Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/mlock.c | |
3 | * | |
4 | * (C) Copyright 1995 Linus Torvalds | |
5 | * (C) Copyright 2002 Christoph Hellwig | |
6 | */ | |
7 | ||
c59ede7b | 8 | #include <linux/capability.h> |
1da177e4 LT |
9 | #include <linux/mman.h> |
10 | #include <linux/mm.h> | |
8703e8a4 | 11 | #include <linux/sched/user.h> |
b291f000 NP |
12 | #include <linux/swap.h> |
13 | #include <linux/swapops.h> | |
14 | #include <linux/pagemap.h> | |
7225522b | 15 | #include <linux/pagevec.h> |
1da177e4 LT |
16 | #include <linux/mempolicy.h> |
17 | #include <linux/syscalls.h> | |
e8edc6e0 | 18 | #include <linux/sched.h> |
b95f1b31 | 19 | #include <linux/export.h> |
b291f000 NP |
20 | #include <linux/rmap.h> |
21 | #include <linux/mmzone.h> | |
22 | #include <linux/hugetlb.h> | |
7225522b VB |
23 | #include <linux/memcontrol.h> |
24 | #include <linux/mm_inline.h> | |
b291f000 NP |
25 | |
26 | #include "internal.h" | |
1da177e4 | 27 | |
7f43add4 | 28 | bool can_do_mlock(void) |
e8edc6e0 | 29 | { |
59e99e5b | 30 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
7f43add4 | 31 | return true; |
a5a6579d | 32 | if (capable(CAP_IPC_LOCK)) |
7f43add4 WX |
33 | return true; |
34 | return false; | |
e8edc6e0 AD |
35 | } |
36 | EXPORT_SYMBOL(can_do_mlock); | |
1da177e4 | 37 | |
b291f000 NP |
38 | /* |
39 | * Mlocked pages are marked with PageMlocked() flag for efficient testing | |
40 | * in vmscan and, possibly, the fault path; and to support semi-accurate | |
41 | * statistics. | |
42 | * | |
43 | * An mlocked page [PageMlocked(page)] is unevictable. As such, it will | |
44 | * be placed on the LRU "unevictable" list, rather than the [in]active lists. | |
45 | * The unevictable list is an LRU sibling list to the [in]active lists. | |
46 | * PageUnevictable is set to indicate the unevictable state. | |
47 | * | |
48 | * When lazy mlocking via vmscan, it is important to ensure that the | |
49 | * vma's VM_LOCKED status is not concurrently being modified, otherwise we | |
50 | * may have mlocked a page that is being munlocked. So lazy mlock must take | |
51 | * the mmap_sem for read, and verify that the vma really is locked | |
52 | * (see mm/rmap.c). | |
53 | */ | |
54 | ||
55 | /* | |
56 | * LRU accounting for clear_page_mlock() | |
57 | */ | |
e6c509f8 | 58 | void clear_page_mlock(struct page *page) |
b291f000 | 59 | { |
e6c509f8 | 60 | if (!TestClearPageMlocked(page)) |
b291f000 | 61 | return; |
b291f000 | 62 | |
8449d21f DR |
63 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
64 | -hpage_nr_pages(page)); | |
5344b7e6 | 65 | count_vm_event(UNEVICTABLE_PGCLEARED); |
b291f000 NP |
66 | if (!isolate_lru_page(page)) { |
67 | putback_lru_page(page); | |
68 | } else { | |
69 | /* | |
8891d6da | 70 | * We lost the race. the page already moved to evictable list. |
b291f000 | 71 | */ |
8891d6da | 72 | if (PageUnevictable(page)) |
5344b7e6 | 73 | count_vm_event(UNEVICTABLE_PGSTRANDED); |
b291f000 NP |
74 | } |
75 | } | |
76 | ||
77 | /* | |
78 | * Mark page as mlocked if not already. | |
79 | * If page on LRU, isolate and putback to move to unevictable list. | |
80 | */ | |
81 | void mlock_vma_page(struct page *page) | |
82 | { | |
57e68e9c | 83 | /* Serialize with page migration */ |
b291f000 NP |
84 | BUG_ON(!PageLocked(page)); |
85 | ||
e90309c9 KS |
86 | VM_BUG_ON_PAGE(PageTail(page), page); |
87 | VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page); | |
88 | ||
5344b7e6 | 89 | if (!TestSetPageMlocked(page)) { |
8449d21f DR |
90 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
91 | hpage_nr_pages(page)); | |
5344b7e6 NP |
92 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
93 | if (!isolate_lru_page(page)) | |
94 | putback_lru_page(page); | |
95 | } | |
b291f000 NP |
96 | } |
97 | ||
01cc2e58 VB |
98 | /* |
99 | * Isolate a page from LRU with optional get_page() pin. | |
100 | * Assumes lru_lock already held and page already pinned. | |
101 | */ | |
102 | static bool __munlock_isolate_lru_page(struct page *page, bool getpage) | |
103 | { | |
104 | if (PageLRU(page)) { | |
