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 | { |
0964730b HD |
61 | int nr_pages; |
62 | ||
e6c509f8 | 63 | if (!TestClearPageMlocked(page)) |
b291f000 | 64 | return; |
b291f000 | 65 | |
0964730b HD |
66 | nr_pages = thp_nr_pages(page); |
67 | mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); | |
68 | count_vm_events(UNEVICTABLE_PGCLEARED, nr_pages); | |
9c4e6b1a SB |
69 | /* |
70 | * The previous TestClearPageMlocked() corresponds to the smp_mb() | |
71 | * in __pagevec_lru_add_fn(). | |
72 | * | |
73 | * See __pagevec_lru_add_fn for more explanation. | |
74 | */ | |
b291f000 NP |
75 | if (!isolate_lru_page(page)) { |
76 | putback_lru_page(page); | |
77 | } else { | |
78 | /* | |
8891d6da | 79 | * We lost the race. the page already moved to evictable list. |
b291f000 | 80 | */ |
8891d6da | 81 | if (PageUnevictable(page)) |
0964730b | 82 | count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages); |
b291f000 NP |
83 | } |
84 | } | |
85 | ||
86 | /* | |
87 | * Mark page as mlocked if not already. | |
88 | * If page on LRU, isolate and putback to move to unevictable list. | |
89 | */ | |
90 | void mlock_vma_page(struct page *page) | |
91 | { | |
57e68e9c | 92 | /* Serialize with page migration */ |
b291f000 NP |
93 | BUG_ON(!PageLocked(page)); |
94 | ||
e90309c9 KS |
95 | VM_BUG_ON_PAGE(PageTail(page), page); |
96 | VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page); | |
97 | ||
5344b7e6 | 98 | if (!TestSetPageMlocked(page)) { |
0964730b HD |
99 | int nr_pages = thp_nr_pages(page); |
100 | ||
101 | mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages); | |
102 | count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages); | |
5344b7e6 NP |
103 | if (!isolate_lru_page(page)) |
104 | putback_lru_page(page); | |
105 | } | |
b291f000 NP |
106 | } |
107 | ||
7225522b VB |
108 | /* |
109 | * Finish munlock after successful page isolation | |
110 | * | |
111 | * Page must be locked. This is a wrapper for try_to_munlock() | |
112 | * and putback_lru_page() with munlock accounting. | |
113 | */ | |
114 | static void __munlock_isolated_page(struct page *page) | |
115 | { | |
7225522b VB |
116 | /* |
117 | * Optimization: if the page was mapped just once, that's our mapping | |
118 | * and we don't need to check all the other vmas. | |
119 | */ | |
120 | if (page_mapcount(page) > 1) | |
192d7232 | 121 | try_to_munlock(page); |
7225522b VB |
122 | |
123 | /* Did try_to_unlock() succeed or punt? */ | |
192d7232 | 124 | if (!PageMlocked(page)) |
0964730b | 125 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, thp_nr_pages(page)); |
7225522b VB |
126 | |
127 | putback_lru_page(page); | |
128 | } | |
129 | ||
130 | /* | |
131 | * Accounting for page isolation fail during munlock | |
132 | * | |
133 | * Performs accounting when page isolation fails in munlock. There is nothing | |
134 | * else to do because it means some other task has already removed the page | |
135 | * from the LRU. putback_lru_page() will take care of removing the page from | |
136 | * the unevictable list, if necessary. vmscan [page_referenced()] will move | |
137 | * the page back to the unevictable list if some other vma has it mlocked. | |
138 | */ | |
139 | static void __munlock_isolation_failed(struct page *page) | |
140 | { | |
0964730b HD |
141 | int nr_pages = thp_nr_pages(page); |
142 | ||
7225522b | 143 | if (PageUnevictable(page)) |
0964730b | 144 | __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages); |
7225522b | 145 | else |
0964730b | 146 | __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages); |
7225522b VB |
147 | } |
148 | ||
6927c1dd LS |
149 | /** |
150 | * munlock_vma_page - munlock a vma page | |
b7701a5f | 151 | * @page: page to be unlocked, either a normal page or THP page head |
