]>
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> |
1da177e4 LT |
17 | #include <linux/mempolicy.h> |
18 | #include <linux/syscalls.h> | |
e8edc6e0 | 19 | #include <linux/sched.h> |
b95f1b31 | 20 | #include <linux/export.h> |
b291f000 NP |
21 | #include <linux/rmap.h> |
22 | #include <linux/mmzone.h> | |
23 | #include <linux/hugetlb.h> | |
7225522b VB |
24 | #include <linux/memcontrol.h> |
25 | #include <linux/mm_inline.h> | |
b291f000 NP |
26 | |
27 | #include "internal.h" | |
1da177e4 | 28 | |
7f43add4 | 29 | bool can_do_mlock(void) |
e8edc6e0 | 30 | { |
59e99e5b | 31 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
7f43add4 | 32 | return true; |
a5a6579d | 33 | if (capable(CAP_IPC_LOCK)) |
7f43add4 WX |
34 | return true; |
35 | return false; | |
e8edc6e0 AD |
36 | } |
37 | EXPORT_SYMBOL(can_do_mlock); | |
1da177e4 | 38 | |
b291f000 NP |
39 | /* |
40 | * Mlocked pages are marked with PageMlocked() flag for efficient testing | |
41 | * in vmscan and, possibly, the fault path; and to support semi-accurate | |
42 | * statistics. | |
43 | * | |
44 | * An mlocked page [PageMlocked(page)] is unevictable. As such, it will | |
45 | * be placed on the LRU "unevictable" list, rather than the [in]active lists. | |
46 | * The unevictable list is an LRU sibling list to the [in]active lists. | |
47 | * PageUnevictable is set to indicate the unevictable state. | |
48 | * | |
49 | * When lazy mlocking via vmscan, it is important to ensure that the | |
50 | * vma's VM_LOCKED status is not concurrently being modified, otherwise we | |
51 | * may have mlocked a page that is being munlocked. So lazy mlock must take | |
c1e8d7c6 | 52 | * the mmap_lock for read, and verify that the vma really is locked |
b291f000 NP |
53 | * (see mm/rmap.c). |
54 | */ | |
55 | ||
56 | /* | |
57 | * LRU accounting for clear_page_mlock() | |
58 | */ | |
e6c509f8 | 59 | void clear_page_mlock(struct page *page) |
b291f000 | 60 | { |
e6c509f8 | 61 | if (!TestClearPageMlocked(page)) |
b291f000 | 62 | return; |
b291f000 | 63 | |
6c357848 | 64 | mod_zone_page_state(page_zone(page), NR_MLOCK, -thp_nr_pages(page)); |
5344b7e6 | 65 | count_vm_event(UNEVICTABLE_PGCLEARED); |
9c4e6b1a SB |
66 | /* |
67 | * The previous TestClearPageMlocked() corresponds to the smp_mb() | |
68 | * in __pagevec_lru_add_fn(). | |
69 | * | |
70 | * See __pagevec_lru_add_fn for more explanation. | |
71 | */ | |
b291f000 NP |
72 | if (!isolate_lru_page(page)) { |
73 | putback_lru_page(page); | |
74 | } else { | |
75 | /* | |
8891d6da | 76 | * We lost the race. the page already moved to evictable list. |
b291f000 | 77 | */ |
8891d6da | 78 | if (PageUnevictable(page)) |
5344b7e6 | 79 | count_vm_event(UNEVICTABLE_PGSTRANDED); |
b291f000 NP |
80 | } |
81 | } | |
82 | ||
83 | /* | |
84 | * Mark page as mlocked if not already. | |
85 | * If page on LRU, isolate and putback to move to unevictable list. | |
86 | */ | |
87 | void mlock_vma_page(struct page *page) | |
88 | { | |
57e68e9c | 89 | /* Serialize with page migration */ |
b291f000 NP |
90 | BUG_ON(!PageLocked(page)); |
91 | ||
e90309c9 KS |
92 | VM_BUG_ON_PAGE(PageTail(page), page); |
93 | VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page); | |
94 | ||
5344b7e6 | 95 | if (!TestSetPageMlocked(page)) { |
8449d21f | 96 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
6c357848 | 97 | thp_nr_pages(page)); |
5344b7e6 NP |
98 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
99 | if (!isolate_lru_page(page)) | |
100 | putback_lru_page(page); | |
101 | } | |
b291f000 NP |
102 | } |
103 | ||
01cc2e58 VB |
104 | /* |
105 | * Isolate a page from LRU with optional get_page() pin. | |
106 | * Assumes lru_lock already held and page already pinned. | |
107 | */ | |
108 | static bool __munlock_isolate_lru_page(struct page *page, bool getpage) | |
109 | { | |
110 | if (PageLRU(page)) { | |
111 | struct lruvec *lruvec; | |
112 | ||
599d0c95 | 113 | lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page)); |
01cc2e58 VB |
114 | if (getpage) |
115 | get_page(page); | |
116 | ClearPageLRU(page); | |
