw them away. The app will be more careful about | |
784 | * data it wants to keep. Be sure to free swap resources too. The | |
21b85b09 MK |
785 | * zap_page_range_single call sets things up for shrink_active_list to actually |
786 | * free these pages later if no one else has touched them in the meantime, | |
1da177e4 | 787 | * although we could add these pages to a global reuse list for |
7e6cbea3 | 788 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
789 | * |
790 | * NB: This interface discards data rather than pushes it out to swap, | |
791 | * as some implementations do. This has performance implications for | |
792 | * applications like large transactional databases which want to discard | |
793 | * pages in anonymous maps after committing to backing store the data | |
794 | * that was kept in them. There is no reason to write this data out to | |
795 | * the swap area if the application is discarding it. | |
796 | * | |
797 | * An interface that causes the system to free clean pages and flush | |
798 | * dirty pages is already available as msync(MS_INVALIDATE). | |
799 | */ | |
230ca982 MR |
800 | static long madvise_dontneed_single_vma(struct vm_area_struct *vma, |
801 | unsigned long start, unsigned long end) | |
802 | { | |
21b85b09 | 803 | zap_page_range_single(vma, start, end - start, NULL); |
230ca982 MR |
804 | return 0; |
805 | } | |
806 | ||
90e7e7f5 MK |
807 | static bool madvise_dontneed_free_valid_vma(struct vm_area_struct *vma, |
808 | unsigned long start, | |
809 | unsigned long *end, | |
810 | int behavior) | |
811 | { | |
9457056a JW |
812 | if (!is_vm_hugetlb_page(vma)) { |
813 | unsigned int forbidden = VM_PFNMAP; | |
814 | ||
815 | if (behavior != MADV_DONTNEED_LOCKED) | |
816 | forbidden |= VM_LOCKED; | |
817 | ||
818 | return !(vma->vm_flags & forbidden); | |
819 | } | |
90e7e7f5 | 820 | |
9457056a | 821 | if (behavior != MADV_DONTNEED && behavior != MADV_DONTNEED_LOCKED) |
90e7e7f5 MK |
822 | return false; |
823 | if (start & ~huge_page_mask(hstate_vma(vma))) | |
824 | return false; | |
825 | ||
8ebe0a5e RR |
826 | /* |
827 | * Madvise callers expect the length to be rounded up to PAGE_SIZE | |
828 | * boundaries, and may be unaware that this VMA uses huge pages. | |
829 | * Avoid unexpected data loss by rounding down the number of | |
830 | * huge pages freed. | |
831 | */ | |
832 | *end = ALIGN_DOWN(*end, huge_page_size(hstate_vma(vma))); | |
833 | ||
90e7e7f5 MK |
834 | return true; |
835 | } | |
836 | ||
230ca982 MR |
837 | static long madvise_dontneed_free(struct vm_area_struct *vma, |
838 | struct vm_area_struct **prev, | |
839 | unsigned long start, unsigned long end, | |
840 | int behavior) | |
1da177e4 | 841 | { |
0726b01e MK |
842 | struct mm_struct *mm = vma->vm_mm; |
843 | ||
05b74384 | 844 | *prev = vma; |
90e7e7f5 | 845 | if (!madvise_dontneed_free_valid_vma(vma, start, &end, behavior)) |
1da177e4 LT |
846 | return -EINVAL; |
847 | ||
8ebe0a5e RR |
848 | if (start == end) |
849 | return 0; | |
850 | ||
70ccb92f | 851 | if (!userfaultfd_remove(vma, start, end)) { |
c1e8d7c6 | 852 | *prev = NULL; /* mmap_lock has been dropped, prev is stale */ |
70ccb92f | 853 | |
0726b01e | 854 | mmap_read_lock(mm); |
92d5df38 | 855 | vma = vma_lookup(mm, start); |
70ccb92f AA |
856 | if (!vma) |
857 | return -ENOMEM; | |
90e7e7f5 MK |
858 | /* |
859 | * Potential end adjustment for hugetlb vma is OK as | |
860 | * the check below keeps end within vma. | |
861 | */ | |
862 | if (!madvise_dontneed_free_valid_vma(vma, start, &end, | |
863 | behavior)) | |
70ccb92f AA |
864 | return -EINVAL; |
865 | if (end > vma->vm_end) { | |