c424be1c VB |
152 | * |
153 | * returns the size of the page as a page mask (0 for normal page, | |
154 | * HPAGE_PMD_NR - 1 for THP head page) | |
b291f000 | 155 | * |
6927c1dd LS |
156 | * called from munlock()/munmap() path with page supposedly on the LRU. |
157 | * When we munlock a page, because the vma where we found the page is being | |
158 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the | |
159 | * page locked so that we can leave it on the unevictable lru list and not | |
160 | * bother vmscan with it. However, to walk the page's rmap list in | |
161 | * try_to_munlock() we must isolate the page from the LRU. If some other | |
162 | * task has removed the page from the LRU, we won't be able to do that. | |
163 | * So we clear the PageMlocked as we might not get another chance. If we | |
164 | * can't isolate the page, we leave it for putback_lru_page() and vmscan | |
165 | * [page_referenced()/try_to_unmap()] to deal with. | |
b291f000 | 166 | */ |
ff6a6da6 | 167 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 168 | { |
7162a1e8 | 169 | int nr_pages; |
ff6a6da6 | 170 | |
57e68e9c | 171 | /* For try_to_munlock() and to serialize with page migration */ |
b291f000 | 172 | BUG_ON(!PageLocked(page)); |
e90309c9 KS |
173 | VM_BUG_ON_PAGE(PageTail(page), page); |
174 | ||
655548bf KS |
175 | if (!TestClearPageMlocked(page)) { |
176 | /* Potentially, PTE-mapped THP: do not skip the rest PTEs */ | |
3db19aa3 | 177 | return 0; |
655548bf | 178 | } |
01cc2e58 | 179 | |
6c357848 | 180 | nr_pages = thp_nr_pages(page); |
3db19aa3 | 181 | mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); |
01cc2e58 | 182 | |
3db19aa3 | 183 | if (!isolate_lru_page(page)) |
01cc2e58 | 184 | __munlock_isolated_page(page); |
3db19aa3 AS |
185 | else |
186 | __munlock_isolation_failed(page); | |
01cc2e58 | 187 | |
c424be1c | 188 | return nr_pages - 1; |
b291f000 NP |
189 | } |
190 | ||
9978ad58 LS |
191 | /* |
192 | * convert get_user_pages() return value to posix mlock() error | |
193 | */ | |
194 | static int __mlock_posix_error_return(long retval) | |
195 | { | |
196 | if (retval == -EFAULT) | |
197 | retval = -ENOMEM; | |
198 | else if (retval == -ENOMEM) | |
199 | retval = -EAGAIN; | |
200 | return retval; | |
b291f000 NP |
201 | } |
202 | ||
56afe477 VB |
203 | /* |
204 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() | |
205 | * | |
206 | * The fast path is available only for evictable pages with single mapping. | |
207 | * Then we can bypass the per-cpu pvec and get better performance. | |
208 | * when mapcount > 1 we need try_to_munlock() which can fail. | |
209 | * when !page_evictable(), we need the full redo logic of putback_lru_page to | |
210 | * avoid leaving evictable page in unevictable list. | |
211 | * | |
212 | * In case of success, @page is added to @pvec and @pgrescued is incremented | |
213 | * in case that the page was previously unevictable. @page is also unlocked. | |
214 | */ | |
215 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, | |
216 | int *pgrescued) | |
217 | { | |
309381fe SL |
218 | VM_BUG_ON_PAGE(PageLRU(page), page); |
219 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
56afe477 VB |
220 | |
221 | if (page_mapcount(page) <= 1 && page_evictable(page)) { | |
222 | pagevec_add(pvec, page); | |
223 | if (TestClearPageUnevictable(page)) | |
224 | (*pgrescued)++; | |
225 | unlock_page(page); | |
226 | return true; | |
227 | } | |
228 | ||
229 | return false; | |
230 | } | |
231 | ||
232 | /* | |
233 | * Putback multiple evictable pages to the LRU | |
234 | * | |
235 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of | |
236 | * the pages might have meanwhile become unevictable but that is OK. | |
