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; | |
183 | struct page *page; | |
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 | ||
1998cc04 | 201 | page = read_swap_cache_async(entry, GFP_HIGHUSER_MOVABLE, |
b243dcbf | 202 | vma, addr, &splug); |
1998cc04 | 203 | if (page) |
09cbfeaf | 204 | put_page(page); |
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; |
1998cc04 | 226 | struct page *page; |
5169b844 | 227 | struct swap_iocb *splug = NULL; |
1998cc04 | 228 | |
e6e88712 MWO |
229 | rcu_read_lock(); |
230 | xas_for_each(&xas, page, end_index) { | |
179d3e4f HD |
231 | unsigned long addr; |
232 | swp_entry_t entry; | |
1998cc04 | 233 | |
e6e88712 | 234 | if (!xa_is_value(page)) |
1998cc04 | 235 | continue; |
179d3e4f | 236 | entry = radix_to_swp_entry(page); |
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 | ||
179d3e4f | 246 | page = read_swap_cache_async(entry, mapping_gfp_mask(mapping), |
b243dcbf | 247 | vma, addr, &splug); |
1998cc04 | 248 | if (page) |
09cbfeaf | 249 | put_page(page); |
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()) { | |
432 | pte_unmap_unlock(start_pte, ptl); | |
433 | cond_resched(); | |
434 | goto restart; | |
435 | } | |
436 | } | |
437 | ||
9c276cc6 MK |
438 | if (pte_none(ptent)) |
439 | continue; | |
440 | ||
441 | if (!pte_present(ptent)) | |
442 | continue; | |
443 | ||
07e8c82b VMO |
444 | folio = vm_normal_folio(vma, addr, ptent); |
445 | if (!folio || folio_is_zone_device(folio)) | |
9c276cc6 MK |
446 | continue; |
447 | ||
448 | /* | |
449 | * Creating a THP page is expensive so split it only if we | |
450 | * are sure it's worth. Split it if we are only owner. | |
451 | */ | |
07e8c82b | 452 | if (folio_test_large(folio)) { |
f3cd4ab0 HD |
453 | int err; |
454 | ||
2f406263 | 455 | if (folio_estimated_sharers(folio) != 1) |
9c276cc6 | 456 | break; |
07e8c82b | 457 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 | 458 | break; |
f3cd4ab0 | 459 | if (!folio_trylock(folio)) |
9c276cc6 | 460 | break; |
f3cd4ab0 HD |
461 | folio_get(folio); |
462 | arch_leave_lazy_mmu_mode(); | |
463 | pte_unmap_unlock(start_pte, ptl); | |
464 | start_pte = NULL; | |
465 | err = split_folio(folio); | |
07e8c82b VMO |
466 | folio_unlock(folio); |
467 | folio_put(folio); | |
f3cd4ab0 HD |
468 | if (err) |
469 | break; | |
470 | start_pte = pte = | |
471 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
472 | if (!start_pte) | |
473 | break; | |
474 | arch_enter_lazy_mmu_mode(); | |
9c276cc6 MK |
475 | pte--; |
476 | addr -= PAGE_SIZE; | |
477 | continue; | |
478 | } | |
479 | ||
58d426a7 | 480 | /* |
07e8c82b VMO |
481 | * Do not interfere with other mappings of this folio and |
482 | * non-LRU folio. | |
58d426a7 | 483 | */ |
07e8c82b | 484 | if (!folio_test_lru(folio) || folio_mapcount(folio) != 1) |
12e967fd MH |
485 | continue; |
486 | ||
07e8c82b | 487 | if (pageout_anon_only_filter && !folio_test_anon(folio)) |
fd3b1bc3 PK |
488 | continue; |
489 | ||
07e8c82b | 490 | VM_BUG_ON_FOLIO(folio_test_large(folio), folio); |
9c276cc6 MK |
491 | |
492 | if (pte_young(ptent)) { | |
493 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
494 | tlb->fullmm); | |
495 | ptent = pte_mkold(ptent); | |
496 | set_pte_at(mm, addr, pte, ptent); | |
497 | tlb_remove_tlb_entry(tlb, pte, addr); | |
498 | } | |
499 | ||
500 | /* | |
07e8c82b VMO |
501 | * We are deactivating a folio for accelerating reclaiming. |
502 | * VM couldn't reclaim the folio unless we clear PG_young. | |
9c276cc6 MK |
503 | * As a side effect, it makes confuse idle-page tracking |
504 | * because they will miss recent referenced history. | |
505 | */ | |
07e8c82b VMO |
506 | folio_clear_referenced(folio); |
507 | folio_test_clear_young(folio); | |
20c897ea CTK |
508 | if (folio_test_active(folio)) |
509 | folio_set_workingset(folio); | |
d616d512 | 510 | if (pageout) { |
be2d5756 | 511 | if (folio_isolate_lru(folio)) { |
07e8c82b VMO |
512 | if (folio_test_unevictable(folio)) |
513 | folio_putback_lru(folio); | |
82072962 | 514 | else |
07e8c82b | 515 | list_add(&folio->lru, &folio_list); |
82072962 | 516 | } |
d616d512 | 517 | } else |
5a9e3474 | 518 | folio_deactivate(folio); |
9c276cc6 MK |
519 | } |
520 | ||
f3cd4ab0 HD |
521 | if (start_pte) { |
522 | arch_leave_lazy_mmu_mode(); | |
523 | pte_unmap_unlock(start_pte, ptl); | |
524 | } | |
d616d512 | 525 | if (pageout) |
07e8c82b | 526 | reclaim_pages(&folio_list); |
9c276cc6 MK |
527 | cond_resched(); |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
532 | static const struct mm_walk_ops cold_walk_ops = { | |
d616d512 | 533 | .pmd_entry = madvise_cold_or_pageout_pte_range, |
49b06385 | 534 | .walk_lock = PGWALK_RDLOCK, |
9c276cc6 MK |
535 | }; |
536 | ||
537 | static void madvise_cold_page_range(struct mmu_gather *tlb, | |
538 | struct vm_area_struct *vma, | |
539 | unsigned long addr, unsigned long end) | |
540 | { | |
d616d512 MK |
541 | struct madvise_walk_private walk_private = { |
542 | .pageout = false, | |
543 | .tlb = tlb, | |
544 | }; | |
545 | ||
9c276cc6 | 546 | tlb_start_vma(tlb, vma); |
d616d512 | 547 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); |
9c276cc6 MK |
548 | tlb_end_vma(tlb, vma); |
549 | } | |
550 | ||
a213e5cf HD |
551 | static inline bool can_madv_lru_vma(struct vm_area_struct *vma) |
552 | { | |
9457056a | 553 | return !(vma->vm_flags & (VM_LOCKED|VM_PFNMAP|VM_HUGETLB)); |
a213e5cf HD |
554 | } |
555 | ||