866 | /* | |
867 | * Don't fail if end > vma->vm_end. If the old | |
f0953a1b | 868 | * vma was split while the mmap_lock was |
70ccb92f | 869 | * released the effect of the concurrent |
230ca982 | 870 | * operation may not cause madvise() to |
70ccb92f AA |
871 | * have an undefined result. There may be an |
872 | * adjacent next vma that we'll walk | |
873 | * next. userfaultfd_remove() will generate an | |
874 | * UFFD_EVENT_REMOVE repetition on the | |
875 | * end-vma->vm_end range, but the manager can | |
876 | * handle a repetition fine. | |
877 | */ | |
878 | end = vma->vm_end; | |
879 | } | |
880 | VM_WARN_ON(start >= end); | |
881 | } | |
230ca982 | 882 | |
9457056a | 883 | if (behavior == MADV_DONTNEED || behavior == MADV_DONTNEED_LOCKED) |
230ca982 MR |
884 | return madvise_dontneed_single_vma(vma, start, end); |
885 | else if (behavior == MADV_FREE) | |
886 | return madvise_free_single_vma(vma, start, end); | |
887 | else | |
888 | return -EINVAL; | |
1da177e4 LT |
889 | } |
890 | ||
4ca9b385 DH |
891 | static long madvise_populate(struct vm_area_struct *vma, |
892 | struct vm_area_struct **prev, | |
893 | unsigned long start, unsigned long end, | |
894 | int behavior) | |
895 | { | |
896 | const bool write = behavior == MADV_POPULATE_WRITE; | |
897 | struct mm_struct *mm = vma->vm_mm; | |
898 | unsigned long tmp_end; | |
899 | int locked = 1; | |
900 | long pages; | |
901 | ||
902 | *prev = vma; | |
903 | ||
904 | while (start < end) { | |
905 | /* | |
906 | * We might have temporarily dropped the lock. For example, | |
907 | * our VMA might have been split. | |
908 | */ | |
909 | if (!vma || start >= vma->vm_end) { | |
531037a0 ML |
910 | vma = vma_lookup(mm, start); |
911 | if (!vma) | |
4ca9b385 DH |
912 | return -ENOMEM; |
913 | } | |
914 | ||
915 | tmp_end = min_t(unsigned long, end, vma->vm_end); | |
916 | /* Populate (prefault) page tables readable/writable. */ | |
917 | pages = faultin_vma_page_range(vma, start, tmp_end, write, | |
918 | &locked); | |
919 | if (!locked) { | |
920 | mmap_read_lock(mm); | |
921 | locked = 1; | |
922 | *prev = NULL; | |
923 | vma = NULL; | |
924 | } | |
925 | if (pages < 0) { | |
926 | switch (pages) { | |
927 | case -EINTR: | |
928 | return -EINTR; | |
eb2faa51 | 929 | case -EINVAL: /* Incompatible mappings / permissions. */ |
4ca9b385 DH |
930 | return -EINVAL; |
931 | case -EHWPOISON: | |
932 | return -EHWPOISON; | |
eb2faa51 DH |
933 | case -EFAULT: /* VM_FAULT_SIGBUS or VM_FAULT_SIGSEGV */ |
934 | return -EFAULT; | |
4ca9b385 DH |
935 | default: |
936 | pr_warn_once("%s: unhandled return value: %ld\n", | |
937 | __func__, pages); | |
938 | fallthrough; | |
939 | case -ENOMEM: | |
940 | return -ENOMEM; | |
941 | } | |
942 | } | |
943 | start += pages * PAGE_SIZE; | |
944 | } | |
945 | return 0; | |
946 | } | |
947 | ||
f6b3ec23 BP |
948 | /* |
949 | * Application wants to free up the pages and associated backing store. | |
950 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
951 | */ |
952 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 953 | struct vm_area_struct **prev, |
f6b3ec23 BP |
954 | unsigned long start, unsigned long end) |
955 | { | |
3f31d075 | 956 | loff_t offset; |
90ed52eb | 957 | int error; |
9ab4233d | 958 | struct file *f; |
0726b01e | 959 | struct mm_struct *mm = vma->vm_mm; |
f6b3ec23 | 960 | |
c1e8d7c6 | 961 | *prev = NULL; /* tell sys_madvise we drop mmap_lock */ |
00e9fa2d | 962 | |
72079ba0 | 963 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
964 | return -EINVAL; |
965 | ||
9ab4233d AL |
966 | f = vma->vm_file; |
967 | ||