237 | */ | |
238 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) | |
239 | { | |
240 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); | |
241 | /* | |
242 | *__pagevec_lru_add() calls release_pages() so we don't call | |
243 | * put_page() explicitly | |
244 | */ | |
245 | __pagevec_lru_add(pvec); | |
246 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
247 | } | |
248 | ||
7225522b VB |
249 | /* |
250 | * Munlock a batch of pages from the same zone | |
251 | * | |
252 | * The work is split to two main phases. First phase clears the Mlocked flag | |
253 | * and attempts to isolate the pages, all under a single zone lru lock. | |
254 | * The second phase finishes the munlock only for pages where isolation | |
255 | * succeeded. | |
256 | * | |
7a8010cd | 257 | * Note that the pagevec may be modified during the process. |
7225522b VB |
258 | */ |
259 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) | |
260 | { | |
261 | int i; | |
262 | int nr = pagevec_count(pvec); | |
70feee0e | 263 | int delta_munlocked = -nr; |
56afe477 | 264 | struct pagevec pvec_putback; |
6168d0da | 265 | struct lruvec *lruvec = NULL; |
56afe477 | 266 | int pgrescued = 0; |
7225522b | 267 | |
86679820 | 268 | pagevec_init(&pvec_putback); |
3b25df93 | 269 | |
7225522b | 270 | /* Phase 1: page isolation */ |
7225522b VB |
271 | for (i = 0; i < nr; i++) { |
272 | struct page *page = pvec->pages[i]; | |
273 | ||
274 | if (TestClearPageMlocked(page)) { | |
7225522b | 275 | /* |
01cc2e58 VB |
276 | * We already have pin from follow_page_mask() |
277 | * so we can spare the get_page() here. | |
7225522b | 278 | */ |
d25b5bd8 | 279 | if (TestClearPageLRU(page)) { |
2a5e4e34 | 280 | lruvec = relock_page_lruvec_irq(page, lruvec); |
13805a88 AS |
281 | del_page_from_lru_list(page, lruvec, |
282 | page_lru(page)); | |
01cc2e58 | 283 | continue; |
13805a88 | 284 | } else |
01cc2e58 | 285 | __munlock_isolation_failed(page); |
70feee0e YX |
286 | } else { |
287 | delta_munlocked++; | |
7225522b | 288 | } |
01cc2e58 VB |
289 | |
290 | /* | |
291 | * We won't be munlocking this page in the next phase | |
292 | * but we still need to release the follow_page_mask() | |
293 | * pin. We cannot do it under lru_lock however. If it's | |
294 | * the last pin, __page_cache_release() would deadlock. | |
295 | */ | |
296 | pagevec_add(&pvec_putback, pvec->pages[i]); | |
297 | pvec->pages[i] = NULL; | |
7225522b | 298 | } |
6168d0da AS |
299 | if (lruvec) { |
300 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); | |
301 | unlock_page_lruvec_irq(lruvec); | |
302 | } else if (delta_munlocked) { | |
303 | mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); | |
304 | } | |
7225522b | 305 | |
3b25df93 VB |
306 | /* Now we can release pins of pages that we are not munlocking */ |
307 | pagevec_release(&pvec_putback); | |
308 | ||
56afe477 | 309 | /* Phase 2: page munlock */ |
7225522b VB |
310 | for (i = 0; i < nr; i++) { |
311 | struct page *page = pvec->pages[i]; | |
312 | ||
313 | if (page) { | |
314 | lock_page(page); | |
56afe477 VB |
315 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
316 | &pgrescued)) { | |
5b40998a VB |
317 | /* |
318 | * Slow path. We don't want to lose the last | |
319 | * pin before unlock_page() | |
320 | */ | |
321 | get_page(page); /* for putback_lru_page() */ | |
56afe477 VB |
322 | __munlock_isolated_page(page); |
323 | unlock_page(page); | |
5b40998a | 324 | put_page(page); /* from follow_page_mask() */ |
56afe477 | 325 | } |
7225522b VB |
326 | } |
327 | } | |
56afe477 | 328 | |
5b40998a VB |
329 | /* |
330 | * Phase 3: page putback for pages that qualified for the fast path | |
331 | * This will also call put_page() to return pin from follow_page_mask() | |
332 | */ | |
