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; |
85ab779e | 144 | VMA_ITERATOR(vmi, mm, start); |
e798c6e8 | 145 | |
5c26f6ac | 146 | if (new_flags == vma->vm_flags && anon_vma_name_eq(anon_vma_name(vma), anon_name)) { |
05b74384 | 147 | *prev = vma; |
ac1e9acc | 148 | return 0; |
05b74384 PM |
149 | } |
150 | ||
94d7d923 LS |
151 | vma = vma_modify_flags_name(&vmi, *prev, vma, start, end, new_flags, |
152 | anon_name); | |
153 | if (IS_ERR(vma)) | |
154 | return PTR_ERR(vma); | |
05b74384 PM |
155 | |
156 | *prev = vma; | |
1da177e4 | 157 | |
60081bf1 SB |
158 | /* vm_flags is protected by the mmap_lock held in write mode. */ |
159 | vma_start_write(vma); | |
1c71222e | 160 | vm_flags_reset(vma, new_flags); |
d09e8ca6 | 161 | if (!vma->vm_file || vma_is_anon_shmem(vma)) { |
5c26f6ac | 162 | error = replace_anon_vma_name(vma, anon_name); |
9a10064f CC |
163 | if (error) |
164 | return error; | |
165 | } | |
f3bc0dba | 166 | |
ac1e9acc | 167 | return 0; |
1da177e4 LT |
168 | } |
169 | ||
1998cc04 SL |
170 | #ifdef CONFIG_SWAP |
171 | static int swapin_walk_pmd_entry(pmd_t *pmd, unsigned long start, | |
f3cd4ab0 | 172 | unsigned long end, struct mm_walk *walk) |
1998cc04 | 173 | { |
1998cc04 | 174 | struct vm_area_struct *vma = walk->private; |
5169b844 | 175 | struct swap_iocb *splug = NULL; |
f3cd4ab0 HD |
176 | pte_t *ptep = NULL; |
177 | spinlock_t *ptl; | |
178 | unsigned long addr; | |
1998cc04 | 179 | |
f3cd4ab0 | 180 | for (addr = start; addr < end; addr += PAGE_SIZE) { |
1998cc04 SL |
181 | pte_t pte; |
182 | swp_entry_t entry; | |
6e03492e | 183 | struct folio *folio; |
1998cc04 | 184 | |
f3cd4ab0 HD |
185 | if (!ptep++) { |
186 | ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
187 | if (!ptep) | |
188 | break; | |
189 | } | |
1998cc04 | 190 | |
c33c7948 | 191 | pte = ptep_get(ptep); |
f7cc67ae | 192 | if (!is_swap_pte(pte)) |
1998cc04 SL |
193 | continue; |
194 | entry = pte_to_swp_entry(pte); | |
195 | if (unlikely(non_swap_entry(entry))) | |
196 | continue; | |
197 | ||
f3cd4ab0 HD |
198 | pte_unmap_unlock(ptep, ptl); |
199 | ptep = NULL; | |
200 | ||
6e03492e | 201 | folio = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, |
b243dcbf | 202 | vma, addr, &splug); |
6e03492e MWO |
203 | if (folio) |
204 | folio_put(folio); | |
1998cc04 | 205 | } |
f3cd4ab0 HD |
206 | |
207 | if (ptep) | |
208 | pte_unmap_unlock(ptep, ptl); | |
5169b844 | 209 | swap_read_unplug(splug); |
de2e5171 | 210 | cond_resched(); |
1998cc04 SL |
211 | |
212 | return 0; | |
213 | } | |
214 | ||
7b86ac33 CH |
215 | static const struct mm_walk_ops swapin_walk_ops = { |
216 | .pmd_entry = swapin_walk_pmd_entry, | |
49b06385 | 217 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 | 218 | }; |
1998cc04 | 219 | |
179d3e4f | 220 | static void shmem_swapin_range(struct vm_area_struct *vma, |
1998cc04 SL |
221 | unsigned long start, unsigned long end, |
222 | struct address_space *mapping) | |
223 | { | |
e6e88712 | 224 | XA_STATE(xas, &mapping->i_pages, linear_page_index(vma, start)); |
179d3e4f | 225 | pgoff_t end_index = linear_page_index(vma, end) - 1; |
6e03492e | 226 | struct folio *folio; |
5169b844 | 227 | struct swap_iocb *splug = NULL; |
1998cc04 | 228 | |
e6e88712 | 229 | rcu_read_lock(); |
6e03492e | 230 | xas_for_each(&xas, folio, end_index) { |
179d3e4f HD |
231 | unsigned long addr; |
232 | swp_entry_t entry; | |
1998cc04 | 233 | |
6e03492e | 234 | if (!xa_is_value(folio)) |
1998cc04 | 235 | continue; |
6e03492e | 236 | entry = radix_to_swp_entry(folio); |
ba6851b4 | 237 | /* There might be swapin error entries in shmem mapping. */ |
179d3e4f | 238 | if (non_swap_entry(entry)) |
ba6851b4 | 239 | continue; |
179d3e4f HD |
240 | |
241 | addr = vma->vm_start + | |
242 | ((xas.xa_index - vma->vm_pgoff) << PAGE_SHIFT); | |
e6e88712 MWO |
243 | xas_pause(&xas); |
244 | rcu_read_unlock(); | |
245 | ||
6e03492e | 246 | folio = read_swap_cache_async(entry, mapping_gfp_mask(mapping), |
b243dcbf | 247 | vma, addr, &splug); |
6e03492e MWO |
248 | if (folio) |
249 | folio_put(folio); | |
e6e88712 MWO |
250 | |
251 | rcu_read_lock(); | |
1998cc04 | 252 | } |
e6e88712 | 253 | rcu_read_unlock(); |
5169b844 | 254 | swap_read_unplug(splug); |
1998cc04 SL |
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 | { | |
0726b01e | 265 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 266 | struct file *file = vma->vm_file; |
692fe624 | 267 | loff_t offset; |
1da177e4 | 268 | |
6ea8d958 | 269 | *prev = vma; |
1998cc04 | 270 | #ifdef CONFIG_SWAP |
97b713ba | 271 | if (!file) { |
7b86ac33 CH |
272 | walk_page_range(vma->vm_mm, start, end, &swapin_walk_ops, vma); |
273 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
1998cc04 SL |
274 | return 0; |
275 | } | |
1998cc04 | 276 | |
97b713ba | 277 | if (shmem_mapping(file->f_mapping)) { |
179d3e4f HD |
278 | shmem_swapin_range(vma, start, end, file->f_mapping); |
279 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
97b713ba CH |
280 | return 0; |
281 | } | |
282 | #else | |
1bef4003 S |
283 | if (!file) |
284 | return -EBADF; | |
97b713ba | 285 | #endif |
1bef4003 | 286 | |
e748dcd0 | 287 | if (IS_DAX(file_inode(file))) { |
fe77ba6f CO |
288 | /* no bad return value, but ignore advice */ |
289 | return 0; | |
290 | } | |
291 | ||
692fe624 JK |
292 | /* |
293 | * Filesystem's fadvise may need to take various locks. We need to | |
294 | * explicitly grab a reference because the vma (and hence the | |
295 | * vma's reference to the file) can go away as soon as we drop | |
c1e8d7c6 | 296 | * mmap_lock. |
692fe624 | 297 | */ |
c1e8d7c6 | 298 | *prev = NULL; /* tell sys_madvise we drop mmap_lock */ |
692fe624 | 299 | get_file(file); |
692fe624 JK |
300 | offset = (loff_t)(start - vma->vm_start) |
301 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
0726b01e | 302 | mmap_read_unlock(mm); |
692fe624 JK |
303 | vfs_fadvise(file, offset, end - start, POSIX_FADV_WILLNEED); |
304 | fput(file); | |
0726b01e | 305 | mmap_read_lock(mm); |
1da177e4 LT |
306 | return 0; |
307 | } | |
308 | ||
fd3b1bc3 PK |
309 | static inline bool can_do_file_pageout(struct vm_area_struct *vma) |
310 | { | |
311 | if (!vma->vm_file) | |
312 | return false; | |
313 | /* | |
314 | * paging out pagecache only for non-anonymous mappings that correspond | |
315 | * to the files the calling process could (if tried) open for writing; | |
316 | * otherwise we'd be including shared non-exclusive mappings, which | |
317 | * opens a side channel. | |
318 | */ | |
01beba79 | 319 | return inode_owner_or_capable(&nop_mnt_idmap, |
fd3b1bc3 PK |