9c276cc6 MK |
556 | static long madvise_cold(struct vm_area_struct *vma, |
557 | struct vm_area_struct **prev, | |
558 | unsigned long start_addr, unsigned long end_addr) | |
559 | { | |
560 | struct mm_struct *mm = vma->vm_mm; | |
561 | struct mmu_gather tlb; | |
562 | ||
563 | *prev = vma; | |
564 | if (!can_madv_lru_vma(vma)) | |
565 | return -EINVAL; | |
566 | ||
567 | lru_add_drain(); | |
a72afd87 | 568 | tlb_gather_mmu(&tlb, mm); |
9c276cc6 | 569 | madvise_cold_page_range(&tlb, vma, start_addr, end_addr); |
ae8eba8b | 570 | tlb_finish_mmu(&tlb); |
9c276cc6 MK |
571 | |
572 | return 0; | |
573 | } | |
574 | ||
1a4e58cc MK |
575 | static void madvise_pageout_page_range(struct mmu_gather *tlb, |
576 | struct vm_area_struct *vma, | |
577 | unsigned long addr, unsigned long end) | |
578 | { | |
d616d512 MK |
579 | struct madvise_walk_private walk_private = { |
580 | .pageout = true, | |
581 | .tlb = tlb, | |
582 | }; | |
583 | ||
1a4e58cc | 584 | tlb_start_vma(tlb, vma); |
d616d512 | 585 | walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); |
1a4e58cc MK |
586 | tlb_end_vma(tlb, vma); |
587 | } | |
588 | ||
1a4e58cc MK |
589 | static long madvise_pageout(struct vm_area_struct *vma, |
590 | struct vm_area_struct **prev, | |
591 | unsigned long start_addr, unsigned long end_addr) | |
592 | { | |
593 | struct mm_struct *mm = vma->vm_mm; | |
594 | struct mmu_gather tlb; | |
595 | ||
596 | *prev = vma; | |
597 | if (!can_madv_lru_vma(vma)) | |
598 | return -EINVAL; | |
599 | ||
fd3b1bc3 PK |
600 | /* |
601 | * If the VMA belongs to a private file mapping, there can be private | |
602 | * dirty pages which can be paged out if even this process is neither | |
603 | * owner nor write capable of the file. We allow private file mappings | |
604 | * further to pageout dirty anon pages. | |
605 | */ | |
606 | if (!vma_is_anonymous(vma) && (!can_do_file_pageout(vma) && | |
607 | (vma->vm_flags & VM_MAYSHARE))) | |
1a4e58cc MK |
608 | return 0; |
609 | ||
610 | lru_add_drain(); | |
a72afd87 | 611 | tlb_gather_mmu(&tlb, mm); |
1a4e58cc | 612 | madvise_pageout_page_range(&tlb, vma, start_addr, end_addr); |
ae8eba8b | 613 | tlb_finish_mmu(&tlb); |
1a4e58cc MK |
614 | |
615 | return 0; | |
616 | } | |
617 | ||
854e9ed0 MK |
618 | static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr, |
619 | unsigned long end, struct mm_walk *walk) | |
620 | ||
621 | { | |
622 | struct mmu_gather *tlb = walk->private; | |
623 | struct mm_struct *mm = tlb->mm; | |
624 | struct vm_area_struct *vma = walk->vma; | |
625 | spinlock_t *ptl; | |
f3cd4ab0 | 626 | pte_t *start_pte, *pte, ptent; |
98b211d6 | 627 | struct folio *folio; |
64b42bc1 | 628 | int nr_swap = 0; |
b8d3c4c3 MK |
629 | unsigned long next; |
630 | ||
631 | next = pmd_addr_end(addr, end); | |
632 | if (pmd_trans_huge(*pmd)) | |
633 | if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) | |
f3cd4ab0 | 634 | return 0; |
854e9ed0 | 635 | |
ed6a7935 | 636 | tlb_change_page_size(tlb, PAGE_SIZE); |
f3cd4ab0 HD |
637 | start_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
638 | if (!start_pte) | |
639 | return 0; | |
3ea27719 | 640 | flush_tlb_batched_pending(mm); |
854e9ed0 MK |
641 | arch_enter_lazy_mmu_mode(); |
642 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
c33c7948 | 643 | ptent = ptep_get(pte); |
854e9ed0 | 644 | |
64b42bc1 | 645 | if (pte_none(ptent)) |
854e9ed0 | 646 | continue; |
64b42bc1 MK |
647 | /* |
648 | * If the pte has swp_entry, just clear page table to | |
649 | * prevent swap-in which is more expensive rather than | |
650 | * (page allocation + zeroing). | |
651 | */ | |
652 | if (!pte_present(ptent)) { | |
653 | swp_entry_t entry; | |
654 | ||
655 | entry = pte_to_swp_entry(ptent); | |
7b49514f ML |
656 | if (!non_swap_entry(entry)) { |
657 | nr_swap--; | |
658 | free_swap_and_cache(entry); | |
659 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
660 | } else if (is_hwpoison_entry(entry) || | |
af19487f | 661 | is_poisoned_swp_entry(entry)) { |
7b49514f ML |
662 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
663 | } | |
64b42bc1 MK |
664 | continue; |
665 | } | |
854e9ed0 | 666 | |
4947ed93 KW |
667 | folio = vm_normal_folio(vma, addr, ptent); |
668 | if (!folio || folio_is_zone_device(folio)) | |
854e9ed0 MK |
669 | continue; |
670 | ||
671 | /* | |
98b211d6 | 672 | * If pmd isn't transhuge but the folio is large and |
854e9ed0 MK |
673 | * is owned by only this process, split it and |
674 | * deactivate all pages. | |
675 | */ | |
98b211d6 | 676 | if (folio_test_large(folio)) { |
f3cd4ab0 HD |
677 | int err; |
678 | ||
0e0e9bd5 | 679 | if (folio_estimated_sharers(folio) != 1) |
f3cd4ab0 HD |
680 | break; |
681 | if (!folio_trylock(folio)) | |
682 | break; | |
98b211d6 | 683 | folio_get(folio); |
f3cd4ab0 HD |
684 | arch_leave_lazy_mmu_mode(); |
685 | pte_unmap_unlock(start_pte, ptl); | |
686 | start_pte = NULL; | |
687 | err = split_folio(folio); | |
98b211d6 MWO |
688 | folio_unlock(folio); |
689 | folio_put(folio); | |
f3cd4ab0 HD |
690 | if (err) |
691 | break; | |
692 | start_pte = pte = | |
693 | pte_offset_map_lock(mm, pmd, addr, &ptl); | |
694 | if (!start_pte) | |
695 | break; | |
696 | arch_enter_lazy_mmu_mode(); | |
854e9ed0 MK |
697 | pte--; |
698 | addr -= PAGE_SIZE; | |
699 | continue; | |
700 | } | |
701 | ||
98b211d6 MWO |
702 | if (folio_test_swapcache(folio) || folio_test_dirty(folio)) { |
703 | if (!folio_trylock(folio)) | |