968 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
969 | return -EINVAL; |
970 | } | |
971 | ||
69cf0fac HD |
972 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
973 | return -EACCES; | |
974 | ||
f6b3ec23 BP |
975 | offset = (loff_t)(start - vma->vm_start) |
976 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 977 | |
9ab4233d | 978 | /* |
9608703e | 979 | * Filesystem's fallocate may need to take i_rwsem. We need to |
9ab4233d AL |
980 | * explicitly grab a reference because the vma (and hence the |
981 | * vma's reference to the file) can go away as soon as we drop | |
c1e8d7c6 | 982 | * mmap_lock. |
9ab4233d AL |
983 | */ |
984 | get_file(f); | |
70ccb92f | 985 | if (userfaultfd_remove(vma, start, end)) { |
c1e8d7c6 | 986 | /* mmap_lock was not released by userfaultfd_remove() */ |
0726b01e | 987 | mmap_read_unlock(mm); |
70ccb92f | 988 | } |
72c72bdf | 989 | error = vfs_fallocate(f, |
3f31d075 HD |
990 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
991 | offset, end - start); | |
9ab4233d | 992 | fput(f); |
0726b01e | 993 | mmap_read_lock(mm); |
90ed52eb | 994 | return error; |
f6b3ec23 BP |
995 | } |
996 | ||
ac1e9acc CC |
997 | /* |
998 | * Apply an madvise behavior to a region of a vma. madvise_update_vma | |
999 | * will handle splitting a vm area into separate areas, each area with its own | |
1000 | * behavior. | |
1001 | */ | |
1002 | static int madvise_vma_behavior(struct vm_area_struct *vma, | |
1003 | struct vm_area_struct **prev, | |
1004 | unsigned long start, unsigned long end, | |
1005 | unsigned long behavior) | |
1006 | { | |
1007 | int error; | |
942341dc | 1008 | struct anon_vma_name *anon_name; |
ac1e9acc CC |
1009 | unsigned long new_flags = vma->vm_flags; |
1010 | ||
1011 | switch (behavior) { | |
1012 | case MADV_REMOVE: | |
1013 | return madvise_remove(vma, prev, start, end); | |
1014 | case MADV_WILLNEED: | |
1015 | return madvise_willneed(vma, prev, start, end); | |
1016 | case MADV_COLD: | |
1017 | return madvise_cold(vma, prev, start, end); | |
1018 | case MADV_PAGEOUT: | |
1019 | return madvise_pageout(vma, prev, start, end); | |
1020 | case MADV_FREE: | |
1021 | case MADV_DONTNEED: | |
9457056a | 1022 | case MADV_DONTNEED_LOCKED: |
ac1e9acc CC |
1023 | return madvise_dontneed_free(vma, prev, start, end, behavior); |
1024 | case MADV_POPULATE_READ: | |
1025 | case MADV_POPULATE_WRITE: | |
1026 | return madvise_populate(vma, prev, start, end, behavior); | |
1027 | case MADV_NORMAL: | |
1028 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
1029 | break; | |
1030 | case MADV_SEQUENTIAL: | |
1031 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; | |
1032 | break; | |
1033 | case MADV_RANDOM: | |
1034 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; | |
1035 | break; | |
1036 | case MADV_DONTFORK: | |
1037 | new_flags |= VM_DONTCOPY; | |
1038 | break; | |
1039 | case MADV_DOFORK: | |
1040 | if (vma->vm_flags & VM_IO) | |
1041 | return -EINVAL; | |
1042 | new_flags &= ~VM_DONTCOPY; | |
1043 | break; | |
1044 | case MADV_WIPEONFORK: | |
1045 | /* MADV_WIPEONFORK is only supported on anonymous memory. */ | |
1046 | if (vma->vm_file || vma->vm_flags & VM_SHARED) | |
1047 | return -EINVAL; | |
1048 | new_flags |= VM_WIPEONFORK; | |
1049 | break; | |
1050 | case MADV_KEEPONFORK: | |
1051 | new_flags &= ~VM_WIPEONFORK; | |
1052 | break; | |
1053 | case MADV_DONTDUMP: | |
1054 | new_flags |= VM_DONTDUMP; | |
1055 | break; | |
1056 | case MADV_DODUMP: | |
1057 | if (!is_vm_hugetlb_page(vma) && new_flags & VM_SPECIAL) | |
1058 | return -EINVAL; | |
1059 | new_flags &= ~VM_DONTDUMP; | |
1060 | break; | |
