]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/mm/mlock.c | |
4 | * | |
5 | * (C) Copyright 1995 Linus Torvalds | |
6 | * (C) Copyright 2002 Christoph Hellwig | |
7 | */ | |
8 | ||
c59ede7b | 9 | #include <linux/capability.h> |
1da177e4 LT |
10 | #include <linux/mman.h> |
11 | #include <linux/mm.h> | |
8703e8a4 | 12 | #include <linux/sched/user.h> |
b291f000 NP |
13 | #include <linux/swap.h> |
14 | #include <linux/swapops.h> | |
15 | #include <linux/pagemap.h> | |
7225522b | 16 | #include <linux/pagevec.h> |
34b67923 | 17 | #include <linux/pagewalk.h> |
1da177e4 LT |
18 | #include <linux/mempolicy.h> |
19 | #include <linux/syscalls.h> | |
e8edc6e0 | 20 | #include <linux/sched.h> |
b95f1b31 | 21 | #include <linux/export.h> |
b291f000 NP |
22 | #include <linux/rmap.h> |
23 | #include <linux/mmzone.h> | |
24 | #include <linux/hugetlb.h> | |
7225522b VB |
25 | #include <linux/memcontrol.h> |
26 | #include <linux/mm_inline.h> | |
1507f512 | 27 | #include <linux/secretmem.h> |
b291f000 NP |
28 | |
29 | #include "internal.h" | |
1da177e4 | 30 | |
90d07210 | 31 | struct mlock_fbatch { |
adb11e78 | 32 | local_lock_t lock; |
90d07210 | 33 | struct folio_batch fbatch; |
adb11e78 SAS |
34 | }; |
35 | ||
90d07210 | 36 | static DEFINE_PER_CPU(struct mlock_fbatch, mlock_fbatch) = { |
adb11e78 SAS |
37 | .lock = INIT_LOCAL_LOCK(lock), |
38 | }; | |
2fbb0c10 | 39 | |
7f43add4 | 40 | bool can_do_mlock(void) |
e8edc6e0 | 41 | { |
59e99e5b | 42 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
7f43add4 | 43 | return true; |
a5a6579d | 44 | if (capable(CAP_IPC_LOCK)) |
7f43add4 WX |
45 | return true; |
46 | return false; | |
e8edc6e0 AD |
47 | } |
48 | EXPORT_SYMBOL(can_do_mlock); | |
1da177e4 | 49 | |
b291f000 | 50 | /* |
90d07210 | 51 | * Mlocked folios are marked with the PG_mlocked flag for efficient testing |
b291f000 NP |
52 | * in vmscan and, possibly, the fault path; and to support semi-accurate |
53 | * statistics. | |
54 | * | |
90d07210 LS |
55 | * An mlocked folio [folio_test_mlocked(folio)] is unevictable. As such, it |
56 | * will be ostensibly placed on the LRU "unevictable" list (actually no such | |
57 | * list exists), rather than the [in]active lists. PG_unevictable is set to | |
58 | * indicate the unevictable state. | |
b291f000 NP |
59 | */ |
60 | ||
90d07210 | 61 | static struct lruvec *__mlock_folio(struct folio *folio, struct lruvec *lruvec) |
b291f000 | 62 | { |
2fbb0c10 | 63 | /* There is nothing more we can do while it's off LRU */ |
90d07210 | 64 | if (!folio_test_clear_lru(folio)) |
2fbb0c10 | 65 | return lruvec; |
0964730b | 66 | |
90d07210 | 67 | lruvec = folio_lruvec_relock_irq(folio, lruvec); |
b291f000 | 68 | |
90d07210 | 69 | if (unlikely(folio_evictable(folio))) { |
b291f000 | 70 | /* |
90d07210 LS |
71 | * This is a little surprising, but quite possible: PG_mlocked |
72 | * must have got cleared already by another CPU. Could this | |
73 | * folio be unevictable? I'm not sure, but move it now if so. | |
b291f000 | 74 | */ |
90d07210 LS |
75 | if (folio_test_unevictable(folio)) { |
76 | lruvec_del_folio(lruvec, folio); | |
77 | folio_clear_unevictable(folio); | |
78 | lruvec_add_folio(lruvec, folio); | |
79 | ||
2fbb0c10 | 80 | __count_vm_events(UNEVICTABLE_PGRESCUED, |
90d07210 | 81 | folio_nr_pages(folio)); |
2fbb0c10 HD |
82 | } |
83 | goto out; | |
b291f000 | 84 | } |
07ca7606 | 85 | |
90d07210 LS |
86 | if (folio_test_unevictable(folio)) { |
87 | if (folio_test_mlocked(folio)) | |
88 | folio->mlock_count++; | |
07ca7606 | 89 | goto out; |
5344b7e6 | 90 | } |
07ca7606 | 91 | |
90d07210 LS |
92 | lruvec_del_folio(lruvec, folio); |
93 | folio_clear_active(folio); | |
94 | folio_set_unevictable(folio); | |
95 | folio->mlock_count = !!folio_test_mlocked(folio); | |
96 | lruvec_add_folio(lruvec, folio); | |
97 | __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio)); | |
07ca7606 | 98 | out: |
90d07210 | 99 | folio_set_lru(folio); |
2fbb0c10 | 100 | return lruvec; |
b291f000 NP |
101 | } |
102 | ||
90d07210 | 103 | static struct lruvec *__mlock_new_folio(struct folio *folio, struct lruvec *lruvec) |