854e9ed0 MK |
704 | continue; |
705 | /* | |
98b211d6 MWO |
706 | * If folio is shared with others, we mustn't clear |
707 | * the folio's dirty flag. | |
854e9ed0 | 708 | */ |
98b211d6 MWO |
709 | if (folio_mapcount(folio) != 1) { |
710 | folio_unlock(folio); | |
854e9ed0 MK |
711 | continue; |
712 | } | |
713 | ||
98b211d6 MWO |
714 | if (folio_test_swapcache(folio) && |
715 | !folio_free_swap(folio)) { | |
716 | folio_unlock(folio); | |
854e9ed0 MK |
717 | continue; |
718 | } | |
719 | ||
98b211d6 MWO |
720 | folio_clear_dirty(folio); |
721 | folio_unlock(folio); | |
854e9ed0 MK |
722 | } |
723 | ||
724 | if (pte_young(ptent) || pte_dirty(ptent)) { | |
725 | /* | |
726 | * Some of architecture(ex, PPC) don't update TLB | |
727 | * with set_pte_at and tlb_remove_tlb_entry so for | |
728 | * the portability, remap the pte with old|clean | |
729 | * after pte clearing. | |
730 | */ | |
731 | ptent = ptep_get_and_clear_full(mm, addr, pte, | |
732 | tlb->fullmm); | |
733 | ||
734 | ptent = pte_mkold(ptent); | |
735 | ptent = pte_mkclean(ptent); | |
736 | set_pte_at(mm, addr, pte, ptent); | |
737 | tlb_remove_tlb_entry(tlb, pte, addr); | |
738 | } | |
6a6fe9eb | 739 | folio_mark_lazyfree(folio); |
854e9ed0 | 740 | } |
f3cd4ab0 | 741 | |
bc0c3357 | 742 | if (nr_swap) |
64b42bc1 | 743 | add_mm_counter(mm, MM_SWAPENTS, nr_swap); |
f3cd4ab0 HD |
744 | if (start_pte) { |
745 | arch_leave_lazy_mmu_mode(); | |
746 | pte_unmap_unlock(start_pte, ptl); | |
747 | } | |
854e9ed0 | 748 | cond_resched(); |
f3cd4ab0 | 749 | |
854e9ed0 MK |
750 | return 0; |
751 | } | |
752 | ||
7b86ac33 CH |
753 | static const struct mm_walk_ops madvise_free_walk_ops = { |
754 | .pmd_entry = madvise_free_pte_range, | |
49b06385 | 755 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 | 756 | }; |
854e9ed0 MK |
757 | |
758 | static int madvise_free_single_vma(struct vm_area_struct *vma, | |
759 | unsigned long start_addr, unsigned long end_addr) | |
760 | { | |
854e9ed0 | 761 | struct mm_struct *mm = vma->vm_mm; |
ac46d4f3 | 762 | struct mmu_notifier_range range; |
854e9ed0 MK |
763 | struct mmu_gather tlb; |
764 | ||
854e9ed0 MK |
765 | /* MADV_FREE works for only anon vma at the moment */ |
766 | if (!vma_is_anonymous(vma)) | |
767 | return -EINVAL; | |
768 | ||
ac46d4f3 JG |
769 | range.start = max(vma->vm_start, start_addr); |
770 | if (range.start >= vma->vm_end) | |
854e9ed0 | 771 | return -EINVAL; |
ac46d4f3 JG |
772 | range.end = min(vma->vm_end, end_addr); |
773 | if (range.end <= vma->vm_start) | |
854e9ed0 | 774 | return -EINVAL; |
7d4a8be0 | 775 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, |
6f4f13e8 | 776 | range.start, range.end); |
854e9ed0 MK |
777 | |
778 | lru_add_drain(); | |
a72afd87 | 779 | tlb_gather_mmu(&tlb, mm); |
854e9ed0 MK |
780 | update_hiwater_rss(mm); |
781 | ||
ac46d4f3 | 782 | mmu_notifier_invalidate_range_start(&range); |
7b86ac33 CH |
783 | tlb_start_vma(&tlb, vma); |
784 | walk_page_range(vma->vm_mm, range.start, range.end, | |
785 | &madvise_free_walk_ops, &tlb); | |
786 | tlb_end_vma(&tlb, vma); | |
ac46d4f3 | 787 | mmu_notifier_invalidate_range_end(&range); |
ae8eba8b | 788 | tlb_finish_mmu(&tlb); |
854e9ed0 MK |
789 | |
790 | return 0; | |
791 | } | |
792 | ||
1da177e4 LT |
793 | /* |
794 | * Application no longer needs these pages. If the pages are dirty, | |
795 | * it's OK to just throw them away. The app will be more careful about | |
796 | * data it wants to keep. Be sure to free swap resources too. The | |
21b85b09 MK |
797 | * zap_page_range_single call sets things up for shrink_active_list to actually |
798 | * free these pages later if no one else has touched them in the meantime, | |
1da177e4 | 799 | * although we could add these pages to a global reuse list for |
7e6cbea3 | 800 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
801 | * |
802 | * NB: This interface discards data rather than pushes it out to swap, | |
803 | * as some implementations do. This has performance implications for | |
804 | * applications like large transactional databases which want to discard | |
805 | * pages in anonymous maps after committing to backing store the data | |
806 | * that was kept in them. There is no reason to write this data out to | |
807 | * the swap area if the application is discarding it. | |
808 | * | |
809 | * An interface that causes the system to free clean pages and flush | |
810 | * dirty pages is already available as msync(MS_INVALIDATE). | |
811 | */ | |
230ca982 MR |
812 | static long madvise_dontneed_single_vma(struct vm_area_struct *vma, |
813 | unsigned long start, unsigned long end) | |
814 | { | |
21b85b09 | 815 | zap_page_range_single(vma, start, end - start, NULL); |
230ca982 MR |
816 | return 0; |
817 | } | |
818 | ||
90e7e7f5 MK |
819 | static bool madvise_dontneed_free_valid_vma(struct vm_area_struct *vma, |
820 | unsigned long start, | |
821 | unsigned long *end, | |
822 | int behavior) | |
823 | { | |
9457056a JW |
824 | if (!is_vm_hugetlb_page(vma)) { |
825 | unsigned int forbidden = VM_PFNMAP; | |
826 | ||
827 | if (behavior != MADV_DONTNEED_LOCKED) | |
828 | forbidden |= VM_LOCKED; | |
829 | ||
830 | return !(vma->vm_flags & forbidden); | |
831 | } | |
90e7e7f5 | 832 | |
9457056a | 833 | if (behavior != MADV_DONTNEED && behavior != MADV_DONTNEED_LOCKED) |
90e7e7f5 MK |
834 | return false; |
835 | if (start & ~huge_page_mask(hstate_vma(vma))) | |
836 | return false; | |