117 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
118 | return true; | |
119 | } | |
120 | ||
121 | return false; | |
122 | } | |
123 | ||
7225522b VB |
124 | /* |
125 | * Finish munlock after successful page isolation | |
126 | * | |
127 | * Page must be locked. This is a wrapper for try_to_munlock() | |
128 | * and putback_lru_page() with munlock accounting. | |
129 | */ | |
130 | static void __munlock_isolated_page(struct page *page) | |
131 | { | |
7225522b VB |
132 | /* |
133 | * Optimization: if the page was mapped just once, that's our mapping | |
134 | * and we don't need to check all the other vmas. | |
135 | */ | |
136 | if (page_mapcount(page) > 1) | |
192d7232 | 137 | try_to_munlock(page); |
7225522b VB |
138 | |
139 | /* Did try_to_unlock() succeed or punt? */ | |
192d7232 | 140 | if (!PageMlocked(page)) |
7225522b VB |
141 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
142 | ||
143 | putback_lru_page(page); | |
144 | } | |
145 | ||
146 | /* | |
147 | * Accounting for page isolation fail during munlock | |
148 | * | |
149 | * Performs accounting when page isolation fails in munlock. There is nothing | |
150 | * else to do because it means some other task has already removed the page | |
151 | * from the LRU. putback_lru_page() will take care of removing the page from | |
152 | * the unevictable list, if necessary. vmscan [page_referenced()] will move | |
153 | * the page back to the unevictable list if some other vma has it mlocked. | |
154 | */ | |
155 | static void __munlock_isolation_failed(struct page *page) | |
156 | { | |
157 | if (PageUnevictable(page)) | |
01cc2e58 | 158 | __count_vm_event(UNEVICTABLE_PGSTRANDED); |
7225522b | 159 | else |
01cc2e58 | 160 | __count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
7225522b VB |
161 | } |
162 | ||
6927c1dd LS |
163 | /** |
164 | * munlock_vma_page - munlock a vma page | |
b7701a5f | 165 | * @page: page to be unlocked, either a normal page or THP page head |
c424be1c VB |
166 | * |
167 | * returns the size of the page as a page mask (0 for normal page, | |
168 | * HPAGE_PMD_NR - 1 for THP head page) | |
b291f000 | 169 | * |
6927c1dd LS |
170 | * called from munlock()/munmap() path with page supposedly on the LRU. |
171 | * When we munlock a page, because the vma where we found the page is being | |
172 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the | |
173 | * page locked so that we can leave it on the unevictable lru list and not | |
174 | * bother vmscan with it. However, to walk the page's rmap list in | |
175 | * try_to_munlock() we must isolate the page from the LRU. If some other | |
176 | * task has removed the page from the LRU, we won't be able to do that. | |
177 | * So we clear the PageMlocked as we might not get another chance. If we | |
178 | * can't isolate the page, we leave it for putback_lru_page() and vmscan | |
179 | * [page_referenced()/try_to_unmap()] to deal with. | |
b291f000 | 180 | */ |
ff6a6da6 | 181 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 182 | { |
7162a1e8 | 183 | int nr_pages; |
f4b7e272 | 184 | pg_data_t *pgdat = page_pgdat(page); |
ff6a6da6 | 185 | |
57e68e9c | 186 | /* For try_to_munlock() and to serialize with page migration */ |
b291f000 NP |
187 | BUG_ON(!PageLocked(page)); |
188 | ||
e90309c9 KS |
189 | VM_BUG_ON_PAGE(PageTail(page), page); |
190 | ||
c424be1c | 191 | /* |
01cc2e58 VB |
192 | * Serialize with any parallel __split_huge_page_refcount() which |
193 | * might otherwise copy PageMlocked to part of the tail pages before | |
6c357848 | 194 | * we clear it in the head page. It also stabilizes thp_nr_pages(). |
c424be1c | 195 | */ |
f4b7e272 | 196 | spin_lock_irq(&pgdat->lru_lock); |
01cc2e58 | 197 | |
655548bf KS |
198 | if (!TestClearPageMlocked(page)) { |
199 | /* Potentially, PTE-mapped THP: do not skip the rest PTEs */ | |
200 | nr_pages = 1; | |
01cc2e58 | 201 | goto unlock_out; |
655548bf | 202 | } |
01cc2e58 | 203 | |
6c357848 | 204 | nr_pages = thp_nr_pages(page); |
f4b7e272 | 205 | __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); |
01cc2e58 VB |