105 | struct lruvec *lruvec; | |
106 | ||
599d0c95 | 107 | lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page)); |
01cc2e58 VB |
108 | if (getpage) |
109 | get_page(page); | |
110 | ClearPageLRU(page); | |
111 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
112 | return true; | |
113 | } | |
114 | ||
115 | return false; | |
116 | } | |
117 | ||
7225522b VB |
118 | /* |
119 | * Finish munlock after successful page isolation | |
120 | * | |
121 | * Page must be locked. This is a wrapper for try_to_munlock() | |
122 | * and putback_lru_page() with munlock accounting. | |
123 | */ | |
124 | static void __munlock_isolated_page(struct page *page) | |
125 | { | |
7225522b VB |
126 | /* |
127 | * Optimization: if the page was mapped just once, that's our mapping | |
128 | * and we don't need to check all the other vmas. | |
129 | */ | |
130 | if (page_mapcount(page) > 1) | |
192d7232 | 131 | try_to_munlock(page); |
7225522b VB |
132 | |
133 | /* Did try_to_unlock() succeed or punt? */ | |
192d7232 | 134 | if (!PageMlocked(page)) |
7225522b VB |
135 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
136 | ||
137 | putback_lru_page(page); | |
138 | } | |
139 | ||
140 | /* | |
141 | * Accounting for page isolation fail during munlock | |
142 | * | |
143 | * Performs accounting when page isolation fails in munlock. There is nothing | |
144 | * else to do because it means some other task has already removed the page | |
145 | * from the LRU. putback_lru_page() will take care of removing the page from | |
146 | * the unevictable list, if necessary. vmscan [page_referenced()] will move | |
147 | * the page back to the unevictable list if some other vma has it mlocked. | |
148 | */ | |
149 | static void __munlock_isolation_failed(struct page *page) | |
150 | { | |
151 | if (PageUnevictable(page)) | |
01cc2e58 | 152 | __count_vm_event(UNEVICTABLE_PGSTRANDED); |
7225522b | 153 | else |
01cc2e58 | 154 | __count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
7225522b VB |
155 | } |
156 | ||
6927c1dd LS |
157 | /** |
158 | * munlock_vma_page - munlock a vma page | |
c424be1c VB |
159 | * @page - page to be unlocked, either a normal page or THP page head |
160 | * | |
161 | * returns the size of the page as a page mask (0 for normal page, | |
162 | * HPAGE_PMD_NR - 1 for THP head page) | |
b291f000 | 163 | * |
6927c1dd LS |
164 | * called from munlock()/munmap() path with page supposedly on the LRU. |
165 | * When we munlock a page, because the vma where we found the page is being | |
166 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the | |
167 | * page locked so that we can leave it on the unevictable lru list and not | |
168 | * bother vmscan with it. However, to walk the page's rmap list in | |
169 | * try_to_munlock() we must isolate the page from the LRU. If some other | |
170 | * task has removed the page from the LRU, we won't be able to do that. | |
171 | * So we clear the PageMlocked as we might not get another chance. If we | |
172 | * can't isolate the page, we leave it for putback_lru_page() and vmscan | |
173 | * [page_referenced()/try_to_unmap()] to deal with. | |
b291f000 | 174 | */ |
ff6a6da6 | 175 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 176 | { |
7162a1e8 | 177 | int nr_pages; |
01cc2e58 | 178 | struct zone *zone = page_zone(page); |
ff6a6da6 | 179 | |
57e68e9c | 180 | /* For try_to_munlock() and to serialize with page migration */ |
b291f000 NP |
181 | BUG_ON(!PageLocked(page)); |
182 | ||
e90309c9 KS |
183 | VM_BUG_ON_PAGE(PageTail(page), page); |
184 | ||
c424be1c | 185 | /* |
01cc2e58 VB |
186 | * Serialize with any parallel __split_huge_page_refcount() which |
187 | * might otherwise copy PageMlocked to part of the tail pages before | |
188 | * we clear it in the head page. It also stabilizes hpage_nr_pages(). | |
c424be1c | 189 | */ |
a52633d8 | 190 | spin_lock_irq(zone_lru_lock(zone)); |
01cc2e58 | 191 | |
655548bf KS |
192 | if (!TestClearPageMlocked(page)) { |
193 | /* Potentially, PTE-mapped THP: do not skip the rest PTEs */ | |
194 | nr_pages = 1; | |
01cc2e58 | 195 | goto unlock_out; |