320 | file_inode(vma->vm_file)) || |
321 | file_permission(vma->vm_file, MAY_WRITE) == 0; | |
322 | } | |
323 | ||
d616d512 MK |
324 | static int madvise_cold_or_pageout_pte_range(pmd_t *pmd, |
325 | unsigned long addr, unsigned long end, | |
326 | struct mm_walk *walk) | |
9c276cc6 | 327 | { |
d616d512 MK |
328 | struct madvise_walk_private *private = walk->private; |
329 | struct mmu_gather *tlb = private->tlb; | |
330 | bool pageout = private->pageout; | |
9c276cc6 MK |
331 | struct mm_struct *mm = tlb->mm; |
332 | struct vm_area_struct *vma = walk->vma; | |
f3cd4ab0 | 333 | pte_t *start_pte, *pte, ptent; |
9c276cc6 | 334 | spinlock_t *ptl; |
07e8c82b VMO |
335 | struct folio *folio = NULL; |
336 | LIST_HEAD(folio_list); | |
fd3b1bc3 | 337 | bool pageout_anon_only_filter; |
b2f557a2 | 338 | unsigned int batch_count = 0; |
d616d512 MK |
339 | |
340 | if (fatal_signal_pending(current)) | |
341 | return -EINTR; | |
9c276cc6 | 342 | |
fd3b1bc3 PK |
343 | pageout_anon_only_filter = pageout && !vma_is_anonymous(vma) && |
344 | !can_do_file_pageout(vma); | |
345 | ||
9c276cc6 MK |
346 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
347 | if (pmd_trans_huge(*pmd)) { | |
348 | pmd_t orig_pmd; | |
349 | unsigned long next = pmd_addr_end(addr, end); | |
350 | ||
351 | tlb_change_page_size(tlb, HPAGE_PMD_SIZE); | |
352 | ptl = pmd_trans_huge_lock(pmd, vma); | |
353 | if (!ptl) | |
354 | return 0; | |
355 | ||
356 | orig_pmd = *pmd; | |
357 | if (is_huge_zero_pmd(orig_pmd)) | |
358 | goto huge_unlock; | |
359 | ||
360 | if (unlikely(!pmd_present(orig_pmd))) { | |
361 | VM_BUG_ON(thp_migration_supported() && | |
362 | !is_pmd_migration_entry(orig_pmd)); | |
363 | goto huge_unlock; | |
364 | } | |
365 | ||
07e8c82b | 366 | folio = pfn_folio(pmd_pfn(orig_pmd)); |
12e967fd | 367 | |
07e8c82b | 368 | /* Do not interfere with other mappings of this folio */ |
2f406263 | 369 | if (folio_estimated_sharers(folio) != 1) |
12e967fd MH |
370 | goto huge_unlock; |
371 | ||
07e8c82b | 372 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 PK |
373 | goto huge_unlock; |
374 | ||
9c276cc6 MK |
375 | if (next - addr != HPAGE_PMD_SIZE) { |
376 | int err; | |
377 | ||
07e8c82b | 378 | folio_get(folio); |
9c276cc6 | 379 | spin_unlock(ptl); |
07e8c82b VMO |
380 | folio_lock(folio); |
381 | err = split_folio(folio); | |
382 | folio_unlock(folio); | |
383 | folio_put(folio); | |
9c276cc6 | 384 | if (!err) |
07e8c82b | 385 | goto regular_folio; |
9c276cc6 MK |
386 | return 0; |
387 | } | |
388 | ||
389 | if (pmd_young(orig_pmd)) { | |
390 | pmdp_invalidate(vma, addr, pmd); | |
391 | orig_pmd = pmd_mkold(orig_pmd); | |
392 | ||
393 | set_pmd_at(mm, addr, pmd, orig_pmd); | |
394 | tlb_remove_pmd_tlb_entry(tlb, pmd, addr); | |
395 | } | |
396 | ||
07e8c82b VMO |
397 | folio_clear_referenced(folio); |
398 | folio_test_clear_young(folio); | |
20c897ea CTK |
399 | if (folio_test_active(folio)) |
400 | folio_set_workingset(folio); | |
d616d512 | 401 | if (pageout) { |
be2d5756 | 402 | if (folio_isolate_lru(folio)) { |
07e8c82b VMO |
403 | if (folio_test_unevictable(folio)) |
404 | folio_putback_lru(folio); | |
82072962 | 405 | else |
07e8c82b | 406 | list_add(&folio->lru, &folio_list); |
82072962 | 407 | } |
d616d512 | 408 | } else |
5a9e3474 | 409 | folio_deactivate(folio); |
9c276cc6 MK |
410 | huge_unlock: |
411 | spin_unlock(ptl); | |
d616d512 | 412 | if (pageout) |
07e8c82b | 413 | reclaim_pages(&folio_list); |
9c276cc6 MK |
414 | return 0; |
415 | } | |
416 | ||
07e8c82b | 417 | regular_folio: |
9c276cc6 MK |
418 | #endif |
419 | tlb_change_page_size(tlb, PAGE_SIZE); | |
b2f557a2 | 420 | restart: |
f3cd4ab0 HD |
421 | start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
422 | if (!start_pte) | |
423 | return 0; | |
9c276cc6 MK |
424 | flush_tlb_batched_pending(mm); |
425 | arch_enter_lazy_mmu_mode(); | |
426 | for (; addr < end; pte++, addr += PAGE_SIZE) { | |
c33c7948 | 427 | ptent = ptep_get(pte); |
9c276cc6 | 428 | |
b2f557a2 JW |
429 | if (++batch_count == SWAP_CLUSTER_MAX) { |
430 | batch_count = 0; | |
431 | if (need_resched()) { | |
4c2da318 | 432 | arch_leave_lazy_mmu_mode(); |
b2f557a2 JW |
433 | pte_unmap_unlock(start_pte, ptl); |
434 | cond_resched(); | |
435 | goto restart; | |
436 | } | |
437 | } | |
438 | ||
9c276cc6 MK |
439 | if (pte_none(ptent)) |
440 | continue; | |
441 | ||
442 | if (!pte_present(ptent)) | |
443 | continue; | |
444 | ||
07e8c82b VMO |
445 | folio = vm_normal_folio(vma, addr, ptent); |
446 | if (!folio || folio_is_zone_device(folio)) | |
9c276cc6 MK |
447 | continue; |
448 | ||
449 | /* | |
450 | * Creating a THP page is expensive so split it only if we | |
451 | * are sure it's worth. Split it if we are only owner. | |
452 | */ | |
07e8c82b | 453 | if (folio_test_large(folio)) { |
f3cd4ab0 HD |
454 | int err; |
455 | ||
2f406263 | 456 | if (folio_estimated_sharers(folio) != 1) |
9c276cc6 | 457 | break; |
07e8c82b | 458 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 | 459 | break; |
f3cd4ab0 | 460 | if (!folio_trylock(folio)) |
9c276cc6 | 461 | break; |
f3cd4ab0 HD |
462 | folio_get(folio); |
463 | arch_leave_lazy_mmu_mode(); | |
464 | pte_unmap_unlock(start_pte, ptl); | |
465 | start_pte = NULL; | |
466 | err = split_folio(folio); | |
07e8c82b VMO |
467 | folio_unlock(folio); |
468 | folio_put(folio); | |
f3cd4ab0 HD |
469 | if (err) |
470 | break; | |
471 | start_pte = pte = | |
472 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
473 | if (!start_pte) | |
474 | break; | |
475 | arch_enter_lazy_mmu_mode(); | |
9c276cc6 MK |
476 | pte--; |
477 | addr -= PAGE_SIZE; | |
478 | continue; | |
479 | } | |
480 | ||
58d426a7 | 481 | /* |
07e8c82b VMO |
482 | * Do not interfere with other mappings of this folio and |
483 | * non-LRU folio. | |
58d426a7 | 484 | */ |
07e8c82b | 485 | if (!folio_test_lru(folio) || folio_mapcount(folio) != 1) |
12e967fd MH |
486 | continue; |
487 | ||
07e8c82b | 488 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 PK |
489 | continue; |
490 | ||
07e8c82b | 491 | VM_BUG_ON_FOLIO(folio_test_large(folio), folio); |
9c276cc6 MK |
492 | |
493 | if (pte_young(ptent)) { | |
494 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
495 | tlb->fullmm); | |
496 | ptent = pte_mkold(ptent); | |
497 | set_pte_at(mm, addr, pte, ptent); | |
498 | tlb_remove_tlb_entry(tlb, pte, addr); | |
499 | } | |
500 | ||
501 | /* | |
07e8c82b VMO |
502 | * We are deactivating a folio for accelerating reclaiming. |
503 | * VM couldn't reclaim the folio unless we clear PG_young. | |
9c276cc6 MK |
504 | * As a side effect, it makes confuse idle-page tracking |