206 | |
207 | if (__munlock_isolate_lru_page(page, true)) { | |
f4b7e272 | 208 | spin_unlock_irq(&pgdat->lru_lock); |
01cc2e58 VB |
209 | __munlock_isolated_page(page); |
210 | goto out; | |
211 | } | |
212 | __munlock_isolation_failed(page); | |
213 | ||
214 | unlock_out: | |
f4b7e272 | 215 | spin_unlock_irq(&pgdat->lru_lock); |
01cc2e58 VB |
216 | |
217 | out: | |
c424be1c | 218 | return nr_pages - 1; |
b291f000 NP |
219 | } |
220 | ||
9978ad58 LS |
221 | /* |
222 | * convert get_user_pages() return value to posix mlock() error | |
223 | */ | |
224 | static int __mlock_posix_error_return(long retval) | |
225 | { | |
226 | if (retval == -EFAULT) | |
227 | retval = -ENOMEM; | |
228 | else if (retval == -ENOMEM) | |
229 | retval = -EAGAIN; | |
230 | return retval; | |
b291f000 NP |
231 | } |
232 | ||
56afe477 VB |
233 | /* |
234 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() | |
235 | * | |
236 | * The fast path is available only for evictable pages with single mapping. | |
237 | * Then we can bypass the per-cpu pvec and get better performance. | |
238 | * when mapcount > 1 we need try_to_munlock() which can fail. | |
239 | * when !page_evictable(), we need the full redo logic of putback_lru_page to | |
240 | * avoid leaving evictable page in unevictable list. | |
241 | * | |
242 | * In case of success, @page is added to @pvec and @pgrescued is incremented | |
243 | * in case that the page was previously unevictable. @page is also unlocked. | |
244 | */ | |
245 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, | |
246 | int *pgrescued) | |
247 | { | |
309381fe SL |
248 | VM_BUG_ON_PAGE(PageLRU(page), page); |
249 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
56afe477 VB |
250 | |
251 | if (page_mapcount(page) <= 1 && page_evictable(page)) { | |
252 | pagevec_add(pvec, page); | |
253 | if (TestClearPageUnevictable(page)) | |
254 | (*pgrescued)++; | |
255 | unlock_page(page); | |
256 | return true; | |
257 | } | |
258 | ||
259 | return false; | |
260 | } | |
261 | ||
262 | /* | |
263 | * Putback multiple evictable pages to the LRU | |
264 | * | |
265 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of | |
266 | * the pages might have meanwhile become unevictable but that is OK. | |
267 | */ | |
268 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) | |
269 | { | |
270 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); | |
271 | /* | |
272 | *__pagevec_lru_add() calls release_pages() so we don't call | |
273 | * put_page() explicitly | |
274 | */ | |
275 | __pagevec_lru_add(pvec); | |
276 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
277 | } | |
278 | ||
7225522b VB |
279 | /* |
280 | * Munlock a batch of pages from the same zone | |
281 | * | |
282 | * The work is split to two main phases. First phase clears the Mlocked flag | |
283 | * and attempts to isolate the pages, all under a single zone lru lock. | |
284 | * The second phase finishes the munlock only for pages where isolation | |
285 | * succeeded. | |
286 | * | |
7a8010cd | 287 | * Note that the pagevec may be modified during the process. |
7225522b VB |
288 | */ |
289 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) | |
290 | { | |
291 | int i; | |
292 | int nr = pagevec_count(pvec); | |
70feee0e | 293 | int delta_munlocked = -nr; |
56afe477 VB |
294 | struct pagevec pvec_putback; |
295 | int pgrescued = 0; | |
7225522b | 296 | |
86679820 | 297 | pagevec_init(&pvec_putback); |
3b25df93 | 298 | |
7225522b | 299 | /* Phase 1: page isolation */ |
f4b7e272 | 300 | spin_lock_irq(&zone->zone_pgdat->lru_lock); |
7225522b VB |
301 | for (i = 0; i < nr; i++) { |
302 | struct page *page = pvec->pages[i]; | |
303 | ||
304 | if (TestClearPageMlocked(page)) { | |
7225522b | 305 | /* |
01cc2e58 VB |
306 | * We already have pin from follow_page_mask() |
307 | * so we can spare the get_page() here. | |
7225522b | 308 | */ |
01cc2e58 VB |
309 | if (__munlock_isolate_lru_page(page, false)) |
310 | continue; | |
311 | else | |
312 | __munlock_isolation_failed(page); | |
70feee0e YX |
313 | } else { |
314 | delta_munlocked++; | |