655548bf | 196 | } |
01cc2e58 | 197 | |
655548bf | 198 | nr_pages = hpage_nr_pages(page); |
01cc2e58 VB |
199 | __mod_zone_page_state(zone, NR_MLOCK, -nr_pages); |
200 | ||
201 | if (__munlock_isolate_lru_page(page, true)) { | |
a52633d8 | 202 | spin_unlock_irq(zone_lru_lock(zone)); |
01cc2e58 VB |
203 | __munlock_isolated_page(page); |
204 | goto out; | |
205 | } | |
206 | __munlock_isolation_failed(page); | |
207 | ||
208 | unlock_out: | |
a52633d8 | 209 | spin_unlock_irq(zone_lru_lock(zone)); |
01cc2e58 VB |
210 | |
211 | out: | |
c424be1c | 212 | return nr_pages - 1; |
b291f000 NP |
213 | } |
214 | ||
9978ad58 LS |
215 | /* |
216 | * convert get_user_pages() return value to posix mlock() error | |
217 | */ | |
218 | static int __mlock_posix_error_return(long retval) | |
219 | { | |
220 | if (retval == -EFAULT) | |
221 | retval = -ENOMEM; | |
222 | else if (retval == -ENOMEM) | |
223 | retval = -EAGAIN; | |
224 | return retval; | |
b291f000 NP |
225 | } |
226 | ||
56afe477 VB |
227 | /* |
228 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() | |
229 | * | |
230 | * The fast path is available only for evictable pages with single mapping. | |
231 | * Then we can bypass the per-cpu pvec and get better performance. | |
232 | * when mapcount > 1 we need try_to_munlock() which can fail. | |
233 | * when !page_evictable(), we need the full redo logic of putback_lru_page to | |
234 | * avoid leaving evictable page in unevictable list. | |
235 | * | |
236 | * In case of success, @page is added to @pvec and @pgrescued is incremented | |
237 | * in case that the page was previously unevictable. @page is also unlocked. | |
238 | */ | |
239 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, | |
240 | int *pgrescued) | |
241 | { | |
309381fe SL |
242 | VM_BUG_ON_PAGE(PageLRU(page), page); |
243 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
56afe477 VB |
244 | |
245 | if (page_mapcount(page) <= 1 && page_evictable(page)) { | |
246 | pagevec_add(pvec, page); | |
247 | if (TestClearPageUnevictable(page)) | |
248 | (*pgrescued)++; | |
249 | unlock_page(page); | |
250 | return true; | |
251 | } | |
252 | ||
253 | return false; | |
254 | } | |
255 | ||
256 | /* | |
257 | * Putback multiple evictable pages to the LRU | |
258 | * | |
259 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of | |
260 | * the pages might have meanwhile become unevictable but that is OK. | |
261 | */ | |
262 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) | |
263 | { | |
264 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); | |
265 | /* | |
266 | *__pagevec_lru_add() calls release_pages() so we don't call | |
267 | * put_page() explicitly | |
268 | */ | |
269 | __pagevec_lru_add(pvec); | |
270 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
271 | } | |
272 | ||
7225522b VB |
273 | /* |
274 | * Munlock a batch of pages from the same zone | |
275 | * | |
276 | * The work is split to two main phases. First phase clears the Mlocked flag | |
277 | * and attempts to isolate the pages, all under a single zone lru lock. | |
278 | * The second phase finishes the munlock only for pages where isolation | |
279 | * succeeded. | |
280 | * | |
7a8010cd | 281 | * Note that the pagevec may be modified during the process. |
7225522b VB |
282 | */ |
283 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) | |
284 | { | |
285 | int i; | |
286 | int nr = pagevec_count(pvec); | |
70feee0e | 287 | int delta_munlocked = -nr; |
56afe477 VB |
288 | struct pagevec pvec_putback; |
289 | int pgrescued = 0; | |
7225522b | 290 | |
3b25df93 VB |
291 | pagevec_init(&pvec_putback, 0); |
292 | ||
7225522b | 293 | /* Phase 1: page isolation */ |
a52633d8 | 294 | spin_lock_irq(zone_lru_lock(zone)); |
7225522b VB |
295 | for (i = 0; i < nr; i++) { |
296 | struct page *page = pvec->pages[i]; | |
297 | ||
298 | if (TestClearPageMlocked(page)) { | |
7225522b | 299 | /* |
01cc2e58 VB |
300 | * We already have pin from follow_page_mask() |
301 | * so we can spare the get_page() here. | |
7225522b | 302 | */ |
01cc2e58 VB |