b291f000 | 104 | { |
90d07210 | 105 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
b291f000 | 106 | |
90d07210 | 107 | lruvec = folio_lruvec_relock_irq(folio, lruvec); |
e90309c9 | 108 | |
2fbb0c10 | 109 | /* As above, this is a little surprising, but possible */ |
90d07210 | 110 | if (unlikely(folio_evictable(folio))) |
2fbb0c10 | 111 | goto out; |
0964730b | 112 | |
90d07210 LS |
113 | folio_set_unevictable(folio); |
114 | folio->mlock_count = !!folio_test_mlocked(folio); | |
115 | __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio)); | |
2fbb0c10 | 116 | out: |
90d07210 LS |
117 | lruvec_add_folio(lruvec, folio); |
118 | folio_set_lru(folio); | |
2fbb0c10 | 119 | return lruvec; |
b291f000 NP |
120 | } |
121 | ||
90d07210 | 122 | static struct lruvec *__munlock_folio(struct folio *folio, struct lruvec *lruvec) |
7225522b | 123 | { |
90d07210 | 124 | int nr_pages = folio_nr_pages(folio); |
2fbb0c10 HD |
125 | bool isolated = false; |
126 | ||
90d07210 | 127 | if (!folio_test_clear_lru(folio)) |
2fbb0c10 | 128 | goto munlock; |
7225522b | 129 | |
2fbb0c10 | 130 | isolated = true; |
90d07210 | 131 | lruvec = folio_lruvec_relock_irq(folio, lruvec); |
7225522b | 132 | |
90d07210 | 133 | if (folio_test_unevictable(folio)) { |
07ca7606 | 134 | /* Then mlock_count is maintained, but might undercount */ |
90d07210 LS |
135 | if (folio->mlock_count) |
136 | folio->mlock_count--; | |
137 | if (folio->mlock_count) | |
07ca7606 HD |
138 | goto out; |
139 | } | |
140 | /* else assume that was the last mlock: reclaim will fix it if not */ | |
141 | ||
2fbb0c10 | 142 | munlock: |
90d07210 LS |
143 | if (folio_test_clear_mlocked(folio)) { |
144 | __zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages); | |
145 | if (isolated || !folio_test_unevictable(folio)) | |
07ca7606 HD |
146 | __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages); |
147 | else | |
148 | __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages); | |
149 | } | |
150 | ||
90d07210 LS |
151 | /* folio_evictable() has to be checked *after* clearing Mlocked */ |
152 | if (isolated && folio_test_unevictable(folio) && folio_evictable(folio)) { | |
153 | lruvec_del_folio(lruvec, folio); | |
154 | folio_clear_unevictable(folio); | |
155 | lruvec_add_folio(lruvec, folio); | |
07ca7606 | 156 | __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); |
7225522b | 157 | } |
07ca7606 | 158 | out: |
2fbb0c10 | 159 | if (isolated) |
90d07210 | 160 | folio_set_lru(folio); |
2fbb0c10 | 161 | return lruvec; |
7225522b VB |
162 | } |
163 | ||
164 | /* | |
90d07210 | 165 | * Flags held in the low bits of a struct folio pointer on the mlock_fbatch. |
7225522b | 166 | */ |
90d07210 LS |
167 | #define LRU_FOLIO 0x1 |
168 | #define NEW_FOLIO 0x2 | |
169 | static inline struct folio *mlock_lru(struct folio *folio) | |
7225522b | 170 | { |
90d07210 | 171 | return (struct folio *)((unsigned long)folio + LRU_FOLIO); |
2fbb0c10 | 172 | } |
0964730b | 173 | |
90d07210 | 174 | static inline struct folio *mlock_new(struct folio *folio) |
2fbb0c10 | 175 | { |
90d07210 | 176 | return (struct folio *)((unsigned long)folio + NEW_FOLIO); |
7225522b VB |
177 | } |
178 | ||
2fbb0c10 | 179 | /* |
90d07210 LS |
180 | * mlock_folio_batch() is derived from folio_batch_move_lru(): perhaps that can |
181 | * make use of such folio pointer flags in future, but for now just keep it for | |
182 | * mlock. We could use three separate folio batches instead, but one feels | |
183 | * better (munlocking a full folio batch does not need to drain mlocking folio | |
184 | * batches first). | |
b291f000 | 185 | */ |
90d07210 | 186 | static void mlock_folio_batch(struct folio_batch *fbatch) |
b291f000 | 187 | { |
2fbb0c10 HD |
188 | struct lruvec *lruvec = NULL; |
189 | unsigned long mlock; | |
90d07210 | 190 | struct folio *folio; |
2fbb0c10 | 191 | int i; |
e90309c9 | 192 | |
90d07210 LS |
193 | for (i = 0; i < folio_batch_count(fbatch); i++) { |
194 | folio = fbatch->folios[i]; | |