505 | * because they will miss recent referenced history. | |
506 | */ | |
07e8c82b VMO |
507 | folio_clear_referenced(folio); |
508 | folio_test_clear_young(folio); | |
20c897ea CTK |
509 | if (folio_test_active(folio)) |
510 | folio_set_workingset(folio); | |
d616d512 | 511 | if (pageout) { |
be2d5756 | 512 | if (folio_isolate_lru(folio)) { |
07e8c82b VMO |
513 | if (folio_test_unevictable(folio)) |
514 | folio_putback_lru(folio); | |
82072962 | 515 | else |
07e8c82b | 516 | list_add(&folio->lru, &folio_list); |
82072962 | 517 | } |
d616d512 | 518 | } else |
5a9e3474 | 519 | folio_deactivate(folio); |
9c276cc6 MK |
520 | } |
521 | ||
f3cd4ab0 HD |
522 | if (start_pte) { |
523 | arch_leave_lazy_mmu_mode(); | |
524 | pte_unmap_unlock(start_pte, ptl); | |
525 | } | |
d616d512 | 526 | if (pageout) |
07e8c82b | 527 | reclaim_pages(&folio_list); |
9c276cc6 MK |
528 | cond_resched(); |
529 | ||
530 | return 0; | |
531 | } | |
532 | ||
533 | static const struct mm_walk_ops cold_walk_ops = { | |
d616d512 | 534 | .pmd_entry = madvise_cold_or_pageout_pte_range, |
49b06385 | 535 | .walk_lock = PGWALK_RDLOCK, |
9c276cc6 MK |
536 | }; |
537 | ||
538 | static void madvise_cold_page_range(struct mmu_gather *tlb, | |
539 | struct vm_area_struct *vma, | |
540 | unsigned long addr, unsigned long end) | |
541 | { | |
d616d512 MK |
542 | struct madvise_walk_private walk_private = { |
543 | .pageout = false, | |
544 | .tlb = tlb, | |
545 | }; | |
546 | ||
9c276cc6 | 547 | tlb_start_vma(tlb, vma); |
d616d512 | 548 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); |
9c276cc6 MK |
549 | tlb_end_vma(tlb, vma); |
550 | } | |
551 | ||
a213e5cf HD |
552 | static inline bool can_madv_lru_vma(struct vm_area_struct *vma) |
553 | { | |
9457056a | 554 | return !(vma->vm_flags & (VM_LOCKED|VM_PFNMAP|VM_HUGETLB)); |
a213e5cf HD |
555 | } |
556 | ||
9c276cc6 MK |
557 | static long madvise_cold(struct vm_area_struct *vma, |
558 | struct vm_area_struct **prev, | |
559 | unsigned long start_addr, unsigned long end_addr) | |
560 | { | |
561 | struct mm_struct *mm = vma->vm_mm; | |
562 | struct mmu_gather tlb; | |
563 | ||
564 | *prev = vma; | |
565 | if (!can_madv_lru_vma(vma)) | |
566 | return -EINVAL; | |
567 | ||
568 | lru_add_drain(); | |
a72afd87 | 569 | tlb_gather_mmu(&tlb, mm); |
9c276cc6 | 570 | madvise_cold_page_range(&tlb, vma, start_addr, end_addr); |
ae8eba8b | 571 | tlb_finish_mmu(&tlb); |
9c276cc6 MK |
572 | |
573 | return 0; | |
574 | } | |
575 | ||
1a4e58cc MK |
576 | static void madvise_pageout_page_range(struct mmu_gather *tlb, |
577 | struct vm_area_struct *vma, | |
578 | unsigned long addr, unsigned long end) | |
579 | { | |
d616d512 MK |
580 | struct madvise_walk_private walk_private = { |
581 | .pageout = true, | |
582 | .tlb = tlb, | |
583 | }; | |
584 | ||
1a4e58cc | 585 | tlb_start_vma(tlb, vma); |
d616d512 | 586 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); |
1a4e58cc MK |
587 | tlb_end_vma(tlb, vma); |
588 | } | |
589 | ||
1a4e58cc MK |
590 | static long madvise_pageout(struct vm_area_struct *vma, |
591 | struct vm_area_struct **prev, | |
592 | unsigned long start_addr, unsigned long end_addr) | |
593 | { | |
594 | struct mm_struct *mm = vma->vm_mm; | |
595 | struct mmu_gather tlb; | |
596 | ||
597 | *prev = vma; | |
598 | if (!can_madv_lru_vma(vma)) | |
599 | return -EINVAL; | |
600 | ||
fd3b1bc3 PK |
601 | /* |
602 | * If the VMA belongs to a private file mapping, there can be private | |
603 | * dirty pages which can be paged out if even this process is neither | |
604 | * owner nor write capable of the file. We allow private file mappings | |
605 | * further to pageout dirty anon pages. | |
606 | */ | |
607 | if (!vma_is_anonymous(vma) && (!can_do_file_pageout(vma) && | |
608 | (vma->vm_flags & VM_MAYSHARE))) | |
1a4e58cc MK |
609 | return 0; |
610 | ||
611 | lru_add_drain(); | |
a72afd87 | 612 | tlb_gather_mmu(&tlb, mm); |
1a4e58cc | 613 | madvise_pageout_page_range(&tlb, vma, start_addr, end_addr); |
ae8eba8b | 614 | tlb_finish_mmu(&tlb); |
1a4e58cc MK |
615 | |
616 | return 0; | |
617 | } | |
618 | ||
854e9ed0 MK |
619 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
620 | unsigned long end, struct mm_walk *walk) | |
621 | ||
622 | { | |
623 | struct mmu_gather *tlb = walk->private; | |
624 | struct mm_struct *mm = tlb->mm; | |
625 | struct vm_area_struct *vma = walk->vma; | |
626 | spinlock_t *ptl; | |
f3cd4ab0 | 627 | pte_t *start_pte, *pte, ptent; |
98b211d6 | 628 | struct folio *folio; |
64b42bc1 | 629 | int nr_swap = 0; |
b8d3c4c3 MK |
630 | unsigned long next; |
631 | ||
632 | next = pmd_addr_end(addr, end); | |
633 | if (pmd_trans_huge(*pmd)) | |
634 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
f3cd4ab0 | 635 | return 0; |
854e9ed0 | 636 | |
ed6a7935 | 637 | tlb_change_page_size(tlb, PAGE_SIZE); |
f3cd4ab0 HD |
638 | start_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
639 | if (!start_pte) | |
640 | return 0; | |
3ea27719 | 641 | flush_tlb_batched_pending(mm); |
854e9ed0 MK |
642 | arch_enter_lazy_mmu_mode(); |
643 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
c33c7948 | 644 | ptent = ptep_get(pte); |
854e9ed0 | 645 | |
64b42bc1 | 646 | if (pte_none(ptent)) |
854e9ed0 | 647 | continue; |
64b42bc1 MK |
648 | /* |
649 | * If the pte has swp_entry, just clear page table to | |
650 | * prevent swap-in which is more expensive rather than | |
651 | * (page allocation + zeroing). | |
652 | */ | |
653 | if (!pte_present(ptent)) { | |
654 | swp_entry_t entry; | |
655 | ||
656 | entry = pte_to_swp_entry(ptent); | |
7b49514f ML |
657 | if (!non_swap_entry(entry)) { |
658 | nr_swap--; | |
659 | free_swap_and_cache(entry); | |
660 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
661 | } else if (is_hwpoison_entry(entry) || | |
af19487f | 662 | is_poisoned_swp_entry(entry)) { |
7b49514f ML |
663 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
664 | } | |
64b42bc1 MK |
665 | continue; |
666 | } | |
854e9ed0 | 667 | |
4947ed93 KW |
668 | folio = vm_normal_folio(vma, addr, ptent); |
669 | if (!folio || folio_is_zone_device(folio)) | |
854e9ed0 MK |
670 | continue; |
671 | ||
672 | /* | |
98b211d6 | 673 | * If pmd isn't transhuge but the folio is large and |
854e9ed0 MK |
674 | * is owned by only this process, split it and |
675 | * deactivate all pages. | |
676 | */ | |
98b211d6 | 677 | if (folio_test_large(folio)) { |
f3cd4ab0 HD |
678 | int err; |
679 | ||
0e0e9bd5 | 680 | if (folio_estimated_sharers(folio) != 1) |
f3cd4ab0 HD |
681 | break; |
682 | if (!folio_trylock(folio)) | |
683 | break; | |