1061 | case MADV_MERGEABLE: | |
1062 | case MADV_UNMERGEABLE: | |
1063 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
1064 | if (error) | |
1065 | goto out; | |
1066 | break; | |
1067 | case MADV_HUGEPAGE: | |
1068 | case MADV_NOHUGEPAGE: | |
1069 | error = hugepage_madvise(vma, &new_flags, behavior); | |
1070 | if (error) | |
1071 | goto out; | |
1072 | break; | |
7d8faaf1 ZK |
1073 | case MADV_COLLAPSE: |
1074 | return madvise_collapse(vma, prev, start, end); | |
ac1e9acc CC |
1075 | } |
1076 | ||
942341dc SB |
1077 | anon_name = anon_vma_name(vma); |
1078 | anon_vma_name_get(anon_name); | |
9a10064f | 1079 | error = madvise_update_vma(vma, prev, start, end, new_flags, |
942341dc SB |
1080 | anon_name); |
1081 | anon_vma_name_put(anon_name); | |
ac1e9acc CC |
1082 | |
1083 | out: | |
1084 | /* | |
1085 | * madvise() returns EAGAIN if kernel resources, such as | |
1086 | * slab, are temporarily unavailable. | |
1087 | */ | |
1088 | if (error == -ENOMEM) | |
1089 | error = -EAGAIN; | |
1090 | return error; | |
1091 | } | |
1092 | ||
9893e49d AK |
1093 | #ifdef CONFIG_MEMORY_FAILURE |
1094 | /* | |
1095 | * Error injection support for memory error handling. | |
1096 | */ | |
97167a76 AK |
1097 | static int madvise_inject_error(int behavior, |
1098 | unsigned long start, unsigned long end) | |
9893e49d | 1099 | { |
d3cd257c | 1100 | unsigned long size; |
97167a76 | 1101 | |
9893e49d AK |
1102 | if (!capable(CAP_SYS_ADMIN)) |
1103 | return -EPERM; | |
97167a76 | 1104 | |
19bfbe22 | 1105 | |
d3cd257c | 1106 | for (; start < end; start += size) { |
23e7b5c2 | 1107 | unsigned long pfn; |
dc7560b4 | 1108 | struct page *page; |
325c4ef5 AM |
1109 | int ret; |
1110 | ||
97167a76 | 1111 | ret = get_user_pages_fast(start, 1, 0, &page); |
9893e49d AK |
1112 | if (ret != 1) |
1113 | return ret; | |
23e7b5c2 | 1114 | pfn = page_to_pfn(page); |
325c4ef5 | 1115 | |
19bfbe22 AM |
1116 | /* |
1117 | * When soft offlining hugepages, after migrating the page | |
1118 | * we dissolve it, therefore in the second loop "page" will | |
d3cd257c | 1119 | * no longer be a compound page. |
19bfbe22 | 1120 | */ |
d3cd257c | 1121 | size = page_size(compound_head(page)); |
19bfbe22 | 1122 | |
97167a76 AK |
1123 | if (behavior == MADV_SOFT_OFFLINE) { |
1124 | pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n", | |
dc7560b4 | 1125 | pfn, start); |
feec24a6 | 1126 | ret = soft_offline_page(pfn, MF_COUNT_INCREASED); |
dc7560b4 OS |
1127 | } else { |
1128 | pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n", | |
1129 | pfn, start); | |
67f22ba7 | 1130 | ret = memory_failure(pfn, MF_COUNT_INCREASED | MF_SW_SIMULATED); |
d1fe111f | 1131 | if (ret == -EOPNOTSUPP) |
1132 | ret = 0; | |
afcf938e | 1133 | } |
23e7b5c2 | 1134 | |
23a003bf NH |
1135 | if (ret) |
1136 | return ret; | |
9893e49d | 1137 | } |
c461ad6a | 1138 | |
325c4ef5 | 1139 | return 0; |
9893e49d AK |
1140 | } |
1141 | #endif | |
1142 | ||
1ecef9ed | 1143 | static bool |
75927af8 NP |
1144 | madvise_behavior_valid(int behavior) |
1145 | { | |
1146 | switch (behavior) { | |
1147 | case MADV_DOFORK: | |
1148 | case MADV_DONTFORK: | |
1149 | case MADV_NORMAL: | |
1150 | case MADV_SEQUENTIAL: | |
1151 | case MADV_RANDOM: | |
1152 | case MADV_REMOVE: | |
1153 | case MADV_WILLNEED: | |
1154 | case MADV_DONTNEED: | |
9457056a | 1155 | case MADV_DONTNEED_LOCKED: |
854e9ed0 | 1156 | case MADV_FREE: |
9c276cc6 | 1157 | case MADV_COLD: |
1a4e58cc | 1158 | case MADV_PAGEOUT: |
4ca9b385 DH |
1159 | case MADV_POPULATE_READ: |
1160 | case MADV_POPULATE_WRITE: | |
f8af4da3 HD |
1161 | #ifdef CONFIG_KSM |