195 | mlock = (unsigned long)folio & (LRU_FOLIO | NEW_FOLIO); | |
196 | folio = (struct folio *)((unsigned long)folio - mlock); | |
197 | fbatch->folios[i] = folio; | |
2fbb0c10 | 198 | |
90d07210 LS |
199 | if (mlock & LRU_FOLIO) |
200 | lruvec = __mlock_folio(folio, lruvec); | |
201 | else if (mlock & NEW_FOLIO) | |
202 | lruvec = __mlock_new_folio(folio, lruvec); | |
2fbb0c10 | 203 | else |
90d07210 | 204 | lruvec = __munlock_folio(folio, lruvec); |
655548bf | 205 | } |
01cc2e58 | 206 | |
2fbb0c10 HD |
207 | if (lruvec) |
208 | unlock_page_lruvec_irq(lruvec); | |
99fbb6bf | 209 | folios_put(fbatch); |
2fbb0c10 | 210 | } |
01cc2e58 | 211 | |
96f97c43 | 212 | void mlock_drain_local(void) |
adb11e78 | 213 | { |
90d07210 | 214 | struct folio_batch *fbatch; |
adb11e78 | 215 | |
90d07210 LS |
216 | local_lock(&mlock_fbatch.lock); |
217 | fbatch = this_cpu_ptr(&mlock_fbatch.fbatch); | |
218 | if (folio_batch_count(fbatch)) | |
219 | mlock_folio_batch(fbatch); | |
220 | local_unlock(&mlock_fbatch.lock); | |
adb11e78 SAS |
221 | } |
222 | ||
96f97c43 | 223 | void mlock_drain_remote(int cpu) |
2fbb0c10 | 224 | { |
90d07210 | 225 | struct folio_batch *fbatch; |
01cc2e58 | 226 | |
adb11e78 | 227 | WARN_ON_ONCE(cpu_online(cpu)); |
90d07210 LS |
228 | fbatch = &per_cpu(mlock_fbatch.fbatch, cpu); |
229 | if (folio_batch_count(fbatch)) | |
230 | mlock_folio_batch(fbatch); | |
b291f000 NP |
231 | } |
232 | ||
96f97c43 | 233 | bool need_mlock_drain(int cpu) |
9978ad58 | 234 | { |
90d07210 | 235 | return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu)); |
b291f000 NP |
236 | } |
237 | ||
2fbb0c10 | 238 | /** |
dcc5d337 MWO |
239 | * mlock_folio - mlock a folio already on (or temporarily off) LRU |
240 | * @folio: folio to be mlocked. | |
56afe477 | 241 | */ |
dcc5d337 | 242 | void mlock_folio(struct folio *folio) |
56afe477 | 243 | { |
90d07210 | 244 | struct folio_batch *fbatch; |
adb11e78 | 245 | |
90d07210 LS |
246 | local_lock(&mlock_fbatch.lock); |
247 | fbatch = this_cpu_ptr(&mlock_fbatch.fbatch); | |
56afe477 | 248 | |
dcc5d337 MWO |
249 | if (!folio_test_set_mlocked(folio)) { |
250 | int nr_pages = folio_nr_pages(folio); | |
2fbb0c10 | 251 | |
dcc5d337 | 252 | zone_stat_mod_folio(folio, NR_MLOCK, nr_pages); |
2fbb0c10 | 253 | __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages); |
56afe477 VB |
254 | } |
255 | ||
dcc5d337 | 256 | folio_get(folio); |
90d07210 | 257 | if (!folio_batch_add(fbatch, mlock_lru(folio)) || |
dcc5d337 | 258 | folio_test_large(folio) || lru_cache_disabled()) |
90d07210 LS |
259 | mlock_folio_batch(fbatch); |
260 | local_unlock(&mlock_fbatch.lock); | |
56afe477 VB |
261 | } |
262 | ||
2fbb0c10 | 263 | /** |
96f97c43 LS |
264 | * mlock_new_folio - mlock a newly allocated folio not yet on LRU |
265 | * @folio: folio to be mlocked, either normal or a THP head. | |
56afe477 | 266 | */ |
96f97c43 | 267 | void mlock_new_folio(struct folio *folio) |
56afe477 | 268 | { |
90d07210 | 269 | struct folio_batch *fbatch; |
90d07210 | 270 | int nr_pages = folio_nr_pages(folio); |
2fbb0c10 | 271 | |
90d07210 LS |
272 | local_lock(&mlock_fbatch.lock); |
273 | fbatch = this_cpu_ptr(&mlock_fbatch.fbatch); | |
274 | folio_set_mlocked(folio); | |
275 | ||
276 | zone_stat_mod_folio(folio, NR_MLOCK, nr_pages); | |
2fbb0c10 HD |
277 | __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages); |
278 | ||
90d07210 LS |
279 | folio_get(folio); |
280 | if (!folio_batch_add(fbatch, mlock_new(folio)) || | |
281 | folio_test_large(folio) || lru_cache_disabled()) | |
282 | mlock_folio_batch(fbatch); | |
283 | local_unlock(&mlock_fbatch.lock); | |
56afe477 VB |
284 | } |
285 | ||
2fbb0c10 | 286 | /** |
96f97c43 LS |
287 | * munlock_folio - munlock a folio |
288 | * @folio: folio to be munlocked, either normal or a THP head. | |
7225522b | 289 | */ |
96f97c43 | 290 | void munlock_folio(struct folio *folio) |
7225522b | 291 | { |
90d07210 | 292 | struct folio_batch *fbatch; |
56afe477 | 293 | |
90d07210 LS |
294 | local_lock(&mlock_fbatch.lock); |
295 | fbatch = this_cpu_ptr(&mlock_fbatch.fbatch); | |
