]>
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 | { | |
126 | int ret = SWAP_AGAIN; | |
127 | ||
128 | /* | |
129 | * Optimization: if the page was mapped just once, that's our mapping | |
130 | * and we don't need to check all the other vmas. | |
131 | */ | |
132 | if (page_mapcount(page) > 1) | |
133 | ret = try_to_munlock(page); | |
134 | ||
135 | /* Did try_to_unlock() succeed or punt? */ | |
136 | if (ret != SWAP_MLOCK) | |
137 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); | |
138 | ||
139 | putback_lru_page(page); | |
140 | } | |
141 | ||
142 | /* | |
143 | * Accounting for page isolation fail during munlock | |
144 | * | |
145 | * Performs accounting when page isolation fails in munlock. There is nothing | |
146 | * else to do because it means some other task has already removed the page | |
147 | * from the LRU. putback_lru_page() will take care of removing the page from | |
148 | * the unevictable list, if necessary. vmscan [page_referenced()] will move | |
149 | * the page back to the unevictable list if some other vma has it mlocked. | |
150 | */ | |
151 | static void __munlock_isolation_failed(struct page *page) | |
152 | { | |
153 | if (PageUnevictable(page)) | |
01cc2e58 | 154 | __count_vm_event(UNEVICTABLE_PGSTRANDED); |
7225522b | 155 | else |
01cc2e58 | 156 | __count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
7225522b VB |
157 | } |
158 | ||
6927c1dd LS |
159 | /** |
160 | * munlock_vma_page - munlock a vma page | |
c424be1c VB |
161 | * @page - page to be unlocked, either a normal page or THP page head |
162 | * | |
163 | * returns the size of the page as a page mask (0 for normal page, | |
164 | * HPAGE_PMD_NR - 1 for THP head page) | |
b291f000 | 165 | * |
6927c1dd LS |
166 | * called from munlock()/munmap() path with page supposedly on the LRU. |
167 | * When we munlock a page, because the vma where we found the page is being | |
168 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the | |
169 | * page locked so that we can leave it on the unevictable lru list and not | |
170 | * bother vmscan with it. However, to walk the page's rmap list in | |
171 | * try_to_munlock() we must isolate the page from the LRU. If some other | |
172 | * task has removed the page from the LRU, we won't be able to do that. | |
173 | * So we clear the PageMlocked as we might not get another chance. If we | |
174 | * can't isolate the page, we leave it for putback_lru_page() and vmscan | |
175 | * [page_referenced()/try_to_unmap()] to deal with. | |
b291f000 | 176 | */ |
ff6a6da6 | 177 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 178 | { |
7162a1e8 | 179 | int nr_pages; |
01cc2e58 | 180 | struct zone *zone = page_zone(page); |
ff6a6da6 | 181 | |
57e68e9c | 182 | /* For try_to_munlock() and to serialize with page migration */ |
b291f000 NP |
183 | BUG_ON(!PageLocked(page)); |
184 | ||
e90309c9 KS |
185 | VM_BUG_ON_PAGE(PageTail(page), page); |
186 | ||
c424be1c | 187 | /* |
01cc2e58 VB |
188 | * Serialize with any parallel __split_huge_page_refcount() which |
189 | * might otherwise copy PageMlocked to part of the tail pages before | |
190 | * we clear it in the head page. It also stabilizes hpage_nr_pages(). | |
c424be1c | 191 | */ |
a52633d8 | 192 | spin_lock_irq(zone_lru_lock(zone)); |
01cc2e58 | 193 | |
655548bf KS |
194 | if (!TestClearPageMlocked(page)) { |
195 | /* Potentially, PTE-mapped THP: do not skip the rest PTEs */ | |
196 | nr_pages = 1; | |
01cc2e58 | 197 | goto unlock_out; |
655548bf | 198 | } |
01cc2e58 | 199 | |
655548bf | 200 | nr_pages = hpage_nr_pages(page); |
01cc2e58 VB |
201 | __mod_zone_page_state(zone, NR_MLOCK, -nr_pages); |
202 | ||
203 | if (__munlock_isolate_lru_page(page, true)) { | |
a52633d8 | 204 | spin_unlock_irq(zone_lru_lock(zone)); |
01cc2e58 VB |
205 | __munlock_isolated_page(page); |
206 | goto out; | |
207 | } | |
208 | __munlock_isolation_failed(page); | |
209 | ||
210 | unlock_out: | |
a52633d8 | 211 | spin_unlock_irq(zone_lru_lock(zone)); |
01cc2e58 VB |