1162 | case MADV_MERGEABLE: | |
1163 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
1164 | #endif |
1165 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1166 | case MADV_HUGEPAGE: | |
a664b2d8 | 1167 | case MADV_NOHUGEPAGE: |
7d8faaf1 | 1168 | case MADV_COLLAPSE: |
f8af4da3 | 1169 | #endif |
accb61fe JB |
1170 | case MADV_DONTDUMP: |
1171 | case MADV_DODUMP: | |
d2cd9ede RR |
1172 | case MADV_WIPEONFORK: |
1173 | case MADV_KEEPONFORK: | |
5e451be7 AK |
1174 | #ifdef CONFIG_MEMORY_FAILURE |
1175 | case MADV_SOFT_OFFLINE: | |
1176 | case MADV_HWPOISON: | |
1177 | #endif | |
1ecef9ed | 1178 | return true; |
75927af8 NP |
1179 | |
1180 | default: | |
1ecef9ed | 1181 | return false; |
75927af8 NP |
1182 | } |
1183 | } | |
3866ea90 | 1184 | |
876b4a18 | 1185 | static bool process_madvise_behavior_valid(int behavior) |
ecb8ac8b MK |
1186 | { |
1187 | switch (behavior) { | |
1188 | case MADV_COLD: | |
1189 | case MADV_PAGEOUT: | |
d5fffc5a | 1190 | case MADV_WILLNEED: |
876b4a18 | 1191 | case MADV_COLLAPSE: |
ecb8ac8b MK |
1192 | return true; |
1193 | default: | |
1194 | return false; | |
1195 | } | |
1196 | } | |
1197 | ||
ac1e9acc CC |
1198 | /* |
1199 | * Walk the vmas in range [start,end), and call the visit function on each one. | |
1200 | * The visit function will get start and end parameters that cover the overlap | |
1201 | * between the current vma and the original range. Any unmapped regions in the | |
1202 | * original range will result in this function returning -ENOMEM while still | |
1203 | * calling the visit function on all of the existing vmas in the range. | |
1204 | * Must be called with the mmap_lock held for reading or writing. | |
1205 | */ | |
1206 | static | |
1207 | int madvise_walk_vmas(struct mm_struct *mm, unsigned long start, | |
1208 | unsigned long end, unsigned long arg, | |
1209 | int (*visit)(struct vm_area_struct *vma, | |
1210 | struct vm_area_struct **prev, unsigned long start, | |
1211 | unsigned long end, unsigned long arg)) | |
1212 | { | |
1213 | struct vm_area_struct *vma; | |
1214 | struct vm_area_struct *prev; | |
1215 | unsigned long tmp; | |
1216 | int unmapped_error = 0; | |
1217 | ||
1218 | /* | |
1219 | * If the interval [start,end) covers some unmapped address | |
1220 | * ranges, just ignore them, but return -ENOMEM at the end. | |
1221 | * - different from the way of handling in mlock etc. | |
1222 | */ | |
1223 | vma = find_vma_prev(mm, start, &prev); | |
1224 | if (vma && start > vma->vm_start) | |
1225 | prev = vma; | |
1226 | ||
1227 | for (;;) { | |
1228 | int error; | |
1229 | ||
1230 | /* Still start < end. */ | |
1231 | if (!vma) | |
1232 | return -ENOMEM; | |
1233 | ||
1234 | /* Here start < (end|vma->vm_end). */ | |
1235 | if (start < vma->vm_start) { | |
1236 | unmapped_error = -ENOMEM; | |
1237 | start = vma->vm_start; | |
1238 | if (start >= end) | |
1239 | break; | |
1240 | } | |
1241 | ||
1242 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ | |
1243 | tmp = vma->vm_end; | |
1244 | if (end < tmp) | |
1245 | tmp = end; | |
1246 | ||
1247 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ | |
1248 | error = visit(vma, &prev, start, tmp, arg); | |
1249 | if (error) | |
1250 | return error; | |
1251 | start = tmp; | |
1252 | if (prev && start < prev->vm_end) | |
1253 | start = prev->vm_end; | |
1254 | if (start >= end) | |
1255 | break; | |
1256 | if (prev) | |
35474818 | 1257 | vma = find_vma(mm, prev->vm_end); |
ac1e9acc CC |
1258 | else /* madvise_remove dropped mmap_lock */ |
1259 | vma = find_vma(mm, start); | |
1260 | } | |
1261 | ||
1262 | return unmapped_error; | |