56afe477 VB |
333 | if (pagevec_count(&pvec_putback)) |
334 | __putback_lru_fast(&pvec_putback, pgrescued); | |
7a8010cd VB |
335 | } |
336 | ||
337 | /* | |
338 | * Fill up pagevec for __munlock_pagevec using pte walk | |
339 | * | |
340 | * The function expects that the struct page corresponding to @start address is | |
341 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. | |
342 | * | |
343 | * The rest of @pvec is filled by subsequent pages within the same pmd and same | |
344 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These | |
345 | * pages also get pinned. | |
346 | * | |
347 | * Returns the address of the next page that should be scanned. This equals | |
348 | * @start + PAGE_SIZE when no page could be added by the pte walk. | |
349 | */ | |
350 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, | |
9472f23c JK |
351 | struct vm_area_struct *vma, struct zone *zone, |
352 | unsigned long start, unsigned long end) | |
7a8010cd VB |
353 | { |
354 | pte_t *pte; | |
355 | spinlock_t *ptl; | |
356 | ||
357 | /* | |
358 | * Initialize pte walk starting at the already pinned page where we | |
eadb41ae | 359 | * are sure that there is a pte, as it was pinned under the same |
c1e8d7c6 | 360 | * mmap_lock write op. |
7a8010cd VB |
361 | */ |
362 | pte = get_locked_pte(vma->vm_mm, start, &ptl); | |
eadb41ae VB |
363 | /* Make sure we do not cross the page table boundary */ |
364 | end = pgd_addr_end(start, end); | |
c2febafc | 365 | end = p4d_addr_end(start, end); |
eadb41ae VB |
366 | end = pud_addr_end(start, end); |
367 | end = pmd_addr_end(start, end); | |
7a8010cd VB |
368 | |
369 | /* The page next to the pinned page is the first we will try to get */ | |
370 | start += PAGE_SIZE; | |
371 | while (start < end) { | |
372 | struct page *page = NULL; | |
373 | pte++; | |
374 | if (pte_present(*pte)) | |
375 | page = vm_normal_page(vma, start, *pte); | |
376 | /* | |
377 | * Break if page could not be obtained or the page's node+zone does not | |
378 | * match | |
379 | */ | |
9472f23c | 380 | if (!page || page_zone(page) != zone) |
7a8010cd | 381 | break; |
56afe477 | 382 | |
e90309c9 KS |
383 | /* |
384 | * Do not use pagevec for PTE-mapped THP, | |
385 | * munlock_vma_pages_range() will handle them. | |
386 | */ | |
387 | if (PageTransCompound(page)) | |
388 | break; | |
389 | ||
7a8010cd VB |
390 | get_page(page); |
391 | /* | |
392 | * Increase the address that will be returned *before* the | |
393 | * eventual break due to pvec becoming full by adding the page | |
394 | */ | |
395 | start += PAGE_SIZE; | |
396 | if (pagevec_add(pvec, page) == 0) | |
397 | break; | |
398 | } | |
399 | pte_unmap_unlock(pte, ptl); | |
400 | return start; | |
7225522b VB |
401 | } |
402 | ||
b291f000 | 403 | /* |
ba470de4 RR |
404 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
405 | * @vma - vma containing range to be munlock()ed. | |
406 | * @start - start address in @vma of the range | |
407 | * @end - end of range in @vma. | |
408 | * | |
409 | * For mremap(), munmap() and exit(). | |
410 | * | |
411 | * Called with @vma VM_LOCKED. | |
412 | * | |
413 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
414 | * deal with this. | |
415 | * | |
416 | * We don't save and restore VM_LOCKED here because pages are | |
417 | * still on lru. In unmap path, pages might be scanned by reclaim | |
418 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
419 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 420 | */ |
ba470de4 | 421 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 422 | unsigned long start, unsigned long end) |
b291f000 | 423 | { |
de60f5f1 | 424 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
408e82b7 | 425 | |
ff6a6da6 | 426 | while (start < end) { |