837 | ||
8ebe0a5e RR |
838 | /* |
839 | * Madvise callers expect the length to be rounded up to PAGE_SIZE | |
840 | * boundaries, and may be unaware that this VMA uses huge pages. | |
841 | * Avoid unexpected data loss by rounding down the number of | |
842 | * huge pages freed. | |
843 | */ | |
844 | *end = ALIGN_DOWN(*end, huge_page_size(hstate_vma(vma))); | |
845 | ||
90e7e7f5 MK |
846 | return true; |
847 | } | |
848 | ||
230ca982 MR |
849 | static long madvise_dontneed_free(struct vm_area_struct *vma, |
850 | struct vm_area_struct **prev, | |
851 | unsigned long start, unsigned long end, | |
852 | int behavior) | |
1da177e4 | 853 | { |
0726b01e MK |
854 | struct mm_struct *mm = vma->vm_mm; |
855 | ||
05b74384 | 856 | *prev = vma; |
90e7e7f5 | 857 | if (!madvise_dontneed_free_valid_vma(vma, start, &end, behavior)) |
1da177e4 LT |
858 | return -EINVAL; |
859 | ||
8ebe0a5e RR |
860 | if (start == end) |
861 | return 0; | |
862 | ||
70ccb92f | 863 | if (!userfaultfd_remove(vma, start, end)) { |
c1e8d7c6 | 864 | *prev = NULL; /* mmap_lock has been dropped, prev is stale */ |
70ccb92f | 865 | |
0726b01e | 866 | mmap_read_lock(mm); |
92d5df38 | 867 | vma = vma_lookup(mm, start); |
70ccb92f AA |
868 | if (!vma) |
869 | return -ENOMEM; | |
90e7e7f5 MK |
870 | /* |
871 | * Potential end adjustment for hugetlb vma is OK as | |
872 | * the check below keeps end within vma. | |
873 | */ | |
874 | if (!madvise_dontneed_free_valid_vma(vma, start, &end, | |
875 | behavior)) | |
70ccb92f AA |
876 | return -EINVAL; |
877 | if (end > vma->vm_end) { | |
878 | /* | |
879 | * Don't fail if end > vma->vm_end. If the old | |
f0953a1b | 880 | * vma was split while the mmap_lock was |
70ccb92f | 881 | * released the effect of the concurrent |
230ca982 | 882 | * operation may not cause madvise() to |
70ccb92f AA |
883 | * have an undefined result. There may be an |
884 | * adjacent next vma that we'll walk | |
885 | * next. userfaultfd_remove() will generate an | |
886 | * UFFD_EVENT_REMOVE repetition on the | |
887 | * end-vma->vm_end range, but the manager can | |
888 | * handle a repetition fine. | |
889 | */ | |
890 | end = vma->vm_end; | |
891 | } | |
892 | VM_WARN_ON(start >= end); | |
893 | } | |
230ca982 | 894 | |
9457056a | 895 | if (behavior == MADV_DONTNEED || behavior == MADV_DONTNEED_LOCKED) |
230ca982 MR |
896 | return madvise_dontneed_single_vma(vma, start, end); |
897 | else if (behavior == MADV_FREE) | |
898 | return madvise_free_single_vma(vma, start, end); | |
899 | else | |
900 | return -EINVAL; | |
1da177e4 LT |
901 | } |
902 | ||
4ca9b385 DH |
903 | static long madvise_populate(struct vm_area_struct *vma, |
904 | struct vm_area_struct **prev, | |
905 | unsigned long start, unsigned long end, | |
906 | int behavior) | |
907 | { | |
908 | const bool write = behavior == MADV_POPULATE_WRITE; | |
909 | struct mm_struct *mm = vma->vm_mm; | |
910 | unsigned long tmp_end; | |
911 | int locked = 1; | |
912 | long pages; | |
913 | ||
914 | *prev = vma; | |
915 | ||
916 | while (start < end) { | |
917 | /* | |
918 | * We might have temporarily dropped the lock. For example, | |
919 | * our VMA might have been split. | |
920 | */ | |
921 | if (!vma || start >= vma->vm_end) { | |
531037a0 ML |
922 | vma = vma_lookup(mm, start); |
923 | if (!vma) | |
4ca9b385 DH |
924 | return -ENOMEM; |
925 | } | |
926 | ||
927 | tmp_end = min_t(unsigned long, end, vma->vm_end); | |
928 | /* Populate (prefault) page tables readable/writable. */ | |
929 | pages = faultin_vma_page_range(vma, start, tmp_end, write, | |
930 | &locked); | |
931 | if (!locked) { | |
932 | mmap_read_lock(mm); | |
933 | locked = 1; | |
934 | *prev = NULL; | |
935 | vma = NULL; | |
936 | } | |
937 | if (pages < 0) { | |
938 | switch (pages) { | |
939 | case -EINTR: | |
940 | return -EINTR; | |
eb2faa51 | 941 | case -EINVAL: /* Incompatible mappings / permissions. */ |
4ca9b385 DH |
942 | return -EINVAL; |
943 | case -EHWPOISON: | |
944 | return -EHWPOISON; | |
eb2faa51 DH |
945 | case -EFAULT: /* VM_FAULT_SIGBUS or VM_FAULT_SIGSEGV */ |
946 | return -EFAULT; | |
4ca9b385 DH |
947 | default: |
948 | pr_warn_once("%s: unhandled return value: %ld\n", | |
949 | __func__, pages); | |
950 | fallthrough; | |
951 | case -ENOMEM: | |
952 | return -ENOMEM; | |
953 | } | |
954 | } | |
955 | start += pages * PAGE_SIZE; | |
956 | } | |
957 | return 0; | |
958 | } | |
959 | ||
f6b3ec23 BP |
960 | /* |
961 | * Application wants to free up the pages and associated backing store. | |
962 | * This is effectively punching a hole into the middle of a file. | |
f6b3ec23 BP |
963 | */ |
964 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 965 | struct vm_area_struct **prev, |
f6b3ec23 BP |
966 | unsigned long start, unsigned long end) |
967 | { | |
3f31d075 | 968 | loff_t offset; |
90ed52eb | 969 | int error; |
9ab4233d | 970 | struct file *f; |
0726b01e | 971 | struct mm_struct *mm = vma->vm_mm; |
f6b3ec23 | 972 | |
c1e8d7c6 | 973 | *prev = NULL; /* tell sys_madvise we drop mmap_lock */ |
00e9fa2d | 974 | |
72079ba0 | 975 | if (vma->vm_flags & VM_LOCKED) |
f6b3ec23 BP |
976 | return -EINVAL; |
977 | ||
9ab4233d AL |
978 | f = vma->vm_file; |
979 | ||
980 | if (!f || !f->f_mapping || !f->f_mapping->host) { | |
f6b3ec23 BP |
981 | return -EINVAL; |
982 | } | |
983 | ||
e8e17ee9 | 984 | if (!vma_is_shared_maywrite(vma)) |
69cf0fac HD |
985 | return -EACCES; |
986 | ||
f6b3ec23 BP |