212 | |
213 | out: | |
c424be1c | 214 | return nr_pages - 1; |
b291f000 NP |
215 | } |
216 | ||
9978ad58 LS |
217 | /* |
218 | * convert get_user_pages() return value to posix mlock() error | |
219 | */ | |
220 | static int __mlock_posix_error_return(long retval) | |
221 | { | |
222 | if (retval == -EFAULT) | |
223 | retval = -ENOMEM; | |
224 | else if (retval == -ENOMEM) | |
225 | retval = -EAGAIN; | |
226 | return retval; | |
b291f000 NP |
227 | } |
228 | ||
56afe477 VB |
229 | /* |
230 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() | |
231 | * | |
232 | * The fast path is available only for evictable pages with single mapping. | |
233 | * Then we can bypass the per-cpu pvec and get better performance. | |
234 | * when mapcount > 1 we need try_to_munlock() which can fail. | |
235 | * when !page_evictable(), we need the full redo logic of putback_lru_page to | |
236 | * avoid leaving evictable page in unevictable list. | |
237 | * | |
238 | * In case of success, @page is added to @pvec and @pgrescued is incremented | |
239 | * in case that the page was previously unevictable. @page is also unlocked. | |
240 | */ | |
241 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, | |
242 | int *pgrescued) | |
243 | { | |
309381fe SL |
244 | VM_BUG_ON_PAGE(PageLRU(page), page); |
245 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
56afe477 VB |
246 | |
247 | if (page_mapcount(page) <= 1 && page_evictable(page)) { | |
248 | pagevec_add(pvec, page); | |
249 | if (TestClearPageUnevictable(page)) | |
250 | (*pgrescued)++; | |
251 | unlock_page(page); | |
252 | return true; | |
253 | } | |
254 | ||
255 | return false; | |
256 | } | |
257 | ||
258 | /* | |
259 | * Putback multiple evictable pages to the LRU | |
260 | * | |
261 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of | |
262 | * the pages might have meanwhile become unevictable but that is OK. | |
263 | */ | |
264 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) | |
265 | { | |
266 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); | |
267 | /* | |
268 | *__pagevec_lru_add() calls release_pages() so we don't call | |
269 | * put_page() explicitly | |
270 | */ | |
271 | __pagevec_lru_add(pvec); | |
272 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
273 | } | |
274 | ||
7225522b VB |
275 | /* |
276 | * Munlock a batch of pages from the same zone | |
277 | * | |
278 | * The work is split to two main phases. First phase clears the Mlocked flag | |
279 | * and attempts to isolate the pages, all under a single zone lru lock. | |
280 | * The second phase finishes the munlock only for pages where isolation | |
281 | * succeeded. | |
282 | * | |
7a8010cd | 283 | * Note that the pagevec may be modified during the process. |
7225522b VB |
284 | */ |
285 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) | |
286 | { | |
287 | int i; | |
288 | int nr = pagevec_count(pvec); | |
3b25df93 | 289 | int delta_munlocked; |
56afe477 VB |
290 | struct pagevec pvec_putback; |
291 | int pgrescued = 0; | |
7225522b | 292 | |
3b25df93 VB |
293 | pagevec_init(&pvec_putback, 0); |
294 | ||
7225522b | 295 | /* Phase 1: page isolation */ |
a52633d8 | 296 | spin_lock_irq(zone_lru_lock(zone)); |
7225522b VB |
297 | for (i = 0; i < nr; i++) { |
298 | struct page *page = pvec->pages[i]; | |
299 | ||
300 | if (TestClearPageMlocked(page)) { | |
7225522b | 301 | /* |
01cc2e58 VB |
302 | * We already have pin from follow_page_mask() |
303 | * so we can spare the get_page() here. | |
7225522b | 304 | */ |
01cc2e58 VB |
305 | if (__munlock_isolate_lru_page(page, false)) |
306 | continue; | |
307 | else | |
308 | __munlock_isolation_failed(page); | |
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 | } |
3b25df93 | 320 | delta_munlocked = -nr + pagevec_count(&pvec_putback); |
1ebb7cc6 | 321 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
a52633d8 | 322 | spin_unlock_irq(zone_lru_lock(zone)); |
7225522b | 323 | |
3b25df93 VB |
324 | /* Now we can release pins of pages that we are not munlocking */ |