5b40998a | 296 | /* |
90d07210 LS |
297 | * folio_test_clear_mlocked(folio) must be left to __munlock_folio(), |
298 | * which will check whether the folio is multiply mlocked. | |
5b40998a | 299 | */ |
90d07210 LS |
300 | folio_get(folio); |
301 | if (!folio_batch_add(fbatch, folio) || | |
302 | folio_test_large(folio) || lru_cache_disabled()) | |
303 | mlock_folio_batch(fbatch); | |
304 | local_unlock(&mlock_fbatch.lock); | |
7a8010cd VB |
305 | } |
306 | ||
dc68badc YF |
307 | static inline unsigned int folio_mlock_step(struct folio *folio, |
308 | pte_t *pte, unsigned long addr, unsigned long end) | |
309 | { | |
f742829d LY |
310 | const fpb_t fpb_flags = FPB_IGNORE_DIRTY | FPB_IGNORE_SOFT_DIRTY; |
311 | unsigned int count = (end - addr) >> PAGE_SHIFT; | |
dc68badc YF |
312 | pte_t ptent = ptep_get(pte); |
313 | ||
314 | if (!folio_test_large(folio)) | |
315 | return 1; | |
316 | ||
f742829d LY |
317 | return folio_pte_batch(folio, addr, pte, ptent, count, fpb_flags, NULL, |
318 | NULL, NULL); | |
dc68badc YF |
319 | } |
320 | ||
321 | static inline bool allow_mlock_munlock(struct folio *folio, | |
322 | struct vm_area_struct *vma, unsigned long start, | |
323 | unsigned long end, unsigned int step) | |
324 | { | |
325 | /* | |
326 | * For unlock, allow munlock large folio which is partially | |
327 | * mapped to VMA. As it's possible that large folio is | |
328 | * mlocked and VMA is split later. | |
329 | * | |
330 | * During memory pressure, such kind of large folio can | |
331 | * be split. And the pages are not in VM_LOCKed VMA | |
332 | * can be reclaimed. | |
333 | */ | |
334 | if (!(vma->vm_flags & VM_LOCKED)) | |
335 | return true; | |
336 | ||
b1454b46 HD |
337 | /* folio_within_range() cannot take KSM, but any small folio is OK */ |
338 | if (!folio_test_large(folio)) | |
339 | return true; | |
340 | ||
dc68badc YF |
341 | /* folio not in range [start, end), skip mlock */ |
342 | if (!folio_within_range(folio, vma, start, end)) | |
343 | return false; | |
344 | ||
345 | /* folio is not fully mapped, skip mlock */ | |
346 | if (step != folio_nr_pages(folio)) | |
347 | return false; | |
348 | ||
349 | return true; | |
350 | } | |
351 | ||
34b67923 HD |
352 | static int mlock_pte_range(pmd_t *pmd, unsigned long addr, |
353 | unsigned long end, struct mm_walk *walk) | |
354 | ||
7a8010cd | 355 | { |
34b67923 | 356 | struct vm_area_struct *vma = walk->vma; |
7a8010cd | 357 | spinlock_t *ptl; |
34b67923 | 358 | pte_t *start_pte, *pte; |
c33c7948 | 359 | pte_t ptent; |
96f97c43 | 360 | struct folio *folio; |
dc68badc YF |
361 | unsigned int step = 1; |
362 | unsigned long start = addr; | |
7a8010cd | 363 | |
34b67923 HD |
364 | ptl = pmd_trans_huge_lock(pmd, vma); |
365 | if (ptl) { | |
366 | if (!pmd_present(*pmd)) | |
367 | goto out; | |
368 | if (is_huge_zero_pmd(*pmd)) | |
369 | goto out; | |
e06d03d5 | 370 | folio = pmd_folio(*pmd); |
82ba975e AP |
371 | if (folio_is_zone_device(folio)) |
372 | goto out; | |
34b67923 | 373 | if (vma->vm_flags & VM_LOCKED) |
96f97c43 | 374 | mlock_folio(folio); |
34b67923 | 375 | else |
96f97c43 | 376 | munlock_folio(folio); |
34b67923 HD |
377 | goto out; |
378 | } | |
56afe477 | 379 | |
34b67923 | 380 | start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
7780d040 HD |
381 | if (!start_pte) { |
382 | walk->action = ACTION_AGAIN; | |
383 | return 0; | |
384 | } | |
dc68badc | 385 | |
34b67923 | 386 | for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) { |
c33c7948 RR |
387 | ptent = ptep_get(pte); |
388 | if (!pte_present(ptent)) | |
34b67923 | 389 | continue; |
c33c7948 | 390 | folio = vm_normal_folio(vma, addr, ptent); |
96f97c43 | 391 | if (!folio || folio_is_zone_device(folio)) |
34b67923 | 392 | continue; |
dc68badc YF |
393 | |
394 | step = folio_mlock_step(folio, pte, addr, end); | |
395 | if (!allow_mlock_munlock(folio, vma, start, end, step)) | |
396 | goto next_entry; | |
397 | ||
34b67923 | 398 | if (vma->vm_flags & VM_LOCKED) |
96f97c43 | 399 | mlock_folio(folio); |
34b67923 | 400 | else |
96f97c43 | 401 | munlock_folio(folio); |