303 | if (__munlock_isolate_lru_page(page, false)) |
304 | continue; | |
305 | else | |
306 | __munlock_isolation_failed(page); | |
70feee0e YX |
307 | } else { |
308 | delta_munlocked++; | |
7225522b | 309 | } |
01cc2e58 VB |
310 | |
311 | /* | |
312 | * We won't be munlocking this page in the next phase | |
313 | * but we still need to release the follow_page_mask() | |
314 | * pin. We cannot do it under lru_lock however. If it's | |
315 | * the last pin, __page_cache_release() would deadlock. | |
316 | */ | |
317 | pagevec_add(&pvec_putback, pvec->pages[i]); | |
318 | pvec->pages[i] = NULL; | |
7225522b | 319 | } |
1ebb7cc6 | 320 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
a52633d8 | 321 | spin_unlock_irq(zone_lru_lock(zone)); |
7225522b | 322 | |
3b25df93 VB |
323 | /* Now we can release pins of pages that we are not munlocking */ |
324 | pagevec_release(&pvec_putback); | |
325 | ||
56afe477 | 326 | /* Phase 2: page munlock */ |
7225522b VB |
327 | for (i = 0; i < nr; i++) { |
328 | struct page *page = pvec->pages[i]; | |
329 | ||
330 | if (page) { | |
331 | lock_page(page); | |
56afe477 VB |
332 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
333 | &pgrescued)) { | |
5b40998a VB |
334 | /* |
335 | * Slow path. We don't want to lose the last | |
336 | * pin before unlock_page() | |
337 | */ | |
338 | get_page(page); /* for putback_lru_page() */ | |
56afe477 VB |
339 | __munlock_isolated_page(page); |
340 | unlock_page(page); | |
5b40998a | 341 | put_page(page); /* from follow_page_mask() */ |
56afe477 | 342 | } |
7225522b VB |
343 | } |
344 | } | |
56afe477 | 345 | |
5b40998a VB |
346 | /* |
347 | * Phase 3: page putback for pages that qualified for the fast path | |
348 | * This will also call put_page() to return pin from follow_page_mask() | |
349 | */ | |
56afe477 VB |
350 | if (pagevec_count(&pvec_putback)) |
351 | __putback_lru_fast(&pvec_putback, pgrescued); | |
7a8010cd VB |
352 | } |
353 | ||
354 | /* | |
355 | * Fill up pagevec for __munlock_pagevec using pte walk | |
356 | * | |
357 | * The function expects that the struct page corresponding to @start address is | |
358 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. | |
359 | * | |
360 | * The rest of @pvec is filled by subsequent pages within the same pmd and same | |
361 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These | |
362 | * pages also get pinned. | |
363 | * | |
364 | * Returns the address of the next page that should be scanned. This equals | |
365 | * @start + PAGE_SIZE when no page could be added by the pte walk. | |
366 | */ | |
367 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, | |
9472f23c JK |
368 | struct vm_area_struct *vma, struct zone *zone, |
369 | unsigned long start, unsigned long end) | |
7a8010cd VB |
370 | { |
371 | pte_t *pte; | |
372 | spinlock_t *ptl; | |
373 | ||
374 | /* | |
375 | * Initialize pte walk starting at the already pinned page where we | |
eadb41ae VB |
376 | * are sure that there is a pte, as it was pinned under the same |
377 | * mmap_sem write op. | |
7a8010cd VB |
378 | */ |
379 | pte = get_locked_pte(vma->vm_mm, start, &ptl); | |
eadb41ae VB |
380 | /* Make sure we do not cross the page table boundary */ |
381 | end = pgd_addr_end(start, end); | |
c2febafc | 382 | end = p4d_addr_end(start, end); |
eadb41ae VB |
383 | end = pud_addr_end(start, end); |
384 | end = pmd_addr_end(start, end); | |
7a8010cd VB |
385 | |
386 | /* The page next to the pinned page is the first we will try to get */ | |
387 | start += PAGE_SIZE; | |
388 | while (start < end) { | |
389 | struct page *page = NULL; | |
390 | pte++; | |
391 | if (pte_present(*pte)) | |
392 | page = vm_normal_page(vma, start, *pte); | |
393 | /* | |
394 | * Break if page could not be obtained or the page's node+zone does not | |
395 | * match | |
396 | */ | |
9472f23c | 397 | if (!page || page_zone(page) != zone) |
7a8010cd | 398 | break; |
56afe477 | 399 | |
e90309c9 KS |
400 | /* |
401 | * Do not use pagevec for PTE-mapped THP, | |
402 | * munlock_vma_pages_range() will handle them. | |
403 | */ | |