987 | offset = (loff_t)(start - vma->vm_start) |
988 | + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
90ed52eb | 989 | |
9ab4233d | 990 | /* |
9608703e | 991 | * Filesystem's fallocate may need to take i_rwsem. We need to |
9ab4233d AL |
992 | * explicitly grab a reference because the vma (and hence the |
993 | * vma's reference to the file) can go away as soon as we drop | |
c1e8d7c6 | 994 | * mmap_lock. |
9ab4233d AL |
995 | */ |
996 | get_file(f); | |
70ccb92f | 997 | if (userfaultfd_remove(vma, start, end)) { |
c1e8d7c6 | 998 | /* mmap_lock was not released by userfaultfd_remove() */ |
0726b01e | 999 | mmap_read_unlock(mm); |
70ccb92f | 1000 | } |
72c72bdf | 1001 | error = vfs_fallocate(f, |
3f31d075 HD |
1002 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1003 | offset, end - start); | |
9ab4233d | 1004 | fput(f); |
0726b01e | 1005 | mmap_read_lock(mm); |
90ed52eb | 1006 | return error; |
f6b3ec23 BP |
1007 | } |
1008 | ||
ac1e9acc CC |
1009 | /* |
1010 | * Apply an madvise behavior to a region of a vma. madvise_update_vma | |
1011 | * will handle splitting a vm area into separate areas, each area with its own | |
1012 | * behavior. | |
1013 | */ | |
1014 | static int madvise_vma_behavior(struct vm_area_struct *vma, | |
1015 | struct vm_area_struct **prev, | |
1016 | unsigned long start, unsigned long end, | |
1017 | unsigned long behavior) | |
1018 | { | |
1019 | int error; | |
942341dc | 1020 | struct anon_vma_name *anon_name; |
ac1e9acc CC |
1021 | unsigned long new_flags = vma->vm_flags; |
1022 | ||
1023 | switch (behavior) { | |
1024 | case MADV_REMOVE: | |
1025 | return madvise_remove(vma, prev, start, end); | |
1026 | case MADV_WILLNEED: | |
1027 | return madvise_willneed(vma, prev, start, end); | |
1028 | case MADV_COLD: | |
1029 | return madvise_cold(vma, prev, start, end); | |
1030 | case MADV_PAGEOUT: | |
1031 | return madvise_pageout(vma, prev, start, end); | |
1032 | case MADV_FREE: | |
1033 | case MADV_DONTNEED: | |
9457056a | 1034 | case MADV_DONTNEED_LOCKED: |
ac1e9acc CC |
1035 | return madvise_dontneed_free(vma, prev, start, end, behavior); |
1036 | case MADV_POPULATE_READ: | |
1037 | case MADV_POPULATE_WRITE: | |
1038 | return madvise_populate(vma, prev, start, end, behavior); | |
1039 | case MADV_NORMAL: | |
1040 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
1041 | break; | |
1042 | case MADV_SEQUENTIAL: | |
1043 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; | |
1044 | break; | |
1045 | case MADV_RANDOM: | |
1046 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; | |
1047 | break; | |
1048 | case MADV_DONTFORK: | |
1049 | new_flags |= VM_DONTCOPY; | |
1050 | break; | |
1051 | case MADV_DOFORK: | |
1052 | if (vma->vm_flags & VM_IO) | |
1053 | return -EINVAL; | |
1054 | new_flags &= ~VM_DONTCOPY; | |
1055 | break; | |
1056 | case MADV_WIPEONFORK: | |
1057 | /* MADV_WIPEONFORK is only supported on anonymous memory. */ | |
1058 | if (vma->vm_file || vma->vm_flags & VM_SHARED) | |
1059 | return -EINVAL; | |
1060 | new_flags |= VM_WIPEONFORK; | |
1061 | break; | |
1062 | case MADV_KEEPONFORK: | |
1063 | new_flags &= ~VM_WIPEONFORK; | |
1064 | break; | |
1065 | case MADV_DONTDUMP: | |
1066 | new_flags |= VM_DONTDUMP; | |
1067 | break; | |
1068 | case MADV_DODUMP: | |
1069 | if (!is_vm_hugetlb_page(vma) && new_flags & VM_SPECIAL) | |
1070 | return -EINVAL; | |
1071 | new_flags &= ~VM_DONTDUMP; | |
1072 | break; | |
1073 | case MADV_MERGEABLE: | |
1074 | case MADV_UNMERGEABLE: | |
1075 | error = ksm_madvise(vma, start, end, behavior, &new_flags); | |
1076 | if (error) | |
1077 | goto out; | |
1078 | break; | |
1079 | case MADV_HUGEPAGE: | |
1080 | case MADV_NOHUGEPAGE: | |
1081 | error = hugepage_madvise(vma, &new_flags, behavior); | |
1082 | if (error) | |
1083 | goto out; | |
1084 | break; | |
7d8faaf1 ZK |
1085 | case MADV_COLLAPSE: |
1086 | return madvise_collapse(vma, prev, start, end); | |
ac1e9acc CC |
1087 | } |
1088 | ||
942341dc SB |
1089 | anon_name = anon_vma_name(vma); |
1090 | anon_vma_name_get(anon_name); | |
9a10064f | 1091 | error = madvise_update_vma(vma, prev, start, end, new_flags, |
942341dc SB |
1092 | anon_name); |
1093 | anon_vma_name_put(anon_name); | |
ac1e9acc CC |
1094 | |
1095 | out: | |
1096 | /* | |
1097 | * madvise() returns EAGAIN if kernel resources, such as | |
1098 | * slab, are temporarily unavailable. | |
1099 | */ | |
1100 | if (error == -ENOMEM) | |
1101 | error = -EAGAIN; | |
1102 | return error; | |
1103 | } | |
1104 | ||
9893e49d AK |
1105 | #ifdef CONFIG_MEMORY_FAILURE |
1106 | /* | |
1107 | * Error injection support for memory error handling. | |
1108 | */ | |
97167a76 AK |
1109 | static int madvise_inject_error(int behavior, |
1110 | unsigned long start, unsigned long end) | |
9893e49d | 1111 | { |
d3cd257c | 1112 | unsigned long size; |
97167a76 | 1113 | |
9893e49d AK |
1114 | if (!capable(CAP_SYS_ADMIN)) |
1115 | return -EPERM; | |
97167a76 | 1116 | |
19bfbe22 | 1117 | |
d3cd257c | 1118 | for (; start < end; start += size) { |
23e7b5c2 | 1119 | unsigned long pfn; |
dc7560b4 | 1120 | struct page *page; |
325c4ef5 AM |
1121 | int ret; |
1122 | ||
97167a76 | 1123 | ret = get_user_pages_fast(start, 1, 0, &page); |
9893e49d AK |
1124 | if (ret != 1) |
1125 | return ret; | |
23e7b5c2 | 1126 | pfn = page_to_pfn(page); |
325c4ef5 | 1127 | |
19bfbe22 AM |
1128 | /* |
1129 | * When soft offlining hugepages, after migrating the page | |