98b211d6 | 684 | folio_get(folio); |
f3cd4ab0 HD |
685 | arch_leave_lazy_mmu_mode(); |
686 | pte_unmap_unlock(start_pte, ptl); | |
687 | start_pte = NULL; | |
688 | err = split_folio(folio); | |
98b211d6 MWO |
689 | folio_unlock(folio); |
690 | folio_put(folio); | |
f3cd4ab0 HD |
691 | if (err) |
692 | break; | |
693 | start_pte = pte = | |
694 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
695 | if (!start_pte) | |
696 | break; | |
697 | arch_enter_lazy_mmu_mode(); | |
854e9ed0 MK |
698 | pte--; |
699 | addr -= PAGE_SIZE; | |
700 | continue; | |
701 | } | |
702 | ||
98b211d6 MWO |
703 | if (folio_test_swapcache(folio) || folio_test_dirty(folio)) { |
704 | if (!folio_trylock(folio)) | |
854e9ed0 MK |
705 | continue; |
706 | /* | |
98b211d6 MWO |
707 | * If folio is shared with others, we mustn't clear |
708 | * the folio's dirty flag. | |
854e9ed0 | 709 | */ |
98b211d6 MWO |
710 | if (folio_mapcount(folio) != 1) { |
711 | folio_unlock(folio); | |
854e9ed0 MK |
712 | continue; |
713 | } | |
714 | ||
98b211d6 MWO |
715 | if (folio_test_swapcache(folio) && |
716 | !folio_free_swap(folio)) { | |
717 | folio_unlock(folio); | |
854e9ed0 MK |
718 | continue; |
719 | } | |
720 | ||
98b211d6 MWO |
721 | folio_clear_dirty(folio); |
722 | folio_unlock(folio); | |
854e9ed0 MK |
723 | } |
724 | ||
725 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
726 | /* | |
727 | * Some of architecture(ex, PPC) don't update TLB | |
728 | * with set_pte_at and tlb_remove_tlb_entry so for | |
729 | * the portability, remap the pte with old|clean | |
730 | * after pte clearing. | |
731 | */ | |
732 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
733 | tlb->fullmm); | |
734 | ||
735 | ptent = pte_mkold(ptent); | |
736 | ptent = pte_mkclean(ptent); | |
737 | set_pte_at(mm, addr, pte, ptent); | |
738 | tlb_remove_tlb_entry(tlb, pte, addr); | |
739 | } | |
6a6fe9eb | 740 | folio_mark_lazyfree(folio); |
854e9ed0 | 741 | } |
f3cd4ab0 | 742 | |
bc0c3357 | 743 | if (nr_swap) |
64b42bc1 | 744 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); |
f3cd4ab0 HD |
745 | if (start_pte) { |
746 | arch_leave_lazy_mmu_mode(); | |
747 | pte_unmap_unlock(start_pte, ptl); | |
748 | } | |
854e9ed0 | 749 | cond_resched(); |
f3cd4ab0 | 750 | |
854e9ed0 MK |
751 | return 0; |
752 | } | |
753 | ||
7b86ac33 CH |
754 | static const struct mm_walk_ops madvise_free_walk_ops = { |
755 | .pmd_entry = madvise_free_pte_range, | |
49b06385 | 756 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 | 757 | }; |
854e9ed0 MK |
758 | |
759 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
760 | unsigned long start_addr, unsigned long end_addr) | |
761 | { | |
854e9ed0 | 762 | struct mm_struct *mm = vma->vm_mm; |
ac46d4f3 | 763 | struct mmu_notifier_range range; |
854e9ed0 MK |
764 | struct mmu_gather tlb; |
765 | ||
854e9ed0 MK |
766 | /* MADV_FREE works for only anon vma at the moment */ |
767 | if (!vma_is_anonymous(vma)) | |
768 | return -EINVAL; | |
769 | ||
ac46d4f3 JG |
770 | range.start = max(vma->vm_start, start_addr); |
771 | if (range.start >= vma->vm_end) | |
854e9ed0 | 772 | return -EINVAL; |
ac46d4f3 JG |
773 | range.end = min(vma->vm_end, end_addr); |
774 | if (range.end <= vma->vm_start) | |
854e9ed0 | 775 | return -EINVAL; |
7d4a8be0 | 776 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, |
6f4f13e8 | 777 | range.start, range.end); |
854e9ed0 MK |
778 | |
779 | lru_add_drain(); | |
a72afd87 | 780 | tlb_gather_mmu(&tlb, mm); |
854e9ed0 MK |
781 | update_hiwater_rss(mm); |
782 | ||
ac46d4f3 | 783 | mmu_notifier_invalidate_range_start(&range); |
7b86ac33 CH |
784 | tlb_start_vma(&tlb, vma); |
785 | walk_page_range(vma->vm_mm, range.start, range.end, | |
786 | &madvise_free_walk_ops, &tlb); | |
787 | tlb_end_vma(&tlb, vma); | |
ac46d4f3 | 788 | mmu_notifier_invalidate_range_end(&range); |
ae8eba8b | 789 | tlb_finish_mmu(&tlb); |
854e9ed0 MK |
790 | |
791 | return 0; | |
792 | } | |
793 | ||
1da177e4 LT |
794 | /* |
795 | * Application no longer needs these pages. If the pages are dirty, | |
796 | * it's OK to just throw them away. The app will be more careful about | |
797 | * data it wants to keep. Be sure to free swap resources too. The | |
21b85b09 MK |
798 | * zap_page_range_single call sets things up for shrink_active_list to actually |
799 | * free these pages later if no one else has touched them in the meantime, | |
1da177e4 | 800 | * although we could add these pages to a global reuse list for |
7e6cbea3 | 801 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
802 | * |
803 | * NB: This interface discards data rather than pushes it out to swap, | |
804 | * as some implementations do. This has performance implications for | |
805 | * applications like large transactional databases which want to discard | |
806 | * pages in anonymous maps after committing to backing store the data | |
807 | * that was kept in them. There is no reason to write this data out to | |
808 | * the swap area if the application is discarding it. | |
809 | * | |
810 | * An interface that causes the system to free clean pages and flush | |
811 | * dirty pages is already available as msync(MS_INVALIDATE). | |
812 | */ | |
230ca982 MR |
813 | static long madvise_dontneed_single_vma(struct vm_area_struct *vma, |
814 | unsigned long start, unsigned long end) | |
815 | { | |
21b85b09 | 816 | zap_page_range_single(vma, start, end - start, NULL); |
230ca982 MR |
817 | return 0; |
818 | } | |
819 | ||
90e7e7f5 MK |
820 | static bool madvise_dontneed_free_valid_vma(struct vm_area_struct *vma, |
821 | unsigned long start, | |
822 | unsigned long *end, | |
823 | int behavior) | |
824 | { | |
9457056a JW |
825 | if (!is_vm_hugetlb_page(vma)) { |
826 | unsigned int forbidden = VM_PFNMAP; | |
827 | ||
828 | if (behavior != MADV_DONTNEED_LOCKED) | |
829 | forbidden |= VM_LOCKED; | |
830 | ||
831 | return !(vma->vm_flags & forbidden); | |
832 | } | |
90e7e7f5 | 833 | |
9457056a | 834 | if (behavior != MADV_DONTNEED && behavior != MADV_DONTNEED_LOCKED) |
90e7e7f5 MK |
835 | return false; |
836 | if (start & ~huge_page_mask(hstate_vma(vma))) | |
837 | return false; | |
838 | ||
8ebe0a5e RR |
839 | /* |
840 | * Madvise callers expect the length to be rounded up to PAGE_SIZE | |
841 | * boundaries, and may be unaware that this VMA uses huge pages. | |
842 | * Avoid unexpected data loss by rounding down the number of | |
843 | * huge pages freed. | |
844 | */ | |
845 | *end = ALIGN_DOWN(*end, huge_page_size(hstate_vma(vma))); | |
846 | ||
90e7e7f5 MK |
847 | return true; |
848 | } | |
849 | ||
230ca982 MR |