404 | if (PageTransCompound(page)) | |
405 | break; | |
406 | ||
7a8010cd VB |
407 | get_page(page); |
408 | /* | |
409 | * Increase the address that will be returned *before* the | |
410 | * eventual break due to pvec becoming full by adding the page | |
411 | */ | |
412 | start += PAGE_SIZE; | |
413 | if (pagevec_add(pvec, page) == 0) | |
414 | break; | |
415 | } | |
416 | pte_unmap_unlock(pte, ptl); | |
417 | return start; | |
7225522b VB |
418 | } |
419 | ||
b291f000 | 420 | /* |
ba470de4 RR |
421 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
422 | * @vma - vma containing range to be munlock()ed. | |
423 | * @start - start address in @vma of the range | |
424 | * @end - end of range in @vma. | |
425 | * | |
426 | * For mremap(), munmap() and exit(). | |
427 | * | |
428 | * Called with @vma VM_LOCKED. | |
429 | * | |
430 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
431 | * deal with this. | |
432 | * | |
433 | * We don't save and restore VM_LOCKED here because pages are | |
434 | * still on lru. In unmap path, pages might be scanned by reclaim | |
435 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
436 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 437 | */ |
ba470de4 | 438 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 439 | unsigned long start, unsigned long end) |
b291f000 | 440 | { |
de60f5f1 | 441 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
408e82b7 | 442 | |
ff6a6da6 | 443 | while (start < end) { |
ab7a5af7 | 444 | struct page *page; |
6ebb4a1b | 445 | unsigned int page_mask = 0; |
c424be1c | 446 | unsigned long page_increm; |
7a8010cd VB |
447 | struct pagevec pvec; |
448 | struct zone *zone; | |
ff6a6da6 | 449 | |
7a8010cd | 450 | pagevec_init(&pvec, 0); |
6e919717 HD |
451 | /* |
452 | * Although FOLL_DUMP is intended for get_dump_page(), | |
453 | * it just so happens that its special treatment of the | |
454 | * ZERO_PAGE (returning an error instead of doing get_page) | |
455 | * suits munlock very well (and if somehow an abnormal page | |
456 | * has sneaked into the range, we won't oops here: great). | |
457 | */ | |
6ebb4a1b | 458 | page = follow_page(vma, start, FOLL_GET | FOLL_DUMP); |
7a8010cd | 459 | |
e90309c9 KS |
460 | if (page && !IS_ERR(page)) { |
461 | if (PageTransTail(page)) { | |
462 | VM_BUG_ON_PAGE(PageMlocked(page), page); | |
463 | put_page(page); /* follow_page_mask() */ | |
464 | } else if (PageTransHuge(page)) { | |
465 | lock_page(page); | |
466 | /* | |
467 | * Any THP page found by follow_page_mask() may | |
468 | * have gotten split before reaching | |
6ebb4a1b KS |
469 | * munlock_vma_page(), so we need to compute |
470 | * the page_mask here instead. | |
e90309c9 KS |
471 | */ |
472 | page_mask = munlock_vma_page(page); | |
473 | unlock_page(page); | |
474 | put_page(page); /* follow_page_mask() */ | |
475 | } else { | |
476 | /* | |
477 | * Non-huge pages are handled in batches via | |
478 | * pagevec. The pin from follow_page_mask() | |
479 | * prevents them from collapsing by THP. | |
480 | */ | |
481 | pagevec_add(&pvec, page); | |
482 | zone = page_zone(page); | |
7a8010cd | 483 | |
e90309c9 KS |
484 | /* |
485 | * Try to fill the rest of pagevec using fast | |
486 | * pte walk. This will also update start to | |
487 | * the next page to process. Then munlock the | |
488 | * pagevec. | |
489 | */ | |
490 | start = __munlock_pagevec_fill(&pvec, vma, | |
9472f23c | 491 | zone, start, end); |
e90309c9 KS |
492 | __munlock_pagevec(&pvec, zone); |
493 | goto next; | |
494 | } | |
408e82b7 | 495 | } |
c424be1c | 496 | page_increm = 1 + page_mask; |
ff6a6da6 | 497 | start += page_increm * PAGE_SIZE; |
7a8010cd | 498 | next: |
408e82b7 HD |
499 | cond_resched(); |
500 | } | |
b291f000 NP |
501 | } |
502 | ||
503 | /* | |
504 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
505 | * | |
506 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
507 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 508 | * populate the ptes. |
b291f000 NP |
509 | * |
510 | * For vmas that pass the filters, merge/split as appropriate. | |
511 | */ | |