ab7a5af7 | 427 | struct page *page; |
6ebb4a1b | 428 | unsigned int page_mask = 0; |
c424be1c | 429 | unsigned long page_increm; |
7a8010cd VB |
430 | struct pagevec pvec; |
431 | struct zone *zone; | |
ff6a6da6 | 432 | |
86679820 | 433 | pagevec_init(&pvec); |
6e919717 HD |
434 | /* |
435 | * Although FOLL_DUMP is intended for get_dump_page(), | |
436 | * it just so happens that its special treatment of the | |
437 | * ZERO_PAGE (returning an error instead of doing get_page) | |
438 | * suits munlock very well (and if somehow an abnormal page | |
439 | * has sneaked into the range, we won't oops here: great). | |
440 | */ | |
6ebb4a1b | 441 | page = follow_page(vma, start, FOLL_GET | FOLL_DUMP); |
7a8010cd | 442 | |
e90309c9 KS |
443 | if (page && !IS_ERR(page)) { |
444 | if (PageTransTail(page)) { | |
445 | VM_BUG_ON_PAGE(PageMlocked(page), page); | |
446 | put_page(page); /* follow_page_mask() */ | |
447 | } else if (PageTransHuge(page)) { | |
448 | lock_page(page); | |
449 | /* | |
450 | * Any THP page found by follow_page_mask() may | |
451 | * have gotten split before reaching | |
6ebb4a1b KS |
452 | * munlock_vma_page(), so we need to compute |
453 | * the page_mask here instead. | |
e90309c9 KS |
454 | */ |
455 | page_mask = munlock_vma_page(page); | |
456 | unlock_page(page); | |
457 | put_page(page); /* follow_page_mask() */ | |
458 | } else { | |
459 | /* | |
460 | * Non-huge pages are handled in batches via | |
461 | * pagevec. The pin from follow_page_mask() | |
462 | * prevents them from collapsing by THP. | |
463 | */ | |
464 | pagevec_add(&pvec, page); | |
465 | zone = page_zone(page); | |
7a8010cd | 466 | |
e90309c9 KS |
467 | /* |
468 | * Try to fill the rest of pagevec using fast | |
469 | * pte walk. This will also update start to | |
470 | * the next page to process. Then munlock the | |
471 | * pagevec. | |
472 | */ | |
473 | start = __munlock_pagevec_fill(&pvec, vma, | |
9472f23c | 474 | zone, start, end); |
e90309c9 KS |
475 | __munlock_pagevec(&pvec, zone); |
476 | goto next; | |
477 | } | |
408e82b7 | 478 | } |
c424be1c | 479 | page_increm = 1 + page_mask; |
ff6a6da6 | 480 | start += page_increm * PAGE_SIZE; |
7a8010cd | 481 | next: |
408e82b7 HD |
482 | cond_resched(); |
483 | } | |
b291f000 NP |
484 | } |
485 | ||
486 | /* | |
487 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
488 | * | |
489 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
490 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 491 | * populate the ptes. |
b291f000 NP |
492 | * |
493 | * For vmas that pass the filters, merge/split as appropriate. | |
494 | */ | |
1da177e4 | 495 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 496 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 497 | { |
b291f000 | 498 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 499 | pgoff_t pgoff; |
b291f000 | 500 | int nr_pages; |
1da177e4 | 501 | int ret = 0; |
ca16d140 | 502 | int lock = !!(newflags & VM_LOCKED); |
b155b4fd | 503 | vm_flags_t old_flags = vma->vm_flags; |
1da177e4 | 504 | |
fed067da | 505 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
e1fb4a08 DJ |
506 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) || |
507 | vma_is_dax(vma)) | |
b0f205c2 EM |
508 | /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */ |
509 | goto out; | |
b291f000 | 510 | |
1da177e4 LT |
511 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
512 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
19a809af AA |
513 | vma->vm_file, pgoff, vma_policy(vma), |
514 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
515 | if (*prev) { |
516 | vma = *prev; | |
517 | goto success; | |
518 | } | |
519 | ||
1da177e4 LT |
520 | if (start != vma->vm_start) { |