dc68badc YF |
402 | |
403 | next_entry: | |
404 | pte += step - 1; | |
405 | addr += (step - 1) << PAGE_SHIFT; | |
7a8010cd | 406 | } |
34b67923 HD |
407 | pte_unmap(start_pte); |
408 | out: | |
409 | spin_unlock(ptl); | |
410 | cond_resched(); | |
411 | return 0; | |
7225522b VB |
412 | } |
413 | ||
b291f000 | 414 | /* |
34b67923 HD |
415 | * mlock_vma_pages_range() - mlock any pages already in the range, |
416 | * or munlock all pages in the range. | |
417 | * @vma - vma containing range to be mlock()ed or munlock()ed | |
ba470de4 | 418 | * @start - start address in @vma of the range |
34b67923 HD |
419 | * @end - end of range in @vma |
420 | * @newflags - the new set of flags for @vma. | |
ba470de4 | 421 | * |
34b67923 HD |
422 | * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED; |
423 | * called for munlock() and munlockall(), to clear VM_LOCKED from @vma. | |
b291f000 | 424 | */ |
34b67923 HD |
425 | static void mlock_vma_pages_range(struct vm_area_struct *vma, |
426 | unsigned long start, unsigned long end, vm_flags_t newflags) | |
b291f000 | 427 | { |
34b67923 HD |
428 | static const struct mm_walk_ops mlock_walk_ops = { |
429 | .pmd_entry = mlock_pte_range, | |
49b06385 | 430 | .walk_lock = PGWALK_WRLOCK_VERIFY, |
34b67923 | 431 | }; |
408e82b7 | 432 | |
34b67923 HD |
433 | /* |
434 | * There is a slight chance that concurrent page migration, | |
435 | * or page reclaim finding a page of this now-VM_LOCKED vma, | |
7efecffb | 436 | * will call mlock_vma_folio() and raise page's mlock_count: |
34b67923 | 437 | * double counting, leaving the page unevictable indefinitely. |
7efecffb | 438 | * Communicate this danger to mlock_vma_folio() with VM_IO, |
34b67923 HD |
439 | * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas. |
440 | * mmap_lock is held in write mode here, so this weird | |
441 | * combination should not be visible to other mmap_lock users; | |
442 | * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED. | |
443 | */ | |
444 | if (newflags & VM_LOCKED) | |
445 | newflags |= VM_IO; | |
60081bf1 | 446 | vma_start_write(vma); |
601c3c29 | 447 | vm_flags_reset_once(vma, newflags); |
ff6a6da6 | 448 | |
34b67923 HD |
449 | lru_add_drain(); |
450 | walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL); | |
451 | lru_add_drain(); | |
408e82b7 | 452 | |
34b67923 HD |
453 | if (newflags & VM_IO) { |
454 | newflags &= ~VM_IO; | |
601c3c29 | 455 | vm_flags_reset_once(vma, newflags); |
408e82b7 | 456 | } |
b291f000 NP |
457 | } |
458 | ||
459 | /* | |
460 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
461 | * | |
462 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
463 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 464 | * populate the ptes. |
b291f000 NP |
465 | * |
466 | * For vmas that pass the filters, merge/split as appropriate. | |
467 | */ | |
37598f5a LH |
468 | static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma, |
469 | struct vm_area_struct **prev, unsigned long start, | |
470 | unsigned long end, vm_flags_t newflags) | |
1da177e4 | 471 | { |
b291f000 | 472 | struct mm_struct *mm = vma->vm_mm; |
b291f000 | 473 | int nr_pages; |
1da177e4 | 474 | int ret = 0; |
34b67923 | 475 | vm_flags_t oldflags = vma->vm_flags; |
1da177e4 | 476 | |
34b67923 | 477 | if (newflags == oldflags || (oldflags & VM_SPECIAL) || |
e1fb4a08 | 478 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) || |
9651fced | 479 | vma_is_dax(vma) || vma_is_secretmem(vma) || (oldflags & VM_DROPPABLE)) |
b0f205c2 EM |
480 | /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */ |
481 | goto out; | |
b291f000 | 482 | |
94d7d923 LS |
483 | vma = vma_modify_flags(vmi, *prev, vma, start, end, newflags); |
484 | if (IS_ERR(vma)) { | |
485 | ret = PTR_ERR(vma); | |
486 | goto out; | |
1da177e4 LT |
487 | } |
488 | ||
b291f000 NP |
489 | /* |
490 | * Keep track of amount of locked VM. | |
491 | */ | |
492 | nr_pages = (end - start) >> PAGE_SHIFT; | |
34b67923 | 493 | if (!(newflags & VM_LOCKED)) |
b291f000 | 494 | nr_pages = -nr_pages; |