1da177e4 | 512 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 513 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 514 | { |
b291f000 | 515 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 516 | pgoff_t pgoff; |
b291f000 | 517 | int nr_pages; |
1da177e4 | 518 | int ret = 0; |
ca16d140 | 519 | int lock = !!(newflags & VM_LOCKED); |
b155b4fd | 520 | vm_flags_t old_flags = vma->vm_flags; |
1da177e4 | 521 | |
fed067da | 522 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
31db58b3 | 523 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
b0f205c2 EM |
524 | /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */ |
525 | goto out; | |
b291f000 | 526 | |
1da177e4 LT |
527 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
528 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
19a809af AA |
529 | vma->vm_file, pgoff, vma_policy(vma), |
530 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
531 | if (*prev) { |
532 | vma = *prev; | |
533 | goto success; | |
534 | } | |
535 | ||
1da177e4 LT |
536 | if (start != vma->vm_start) { |
537 | ret = split_vma(mm, vma, start, 1); | |
538 | if (ret) | |
539 | goto out; | |
540 | } | |
541 | ||
542 | if (end != vma->vm_end) { | |
543 | ret = split_vma(mm, vma, end, 0); | |
544 | if (ret) | |
545 | goto out; | |
546 | } | |
547 | ||
548 | success: | |
b291f000 NP |
549 | /* |
550 | * Keep track of amount of locked VM. | |
551 | */ | |
552 | nr_pages = (end - start) >> PAGE_SHIFT; | |
553 | if (!lock) | |
554 | nr_pages = -nr_pages; | |
b155b4fd SG |
555 | else if (old_flags & VM_LOCKED) |
556 | nr_pages = 0; | |
b291f000 NP |
557 | mm->locked_vm += nr_pages; |
558 | ||
1da177e4 LT |
559 | /* |
560 | * vm_flags is protected by the mmap_sem held in write mode. | |
561 | * It's okay if try_to_unmap_one unmaps a page just after we | |
fc05f566 | 562 | * set VM_LOCKED, populate_vma_page_range will bring it back. |
1da177e4 | 563 | */ |
1da177e4 | 564 | |
fed067da | 565 | if (lock) |
408e82b7 | 566 | vma->vm_flags = newflags; |
fed067da | 567 | else |
408e82b7 | 568 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 569 | |
1da177e4 | 570 | out: |
b291f000 | 571 | *prev = vma; |
1da177e4 LT |
572 | return ret; |
573 | } | |
574 | ||
1aab92ec EM |
575 | static int apply_vma_lock_flags(unsigned long start, size_t len, |
576 | vm_flags_t flags) | |
1da177e4 LT |
577 | { |
578 | unsigned long nstart, end, tmp; | |
579 | struct vm_area_struct * vma, * prev; | |
580 | int error; | |
581 | ||
8fd9e488 | 582 | VM_BUG_ON(offset_in_page(start)); |
fed067da | 583 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
1da177e4 LT |
584 | end = start + len; |
585 | if (end < start) | |
586 | return -EINVAL; | |
587 | if (end == start) | |
588 | return 0; | |
097d5910 | 589 | vma = find_vma(current->mm, start); |
1da177e4 LT |
590 | if (!vma || vma->vm_start > start) |
591 | return -ENOMEM; | |
592 | ||
097d5910 | 593 | prev = vma->vm_prev; |
1da177e4 LT |
594 | if (start > vma->vm_start) |
595 | prev = vma; | |
596 | ||
597 | for (nstart = start ; ; ) { | |
b0f205c2 | 598 | vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
1da177e4 | 599 | |
1aab92ec | 600 | newflags |= flags; |
1da177e4 | 601 | |
1aab92ec | 602 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 LT |
603 | tmp = vma->vm_end; |
604 | if (tmp > end) | |
605 | tmp = end; | |
606 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
607 | if (error) | |
608 | break; | |
609 | nstart = tmp; | |
610 | if (nstart < prev->vm_end) | |
611 | nstart = prev->vm_end; | |
612 | if (nstart >= end) | |
613 | break; | |
614 | ||
615 | vma = prev->vm_next; | |
616 | if (!vma || vma->vm_start != nstart) { | |
617 | error = -ENOMEM; | |
618 | break; | |
619 | } | |
620 | } | |
621 | return error; | |
622 | } | |
623 | ||
0cf2f6f6 SG |
624 | /* |
625 | * Go through vma areas and sum size of mlocked | |
626 | * vma pages, as return value. | |
627 | * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT) | |
628 | * is also counted. | |
629 | * Return value: previously mlocked page counts | |
630 | */ | |