1130 | * we dissolve it, therefore in the second loop "page" will | |
d3cd257c | 1131 | * no longer be a compound page. |
19bfbe22 | 1132 | */ |
d3cd257c | 1133 | size = page_size(compound_head(page)); |
19bfbe22 | 1134 | |
97167a76 AK |
1135 | if (behavior == MADV_SOFT_OFFLINE) { |
1136 | pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n", | |
dc7560b4 | 1137 | pfn, start); |
feec24a6 | 1138 | ret = soft_offline_page(pfn, MF_COUNT_INCREASED); |
dc7560b4 OS |
1139 | } else { |
1140 | pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n", | |
1141 | pfn, start); | |
67f22ba7 | 1142 | ret = memory_failure(pfn, MF_COUNT_INCREASED | MF_SW_SIMULATED); |
d1fe111f | 1143 | if (ret == -EOPNOTSUPP) |
1144 | ret = 0; | |
afcf938e | 1145 | } |
23e7b5c2 | 1146 | |
23a003bf NH |
1147 | if (ret) |
1148 | return ret; | |
9893e49d | 1149 | } |
c461ad6a | 1150 | |
325c4ef5 | 1151 | return 0; |
9893e49d AK |
1152 | } |
1153 | #endif | |
1154 | ||
1ecef9ed | 1155 | static bool |
75927af8 NP |
1156 | madvise_behavior_valid(int behavior) |
1157 | { | |
1158 | switch (behavior) { | |
1159 | case MADV_DOFORK: | |
1160 | case MADV_DONTFORK: | |
1161 | case MADV_NORMAL: | |
1162 | case MADV_SEQUENTIAL: | |
1163 | case MADV_RANDOM: | |
1164 | case MADV_REMOVE: | |
1165 | case MADV_WILLNEED: | |
1166 | case MADV_DONTNEED: | |
9457056a | 1167 | case MADV_DONTNEED_LOCKED: |
854e9ed0 | 1168 | case MADV_FREE: |
9c276cc6 | 1169 | case MADV_COLD: |
1a4e58cc | 1170 | case MADV_PAGEOUT: |
4ca9b385 DH |
1171 | case MADV_POPULATE_READ: |
1172 | case MADV_POPULATE_WRITE: | |
f8af4da3 HD |
1173 | #ifdef CONFIG_KSM |
1174 | case MADV_MERGEABLE: | |
1175 | case MADV_UNMERGEABLE: | |
0af4e98b AA |
1176 | #endif |
1177 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1178 | case MADV_HUGEPAGE: | |
a664b2d8 | 1179 | case MADV_NOHUGEPAGE: |
7d8faaf1 | 1180 | case MADV_COLLAPSE: |
f8af4da3 | 1181 | #endif |
accb61fe JB |
1182 | case MADV_DONTDUMP: |
1183 | case MADV_DODUMP: | |
d2cd9ede RR |
1184 | case MADV_WIPEONFORK: |
1185 | case MADV_KEEPONFORK: | |
5e451be7 AK |
1186 | #ifdef CONFIG_MEMORY_FAILURE |
1187 | case MADV_SOFT_OFFLINE: | |
1188 | case MADV_HWPOISON: | |
1189 | #endif | |
1ecef9ed | 1190 | return true; |
75927af8 NP |
1191 | |
1192 | default: | |
1ecef9ed | 1193 | return false; |
75927af8 NP |
1194 | } |
1195 | } | |
3866ea90 | 1196 | |
876b4a18 | 1197 | static bool process_madvise_behavior_valid(int behavior) |
ecb8ac8b MK |
1198 | { |
1199 | switch (behavior) { | |
1200 | case MADV_COLD: | |
1201 | case MADV_PAGEOUT: | |
d5fffc5a | 1202 | case MADV_WILLNEED: |
876b4a18 | 1203 | case MADV_COLLAPSE: |
ecb8ac8b MK |
1204 | return true; |
1205 | default: | |
1206 | return false; | |
1207 | } | |
1208 | } | |
1209 | ||
ac1e9acc CC |
1210 | /* |
1211 | * Walk the vmas in range [start,end), and call the visit function on each one. | |
1212 | * The visit function will get start and end parameters that cover the overlap | |
1213 | * between the current vma and the original range. Any unmapped regions in the | |
1214 | * original range will result in this function returning -ENOMEM while still | |
1215 | * calling the visit function on all of the existing vmas in the range. | |
1216 | * Must be called with the mmap_lock held for reading or writing. | |
1217 | */ | |
1218 | static | |
1219 | int madvise_walk_vmas(struct mm_struct *mm, unsigned long start, | |
1220 | unsigned long end, unsigned long arg, | |
1221 | int (*visit)(struct vm_area_struct *vma, | |
1222 | struct vm_area_struct **prev, unsigned long start, | |
1223 | unsigned long end, unsigned long arg)) | |
1224 | { | |
1225 | struct vm_area_struct *vma; | |
1226 | struct vm_area_struct *prev; | |
1227 | unsigned long tmp; | |
1228 | int unmapped_error = 0; | |
1229 | ||
1230 | /* | |
1231 | * If the interval [start,end) covers some unmapped address | |
1232 | * ranges, just ignore them, but return -ENOMEM at the end. | |
1233 | * - different from the way of handling in mlock etc. | |
1234 | */ | |
1235 | vma = find_vma_prev(mm, start, &prev); | |
1236 | if (vma && start > vma->vm_start) | |
1237 | prev = vma; | |
1238 | ||
1239 | for (;;) { | |
1240 | int error; | |
1241 | ||
1242 | /* Still start < end. */ | |
1243 | if (!vma) | |
1244 | return -ENOMEM; | |
1245 | ||
1246 | /* Here start < (end|vma->vm_end). */ | |
1247 | if (start < vma->vm_start) { | |
1248 | unmapped_error = -ENOMEM; | |
1249 | start = vma->vm_start; | |
1250 | if (start >= end) | |
1251 | break; | |
1252 | } | |
1253 | ||
1254 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ | |
1255 | tmp = vma->vm_end; | |
1256 | if (end < tmp) | |
1257 | tmp = end; | |
1258 | ||
1259 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ | |
1260 | error = visit(vma, &prev, start, tmp, arg); | |
1261 | if (error) | |
1262 | return error; | |
1263 | start = tmp; | |
1264 | if (prev && start < prev->vm_end) | |
1265 | start = prev->vm_end; | |
1266 | if (start >= end) | |
1267 | break; | |
1268 | if (prev) | |
35474818 | 1269 | vma = find_vma(mm, prev->vm_end); |
ac1e9acc CC |
1270 | else /* madvise_remove dropped mmap_lock */ |
1271 | vma = find_vma(mm, start); | |
1272 | } | |
1273 | ||
1274 | return unmapped_error; | |
1275 | } | |
1276 | ||
9a10064f CC |
1277 | #ifdef CONFIG_ANON_VMA_NAME |
1278 | static int madvise_vma_anon_name(struct vm_area_struct *vma, | |
1279 | struct vm_area_struct **prev, | |
1280 | unsigned long start, unsigned long end, | |