1263 | } | |
1264 | ||
9a10064f CC |
1265 | #ifdef CONFIG_ANON_VMA_NAME |
1266 | static int madvise_vma_anon_name(struct vm_area_struct *vma, | |
1267 | struct vm_area_struct **prev, | |
1268 | unsigned long start, unsigned long end, | |
5c26f6ac | 1269 | unsigned long anon_name) |
9a10064f CC |
1270 | { |
1271 | int error; | |
1272 | ||
1273 | /* Only anonymous mappings can be named */ | |
d09e8ca6 | 1274 | if (vma->vm_file && !vma_is_anon_shmem(vma)) |
9a10064f CC |
1275 | return -EBADF; |
1276 | ||
1277 | error = madvise_update_vma(vma, prev, start, end, vma->vm_flags, | |
5c26f6ac | 1278 | (struct anon_vma_name *)anon_name); |
9a10064f CC |
1279 | |
1280 | /* | |
1281 | * madvise() returns EAGAIN if kernel resources, such as | |
1282 | * slab, are temporarily unavailable. | |
1283 | */ | |
1284 | if (error == -ENOMEM) | |
1285 | error = -EAGAIN; | |
1286 | return error; | |
1287 | } | |
1288 | ||
1289 | int madvise_set_anon_name(struct mm_struct *mm, unsigned long start, | |
5c26f6ac | 1290 | unsigned long len_in, struct anon_vma_name *anon_name) |
9a10064f CC |
1291 | { |
1292 | unsigned long end; | |
1293 | unsigned long len; | |
1294 | ||
1295 | if (start & ~PAGE_MASK) | |
1296 | return -EINVAL; | |
1297 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; | |
1298 | ||
1299 | /* Check to see whether len was rounded up from small -ve to zero */ | |
1300 | if (len_in && !len) | |
1301 | return -EINVAL; | |
1302 | ||
1303 | end = start + len; | |
1304 | if (end < start) | |
1305 | return -EINVAL; | |
1306 | ||
1307 | if (end == start) | |
1308 | return 0; | |
1309 | ||
5c26f6ac | 1310 | return madvise_walk_vmas(mm, start, end, (unsigned long)anon_name, |
9a10064f CC |
1311 | madvise_vma_anon_name); |
1312 | } | |
1313 | #endif /* CONFIG_ANON_VMA_NAME */ | |
1da177e4 LT |
1314 | /* |
1315 | * The madvise(2) system call. | |
1316 | * | |
1317 | * Applications can use madvise() to advise the kernel how it should | |
1318 | * handle paging I/O in this VM area. The idea is to help the kernel | |
1319 | * use appropriate read-ahead and caching techniques. The information | |
1320 | * provided is advisory only, and can be safely disregarded by the | |
1321 | * kernel without affecting the correct operation of the application. | |
1322 | * | |
1323 | * behavior values: | |
1324 | * MADV_NORMAL - the default behavior is to read clusters. This | |
1325 | * results in some read-ahead and read-behind. | |
1326 | * MADV_RANDOM - the system should read the minimum amount of data | |
1327 | * on any access, since it is unlikely that the appli- | |
1328 | * cation will need more than what it asks for. | |
1329 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
1330 | * once, so they can be aggressively read ahead, and | |
1331 | * can be freed soon after they are accessed. | |
1332 | * MADV_WILLNEED - the application is notifying the system to read | |
1333 | * some pages ahead. | |
1334 | * MADV_DONTNEED - the application is finished with the given range, | |
1335 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
1336 | * MADV_FREE - the application marks pages in the given range as lazy free, |
1337 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
1338 | * MADV_REMOVE - the application wants to free up the given range of |
1339 | * pages and associated backing store. | |
3866ea90 HD |
1340 | * MADV_DONTFORK - omit this area from child's address space when forking: |
1341 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
1342 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
c02c3009 YS |