7225522b | 315 | } |
01cc2e58 VB |
316 | |
317 | /* | |
318 | * We won't be munlocking this page in the next phase | |
319 | * but we still need to release the follow_page_mask() | |
320 | * pin. We cannot do it under lru_lock however. If it's | |
321 | * the last pin, __page_cache_release() would deadlock. | |
322 | */ | |
323 | pagevec_add(&pvec_putback, pvec->pages[i]); | |
324 | pvec->pages[i] = NULL; | |
7225522b | 325 | } |
1ebb7cc6 | 326 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
f4b7e272 | 327 | spin_unlock_irq(&zone->zone_pgdat->lru_lock); |
7225522b | 328 | |
3b25df93 VB |
329 | /* Now we can release pins of pages that we are not munlocking */ |
330 | pagevec_release(&pvec_putback); | |
331 | ||
56afe477 | 332 | /* Phase 2: page munlock */ |
7225522b VB |
333 | for (i = 0; i < nr; i++) { |
334 | struct page *page = pvec->pages[i]; | |
335 | ||
336 | if (page) { | |
337 | lock_page(page); | |
56afe477 VB |
338 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
339 | &pgrescued)) { | |
5b40998a VB |
340 | /* |
341 | * Slow path. We don't want to lose the last | |
342 | * pin before unlock_page() | |
343 | */ | |
344 | get_page(page); /* for putback_lru_page() */ | |
56afe477 VB |
345 | __munlock_isolated_page(page); |
346 | unlock_page(page); | |
5b40998a | 347 | put_page(page); /* from follow_page_mask() */ |
56afe477 | 348 | } |
7225522b VB |
349 | } |
350 | } | |
56afe477 | 351 | |
5b40998a VB |
352 | /* |
353 | * Phase 3: page putback for pages that qualified for the fast path | |
354 | * This will also call put_page() to return pin from follow_page_mask() | |
355 | */ | |
56afe477 VB |
356 | if (pagevec_count(&pvec_putback)) |
357 | __putback_lru_fast(&pvec_putback, pgrescued); | |
7a8010cd VB |
358 | } |
359 | ||
360 | /* | |
361 | * Fill up pagevec for __munlock_pagevec using pte walk | |
362 | * | |
363 | * The function expects that the struct page corresponding to @start address is | |
364 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. | |
365 | * | |
366 | * The rest of @pvec is filled by subsequent pages within the same pmd and same | |
367 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These | |
368 | * pages also get pinned. | |
369 | * | |
370 | * Returns the address of the next page that should be scanned. This equals | |
371 | * @start + PAGE_SIZE when no page could be added by the pte walk. | |
372 | */ | |
373 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, | |
9472f23c JK |
374 | struct vm_area_struct *vma, struct zone *zone, |
375 | unsigned long start, unsigned long end) | |
7a8010cd VB |
376 | { |
377 | pte_t *pte; | |
378 | spinlock_t *ptl; | |
379 | ||
380 | /* | |
381 | * Initialize pte walk starting at the already pinned page where we | |
eadb41ae | 382 | * are sure that there is a pte, as it was pinned under the same |
c1e8d7c6 | 383 | * mmap_lock write op. |
7a8010cd VB |
384 | */ |
385 | pte = get_locked_pte(vma->vm_mm, start, &ptl); | |
eadb41ae VB |
386 | /* Make sure we do not cross the page table boundary */ |
387 | end = pgd_addr_end(start, end); | |
c2febafc | 388 | end = p4d_addr_end(start, end); |
eadb41ae VB |
389 | end = pud_addr_end(start, end); |
390 | end = pmd_addr_end(start, end); | |
7a8010cd VB |
391 | |
392 | /* The page next to the pinned page is the first we will try to get */ | |
393 | start += PAGE_SIZE; | |
394 | while (start < end) { | |
395 | struct page *page = NULL; | |
396 | pte++; | |
397 | if (pte_present(*pte)) | |
398 | page = vm_normal_page(vma, start, *pte); | |
399 | /* | |
400 | * Break if page could not be obtained or the page's node+zone does not | |
401 | * match | |
402 | */ | |
9472f23c | 403 | if (!page || page_zone(page) != zone) |
7a8010cd | 404 | break; |
56afe477 | 405 | |
e90309c9 KS |
406 | /* |
407 | * Do not use pagevec for PTE-mapped THP, | |
408 | * munlock_vma_pages_range() will handle them. | |
409 | */ | |
410 | if (PageTransCompound(page)) | |
411 | break; | |
412 | ||
7a8010cd VB |
413 | get_page(page); |
414 | /* | |
415 | * Increase the address that will be returned *before* the | |