5c26f6ac | 1281 | unsigned long anon_name) |
9a10064f CC |
1282 | { |
1283 | int error; | |
1284 | ||
1285 | /* Only anonymous mappings can be named */ | |
d09e8ca6 | 1286 | if (vma->vm_file && !vma_is_anon_shmem(vma)) |
9a10064f CC |
1287 | return -EBADF; |
1288 | ||
1289 | error = madvise_update_vma(vma, prev, start, end, vma->vm_flags, | |
5c26f6ac | 1290 | (struct anon_vma_name *)anon_name); |
9a10064f CC |
1291 | |
1292 | /* | |
1293 | * madvise() returns EAGAIN if kernel resources, such as | |
1294 | * slab, are temporarily unavailable. | |
1295 | */ | |
1296 | if (error == -ENOMEM) | |
1297 | error = -EAGAIN; | |
1298 | return error; | |
1299 | } | |
1300 | ||
1301 | int madvise_set_anon_name(struct mm_struct *mm, unsigned long start, | |
5c26f6ac | 1302 | unsigned long len_in, struct anon_vma_name *anon_name) |
9a10064f CC |
1303 | { |
1304 | unsigned long end; | |
1305 | unsigned long len; | |
1306 | ||
1307 | if (start & ~PAGE_MASK) | |
1308 | return -EINVAL; | |
1309 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; | |
1310 | ||
1311 | /* Check to see whether len was rounded up from small -ve to zero */ | |
1312 | if (len_in && !len) | |
1313 | return -EINVAL; | |
1314 | ||
1315 | end = start + len; | |
1316 | if (end < start) | |
1317 | return -EINVAL; | |
1318 | ||
1319 | if (end == start) | |
1320 | return 0; | |
1321 | ||
5c26f6ac | 1322 | return madvise_walk_vmas(mm, start, end, (unsigned long)anon_name, |
9a10064f CC |
1323 | madvise_vma_anon_name); |
1324 | } | |
1325 | #endif /* CONFIG_ANON_VMA_NAME */ | |
1da177e4 LT |
1326 | /* |
1327 | * The madvise(2) system call. | |
1328 | * | |
1329 | * Applications can use madvise() to advise the kernel how it should | |
1330 | * handle paging I/O in this VM area. The idea is to help the kernel | |
1331 | * use appropriate read-ahead and caching techniques. The information | |
1332 | * provided is advisory only, and can be safely disregarded by the | |
1333 | * kernel without affecting the correct operation of the application. | |
1334 | * | |
1335 | * behavior values: | |
1336 | * MADV_NORMAL - the default behavior is to read clusters. This | |
1337 | * results in some read-ahead and read-behind. | |
1338 | * MADV_RANDOM - the system should read the minimum amount of data | |
1339 | * on any access, since it is unlikely that the appli- | |
1340 | * cation will need more than what it asks for. | |
1341 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
1342 | * once, so they can be aggressively read ahead, and | |
1343 | * can be freed soon after they are accessed. | |
1344 | * MADV_WILLNEED - the application is notifying the system to read | |
1345 | * some pages ahead. | |
1346 | * MADV_DONTNEED - the application is finished with the given range, | |
1347 | * so the kernel can free resources associated with it. | |
d7206a70 NH |
1348 | * MADV_FREE - the application marks pages in the given range as lazy free, |
1349 | * where actual purges are postponed until memory pressure happens. | |
f6b3ec23 BP |
1350 | * MADV_REMOVE - the application wants to free up the given range of |
1351 | * pages and associated backing store. | |
3866ea90 HD |
1352 | * MADV_DONTFORK - omit this area from child's address space when forking: |
1353 | * typically, to avoid COWing pages pinned by get_user_pages(). | |
1354 | * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking. | |
c02c3009 YS |
1355 | * MADV_WIPEONFORK - present the child process with zero-filled memory in this |
1356 | * range after a fork. | |
1357 | * MADV_KEEPONFORK - undo the effect of MADV_WIPEONFORK | |
d7206a70 NH |
1358 | * MADV_HWPOISON - trigger memory error handler as if the given memory range |
1359 | * were corrupted by unrecoverable hardware memory failure. | |
1360 | * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory. | |
f8af4da3 HD |
1361 | * MADV_MERGEABLE - the application recommends that KSM try to merge pages in |
1362 | * this area with pages of identical content from other such areas. | |
1363 | * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others. | |
d7206a70 NH |
1364 | * MADV_HUGEPAGE - the application wants to back the given range by transparent |
1365 | * huge pages in the future. Existing pages might be coalesced and | |
1366 | * new pages might be allocated as THP. | |
1367 | * MADV_NOHUGEPAGE - mark the given range as not worth being backed by | |
1368 | * transparent huge pages so the existing pages will not be | |
1369 | * coalesced into THP and new pages will not be allocated as THP. | |
7d8faaf1 | 1370 | * MADV_COLLAPSE - synchronously coalesce pages into new THP. |
d7206a70 NH |
1371 | * MADV_DONTDUMP - the application wants to prevent pages in the given range |
1372 | * from being included in its core dump. | |
1373 | * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump. | |
ecb8ac8b MK |
1374 | * MADV_COLD - the application is not expected to use this memory soon, |
1375 | * deactivate pages in this range so that they can be reclaimed | |
f0953a1b | 1376 | * easily if memory pressure happens. |
ecb8ac8b MK |
1377 | * MADV_PAGEOUT - the application is not expected to use this memory soon, |
1378 | * page out the pages in this range immediately. | |
4ca9b385 DH |
1379 | * MADV_POPULATE_READ - populate (prefault) page tables readable by |
1380 | * triggering read faults if required | |
1381 | * MADV_POPULATE_WRITE - populate (prefault) page tables writable by | |