521 | ret = split_vma(mm, vma, start, 1); | |
522 | if (ret) | |
523 | goto out; | |
524 | } | |
525 | ||
526 | if (end != vma->vm_end) { | |
527 | ret = split_vma(mm, vma, end, 0); | |
528 | if (ret) | |
529 | goto out; | |
530 | } | |
531 | ||
532 | success: | |
b291f000 NP |
533 | /* |
534 | * Keep track of amount of locked VM. | |
535 | */ | |
536 | nr_pages = (end - start) >> PAGE_SHIFT; | |
537 | if (!lock) | |
538 | nr_pages = -nr_pages; | |
b155b4fd SG |
539 | else if (old_flags & VM_LOCKED) |
540 | nr_pages = 0; | |
b291f000 NP |
541 | mm->locked_vm += nr_pages; |
542 | ||
1da177e4 | 543 | /* |
c1e8d7c6 | 544 | * vm_flags is protected by the mmap_lock held in write mode. |
1da177e4 | 545 | * It's okay if try_to_unmap_one unmaps a page just after we |
fc05f566 | 546 | * set VM_LOCKED, populate_vma_page_range will bring it back. |
1da177e4 | 547 | */ |
1da177e4 | 548 | |
fed067da | 549 | if (lock) |
408e82b7 | 550 | vma->vm_flags = newflags; |
fed067da | 551 | else |
408e82b7 | 552 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 553 | |
1da177e4 | 554 | out: |
b291f000 | 555 | *prev = vma; |
1da177e4 LT |
556 | return ret; |
557 | } | |
558 | ||
1aab92ec EM |
559 | static int apply_vma_lock_flags(unsigned long start, size_t len, |
560 | vm_flags_t flags) | |
1da177e4 LT |
561 | { |
562 | unsigned long nstart, end, tmp; | |
563 | struct vm_area_struct * vma, * prev; | |
564 | int error; | |
565 | ||
8fd9e488 | 566 | VM_BUG_ON(offset_in_page(start)); |
fed067da | 567 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
1da177e4 LT |
568 | end = start + len; |
569 | if (end < start) | |
570 | return -EINVAL; | |
571 | if (end == start) | |
572 | return 0; | |
097d5910 | 573 | vma = find_vma(current->mm, start); |
1da177e4 LT |
574 | if (!vma || vma->vm_start > start) |
575 | return -ENOMEM; | |
576 | ||
097d5910 | 577 | prev = vma->vm_prev; |
1da177e4 LT |
578 | if (start > vma->vm_start) |
579 | prev = vma; | |
580 | ||
581 | for (nstart = start ; ; ) { | |
b0f205c2 | 582 | vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
1da177e4 | 583 | |
1aab92ec | 584 | newflags |= flags; |
1da177e4 | 585 | |
1aab92ec | 586 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 LT |
587 | tmp = vma->vm_end; |
588 | if (tmp > end) | |
589 | tmp = end; | |
590 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
591 | if (error) | |
592 | break; | |
593 | nstart = tmp; | |
594 | if (nstart < prev->vm_end) | |
595 | nstart = prev->vm_end; | |
596 | if (nstart >= end) | |
597 | break; | |
598 | ||
599 | vma = prev->vm_next; | |
600 | if (!vma || vma->vm_start != nstart) { | |
601 | error = -ENOMEM; | |
602 | break; | |
603 | } | |
604 | } | |
605 | return error; | |
606 | } | |
607 | ||
0cf2f6f6 SG |
608 | /* |
609 | * Go through vma areas and sum size of mlocked | |
610 | * vma pages, as return value. | |
611 | * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT) | |
612 | * is also counted. | |
613 | * Return value: previously mlocked page counts | |
614 | */ | |
0874bb49 | 615 | static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm, |
0cf2f6f6 SG |
616 | unsigned long start, size_t len) |
617 | { | |
618 | struct vm_area_struct *vma; | |
0874bb49 | 619 | unsigned long count = 0; |
0cf2f6f6 SG |
620 | |
621 | if (mm == NULL) | |
622 | mm = current->mm; | |
623 | ||
624 | vma = find_vma(mm, start); | |
625 | if (vma == NULL) | |
626 | vma = mm->mmap; | |
627 | ||
628 | for (; vma ; vma = vma->vm_next) { | |
629 | if (start >= vma->vm_end) | |
630 | continue; | |
631 | if (start + len <= vma->vm_start) | |
632 | break; | |
633 | if (vma->vm_flags & VM_LOCKED) { | |
634 | if (start > vma->vm_start) | |
635 | count -= (start - vma->vm_start); | |
636 | if (start + len < vma->vm_end) { | |