1343 | * MADV_WIPEONFORK - present the child process with zero-filled memory in this |
1344 | * range after a fork. | |
1345 | * MADV_KEEPONFORK - undo the effect of MADV_WIPEONFORK | |
d7206a70 NH |
1346 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
1347 | * were corrupted by unrecoverable hardware memory failure. | |
1348 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
1349 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
1350 | * this area with pages of identical content from other such areas. | |
1351 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
1352 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
1353 | * huge pages in the future. Existing pages might be coalesced and | |
1354 | * new pages might be allocated as THP. | |
1355 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
1356 | * transparent huge pages so the existing pages will not be | |
1357 | * coalesced into THP and new pages will not be allocated as THP. | |
7d8faaf1 | 1358 | * MADV_COLLAPSE - synchronously coalesce pages into new THP. |
d7206a70 NH |
1359 | * MADV_DONTDUMP - the application wants to prevent pages in the given range |
1360 | * from being included in its core dump. | |
1361 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
ecb8ac8b MK |
1362 | * MADV_COLD - the application is not expected to use this memory soon, |
1363 | * deactivate pages in this range so that they can be reclaimed | |
f0953a1b | 1364 | * easily if memory pressure happens. |
ecb8ac8b MK |
1365 | * MADV_PAGEOUT - the application is not expected to use this memory soon, |
1366 | * page out the pages in this range immediately. | |
4ca9b385 DH |
1367 | * MADV_POPULATE_READ - populate (prefault) page tables readable by |
1368 | * triggering read faults if required | |
1369 | * MADV_POPULATE_WRITE - populate (prefault) page tables writable by | |
1370 | * triggering write faults if required | |
1da177e4 LT |
1371 | * |
1372 | * return values: | |
1373 | * zero - success | |
1374 | * -EINVAL - start + len < 0, start is not page-aligned, | |
1375 | * "behavior" is not a valid value, or application | |
c02c3009 YS |
1376 | * is attempting to release locked or shared pages, |
1377 | * or the specified address range includes file, Huge TLB, | |
1378 | * MAP_SHARED or VMPFNMAP range. | |
1da177e4 LT |
1379 | * -ENOMEM - addresses in the specified range are not currently |
1380 | * mapped, or are outside the AS of the process. | |
1381 | * -EIO - an I/O error occurred while paging in data. | |
1382 | * -EBADF - map exists, but area maps something that isn't a file. | |
1383 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
1384 | */ | |
0726b01e | 1385 | int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior) |
1da177e4 | 1386 | { |
ac1e9acc CC |
1387 | unsigned long end; |
1388 | int error; | |
f7977793 | 1389 | int write; |
1da177e4 | 1390 | size_t len; |
1998cc04 | 1391 | struct blk_plug plug; |
1da177e4 | 1392 | |
75927af8 | 1393 | if (!madvise_behavior_valid(behavior)) |
ac1e9acc | 1394 | return -EINVAL; |
75927af8 | 1395 | |
df6c6500 | 1396 | if (!PAGE_ALIGNED(start)) |
ac1e9acc | 1397 | return -EINVAL; |
df6c6500 | 1398 | len = PAGE_ALIGN(len_in); |
1da177e4 LT |
1399 | |
1400 | /* Check to see whether len was rounded up from small -ve to zero */ | |
1401 | if (len_in && !len) | |
ac1e9acc | 1402 | return -EINVAL; |
1da177e4 LT |
1403 | |
1404 | end = start + len; | |
1405 | if (end < start) | |
ac1e9acc | 1406 | return -EINVAL; |
1da177e4 | 1407 | |
1da177e4 | 1408 | if (end == start) |
ac1e9acc | 1409 | return 0; |
84d96d89 | 1410 | |
5e451be7 AK |
1411 | #ifdef CONFIG_MEMORY_FAILURE |