34b67923 | 495 | else if (oldflags & VM_LOCKED) |
b155b4fd | 496 | nr_pages = 0; |
b291f000 NP |
497 | mm->locked_vm += nr_pages; |
498 | ||
1da177e4 | 499 | /* |
c1e8d7c6 | 500 | * vm_flags is protected by the mmap_lock held in write mode. |
1da177e4 | 501 | * It's okay if try_to_unmap_one unmaps a page just after we |
fc05f566 | 502 | * set VM_LOCKED, populate_vma_page_range will bring it back. |
1da177e4 | 503 | */ |
34b67923 HD |
504 | if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) { |
505 | /* No work to do, and mlocking twice would be wrong */ | |
60081bf1 | 506 | vma_start_write(vma); |
1c71222e | 507 | vm_flags_reset(vma, newflags); |
34b67923 HD |
508 | } else { |
509 | mlock_vma_pages_range(vma, start, end, newflags); | |
510 | } | |
1da177e4 | 511 | out: |
b291f000 | 512 | *prev = vma; |
1da177e4 LT |
513 | return ret; |
514 | } | |
515 | ||
1aab92ec EM |
516 | static int apply_vma_lock_flags(unsigned long start, size_t len, |
517 | vm_flags_t flags) | |
1da177e4 LT |
518 | { |
519 | unsigned long nstart, end, tmp; | |
68d68ff6 | 520 | struct vm_area_struct *vma, *prev; |
37598f5a | 521 | VMA_ITERATOR(vmi, current->mm, start); |
1da177e4 | 522 | |
8fd9e488 | 523 | VM_BUG_ON(offset_in_page(start)); |
fed067da | 524 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
1da177e4 LT |
525 | end = start + len; |
526 | if (end < start) | |
527 | return -EINVAL; | |
528 | if (end == start) | |
529 | return 0; | |
37598f5a | 530 | vma = vma_iter_load(&vmi); |
33108b05 | 531 | if (!vma) |
1da177e4 LT |
532 | return -ENOMEM; |
533 | ||
37598f5a | 534 | prev = vma_prev(&vmi); |
1da177e4 LT |
535 | if (start > vma->vm_start) |
536 | prev = vma; | |
537 | ||
37598f5a LH |
538 | nstart = start; |
539 | tmp = vma->vm_start; | |
540 | for_each_vma_range(vmi, vma, end) { | |
2658f94d | 541 | int error; |
37598f5a | 542 | vm_flags_t newflags; |
1da177e4 | 543 | |
37598f5a LH |
544 | if (vma->vm_start != tmp) |
545 | return -ENOMEM; | |
1da177e4 | 546 | |
e430a95a | 547 | newflags = vma->vm_flags & ~VM_LOCKED_MASK; |
37598f5a | 548 | newflags |= flags; |
1aab92ec | 549 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 LT |
550 | tmp = vma->vm_end; |
551 | if (tmp > end) | |
552 | tmp = end; | |
37598f5a | 553 | error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags); |
1da177e4 | 554 | if (error) |
2658f94d LH |
555 | return error; |
556 | tmp = vma_iter_end(&vmi); | |
1da177e4 | 557 | nstart = tmp; |
1da177e4 | 558 | } |
37598f5a | 559 | |
2658f94d | 560 | if (tmp < end) |
37598f5a LH |
561 | return -ENOMEM; |
562 | ||
2658f94d | 563 | return 0; |
1da177e4 LT |
564 | } |
565 | ||
0cf2f6f6 SG |
566 | /* |
567 | * Go through vma areas and sum size of mlocked | |
568 | * vma pages, as return value. | |
569 | * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT) | |
570 | * is also counted. | |
571 | * Return value: previously mlocked page counts | |
572 | */ | |
0874bb49 | 573 | static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm, |
0cf2f6f6 SG |
574 | unsigned long start, size_t len) |
575 | { | |
576 | struct vm_area_struct *vma; | |
0874bb49 | 577 | unsigned long count = 0; |
33108b05 MWO |
578 | unsigned long end; |
579 | VMA_ITERATOR(vmi, mm, start); | |
0cf2f6f6 | 580 | |
33108b05 MWO |
581 | /* Don't overflow past ULONG_MAX */ |
582 | if (unlikely(ULONG_MAX - len < start)) | |
583 | end = ULONG_MAX; | |
584 | else | |
585 | end = start + len; | |
66071896 | 586 | |
33108b05 | 587 | for_each_vma_range(vmi, vma, end) { |
0cf2f6f6 SG |
588 | if (vma->vm_flags & VM_LOCKED) { |
589 | if (start > vma->vm_start) | |
590 | count -= (start - vma->vm_start); | |
33108b05 MWO |
591 | if (end < vma->vm_end) { |
592 | count += end - vma->vm_start; | |
0cf2f6f6 SG |
593 | break; |
594 | } | |
595 | count += vma->vm_end - vma->vm_start; | |
596 | } | |
597 | } | |
598 | ||
599 | return count >> PAGE_SHIFT; | |
600 | } | |
601 | ||
ebcbc6ea HD |
602 | /* |
603 | * convert get_user_pages() return value to posix mlock() error | |
604 | */ | |
605 | static int __mlock_posix_error_return(long retval) | |