416 | * eventual break due to pvec becoming full by adding the page | |
417 | */ | |
418 | start += PAGE_SIZE; | |
419 | if (pagevec_add(pvec, page) == 0) | |
420 | break; | |
421 | } | |
422 | pte_unmap_unlock(pte, ptl); | |
423 | return start; | |
7225522b VB |
424 | } |
425 | ||
b291f000 | 426 | /* |
ba470de4 RR |
427 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
428 | * @vma - vma containing range to be munlock()ed. | |
429 | * @start - start address in @vma of the range | |
430 | * @end - end of range in @vma. | |
431 | * | |
432 | * For mremap(), munmap() and exit(). | |
433 | * | |
434 | * Called with @vma VM_LOCKED. | |
435 | * | |
436 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
437 | * deal with this. | |
438 | * | |
439 | * We don't save and restore VM_LOCKED here because pages are | |
440 | * still on lru. In unmap path, pages might be scanned by reclaim | |
441 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
442 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 443 | */ |
ba470de4 | 444 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 445 | unsigned long start, unsigned long end) |
b291f000 | 446 | { |
de60f5f1 | 447 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
408e82b7 | 448 | |
ff6a6da6 | 449 | while (start < end) { |
ab7a5af7 | 450 | struct page *page; |
6ebb4a1b | 451 | unsigned int page_mask = 0; |
c424be1c | 452 | unsigned long page_increm; |
7a8010cd VB |
453 | struct pagevec pvec; |
454 | struct zone *zone; | |
ff6a6da6 | 455 | |
86679820 | 456 | pagevec_init(&pvec); |
6e919717 HD |
457 | /* |
458 | * Although FOLL_DUMP is intended for get_dump_page(), | |
459 | * it just so happens that its special treatment of the | |
460 | * ZERO_PAGE (returning an error instead of doing get_page) | |
461 | * suits munlock very well (and if somehow an abnormal page | |
462 | * has sneaked into the range, we won't oops here: great). | |
463 | */ | |
6ebb4a1b | 464 | page = follow_page(vma, start, FOLL_GET | FOLL_DUMP); |
7a8010cd | 465 | |
e90309c9 KS |
466 | if (page && !IS_ERR(page)) { |
467 | if (PageTransTail(page)) { | |
468 | VM_BUG_ON_PAGE(PageMlocked(page), page); | |
469 | put_page(page); /* follow_page_mask() */ | |
470 | } else if (PageTransHuge(page)) { | |
471 | lock_page(page); | |
472 | /* | |
473 | * Any THP page found by follow_page_mask() may | |
474 | * have gotten split before reaching | |
6ebb4a1b KS |
475 | * munlock_vma_page(), so we need to compute |
476 | * the page_mask here instead. | |
e90309c9 KS |
477 | */ |
478 | page_mask = munlock_vma_page(page); | |
479 | unlock_page(page); | |
480 | put_page(page); /* follow_page_mask() */ | |
481 | } else { | |
482 | /* | |
483 | * Non-huge pages are handled in batches via | |
484 | * pagevec. The pin from follow_page_mask() | |
485 | * prevents them from collapsing by THP. | |
486 | */ | |
487 | pagevec_add(&pvec, page); | |
488 | zone = page_zone(page); | |
7a8010cd | 489 | |
e90309c9 KS |
490 | /* |
491 | * Try to fill the rest of pagevec using fast | |
492 | * pte walk. This will also update start to | |
493 | * the next page to process. Then munlock the | |
494 | * pagevec. | |
495 | */ | |
496 | start = __munlock_pagevec_fill(&pvec, vma, | |
9472f23c | 497 | zone, start, end); |
e90309c9 KS |
498 | __munlock_pagevec(&pvec, zone); |
499 | goto next; | |
500 | } | |
408e82b7 | 501 | } |
c424be1c | 502 | page_increm = 1 + page_mask; |
ff6a6da6 | 503 | start += page_increm * PAGE_SIZE; |
7a8010cd | 504 | next: |
408e82b7 HD |
505 | cond_resched(); |
506 | } | |
b291f000 NP |
507 | } |
508 | ||
509 | /* | |
510 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
511 | * | |
512 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
513 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 514 | * populate the ptes. |
b291f000 NP |
515 | * |
516 | * For vmas that pass the filters, merge/split as appropriate. | |
517 | */ | |
1da177e4 | 518 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 519 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 520 | { |