1412 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) | |
1413 | return madvise_inject_error(behavior, start, start + len_in); | |
1414 | #endif | |
1415 | ||
84d96d89 | 1416 | write = madvise_need_mmap_write(behavior); |
dc0ef0df | 1417 | if (write) { |
0726b01e | 1418 | if (mmap_write_lock_killable(mm)) |
dc0ef0df MH |
1419 | return -EINTR; |
1420 | } else { | |
0726b01e | 1421 | mmap_read_lock(mm); |
dc0ef0df | 1422 | } |
1da177e4 | 1423 | |
428e106a KS |
1424 | start = untagged_addr_remote(mm, start); |
1425 | end = start + len; | |
1426 | ||
1998cc04 | 1427 | blk_start_plug(&plug); |
ac1e9acc CC |
1428 | error = madvise_walk_vmas(mm, start, end, behavior, |
1429 | madvise_vma_behavior); | |
84d96d89 | 1430 | blk_finish_plug(&plug); |
f7977793 | 1431 | if (write) |
0726b01e | 1432 | mmap_write_unlock(mm); |
0a27a14a | 1433 | else |
0726b01e | 1434 | mmap_read_unlock(mm); |
0a27a14a | 1435 | |
1da177e4 LT |
1436 | return error; |
1437 | } | |
db08ca25 JA |
1438 | |
1439 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) | |
1440 | { | |
0726b01e | 1441 | return do_madvise(current->mm, start, len_in, behavior); |
db08ca25 | 1442 | } |
ecb8ac8b MK |
1443 | |
1444 | SYSCALL_DEFINE5(process_madvise, int, pidfd, const struct iovec __user *, vec, | |
1445 | size_t, vlen, int, behavior, unsigned int, flags) | |
1446 | { | |
1447 | ssize_t ret; | |
95e49cf8 | 1448 | struct iovec iovstack[UIO_FASTIOV]; |
ecb8ac8b MK |
1449 | struct iovec *iov = iovstack; |
1450 | struct iov_iter iter; | |
ecb8ac8b MK |
1451 | struct task_struct *task; |
1452 | struct mm_struct *mm; | |
1453 | size_t total_len; | |
1454 | unsigned int f_flags; | |
1455 | ||
1456 | if (flags != 0) { | |
1457 | ret = -EINVAL; | |
1458 | goto out; | |
1459 | } | |
1460 | ||
de4eda9d | 1461 | ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter); |
ecb8ac8b MK |
1462 | if (ret < 0) |
1463 | goto out; | |
1464 | ||
ee9955d6 CB |
1465 | task = pidfd_get_task(pidfd, &f_flags); |
1466 | if (IS_ERR(task)) { | |
1467 | ret = PTR_ERR(task); | |
ecb8ac8b MK |
1468 | goto free_iov; |
1469 | } | |
1470 | ||
a68a0262 | 1471 | if (!process_madvise_behavior_valid(behavior)) { |
ecb8ac8b MK |
1472 | ret = -EINVAL; |
1473 | goto release_task; | |
1474 | } | |
1475 | ||
96cfe2c0 SB |
1476 | /* Require PTRACE_MODE_READ to avoid leaking ASLR metadata. */ |
1477 | mm = mm_access(task, PTRACE_MODE_READ_FSCREDS); | |
ecb8ac8b MK |
1478 | if (IS_ERR_OR_NULL(mm)) { |
1479 | ret = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH; | |
1480 | goto release_task; | |
1481 | } | |
1482 | ||
96cfe2c0 SB |
1483 | /* |
1484 | * Require CAP_SYS_NICE for influencing process performance. Note that | |
1485 | * only non-destructive hints are currently supported. | |
1486 | */ | |
1487 | if (!capable(CAP_SYS_NICE)) { | |
1488 | ret = -EPERM; | |
1489 | goto release_mm; | |
1490 | } | |
1491 | ||
ecb8ac8b MK |
1492 | total_len = iov_iter_count(&iter); |
1493 | ||
1494 | while (iov_iter_count(&iter)) { | |
95e49cf8 JA |
1495 | ret = do_madvise(mm, (unsigned long)iter_iov_addr(&iter), |
1496 | iter_iov_len(&iter), behavior); | |
e6b0a7b3 | 1497 | if (ret < 0) |
ecb8ac8b | 1498 | break; |
95e49cf8 | 1499 | iov_iter_advance(&iter, iter_iov_len(&iter)); |
ecb8ac8b MK |
1500 | } |
1501 | ||
5bd009c7 | 1502 | ret = (total_len - iov_iter_count(&iter)) ? : ret; |
ecb8ac8b | 1503 | |
96cfe2c0 | 1504 | release_mm: |
ecb8ac8b | 1505 | mmput(mm); |
ecb8ac8b MK |
1506 | release_task: |
1507 | put_task_struct(task); | |
ecb8ac8b MK |
1508 | free_iov: |
1509 | kfree(iov); | |
1510 | out: | |
1511 | return ret; | |
1512 | } |