850 | static long madvise_dontneed_free(struct vm_area_struct *vma, |
851 | struct vm_area_struct **prev, | |
852 | unsigned long start, unsigned long end, | |
853 | int behavior) | |
1da177e4 | 854 | { |
0726b01e MK |
855 | struct mm_struct *mm = vma->vm_mm; |
856 | ||
05b74384 | 857 | *prev = vma; |
90e7e7f5 | 858 | if (!madvise_dontneed_free_valid_vma(vma, start, &end, behavior)) |
1da177e4 LT |
859 | return -EINVAL; |
860 | ||
8ebe0a5e RR |
861 | if (start == end) |
862 | return 0; | |
863 | ||
70ccb92f | 864 | if (!userfaultfd_remove(vma, start, end)) { |
c1e8d7c6 | 865 | *prev = NULL; /* mmap_lock has been dropped, prev is stale */ |
70ccb92f | 866 | |
0726b01e | 867 | mmap_read_lock(mm); |
92d5df38 | 868 | vma = vma_lookup(mm, start); |
70ccb92f AA |
869 | if (!vma) |
870 | return -ENOMEM; | |
90e7e7f5 MK |
871 | /* |
872 | * Potential end adjustment for hugetlb vma is OK as | |
873 | * the check below keeps end within vma. | |
874 | */ | |
875 | if (!madvise_dontneed_free_valid_vma(vma, start, &end, | |
876 | behavior)) | |
70ccb92f AA |
877 | return -EINVAL; |
878 | if (end > vma->vm_end) { | |
879 | /* | |
880 | * Don't fail if end > vma->vm_end. If the old | |
f0953a1b | 881 | * vma was split while the mmap_lock was |
70ccb92f | 882 | * released the effect of the concurrent |
230ca982 | 883 | * operation may not cause madvise() to |
70ccb92f AA |
884 | * have an undefined result. There may be an |
885 | * adjacent next vma that we'll walk | |
886 | * next. userfaultfd_remove() will generate an | |
887 | * UFFD_EVENT_REMOVE repetition on the | |
888 | * end-vma->vm_end range, but the manager can | |
889 | * handle a repetition fine. | |
890 | */ | |
891 | end = vma->vm_end; | |
892 | } | |
893 | VM_WARN_ON(start >= end); | |
894 | } | |
230ca982 | 895 | |
9457056a | 896 | if (behavior == MADV_DONTNEED || behavior == MADV_DONTNEED_LOCKED) |
230ca982 MR |
897 | return madvise_dontneed_single_vma(vma, start, end); |
898 | else if (behavior == MADV_FREE) | |
899 | return madvise_free_single_vma(vma, start, end); | |
900 | else | |
901 | return -EINVAL; | |
1da177e4 LT |
902 | } |
903 | ||
4ca9b385 DH |
904 | static long madvise_populate(struct vm_area_struct *vma, |
905 | struct vm_area_struct **prev, | |
906 | unsigned long start, unsigned long end, | |
907 | int behavior) | |
908 | { | |
909 | const bool write = behavior == MADV_POPULATE_WRITE; | |
910 | struct mm_struct *mm = vma->vm_mm; | |
911 | unsigned long tmp_end; | |
912 | int locked = 1; | |
913 | long pages; | |
914 | ||
915 | *prev = vma; | |
916 | ||
917 | while (start < end) { | |
918 | /* | |
919 | * We might have temporarily dropped the lock. For example, | |
920 | * our VMA might have been split. | |
921 | */ | |
922 | if (!vma || start >= vma->vm_end) { | |
531037a0 ML |
923 | vma = vma_lookup(mm, start); |
924 | if (!vma) | |
4ca9b385 DH |
925 | return -ENOMEM; |
926 | } | |
927 | ||
928 | tmp_end = min_t(unsigned long, end, vma->vm_end); | |
929 | /* Populate (prefault) page tables readable/writable. */ | |
930 | pages = faultin_vma_page_range(vma, start, tmp_end, write, | |
931 | &locked); | |
932 | if (!locked) { | |
933 | mmap_read_lock(mm); | |
934 | locked = 1; | |
935 | *prev = NULL; | |
936 | vma = NULL; | |
937 | } | |
938 | if (pages < 0) { | |
939 | switch (pages) { | |
940 | case -EINTR: | |
941 | return -EINTR; | |
eb2faa51 | 942 | case -EINVAL: /* Incompatible mappings / permissions. */ |
4ca9b385 DH |
943 | return -EINVAL; |
944 | case -EHWPOISON: | |
945 | return -EHWPOISON; | |
eb2faa51 DH |
946 | case -EFAULT: /* VM_FAULT_SIGBUS or VM_FAULT_SIGSEGV */ |
947 | return -EFAULT; | |
4ca9b385 DH |
948 | default: |
949 | pr_warn_once("%s: unhandled return value: %ld\n", | |
950 | __func__, pages); | |
951 | fallthrough; | |
952 | case -ENOMEM: | |
953 | return -ENOMEM; | |
954 | } | |
955 | } | |
956 | start += pages * PAGE_SIZE; | |
957 | } | |
958 | return 0; | |
959 | } | |
960 | ||
f6b3ec23 BP |
961 | /* |
962 | * Application wants to free up the pages and associated backing store. | |
963 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
964 | */ |
965 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 966 | struct vm_area_struct **prev, |
f6b3ec23 BP |
967 | unsigned long start, unsigned long end) |
968 | { | |
3f31d075 | 969 | loff_t offset; |
90ed52eb | 970 | int error; |
9ab4233d | 971 | struct file *f; |
0726b01e | 972 | struct mm_struct *mm = vma->vm_mm; |
f6b3ec23 | 973 | |
c1e8d7c6 | 974 | *prev = NULL; /* tell sys_madvise we drop mmap_lock */ |
00e9fa2d | 975 | |
72079ba0 | 976 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
977 | return -EINVAL; |
978 | ||
9ab4233d AL |
979 | f = vma->vm_file; |
980 | ||
981 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
982 | return -EINVAL; |
983 | } | |
984 | ||
e8e17ee9 | 985 | if (!vma_is_shared_maywrite(vma)) |
69cf0fac HD |
986 | return -EACCES; |
987 | ||
f6b3ec23 BP |
988 | offset = (loff_t)(start - vma->vm_start) |
989 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 990 | |
9ab4233d | 991 | /* |
9608703e | 992 | * Filesystem's fallocate may need to take i_rwsem. We need to |
9ab4233d AL |
993 | * explicitly grab a reference because the vma (and hence the |
994 | * vma's reference to the file) can go away as soon as we drop | |
c1e8d7c6 | 995 | * mmap_lock. |
9ab4233d AL |
996 | */ |
997 | get_file(f); | |
70ccb92f | 998 | if (userfaultfd_remove(vma, start, end)) { |
c1e8d7c6 | 999 | /* mmap_lock was not released by userfaultfd_remove() */ |
0726b01e | 1000 | mmap_read_unlock(mm); |
70ccb92f | 1001 | } |
72c72bdf | 1002 | error = vfs_fallocate(f, |
3f31d075 HD |
1003 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1004 | offset, end - start); | |
9ab4233d | 1005 | fput(f); |
0726b01e | 1006 | mmap_read_lock(mm); |
90ed52eb | 1007 | return error; |
f6b3ec23 BP |
1008 | } |
1009 | ||
ac1e9acc CC |
1010 | /* |
1011 | * Apply an madvise behavior to a region of a vma. madvise_update_vma | |
1012 | * will handle splitting a vm area into separate areas, each area with its own | |
1013 | * behavior. | |
1014 | */ | |
1015 | static int madvise_vma_behavior(struct vm_area_struct *vma, | |
1016 | struct vm_area_struct **prev, | |
1017 | unsigned long start, unsigned long end, | |
1018 | unsigned long behavior) | |
1019 | { | |
1020 | int error; | |
942341dc | 1021 | struct anon_vma_name *anon_name; |
ac1e9acc CC |
1022 | unsigned long new_flags = vma->vm_flags; |
1023 | ||
1024 | switch (behavior) { | |
1025 | case MADV_REMOVE: | |
1026 | return madvise_remove(vma, prev, start, end); | |
1027 | case MADV_WILLNEED: | |