b291f000 | 521 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 522 | pgoff_t pgoff; |
b291f000 | 523 | int nr_pages; |
1da177e4 | 524 | int ret = 0; |
ca16d140 | 525 | int lock = !!(newflags & VM_LOCKED); |
b155b4fd | 526 | vm_flags_t old_flags = vma->vm_flags; |
1da177e4 | 527 | |
fed067da | 528 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
e1fb4a08 DJ |
529 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) || |
530 | vma_is_dax(vma)) | |
b0f205c2 EM |
531 | /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */ |
532 | goto out; | |
b291f000 | 533 | |
1da177e4 LT |
534 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
535 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
19a809af AA |
536 | vma->vm_file, pgoff, vma_policy(vma), |
537 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
538 | if (*prev) { |
539 | vma = *prev; | |
540 | goto success; | |
541 | } | |
542 | ||
1da177e4 LT |
543 | if (start != vma->vm_start) { |
544 | ret = split_vma(mm, vma, start, 1); | |
545 | if (ret) | |
546 | goto out; | |
547 | } | |
548 | ||
549 | if (end != vma->vm_end) { | |
550 | ret = split_vma(mm, vma, end, 0); | |
551 | if (ret) | |
552 | goto out; | |
553 | } | |
554 | ||
555 | success: | |
b291f000 NP |
556 | /* |
557 | * Keep track of amount of locked VM. | |
558 | */ | |
559 | nr_pages = (end - start) >> PAGE_SHIFT; | |
560 | if (!lock) | |
561 | nr_pages = -nr_pages; | |
b155b4fd SG |
562 | else if (old_flags & VM_LOCKED) |
563 | nr_pages = 0; | |
b291f000 NP |
564 | mm->locked_vm += nr_pages; |
565 | ||
1da177e4 | 566 | /* |
c1e8d7c6 | 567 | * vm_flags is protected by the mmap_lock held in write mode. |
1da177e4 | 568 | * It's okay if try_to_unmap_one unmaps a page just after we |
fc05f566 | 569 | * set VM_LOCKED, populate_vma_page_range will bring it back. |
1da177e4 | 570 | */ |
1da177e4 | 571 | |
fed067da | 572 | if (lock) |
408e82b7 | 573 | vma->vm_flags = newflags; |
fed067da | 574 | else |
408e82b7 | 575 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 576 | |
1da177e4 | 577 | out: |
b291f000 | 578 | *prev = vma; |
1da177e4 LT |
579 | return ret; |
580 | } | |
581 | ||
1aab92ec EM |
582 | static int apply_vma_lock_flags(unsigned long start, size_t len, |
583 | vm_flags_t flags) | |
1da177e4 LT |
584 | { |
585 | unsigned long nstart, end, tmp; | |
586 | struct vm_area_struct * vma, * prev; | |
587 | int error; | |
588 | ||
8fd9e488 | 589 | VM_BUG_ON(offset_in_page(start)); |
fed067da | 590 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
1da177e4 LT |
591 | end = start + len; |
592 | if (end < start) | |
593 | return -EINVAL; | |
594 | if (end == start) | |
595 | return 0; | |
097d5910 | 596 | vma = find_vma(current->mm, start); |
1da177e4 LT |
597 | if (!vma || vma->vm_start > start) |
598 | return -ENOMEM; | |
599 | ||
097d5910 | 600 | prev = vma->vm_prev; |
1da177e4 LT |
601 | if (start > vma->vm_start) |
602 | prev = vma; | |
603 | ||
604 | for (nstart = start ; ; ) { | |
b0f205c2 | 605 | vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
1da177e4 | 606 | |
1aab92ec | 607 | newflags |= flags; |
1da177e4 | 608 | |
1aab92ec | 609 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 LT |
610 | tmp = vma->vm_end; |
611 | if (tmp > end) | |
612 | tmp = end; | |
613 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
614 | if (error) | |
615 | break; | |
616 | nstart = tmp; | |
617 | if (nstart < prev->vm_end) | |
618 | nstart = prev->vm_end; | |
619 | if (nstart >= end) | |
620 | break; | |
621 | ||
622 | vma = prev->vm_next; | |
623 | if (!vma || vma->vm_start != nstart) { | |
624 | error = -ENOMEM; | |
625 | break; | |
626 | } | |
627 | } | |
628 | return error; | |
629 | } | |
630 | ||
0cf2f6f6 SG |
631 | /* |
632 | * Go through vma areas and sum size of mlocked | |
633 | * vma pages, as return value. | |
634 | * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT) | |
635 | * is also counted. | |
636 | * Return value: previously mlocked page counts | |
637 | */ | |
0874bb49 | 638 | static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm, |
0cf2f6f6 SG |