631 | static int count_mm_mlocked_page_nr(struct mm_struct *mm, | |
632 | unsigned long start, size_t len) | |
633 | { | |
634 | struct vm_area_struct *vma; | |
635 | int count = 0; | |
636 | ||
637 | if (mm == NULL) | |
638 | mm = current->mm; | |
639 | ||
640 | vma = find_vma(mm, start); | |
641 | if (vma == NULL) | |
642 | vma = mm->mmap; | |
643 | ||
644 | for (; vma ; vma = vma->vm_next) { | |
645 | if (start >= vma->vm_end) | |
646 | continue; | |
647 | if (start + len <= vma->vm_start) | |
648 | break; | |
649 | if (vma->vm_flags & VM_LOCKED) { | |
650 | if (start > vma->vm_start) | |
651 | count -= (start - vma->vm_start); | |
652 | if (start + len < vma->vm_end) { | |
653 | count += start + len - vma->vm_start; | |
654 | break; | |
655 | } | |
656 | count += vma->vm_end - vma->vm_start; | |
657 | } | |
658 | } | |
659 | ||
660 | return count >> PAGE_SHIFT; | |
661 | } | |
662 | ||
dc0ef0df | 663 | static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags) |
1da177e4 LT |
664 | { |
665 | unsigned long locked; | |
666 | unsigned long lock_limit; | |
667 | int error = -ENOMEM; | |
668 | ||
669 | if (!can_do_mlock()) | |
670 | return -EPERM; | |
671 | ||
8891d6da KM |
672 | lru_add_drain_all(); /* flush pagevec */ |
673 | ||
8fd9e488 | 674 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 LT |
675 | start &= PAGE_MASK; |
676 | ||
59e99e5b | 677 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 | 678 | lock_limit >>= PAGE_SHIFT; |
1f1cd705 DB |
679 | locked = len >> PAGE_SHIFT; |
680 | ||
dc0ef0df MH |
681 | if (down_write_killable(¤t->mm->mmap_sem)) |
682 | return -EINTR; | |
1f1cd705 DB |
683 | |
684 | locked += current->mm->locked_vm; | |
0cf2f6f6 SG |
685 | if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) { |
686 | /* | |
687 | * It is possible that the regions requested intersect with | |
688 | * previously mlocked areas, that part area in "mm->locked_vm" | |
689 | * should not be counted to new mlock increment count. So check | |
690 | * and adjust locked count if necessary. | |
691 | */ | |
692 | locked -= count_mm_mlocked_page_nr(current->mm, | |
693 | start, len); | |
694 | } | |
1da177e4 LT |
695 | |
696 | /* check against resource limits */ | |
697 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
1aab92ec | 698 | error = apply_vma_lock_flags(start, len, flags); |
1f1cd705 | 699 | |
1da177e4 | 700 | up_write(¤t->mm->mmap_sem); |
c561259c KS |
701 | if (error) |
702 | return error; | |
703 | ||
704 | error = __mm_populate(start, len, 0); | |
705 | if (error) | |
706 | return __mlock_posix_error_return(error); | |
707 | return 0; | |
1da177e4 LT |
708 | } |
709 | ||
1aab92ec EM |
710 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
711 | { | |
712 | return do_mlock(start, len, VM_LOCKED); | |
713 | } | |
714 | ||
a8ca5d0e EM |
715 | SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags) |
716 | { | |
b0f205c2 EM |
717 | vm_flags_t vm_flags = VM_LOCKED; |
718 | ||
719 | if (flags & ~MLOCK_ONFAULT) | |
a8ca5d0e EM |
720 | return -EINVAL; |
721 | ||
b0f205c2 EM |
722 | if (flags & MLOCK_ONFAULT) |
723 | vm_flags |= VM_LOCKONFAULT; | |
724 | ||
725 | return do_mlock(start, len, vm_flags); | |
a8ca5d0e EM |
726 | } |
727 | ||
6a6160a7 | 728 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
729 | { |
730 | int ret; | |
731 | ||
8fd9e488 | 732 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 | 733 | start &= PAGE_MASK; |
1f1cd705 | 734 | |
dc0ef0df MH |
735 | if (down_write_killable(¤t->mm->mmap_sem)) |
736 | return -EINTR; | |
1aab92ec | 737 | ret = apply_vma_lock_flags(start, len, 0); |
1da177e4 | 738 | up_write(¤t->mm->mmap_sem); |
1f1cd705 | 739 | |
1da177e4 LT |
740 | return ret; |
741 | } | |
742 | ||
b0f205c2 EM |
743 | /* |
744 | * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall) | |
745 | * and translate into the appropriate modifications to mm->def_flags and/or the | |
746 | * flags for all current VMAs. | |
747 | * | |
748 | * There are a couple of subtleties with this. If mlockall() is called multiple | |
749 | * times with different flags, the values do not necessarily stack. If mlockall | |