1028 | return madvise_willneed(vma, prev, start, end); | |
1029 | case MADV_COLD: | |
1030 | return madvise_cold(vma, prev, start, end); | |
1031 | case MADV_PAGEOUT: | |
1032 | return madvise_pageout(vma, prev, start, end); | |
1033 | case MADV_FREE: | |
1034 | case MADV_DONTNEED: | |
9457056a | 1035 | case MADV_DONTNEED_LOCKED: |
ac1e9acc CC |
1036 | return madvise_dontneed_free(vma, prev, start, end, behavior); |
1037 | case MADV_POPULATE_READ: | |
1038 | case MADV_POPULATE_WRITE: | |
1039 | return madvise_populate(vma, prev, start, end, behavior); | |
1040 | case MADV_NORMAL: | |
1041 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
1042 | break; | |
1043 | case MADV_SEQUENTIAL: | |
1044 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; | |
1045 | break; | |
1046 | case MADV_RANDOM: | |
1047 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; | |
1048 | break; | |
1049 | case MADV_DONTFORK: | |
1050 | new_flags |= VM_DONTCOPY; | |
1051 | break; | |
1052 | case MADV_DOFORK: | |
1053 | if (vma->vm_flags & VM_IO) | |
1054 | return -EINVAL; | |
1055 | new_flags &= ~VM_DONTCOPY; | |
1056 | break; | |
1057 | case MADV_WIPEONFORK: | |
1058 | /* MADV_WIPEONFORK is only supported on anonymous memory. */ | |
1059 | if (vma->vm_file || vma->vm_flags & VM_SHARED) | |
1060 | return -EINVAL; | |
1061 | new_flags |= VM_WIPEONFORK; | |
1062 | break; | |
1063 | case MADV_KEEPONFORK: | |
1064 | new_flags &= ~VM_WIPEONFORK; | |
1065 | break; | |
1066 | case MADV_DONTDUMP: | |
1067 | new_flags |= VM_DONTDUMP; | |
1068 | break; | |
1069 | case MADV_DODUMP: | |
1070 | if (!is_vm_hugetlb_page(vma) && new_flags & VM_SPECIAL) | |
1071 | return -EINVAL; | |
1072 | new_flags &= ~VM_DONTDUMP; | |
1073 | break; | |
1074 | case MADV_MERGEABLE: | |
1075 | case MADV_UNMERGEABLE: | |
1076 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
1077 | if (error) | |
1078 | goto out; | |
1079 | break; | |
1080 | case MADV_HUGEPAGE: | |
1081 | case MADV_NOHUGEPAGE: | |
1082 | error = hugepage_madvise(vma, &new_flags, behavior); | |
1083 | if (error) | |
1084 | goto out; | |
1085 | break; | |
7d8faaf1 ZK |
1086 | case MADV_COLLAPSE: |
1087 | return madvise_collapse(vma, prev, start, end); | |
ac1e9acc CC |
1088 | } |
1089 | ||
942341dc SB |
1090 | anon_name = anon_vma_name(vma); |
1091 | anon_vma_name_get(anon_name); | |
9a10064f | 1092 | error = madvise_update_vma(vma, prev, start, end, new_flags, |
942341dc SB |
1093 | anon_name); |
1094 | anon_vma_name_put(anon_name); | |
ac1e9acc CC |
1095 | |
1096 | out: | |
1097 | /* | |
1098 | * madvise() returns EAGAIN if kernel resources, such as | |
1099 | * slab, are temporarily unavailable. | |
1100 | */ | |
1101 | if (error == -ENOMEM) | |
1102 | error = -EAGAIN; | |
1103 | return error; | |
1104 | } | |
1105 | ||
9893e49d AK |
1106 | #ifdef CONFIG_MEMORY_FAILURE |
1107 | /* | |
1108 | * Error injection support for memory error handling. | |
1109 | */ | |
97167a76 AK |
1110 | static int madvise_inject_error(int behavior, |
1111 | unsigned long start, unsigned long end) | |
9893e49d | 1112 | { |
d3cd257c | 1113 | unsigned long size; |
97167a76 | 1114 | |
9893e49d AK |
1115 | if (!capable(CAP_SYS_ADMIN)) |
1116 | return -EPERM; | |
97167a76 | 1117 | |
19bfbe22 | 1118 | |
d3cd257c | 1119 | for (; start < end; start += size) { |
23e7b5c2 | 1120 | unsigned long pfn; |
dc7560b4 | 1121 | struct page *page; |
325c4ef5 AM |
1122 | int ret; |
1123 | ||
97167a76 | 1124 | ret = get_user_pages_fast(start, 1, 0, &page); |
9893e49d AK |
1125 | if (ret != 1) |
1126 | return ret; | |
23e7b5c2 | 1127 | pfn = page_to_pfn(page); |
325c4ef5 | 1128 | |
19bfbe22 AM |
1129 | /* |
1130 | * When soft offlining hugepages, after migrating the page | |
1131 | * we dissolve it, therefore in the second loop "page" will | |
d3cd257c | 1132 | * no longer be a compound page. |
19bfbe22 | 1133 | */ |
d3cd257c | 1134 | size = page_size(compound_head(page)); |
19bfbe22 | 1135 | |
97167a76 AK |
1136 | if (behavior == MADV_SOFT_OFFLINE) { |
1137 | pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n", | |
dc7560b4 | 1138 | pfn, start); |
feec24a6 | 1139 | ret = soft_offline_page(pfn, MF_COUNT_INCREASED); |
dc7560b4 OS |
1140 | } else { |
1141 | pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n", | |
1142 | pfn, start); | |
67f22ba7 | 1143 | ret = memory_failure(pfn, MF_COUNT_INCREASED | MF_SW_SIMULATED); |
d1fe111f | 1144 | if (ret == -EOPNOTSUPP) |
1145 | ret = 0; | |
afcf938e | 1146 | } |
23e7b5c2 | 1147 | |
23a003bf NH |
1148 | if (ret) |
1149 | return ret; | |
9893e49d | 1150 | } |
c461ad6a | 1151 | |
325c4ef5 | 1152 | return 0; |
9893e49d AK |
1153 | } |
1154 | #endif | |
1155 | ||
1ecef9ed | 1156 | static bool |
75927af8 NP |
1157 | madvise_behavior_valid(int behavior) |
1158 | { | |
1159 | switch (behavior) { | |
1160 | case MADV_DOFORK: | |
1161 | case MADV_DONTFORK: | |
1162 | case MADV_NORMAL: | |
1163 | case MADV_SEQUENTIAL: | |
1164 | case MADV_RANDOM: | |
1165 | case MADV_REMOVE: | |
1166 | case MADV_WILLNEED: | |
1167 | case MADV_DONTNEED: | |
9457056a | 1168 | case MADV_DONTNEED_LOCKED: |
854e9ed0 | 1169 | case MADV_FREE: |
9c276cc6 | 1170 | case MADV_COLD: |
1a4e58cc | 1171 | case MADV_PAGEOUT: |
4ca9b385 DH |
1172 | case MADV_POPULATE_READ: |
1173 | case MADV_POPULATE_WRITE: | |
f8af4da3 HD |
1174 | #ifdef CONFIG_KSM |
1175 | case MADV_MERGEABLE: | |
1176 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
1177 | #endif |
1178 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1179 | case MADV_HUGEPAGE: | |
a664b2d8 | 1180 | case MADV_NOHUGEPAGE: |
7d8faaf1 | 1181 | case MADV_COLLAPSE: |
f8af4da3 | 1182 | #endif |
accb61fe JB |
1183 | case MADV_DONTDUMP: |
1184 | case MADV_DODUMP: | |
d2cd9ede RR |
1185 | case MADV_WIPEONFORK: |
1186 | case MADV_KEEPONFORK: | |
5e451be7 AK |
1187 | #ifdef CONFIG_MEMORY_FAILURE |
1188 | case MADV_SOFT_OFFLINE: | |
1189 | case MADV_HWPOISON: | |
1190 | #endif | |
1ecef9ed | 1191 | return true; |
75927af8 NP |
1192 | |
1193 | default: | |
1ecef9ed | 1194 | return false; |
75927af8 NP |
1195 | } |
1196 | } | |
3866ea90 | 1197 | |
876b4a18 | 1198 | static bool process_madvise_behavior_valid(int behavior) |
ecb8ac8b MK |
1199 | { |
1200 | switch (behavior) { | |
1201 | case MADV_COLD: | |
1202 | case MADV_PAGEOUT: | |
d5fffc5a | 1203 | case MADV_WILLNEED: |
876b4a18 | 1204 | case MADV_COLLAPSE: |
ecb8ac8b MK |
1205 | return true; |
1206 | default: | |
1207 | return false; | |
1208 | } | |
1209 | } | |
1210 | ||
ac1e9acc CC |
1211 | /* |
1212 | * Walk the vmas in range [start,end), and call the visit function on each one. | |