639 | unsigned long start, size_t len) |
640 | { | |
641 | struct vm_area_struct *vma; | |
0874bb49 | 642 | unsigned long count = 0; |
0cf2f6f6 SG |
643 | |
644 | if (mm == NULL) | |
645 | mm = current->mm; | |
646 | ||
647 | vma = find_vma(mm, start); | |
648 | if (vma == NULL) | |
649 | vma = mm->mmap; | |
650 | ||
651 | for (; vma ; vma = vma->vm_next) { | |
652 | if (start >= vma->vm_end) | |
653 | continue; | |
654 | if (start + len <= vma->vm_start) | |
655 | break; | |
656 | if (vma->vm_flags & VM_LOCKED) { | |
657 | if (start > vma->vm_start) | |
658 | count -= (start - vma->vm_start); | |
659 | if (start + len < vma->vm_end) { | |
660 | count += start + len - vma->vm_start; | |
661 | break; | |
662 | } | |
663 | count += vma->vm_end - vma->vm_start; | |
664 | } | |
665 | } | |
666 | ||
667 | return count >> PAGE_SHIFT; | |
668 | } | |
669 | ||
dc0ef0df | 670 | static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags) |
1da177e4 LT |
671 | { |
672 | unsigned long locked; | |
673 | unsigned long lock_limit; | |
674 | int error = -ENOMEM; | |
675 | ||
057d3389 AK |
676 | start = untagged_addr(start); |
677 | ||
1da177e4 LT |
678 | if (!can_do_mlock()) |
679 | return -EPERM; | |
680 | ||
8fd9e488 | 681 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 LT |
682 | start &= PAGE_MASK; |
683 | ||
59e99e5b | 684 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 | 685 | lock_limit >>= PAGE_SHIFT; |
1f1cd705 DB |
686 | locked = len >> PAGE_SHIFT; |
687 | ||
d8ed45c5 | 688 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 689 | return -EINTR; |
1f1cd705 DB |
690 | |
691 | locked += current->mm->locked_vm; | |
0cf2f6f6 SG |
692 | if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) { |
693 | /* | |
694 | * It is possible that the regions requested intersect with | |
695 | * previously mlocked areas, that part area in "mm->locked_vm" | |
696 | * should not be counted to new mlock increment count. So check | |
697 | * and adjust locked count if necessary. | |
698 | */ | |
699 | locked -= count_mm_mlocked_page_nr(current->mm, | |
700 | start, len); | |
701 | } | |
1da177e4 LT |
702 | |
703 | /* check against resource limits */ | |
704 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
1aab92ec | 705 | error = apply_vma_lock_flags(start, len, flags); |
1f1cd705 | 706 | |
d8ed45c5 | 707 | mmap_write_unlock(current->mm); |
c561259c KS |
708 | if (error) |
709 | return error; | |
710 | ||
711 | error = __mm_populate(start, len, 0); | |
712 | if (error) | |
713 | return __mlock_posix_error_return(error); | |
714 | return 0; | |
1da177e4 LT |
715 | } |
716 | ||
1aab92ec EM |
717 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
718 | { | |
719 | return do_mlock(start, len, VM_LOCKED); | |
720 | } | |
721 | ||
a8ca5d0e EM |
722 | SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags) |
723 | { | |
b0f205c2 EM |
724 | vm_flags_t vm_flags = VM_LOCKED; |
725 | ||
726 | if (flags & ~MLOCK_ONFAULT) | |
a8ca5d0e EM |
727 | return -EINVAL; |
728 | ||
b0f205c2 EM |
729 | if (flags & MLOCK_ONFAULT) |
730 | vm_flags |= VM_LOCKONFAULT; | |
731 | ||
732 | return do_mlock(start, len, vm_flags); | |
a8ca5d0e EM |
733 | } |
734 | ||
6a6160a7 | 735 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
736 | { |
737 | int ret; | |
738 | ||
057d3389 AK |
739 | start = untagged_addr(start); |
740 | ||
8fd9e488 | 741 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 | 742 | start &= PAGE_MASK; |
1f1cd705 | 743 | |
d8ed45c5 | 744 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 745 | return -EINTR; |
1aab92ec | 746 | ret = apply_vma_lock_flags(start, len, 0); |
d8ed45c5 | 747 | mmap_write_unlock(current->mm); |
1f1cd705 | 748 | |
1da177e4 LT |
749 | return ret; |
750 | } | |
751 | ||
b0f205c2 EM |
752 | /* |
753 | * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall) | |
754 | * and translate into the appropriate modifications to mm->def_flags and/or the | |
755 | * flags for all current VMAs. | |
756 | * | |
757 | * There are a couple of subtleties with this. If mlockall() is called multiple | |