637 | count += start + len - vma->vm_start; | |
638 | break; | |
639 | } | |
640 | count += vma->vm_end - vma->vm_start; | |
641 | } | |
642 | } | |
643 | ||
644 | return count >> PAGE_SHIFT; | |
645 | } | |
646 | ||
dc0ef0df | 647 | static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags) |
1da177e4 LT |
648 | { |
649 | unsigned long locked; | |
650 | unsigned long lock_limit; | |
651 | int error = -ENOMEM; | |
652 | ||
057d3389 AK |
653 | start = untagged_addr(start); |
654 | ||
1da177e4 LT |
655 | if (!can_do_mlock()) |
656 | return -EPERM; | |
657 | ||
8fd9e488 | 658 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 LT |
659 | start &= PAGE_MASK; |
660 | ||
59e99e5b | 661 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 | 662 | lock_limit >>= PAGE_SHIFT; |
1f1cd705 DB |
663 | locked = len >> PAGE_SHIFT; |
664 | ||
d8ed45c5 | 665 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 666 | return -EINTR; |
1f1cd705 DB |
667 | |
668 | locked += current->mm->locked_vm; | |
0cf2f6f6 SG |
669 | if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) { |
670 | /* | |
671 | * It is possible that the regions requested intersect with | |
672 | * previously mlocked areas, that part area in "mm->locked_vm" | |
673 | * should not be counted to new mlock increment count. So check | |
674 | * and adjust locked count if necessary. | |
675 | */ | |
676 | locked -= count_mm_mlocked_page_nr(current->mm, | |
677 | start, len); | |
678 | } | |
1da177e4 LT |
679 | |
680 | /* check against resource limits */ | |
681 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
1aab92ec | 682 | error = apply_vma_lock_flags(start, len, flags); |
1f1cd705 | 683 | |
d8ed45c5 | 684 | mmap_write_unlock(current->mm); |
c561259c KS |
685 | if (error) |
686 | return error; | |
687 | ||
688 | error = __mm_populate(start, len, 0); | |
689 | if (error) | |
690 | return __mlock_posix_error_return(error); | |
691 | return 0; | |
1da177e4 LT |
692 | } |
693 | ||
1aab92ec EM |
694 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
695 | { | |
696 | return do_mlock(start, len, VM_LOCKED); | |
697 | } | |
698 | ||
a8ca5d0e EM |
699 | SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags) |
700 | { | |
b0f205c2 EM |
701 | vm_flags_t vm_flags = VM_LOCKED; |
702 | ||
703 | if (flags & ~MLOCK_ONFAULT) | |
a8ca5d0e EM |
704 | return -EINVAL; |
705 | ||
b0f205c2 EM |
706 | if (flags & MLOCK_ONFAULT) |
707 | vm_flags |= VM_LOCKONFAULT; | |
708 | ||
709 | return do_mlock(start, len, vm_flags); | |
a8ca5d0e EM |
710 | } |
711 | ||
6a6160a7 | 712 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
713 | { |
714 | int ret; | |
715 | ||
057d3389 AK |
716 | start = untagged_addr(start); |
717 | ||
8fd9e488 | 718 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 | 719 | start &= PAGE_MASK; |
1f1cd705 | 720 | |
d8ed45c5 | 721 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 722 | return -EINTR; |
1aab92ec | 723 | ret = apply_vma_lock_flags(start, len, 0); |
d8ed45c5 | 724 | mmap_write_unlock(current->mm); |
1f1cd705 | 725 | |
1da177e4 LT |
726 | return ret; |
727 | } | |
728 | ||
b0f205c2 EM |
729 | /* |
730 | * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall) | |
731 | * and translate into the appropriate modifications to mm->def_flags and/or the | |
732 | * flags for all current VMAs. | |
733 | * | |
734 | * There are a couple of subtleties with this. If mlockall() is called multiple | |
735 | * times with different flags, the values do not necessarily stack. If mlockall | |
736 | * is called once including the MCL_FUTURE flag and then a second time without | |
737 | * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags. | |
738 | */ | |
1aab92ec | 739 | static int apply_mlockall_flags(int flags) |