1213 | * The visit function will get start and end parameters that cover the overlap | |
1214 | * between the current vma and the original range. Any unmapped regions in the | |
1215 | * original range will result in this function returning -ENOMEM while still | |
1216 | * calling the visit function on all of the existing vmas in the range. | |
1217 | * Must be called with the mmap_lock held for reading or writing. | |
1218 | */ | |
1219 | static | |
1220 | int madvise_walk_vmas(struct mm_struct *mm, unsigned long start, | |
1221 | unsigned long end, unsigned long arg, | |
1222 | int (*visit)(struct vm_area_struct *vma, | |
1223 | struct vm_area_struct **prev, unsigned long start, | |
1224 | unsigned long end, unsigned long arg)) | |
1225 | { | |
1226 | struct vm_area_struct *vma; | |
1227 | struct vm_area_struct *prev; | |
1228 | unsigned long tmp; | |
1229 | int unmapped_error = 0; | |
1230 | ||
1231 | /* | |
1232 | * If the interval [start,end) covers some unmapped address | |
1233 | * ranges, just ignore them, but return -ENOMEM at the end. | |
1234 | * - different from the way of handling in mlock etc. | |
1235 | */ | |
1236 | vma = find_vma_prev(mm, start, &prev); | |
1237 | if (vma && start > vma->vm_start) | |
1238 | prev = vma; | |
1239 | ||
1240 | for (;;) { | |
1241 | int error; | |
1242 | ||
1243 | /* Still start < end. */ | |
1244 | if (!vma) | |
1245 | return -ENOMEM; | |
1246 | ||
1247 | /* Here start < (end|vma->vm_end). */ | |
1248 | if (start < vma->vm_start) { | |
1249 | unmapped_error = -ENOMEM; | |
1250 | start = vma->vm_start; | |
1251 | if (start >= end) | |
1252 | break; | |
1253 | } | |
1254 | ||
1255 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ | |
1256 | tmp = vma->vm_end; | |
1257 | if (end < tmp) | |
1258 | tmp = end; | |
1259 | ||
1260 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ | |
1261 | error = visit(vma, &prev, start, tmp, arg); | |
1262 | if (error) | |
1263 | return error; | |
1264 | start = tmp; | |
1265 | if (prev && start < prev->vm_end) | |
1266 | start = prev->vm_end; | |
1267 | if (start >= end) | |
1268 | break; | |
1269 | if (prev) | |
35474818 | 1270 | vma = find_vma(mm, prev->vm_end); |
ac1e9acc CC |
1271 | else /* madvise_remove dropped mmap_lock */ |
1272 | vma = find_vma(mm, start); | |
1273 | } | |
1274 | ||
1275 | return unmapped_error; | |
1276 | } | |
1277 | ||
9a10064f CC |
1278 | #ifdef CONFIG_ANON_VMA_NAME |
1279 | static int madvise_vma_anon_name(struct vm_area_struct *vma, | |
1280 | struct vm_area_struct **prev, | |
1281 | unsigned long start, unsigned long end, | |
5c26f6ac | 1282 | unsigned long anon_name) |
9a10064f CC |
1283 | { |
1284 | int error; | |
1285 | ||
1286 | /* Only anonymous mappings can be named */ | |
d09e8ca6 | 1287 | if (vma->vm_file && !vma_is_anon_shmem(vma)) |
9a10064f CC |
1288 | return -EBADF; |
1289 | ||
1290 | error = madvise_update_vma(vma, prev, start, end, vma->vm_flags, | |
5c26f6ac | 1291 | (struct anon_vma_name *)anon_name); |
9a10064f CC |
1292 | |
1293 | /* | |
1294 | * madvise() returns EAGAIN if kernel resources, such as | |
1295 | * slab, are temporarily unavailable. | |
1296 | */ | |
1297 | if (error == -ENOMEM) | |
1298 | error = -EAGAIN; | |
1299 | return error; | |
1300 | } | |
1301 | ||
1302 | int madvise_set_anon_name(struct mm_struct *mm, unsigned long start, | |
5c26f6ac | 1303 | unsigned long len_in, struct anon_vma_name *anon_name) |
9a10064f CC |
1304 | { |
1305 | unsigned long end; | |
1306 | unsigned long len; | |
1307 | ||
1308 | if (start & ~PAGE_MASK) | |
1309 | return -EINVAL; | |
1310 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; | |
1311 | ||
1312 | /* Check to see whether len was rounded up from small -ve to zero */ | |
1313 | if (len_in && !len) | |
1314 | return -EINVAL; | |
1315 | ||
1316 | end = start + len; | |
1317 | if (end < start) | |
1318 | return -EINVAL; | |
1319 | ||
1320 | if (end == start) | |
1321 | return 0; | |
1322 | ||
5c26f6ac | 1323 | return madvise_walk_vmas(mm, start, end, (unsigned long)anon_name, |
9a10064f CC |
1324 | madvise_vma_anon_name); |
1325 | } | |
1326 | #endif /* CONFIG_ANON_VMA_NAME */ | |
1da177e4 LT |
1327 | /* |
1328 | * The madvise(2) system call. | |
1329 | * | |
1330 | * Applications can use madvise() to advise the kernel how it should | |
1331 | * handle paging I/O in this VM area. The idea is to help the kernel | |
1332 | * use appropriate read-ahead and caching techniques. The information | |
1333 | * provided is advisory only, and can be safely disregarded by the | |
1334 | * kernel without affecting the correct operation of the application. | |
1335 | * | |
1336 | * behavior values: | |
1337 | * MADV_NORMAL - the default behavior is to read clusters. This | |
1338 | * results in some read-ahead and read-behind. | |
1339 | * MADV_RANDOM - the system should read the minimum amount of data | |
1340 | * on any access, since it is unlikely that the appli- | |
1341 | * cation will need more than what it asks for. | |
1342 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
1343 | * once, so they can be aggressively read ahead, and | |
1344 | * can be freed soon after they are accessed. | |
1345 | * MADV_WILLNEED - the application is notifying the system to read | |
1346 | * some pages ahead. | |
1347 | * MADV_DONTNEED - the application is finished with the given range, | |
1348 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
1349 | * MADV_FREE - the application marks pages in the given range as lazy free, |
1350 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
1351 | * MADV_REMOVE - the application wants to free up the given range of |
1352 | * pages and associated backing store. | |
3866ea90 HD |
1353 | * MADV_DONTFORK - omit this area from child's address space when forking: |
1354 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
1355 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
c02c3009 YS |
1356 | * MADV_WIPEONFORK - present the child process with zero-filled memory in this |
1357 | * range after a fork. | |
1358 | * MADV_KEEPONFORK - undo the effect of MADV_WIPEONFORK | |
d7206a70 NH |
1359 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
1360 | * were corrupted by unrecoverable hardware memory failure. | |
1361 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
1362 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
1363 | * this area with pages of identical content from other such areas. | |
1364 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
1365 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