758 | * times with different flags, the values do not necessarily stack. If mlockall | |
759 | * is called once including the MCL_FUTURE flag and then a second time without | |
760 | * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags. | |
761 | */ | |
1aab92ec | 762 | static int apply_mlockall_flags(int flags) |
1da177e4 LT |
763 | { |
764 | struct vm_area_struct * vma, * prev = NULL; | |
b0f205c2 | 765 | vm_flags_t to_add = 0; |
1da177e4 | 766 | |
b0f205c2 EM |
767 | current->mm->def_flags &= VM_LOCKED_CLEAR_MASK; |
768 | if (flags & MCL_FUTURE) { | |
09a9f1d2 | 769 | current->mm->def_flags |= VM_LOCKED; |
1aab92ec | 770 | |
b0f205c2 EM |
771 | if (flags & MCL_ONFAULT) |
772 | current->mm->def_flags |= VM_LOCKONFAULT; | |
773 | ||
774 | if (!(flags & MCL_CURRENT)) | |
775 | goto out; | |
776 | } | |
777 | ||
778 | if (flags & MCL_CURRENT) { | |
779 | to_add |= VM_LOCKED; | |
780 | if (flags & MCL_ONFAULT) | |
781 | to_add |= VM_LOCKONFAULT; | |
782 | } | |
1da177e4 LT |
783 | |
784 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 785 | vm_flags_t newflags; |
1da177e4 | 786 | |
b0f205c2 EM |
787 | newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
788 | newflags |= to_add; | |
1da177e4 LT |
789 | |
790 | /* Ignore errors */ | |
791 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
50d4fb78 | 792 | cond_resched(); |
1da177e4 LT |
793 | } |
794 | out: | |
795 | return 0; | |
796 | } | |
797 | ||
3480b257 | 798 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
799 | { |
800 | unsigned long lock_limit; | |
86d2adcc | 801 | int ret; |
1da177e4 | 802 | |
dedca635 PS |
803 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) || |
804 | flags == MCL_ONFAULT) | |
86d2adcc | 805 | return -EINVAL; |
1da177e4 | 806 | |
1da177e4 | 807 | if (!can_do_mlock()) |
86d2adcc | 808 | return -EPERM; |
1da177e4 | 809 | |
59e99e5b | 810 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
811 | lock_limit >>= PAGE_SHIFT; |
812 | ||
d8ed45c5 | 813 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 814 | return -EINTR; |
1f1cd705 | 815 | |
dc0ef0df | 816 | ret = -ENOMEM; |
1da177e4 LT |
817 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
818 | capable(CAP_IPC_LOCK)) | |
1aab92ec | 819 | ret = apply_mlockall_flags(flags); |
d8ed45c5 | 820 | mmap_write_unlock(current->mm); |
bebeb3d6 ML |
821 | if (!ret && (flags & MCL_CURRENT)) |
822 | mm_populate(0, TASK_SIZE); | |
86d2adcc | 823 | |
1da177e4 LT |
824 | return ret; |
825 | } | |
826 | ||
3480b257 | 827 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
828 | { |
829 | int ret; | |
830 | ||
d8ed45c5 | 831 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 832 | return -EINTR; |
1aab92ec | 833 | ret = apply_mlockall_flags(0); |
d8ed45c5 | 834 | mmap_write_unlock(current->mm); |
1da177e4 LT |
835 | return ret; |
836 | } | |
837 | ||
838 | /* | |
839 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
840 | * shm segments) get accounted against the user_struct instead. | |
841 | */ | |
842 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
843 | ||
844 | int user_shm_lock(size_t size, struct user_struct *user) | |
845 | { | |
846 | unsigned long lock_limit, locked; | |
847 | int allowed = 0; | |
848 | ||
849 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 850 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
851 | if (lock_limit == RLIM_INFINITY) |
852 | allowed = 1; | |
1da177e4 LT |
853 | lock_limit >>= PAGE_SHIFT; |
854 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
855 | if (!allowed && |
856 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
857 | goto out; |
858 | get_uid(user); | |
859 | user->locked_shm += locked; | |
860 | allowed = 1; | |
861 | out: | |
862 | spin_unlock(&shmlock_user_lock); | |
863 | return allowed; | |
864 | } | |
865 | ||
866 | void user_shm_unlock(size_t size, struct user_struct *user) | |
867 | { | |
868 | spin_lock(&shmlock_user_lock); | |
869 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
870 | spin_unlock(&shmlock_user_lock); | |
871 | free_uid(user); | |
872 | } |