750 | * is called once including the MCL_FUTURE flag and then a second time without | |
751 | * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags. | |
752 | */ | |
1aab92ec | 753 | static int apply_mlockall_flags(int flags) |
1da177e4 LT |
754 | { |
755 | struct vm_area_struct * vma, * prev = NULL; | |
b0f205c2 | 756 | vm_flags_t to_add = 0; |
1da177e4 | 757 | |
b0f205c2 EM |
758 | current->mm->def_flags &= VM_LOCKED_CLEAR_MASK; |
759 | if (flags & MCL_FUTURE) { | |
09a9f1d2 | 760 | current->mm->def_flags |= VM_LOCKED; |
1aab92ec | 761 | |
b0f205c2 EM |
762 | if (flags & MCL_ONFAULT) |
763 | current->mm->def_flags |= VM_LOCKONFAULT; | |
764 | ||
765 | if (!(flags & MCL_CURRENT)) | |
766 | goto out; | |
767 | } | |
768 | ||
769 | if (flags & MCL_CURRENT) { | |
770 | to_add |= VM_LOCKED; | |
771 | if (flags & MCL_ONFAULT) | |
772 | to_add |= VM_LOCKONFAULT; | |
773 | } | |
1da177e4 LT |
774 | |
775 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 776 | vm_flags_t newflags; |
1da177e4 | 777 | |
b0f205c2 EM |
778 | newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
779 | newflags |= to_add; | |
1da177e4 LT |
780 | |
781 | /* Ignore errors */ | |
782 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
bde6c3aa | 783 | cond_resched_rcu_qs(); |
1da177e4 LT |
784 | } |
785 | out: | |
786 | return 0; | |
787 | } | |
788 | ||
3480b257 | 789 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
790 | { |
791 | unsigned long lock_limit; | |
86d2adcc | 792 | int ret; |
1da177e4 | 793 | |
b0f205c2 | 794 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT))) |
86d2adcc | 795 | return -EINVAL; |
1da177e4 | 796 | |
1da177e4 | 797 | if (!can_do_mlock()) |
86d2adcc | 798 | return -EPERM; |
1da177e4 | 799 | |
df9d6985 CL |
800 | if (flags & MCL_CURRENT) |
801 | lru_add_drain_all(); /* flush pagevec */ | |
8891d6da | 802 | |
59e99e5b | 803 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
804 | lock_limit >>= PAGE_SHIFT; |
805 | ||
dc0ef0df MH |
806 | if (down_write_killable(¤t->mm->mmap_sem)) |
807 | return -EINTR; | |
1f1cd705 | 808 | |
dc0ef0df | 809 | ret = -ENOMEM; |
1da177e4 LT |
810 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
811 | capable(CAP_IPC_LOCK)) | |
1aab92ec | 812 | ret = apply_mlockall_flags(flags); |
1da177e4 | 813 | up_write(¤t->mm->mmap_sem); |
bebeb3d6 ML |
814 | if (!ret && (flags & MCL_CURRENT)) |
815 | mm_populate(0, TASK_SIZE); | |
86d2adcc | 816 | |
1da177e4 LT |
817 | return ret; |
818 | } | |
819 | ||
3480b257 | 820 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
821 | { |
822 | int ret; | |
823 | ||
dc0ef0df MH |
824 | if (down_write_killable(¤t->mm->mmap_sem)) |
825 | return -EINTR; | |
1aab92ec | 826 | ret = apply_mlockall_flags(0); |
1da177e4 LT |
827 | up_write(¤t->mm->mmap_sem); |
828 | return ret; | |
829 | } | |
830 | ||
831 | /* | |
832 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
833 | * shm segments) get accounted against the user_struct instead. | |
834 | */ | |
835 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
836 | ||
837 | int user_shm_lock(size_t size, struct user_struct *user) | |
838 | { | |
839 | unsigned long lock_limit, locked; | |
840 | int allowed = 0; | |
841 | ||
842 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 843 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
844 | if (lock_limit == RLIM_INFINITY) |
845 | allowed = 1; | |
1da177e4 LT |
846 | lock_limit >>= PAGE_SHIFT; |
847 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
848 | if (!allowed && |
849 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
850 | goto out; |
851 | get_uid(user); | |
852 | user->locked_shm += locked; | |
853 | allowed = 1; | |
854 | out: | |
855 | spin_unlock(&shmlock_user_lock); | |
856 | return allowed; | |
857 | } | |
858 | ||
859 | void user_shm_unlock(size_t size, struct user_struct *user) | |
860 | { | |
861 | spin_lock(&shmlock_user_lock); | |
862 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
863 | spin_unlock(&shmlock_user_lock); | |
864 | free_uid(user); | |
865 | } |