606 | { | |
607 | if (retval == -EFAULT) | |
608 | retval = -ENOMEM; | |
609 | else if (retval == -ENOMEM) | |
610 | retval = -EAGAIN; | |
611 | return retval; | |
612 | } | |
613 | ||
dc0ef0df | 614 | static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags) |
1da177e4 LT |
615 | { |
616 | unsigned long locked; | |
617 | unsigned long lock_limit; | |
618 | int error = -ENOMEM; | |
619 | ||
057d3389 AK |
620 | start = untagged_addr(start); |
621 | ||
1da177e4 LT |
622 | if (!can_do_mlock()) |
623 | return -EPERM; | |
624 | ||
8fd9e488 | 625 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 LT |
626 | start &= PAGE_MASK; |
627 | ||
59e99e5b | 628 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 | 629 | lock_limit >>= PAGE_SHIFT; |
1f1cd705 DB |
630 | locked = len >> PAGE_SHIFT; |
631 | ||
d8ed45c5 | 632 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 633 | return -EINTR; |
1f1cd705 DB |
634 | |
635 | locked += current->mm->locked_vm; | |
0cf2f6f6 SG |
636 | if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) { |
637 | /* | |
638 | * It is possible that the regions requested intersect with | |
639 | * previously mlocked areas, that part area in "mm->locked_vm" | |
640 | * should not be counted to new mlock increment count. So check | |
641 | * and adjust locked count if necessary. | |
642 | */ | |
643 | locked -= count_mm_mlocked_page_nr(current->mm, | |
644 | start, len); | |
645 | } | |
1da177e4 LT |
646 | |
647 | /* check against resource limits */ | |
648 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
1aab92ec | 649 | error = apply_vma_lock_flags(start, len, flags); |
1f1cd705 | 650 | |
d8ed45c5 | 651 | mmap_write_unlock(current->mm); |
c561259c KS |
652 | if (error) |
653 | return error; | |
654 | ||
655 | error = __mm_populate(start, len, 0); | |
656 | if (error) | |
657 | return __mlock_posix_error_return(error); | |
658 | return 0; | |
1da177e4 LT |
659 | } |
660 | ||
1aab92ec EM |
661 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
662 | { | |
663 | return do_mlock(start, len, VM_LOCKED); | |
664 | } | |
665 | ||
a8ca5d0e EM |
666 | SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags) |
667 | { | |
b0f205c2 EM |
668 | vm_flags_t vm_flags = VM_LOCKED; |
669 | ||
670 | if (flags & ~MLOCK_ONFAULT) | |
a8ca5d0e EM |
671 | return -EINVAL; |
672 | ||
b0f205c2 EM |
673 | if (flags & MLOCK_ONFAULT) |
674 | vm_flags |= VM_LOCKONFAULT; | |
675 | ||
676 | return do_mlock(start, len, vm_flags); | |
a8ca5d0e EM |
677 | } |
678 | ||
6a6160a7 | 679 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
680 | { |
681 | int ret; | |
682 | ||
057d3389 AK |
683 | start = untagged_addr(start); |
684 | ||
8fd9e488 | 685 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 | 686 | start &= PAGE_MASK; |
1f1cd705 | 687 | |
d8ed45c5 | 688 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 689 | return -EINTR; |
1aab92ec | 690 | ret = apply_vma_lock_flags(start, len, 0); |
d8ed45c5 | 691 | mmap_write_unlock(current->mm); |
1f1cd705 | 692 | |
1da177e4 LT |
693 | return ret; |
694 | } | |
695 | ||
b0f205c2 EM |
696 | /* |
697 | * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall) | |
698 | * and translate into the appropriate modifications to mm->def_flags and/or the | |
699 | * flags for all current VMAs. | |
700 | * | |
701 | * There are a couple of subtleties with this. If mlockall() is called multiple | |
702 | * times with different flags, the values do not necessarily stack. If mlockall | |
703 | * is called once including the MCL_FUTURE flag and then a second time without | |
704 | * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags. | |
705 | */ | |
1aab92ec | 706 | static int apply_mlockall_flags(int flags) |
1da177e4 | 707 | { |
37598f5a | 708 | VMA_ITERATOR(vmi, current->mm, 0); |
68d68ff6 | 709 | struct vm_area_struct *vma, *prev = NULL; |
b0f205c2 | 710 | vm_flags_t to_add = 0; |
1da177e4 | 711 | |
e430a95a | 712 | current->mm->def_flags &= ~VM_LOCKED_MASK; |
b0f205c2 | 713 | if (flags & MCL_FUTURE) { |