1366 | * huge pages in the future. Existing pages might be coalesced and | |
1367 | * new pages might be allocated as THP. | |
1368 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
1369 | * transparent huge pages so the existing pages will not be | |
1370 | * coalesced into THP and new pages will not be allocated as THP. | |
7d8faaf1 | 1371 | * MADV_COLLAPSE - synchronously coalesce pages into new THP. |
d7206a70 NH |
1372 | * MADV_DONTDUMP - the application wants to prevent pages in the given range |
1373 | * from being included in its core dump. | |
1374 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
ecb8ac8b MK |
1375 | * MADV_COLD - the application is not expected to use this memory soon, |
1376 | * deactivate pages in this range so that they can be reclaimed | |
f0953a1b | 1377 | * easily if memory pressure happens. |
ecb8ac8b MK |
1378 | * MADV_PAGEOUT - the application is not expected to use this memory soon, |
1379 | * page out the pages in this range immediately. | |
4ca9b385 DH |
1380 | * MADV_POPULATE_READ - populate (prefault) page tables readable by |
1381 | * triggering read faults if required | |
1382 | * MADV_POPULATE_WRITE - populate (prefault) page tables writable by | |
1383 | * triggering write faults if required | |
1da177e4 LT |
1384 | * |
1385 | * return values: | |
1386 | * zero - success | |
1387 | * -EINVAL - start + len < 0, start is not page-aligned, | |
1388 | * "behavior" is not a valid value, or application | |
c02c3009 YS |
1389 | * is attempting to release locked or shared pages, |
1390 | * or the specified address range includes file, Huge TLB, | |
1391 | * MAP_SHARED or VMPFNMAP range. | |
1da177e4 LT |
1392 | * -ENOMEM - addresses in the specified range are not currently |
1393 | * mapped, or are outside the AS of the process. | |
1394 | * -EIO - an I/O error occurred while paging in data. | |
1395 | * -EBADF - map exists, but area maps something that isn't a file. | |
1396 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
1397 | */ | |
0726b01e | 1398 | int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior) |
1da177e4 | 1399 | { |
ac1e9acc CC |
1400 | unsigned long end; |
1401 | int error; | |
f7977793 | 1402 | int write; |
1da177e4 | 1403 | size_t len; |
1998cc04 | 1404 | struct blk_plug plug; |
1da177e4 | 1405 | |
75927af8 | 1406 | if (!madvise_behavior_valid(behavior)) |
ac1e9acc | 1407 | return -EINVAL; |
75927af8 | 1408 | |
df6c6500 | 1409 | if (!PAGE_ALIGNED(start)) |
ac1e9acc | 1410 | return -EINVAL; |
df6c6500 | 1411 | len = PAGE_ALIGN(len_in); |
1da177e4 LT |
1412 | |
1413 | /* Check to see whether len was rounded up from small -ve to zero */ | |
1414 | if (len_in && !len) | |
ac1e9acc | 1415 | return -EINVAL; |
1da177e4 LT |
1416 | |
1417 | end = start + len; | |
1418 | if (end < start) | |
ac1e9acc | 1419 | return -EINVAL; |
1da177e4 | 1420 | |
1da177e4 | 1421 | if (end == start) |
ac1e9acc | 1422 | return 0; |
84d96d89 | 1423 | |
5e451be7 AK |
1424 | #ifdef CONFIG_MEMORY_FAILURE |
1425 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) | |
1426 | return madvise_inject_error(behavior, start, start + len_in); | |
1427 | #endif | |
1428 | ||
84d96d89 | 1429 | write = madvise_need_mmap_write(behavior); |
dc0ef0df | 1430 | if (write) { |
0726b01e | 1431 | if (mmap_write_lock_killable(mm)) |
dc0ef0df MH |
1432 | return -EINTR; |
1433 | } else { | |
0726b01e | 1434 | mmap_read_lock(mm); |
dc0ef0df | 1435 | } |
1da177e4 | 1436 | |
428e106a KS |
1437 | start = untagged_addr_remote(mm, start); |
1438 | end = start + len; | |
1439 | ||
1998cc04 | 1440 | blk_start_plug(&plug); |
ac1e9acc CC |
1441 | error = madvise_walk_vmas(mm, start, end, behavior, |
1442 | madvise_vma_behavior); | |
84d96d89 | 1443 | blk_finish_plug(&plug); |
f7977793 | 1444 | if (write) |
0726b01e | 1445 | mmap_write_unlock(mm); |
0a27a14a | 1446 | else |
0726b01e | 1447 | mmap_read_unlock(mm); |
0a27a14a | 1448 | |
1da177e4 LT |
1449 | return error; |
1450 | } | |
db08ca25 JA |
1451 | |
1452 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) | |
1453 | { | |
0726b01e | 1454 | return do_madvise(current->mm, start, len_in, behavior); |
db08ca25 | 1455 | } |
ecb8ac8b MK |
1456 | |
1457 | SYSCALL_DEFINE5(process_madvise, int, pidfd, const struct iovec __user *, vec, | |
1458 | size_t, vlen, int, behavior, unsigned int, flags) | |
1459 | { | |
1460 | ssize_t ret; | |
95e49cf8 | 1461 | struct iovec iovstack[UIO_FASTIOV]; |
ecb8ac8b MK |
1462 | struct iovec *iov = iovstack; |
1463 | struct iov_iter iter; | |
ecb8ac8b MK |
1464 | struct task_struct *task; |
1465 | struct mm_struct *mm; | |
1466 | size_t total_len; | |
1467 | unsigned int f_flags; | |
1468 | ||
1469 | if (flags != 0) { | |
1470 | ret = -EINVAL; | |
1471 | goto out; | |
1472 | } | |
1473 | ||
de4eda9d | 1474 | ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter); |
ecb8ac8b MK |
1475 | if (ret < 0) |
1476 | goto out; | |
1477 | ||
ee9955d6 CB |
1478 | task = pidfd_get_task(pidfd, &f_flags); |
1479 | if (IS_ERR(task)) { | |
1480 | ret = PTR_ERR(task); | |
ecb8ac8b MK |
1481 | goto free_iov; |
1482 | } | |
1483 | ||
a68a0262 | 1484 | if (!process_madvise_behavior_valid(behavior)) { |
ecb8ac8b MK |
1485 | ret = -EINVAL; |
1486 | goto release_task; | |
1487 | } | |
1488 | ||
96cfe2c0 SB |
1489 | /* Require PTRACE_MODE_READ to avoid leaking ASLR metadata. */ |
1490 | mm = mm_access(task, PTRACE_MODE_READ_FSCREDS); | |
ecb8ac8b MK |
1491 | if (IS_ERR_OR_NULL(mm)) { |
1492 | ret = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH; | |
1493 | goto release_task; | |
1494 | } | |
1495 | ||
96cfe2c0 SB |
1496 | /* |
1497 | * Require CAP_SYS_NICE for influencing process performance. Note that | |
1498 | * only non-destructive hints are currently supported. | |
1499 | */ | |
1500 | if (!capable(CAP_SYS_NICE)) { | |
1501 | ret = -EPERM; | |
1502 | goto release_mm; | |
1503 | } | |
1504 | ||
ecb8ac8b MK |
1505 | total_len = iov_iter_count(&iter); |
1506 | ||
1507 | while (iov_iter_count(&iter)) { | |
95e49cf8 JA |
1508 | ret = do_madvise(mm, (unsigned long)iter_iov_addr(&iter), |
1509 | iter_iov_len(&iter), behavior); | |
e6b0a7b3 | 1510 | if (ret < 0) |
ecb8ac8b | 1511 | break; |
95e49cf8 | 1512 | iov_iter_advance(&iter, iter_iov_len(&iter)); |
ecb8ac8b MK |
1513 | } |
1514 | ||
5bd009c7 | 1515 | ret = (total_len - iov_iter_count(&iter)) ? : ret; |
ecb8ac8b | 1516 | |
96cfe2c0 | 1517 | release_mm: |
ecb8ac8b | 1518 | mmput(mm); |
ecb8ac8b MK |
1519 | release_task: |
1520 | put_task_struct(task); | |
ecb8ac8b MK |
1521 | free_iov: |
1522 | kfree(iov); | |
1523 | out: | |
1524 | return ret; | |
1525 | } |