1382 | * triggering write faults if required | |
1da177e4 LT |
1383 | * |
1384 | * return values: | |
1385 | * zero - success | |
1386 | * -EINVAL - start + len < 0, start is not page-aligned, | |
1387 | * "behavior" is not a valid value, or application | |
c02c3009 YS |
1388 | * is attempting to release locked or shared pages, |
1389 | * or the specified address range includes file, Huge TLB, | |
1390 | * MAP_SHARED or VMPFNMAP range. | |
1da177e4 LT |
1391 | * -ENOMEM - addresses in the specified range are not currently |
1392 | * mapped, or are outside the AS of the process. | |
1393 | * -EIO - an I/O error occurred while paging in data. | |
1394 | * -EBADF - map exists, but area maps something that isn't a file. | |
1395 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
1396 | */ | |
0726b01e | 1397 | int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior) |
1da177e4 | 1398 | { |
ac1e9acc CC |
1399 | unsigned long end; |
1400 | int error; | |
f7977793 | 1401 | int write; |
1da177e4 | 1402 | size_t len; |
1998cc04 | 1403 | struct blk_plug plug; |
1da177e4 | 1404 | |
75927af8 | 1405 | if (!madvise_behavior_valid(behavior)) |
ac1e9acc | 1406 | return -EINVAL; |
75927af8 | 1407 | |
df6c6500 | 1408 | if (!PAGE_ALIGNED(start)) |
ac1e9acc | 1409 | return -EINVAL; |
df6c6500 | 1410 | len = PAGE_ALIGN(len_in); |
1da177e4 LT |
1411 | |
1412 | /* Check to see whether len was rounded up from small -ve to zero */ | |
1413 | if (len_in && !len) | |
ac1e9acc | 1414 | return -EINVAL; |
1da177e4 LT |
1415 | |
1416 | end = start + len; | |
1417 | if (end < start) | |
ac1e9acc | 1418 | return -EINVAL; |
1da177e4 | 1419 | |
1da177e4 | 1420 | if (end == start) |
ac1e9acc | 1421 | return 0; |
84d96d89 | 1422 | |
5e451be7 AK |
1423 | #ifdef CONFIG_MEMORY_FAILURE |
1424 | if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE) | |
1425 | return madvise_inject_error(behavior, start, start + len_in); | |
1426 | #endif | |
1427 | ||
84d96d89 | 1428 | write = madvise_need_mmap_write(behavior); |
dc0ef0df | 1429 | if (write) { |
0726b01e | 1430 | if (mmap_write_lock_killable(mm)) |
dc0ef0df MH |
1431 | return -EINTR; |
1432 | } else { | |
0726b01e | 1433 | mmap_read_lock(mm); |
dc0ef0df | 1434 | } |
1da177e4 | 1435 | |
428e106a KS |
1436 | start = untagged_addr_remote(mm, start); |
1437 | end = start + len; | |
1438 | ||
1998cc04 | 1439 | blk_start_plug(&plug); |
ac1e9acc CC |
1440 | error = madvise_walk_vmas(mm, start, end, behavior, |
1441 | madvise_vma_behavior); | |
84d96d89 | 1442 | blk_finish_plug(&plug); |
f7977793 | 1443 | if (write) |
0726b01e | 1444 | mmap_write_unlock(mm); |
0a27a14a | 1445 | else |
0726b01e | 1446 | mmap_read_unlock(mm); |
0a27a14a | 1447 | |
1da177e4 LT |
1448 | return error; |
1449 | } | |
db08ca25 JA |
1450 | |
1451 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) | |
1452 | { | |
0726b01e | 1453 | return do_madvise(current->mm, start, len_in, behavior); |
db08ca25 | 1454 | } |
ecb8ac8b MK |
1455 | |
1456 | SYSCALL_DEFINE5(process_madvise, int, pidfd, const struct iovec __user *, vec, | |
1457 | size_t, vlen, int, behavior, unsigned int, flags) | |
1458 | { | |
1459 | ssize_t ret; | |
95e49cf8 | 1460 | struct iovec iovstack[UIO_FASTIOV]; |
ecb8ac8b MK |
1461 | struct iovec *iov = iovstack; |
1462 | struct iov_iter iter; | |
ecb8ac8b MK |
1463 | struct task_struct *task; |
1464 | struct mm_struct *mm; | |
1465 | size_t total_len; | |
1466 | unsigned int f_flags; | |
1467 | ||
1468 | if (flags != 0) { | |
1469 | ret = -EINVAL; | |
1470 | goto out; | |
1471 | } | |
1472 | ||
de4eda9d | 1473 | ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter); |
ecb8ac8b MK |
1474 | if (ret < 0) |
1475 | goto out; | |
1476 | ||
ee9955d6 CB |
1477 | task = pidfd_get_task(pidfd, &f_flags); |
1478 | if (IS_ERR(task)) { | |
1479 | ret = PTR_ERR(task); | |
ecb8ac8b MK |
1480 | goto free_iov; |
1481 | } | |
1482 | ||
a68a0262 | 1483 | if (!process_madvise_behavior_valid(behavior)) { |
ecb8ac8b MK |
1484 | ret = -EINVAL; |
1485 | goto release_task; | |
1486 | } | |
1487 | ||
96cfe2c0 SB |
1488 | /* Require PTRACE_MODE_READ to avoid leaking ASLR metadata. */ |
1489 | mm = mm_access(task, PTRACE_MODE_READ_FSCREDS); | |
ecb8ac8b MK |
1490 | if (IS_ERR_OR_NULL(mm)) { |
1491 | ret = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH; | |
1492 | goto release_task; | |
1493 | } | |
1494 | ||
96cfe2c0 SB |
1495 | /* |
1496 | * Require CAP_SYS_NICE for influencing process performance. Note that | |
1497 | * only non-destructive hints are currently supported. | |
1498 | */ | |
1499 | if (!capable(CAP_SYS_NICE)) { | |
1500 | ret = -EPERM; | |
1501 | goto release_mm; | |
1502 | } | |
1503 | ||
ecb8ac8b MK |
1504 | total_len = iov_iter_count(&iter); |
1505 | ||
1506 | while (iov_iter_count(&iter)) { | |
95e49cf8 JA |
1507 | ret = do_madvise(mm, (unsigned long)iter_iov_addr(&iter), |
1508 | iter_iov_len(&iter), behavior); | |
e6b0a7b3 | 1509 | if (ret < 0) |
ecb8ac8b | 1510 | break; |
95e49cf8 | 1511 | iov_iter_advance(&iter, iter_iov_len(&iter)); |
ecb8ac8b MK |
1512 | } |
1513 | ||
5bd009c7 | 1514 | ret = (total_len - iov_iter_count(&iter)) ? : ret; |
ecb8ac8b | 1515 | |
96cfe2c0 | 1516 | release_mm: |
ecb8ac8b | 1517 | mmput(mm); |
ecb8ac8b MK |
1518 | release_task: |
1519 | put_task_struct(task); | |
ecb8ac8b MK |
1520 | free_iov: |
1521 | kfree(iov); | |
1522 | out: | |
1523 | return ret; | |
1524 | } |