09a9f1d2 | 714 | current->mm->def_flags |= VM_LOCKED; |
1aab92ec | 715 | |
b0f205c2 EM |
716 | if (flags & MCL_ONFAULT) |
717 | current->mm->def_flags |= VM_LOCKONFAULT; | |
718 | ||
719 | if (!(flags & MCL_CURRENT)) | |
720 | goto out; | |
721 | } | |
722 | ||
723 | if (flags & MCL_CURRENT) { | |
724 | to_add |= VM_LOCKED; | |
725 | if (flags & MCL_ONFAULT) | |
726 | to_add |= VM_LOCKONFAULT; | |
727 | } | |
1da177e4 | 728 | |
37598f5a | 729 | for_each_vma(vmi, vma) { |
faa242b1 | 730 | int error; |
ca16d140 | 731 | vm_flags_t newflags; |
1da177e4 | 732 | |
e430a95a | 733 | newflags = vma->vm_flags & ~VM_LOCKED_MASK; |
b0f205c2 | 734 | newflags |= to_add; |
1da177e4 | 735 | |
faa242b1 WY |
736 | error = mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end, |
737 | newflags); | |
738 | /* Ignore errors, but prev needs fixing up. */ | |
739 | if (error) | |
740 | prev = vma; | |
50d4fb78 | 741 | cond_resched(); |
1da177e4 LT |
742 | } |
743 | out: | |
744 | return 0; | |
745 | } | |
746 | ||
3480b257 | 747 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
748 | { |
749 | unsigned long lock_limit; | |
86d2adcc | 750 | int ret; |
1da177e4 | 751 | |
dedca635 PS |
752 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) || |
753 | flags == MCL_ONFAULT) | |
86d2adcc | 754 | return -EINVAL; |
1da177e4 | 755 | |
1da177e4 | 756 | if (!can_do_mlock()) |
86d2adcc | 757 | return -EPERM; |
1da177e4 | 758 | |
59e99e5b | 759 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
760 | lock_limit >>= PAGE_SHIFT; |
761 | ||
d8ed45c5 | 762 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 763 | return -EINTR; |
1f1cd705 | 764 | |
dc0ef0df | 765 | ret = -ENOMEM; |
1da177e4 LT |
766 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
767 | capable(CAP_IPC_LOCK)) | |
1aab92ec | 768 | ret = apply_mlockall_flags(flags); |
d8ed45c5 | 769 | mmap_write_unlock(current->mm); |
bebeb3d6 ML |
770 | if (!ret && (flags & MCL_CURRENT)) |
771 | mm_populate(0, TASK_SIZE); | |
86d2adcc | 772 | |
1da177e4 LT |
773 | return ret; |
774 | } | |
775 | ||
3480b257 | 776 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
777 | { |
778 | int ret; | |
779 | ||
d8ed45c5 | 780 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 781 | return -EINTR; |
1aab92ec | 782 | ret = apply_mlockall_flags(0); |
d8ed45c5 | 783 | mmap_write_unlock(current->mm); |
1da177e4 LT |
784 | return ret; |
785 | } | |
786 | ||
787 | /* | |
788 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
789 | * shm segments) get accounted against the user_struct instead. | |
790 | */ | |
791 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
792 | ||
d7c9e99a | 793 | int user_shm_lock(size_t size, struct ucounts *ucounts) |
1da177e4 LT |
794 | { |
795 | unsigned long lock_limit, locked; | |
d7c9e99a | 796 | long memlock; |
1da177e4 LT |
797 | int allowed = 0; |
798 | ||
799 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 800 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
e97824ff ML |
801 | if (lock_limit != RLIM_INFINITY) |
802 | lock_limit >>= PAGE_SHIFT; | |
1da177e4 | 803 | spin_lock(&shmlock_user_lock); |
d7c9e99a AG |
804 | memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked); |
805 | ||
e97824ff | 806 | if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) { |
d7c9e99a AG |
807 | dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked); |
808 | goto out; | |
809 | } | |
810 | if (!get_ucounts(ucounts)) { | |
811 | dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked); | |
5c2a956c | 812 | allowed = 0; |
1da177e4 | 813 | goto out; |
d7c9e99a | 814 | } |
1da177e4 LT |
815 | allowed = 1; |
816 | out: | |
817 | spin_unlock(&shmlock_user_lock); | |
818 | return allowed; | |
819 | } | |
820 | ||
d7c9e99a | 821 | void user_shm_unlock(size_t size, struct ucounts *ucounts) |
1da177e4 LT |
822 | { |
823 | spin_lock(&shmlock_user_lock); | |
d7c9e99a | 824 | dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT); |
1da177e4 | 825 | spin_unlock(&shmlock_user_lock); |
d7c9e99a | 826 | put_ucounts(ucounts); |
1da177e4 | 827 | } |