1da177e4 LT |
740 | { |
741 | struct vm_area_struct * vma, * prev = NULL; | |
b0f205c2 | 742 | vm_flags_t to_add = 0; |
1da177e4 | 743 | |
b0f205c2 EM |
744 | current->mm->def_flags &= VM_LOCKED_CLEAR_MASK; |
745 | if (flags & MCL_FUTURE) { | |
09a9f1d2 | 746 | current->mm->def_flags |= VM_LOCKED; |
1aab92ec | 747 | |
b0f205c2 EM |
748 | if (flags & MCL_ONFAULT) |
749 | current->mm->def_flags |= VM_LOCKONFAULT; | |
750 | ||
751 | if (!(flags & MCL_CURRENT)) | |
752 | goto out; | |
753 | } | |
754 | ||
755 | if (flags & MCL_CURRENT) { | |
756 | to_add |= VM_LOCKED; | |
757 | if (flags & MCL_ONFAULT) | |
758 | to_add |= VM_LOCKONFAULT; | |
759 | } | |
1da177e4 LT |
760 | |
761 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 762 | vm_flags_t newflags; |
1da177e4 | 763 | |
b0f205c2 EM |
764 | newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
765 | newflags |= to_add; | |
1da177e4 LT |
766 | |
767 | /* Ignore errors */ | |
768 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
50d4fb78 | 769 | cond_resched(); |
1da177e4 LT |
770 | } |
771 | out: | |
772 | return 0; | |
773 | } | |
774 | ||
3480b257 | 775 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
776 | { |
777 | unsigned long lock_limit; | |
86d2adcc | 778 | int ret; |
1da177e4 | 779 | |
dedca635 PS |
780 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) || |
781 | flags == MCL_ONFAULT) | |
86d2adcc | 782 | return -EINVAL; |
1da177e4 | 783 | |
1da177e4 | 784 | if (!can_do_mlock()) |
86d2adcc | 785 | return -EPERM; |
1da177e4 | 786 | |
59e99e5b | 787 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
788 | lock_limit >>= PAGE_SHIFT; |
789 | ||
d8ed45c5 | 790 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 791 | return -EINTR; |
1f1cd705 | 792 | |
dc0ef0df | 793 | ret = -ENOMEM; |
1da177e4 LT |
794 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
795 | capable(CAP_IPC_LOCK)) | |
1aab92ec | 796 | ret = apply_mlockall_flags(flags); |
d8ed45c5 | 797 | mmap_write_unlock(current->mm); |
bebeb3d6 ML |
798 | if (!ret && (flags & MCL_CURRENT)) |
799 | mm_populate(0, TASK_SIZE); | |
86d2adcc | 800 | |
1da177e4 LT |
801 | return ret; |
802 | } | |
803 | ||
3480b257 | 804 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
805 | { |
806 | int ret; | |
807 | ||
d8ed45c5 | 808 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 809 | return -EINTR; |
1aab92ec | 810 | ret = apply_mlockall_flags(0); |
d8ed45c5 | 811 | mmap_write_unlock(current->mm); |
1da177e4 LT |
812 | return ret; |
813 | } | |
814 | ||
815 | /* | |
816 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
817 | * shm segments) get accounted against the user_struct instead. | |
818 | */ | |
819 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
820 | ||
821 | int user_shm_lock(size_t size, struct user_struct *user) | |
822 | { | |
823 | unsigned long lock_limit, locked; | |
824 | int allowed = 0; | |
825 | ||
826 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 827 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
828 | if (lock_limit == RLIM_INFINITY) |
829 | allowed = 1; | |
1da177e4 LT |
830 | lock_limit >>= PAGE_SHIFT; |
831 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
832 | if (!allowed && |
833 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
834 | goto out; |
835 | get_uid(user); | |
836 | user->locked_shm += locked; | |
837 | allowed = 1; | |
838 | out: | |
839 | spin_unlock(&shmlock_user_lock); | |
840 | return allowed; | |
841 | } | |
842 | ||
843 | void user_shm_unlock(size_t size, struct user_struct *user) | |
844 | { | |
845 | spin_lock(&shmlock_user_lock); | |
846 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
847 | spin_unlock(&shmlock_user_lock); | |
848 | free_uid(user); | |
849 | } |