325 | pagevec_release(&pvec_putback); | |
326 | ||
56afe477 | 327 | /* Phase 2: page munlock */ |
7225522b VB |
328 | for (i = 0; i < nr; i++) { |
329 | struct page *page = pvec->pages[i]; | |
330 | ||
331 | if (page) { | |
332 | lock_page(page); | |
56afe477 VB |
333 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
334 | &pgrescued)) { | |
5b40998a VB |
335 | /* |
336 | * Slow path. We don't want to lose the last | |
337 | * pin before unlock_page() | |
338 | */ | |
339 | get_page(page); /* for putback_lru_page() */ | |
56afe477 VB |
340 | __munlock_isolated_page(page); |
341 | unlock_page(page); | |
5b40998a | 342 | put_page(page); /* from follow_page_mask() */ |
56afe477 | 343 | } |
7225522b VB |
344 | } |
345 | } | |
56afe477 | 346 | |
5b40998a VB |
347 | /* |
348 | * Phase 3: page putback for pages that qualified for the fast path | |
349 | * This will also call put_page() to return pin from follow_page_mask() | |
350 | */ | |
56afe477 VB |
351 | if (pagevec_count(&pvec_putback)) |
352 | __putback_lru_fast(&pvec_putback, pgrescued); | |
7a8010cd VB |
353 | } |
354 | ||
355 | /* | |
356 | * Fill up pagevec for __munlock_pagevec using pte walk | |
357 | * | |
358 | * The function expects that the struct page corresponding to @start address is | |
359 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. | |
360 | * | |
361 | * The rest of @pvec is filled by subsequent pages within the same pmd and same | |
362 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These | |
363 | * pages also get pinned. | |
364 | * | |
365 | * Returns the address of the next page that should be scanned. This equals | |
366 | * @start + PAGE_SIZE when no page could be added by the pte walk. | |
367 | */ | |
368 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, | |
369 | struct vm_area_struct *vma, int zoneid, unsigned long start, | |
370 | unsigned long end) | |
371 | { | |
372 | pte_t *pte; | |
373 | spinlock_t *ptl; | |
374 | ||
375 | /* | |
376 | * Initialize pte walk starting at the already pinned page where we | |
eadb41ae VB |
377 | * are sure that there is a pte, as it was pinned under the same |
378 | * mmap_sem write op. | |
7a8010cd VB |
379 | */ |
380 | pte = get_locked_pte(vma->vm_mm, start, &ptl); | |
eadb41ae VB |
381 | /* Make sure we do not cross the page table boundary */ |
382 | end = pgd_addr_end(start, end); | |
c2febafc | 383 | end = p4d_addr_end(start, end); |
eadb41ae VB |
384 | end = pud_addr_end(start, end); |
385 | end = pmd_addr_end(start, end); | |
7a8010cd VB |
386 | |
387 | /* The page next to the pinned page is the first we will try to get */ | |
388 | start += PAGE_SIZE; | |
389 | while (start < end) { | |
390 | struct page *page = NULL; | |
391 | pte++; | |
392 | if (pte_present(*pte)) | |
393 | page = vm_normal_page(vma, start, *pte); | |
394 | /* | |
395 | * Break if page could not be obtained or the page's node+zone does not | |
396 | * match | |
397 | */ | |
398 | if (!page || page_zone_id(page) != zoneid) | |
399 | break; | |
56afe477 | 400 | |
e90309c9 KS |
401 | /* |
402 | * Do not use pagevec for PTE-mapped THP, | |
403 | * munlock_vma_pages_range() will handle them. | |
404 | */ | |
405 | if (PageTransCompound(page)) | |
406 | break; | |
407 | ||
7a8010cd VB |
408 | get_page(page); |
409 | /* | |
410 | * Increase the address that will be returned *before* the | |
411 | * eventual break due to pvec becoming full by adding the page | |
412 | */ | |
413 | start += PAGE_SIZE; | |
414 | if (pagevec_add(pvec, page) == 0) | |
415 | break; | |
416 | } | |
417 | pte_unmap_unlock(pte, ptl); | |
418 | return start; | |
7225522b VB |
419 | } |
420 | ||
b291f000 | 421 | /* |
ba470de4 RR |
422 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
423 | * @vma - vma containing range to be munlock()ed. | |
424 | * @start - start address in @vma of the range | |
425 | * @end - end of range in @vma. | |
426 | * | |
427 | * For mremap(), munmap() and exit(). | |
428 | * | |
429 | * Called with @vma VM_LOCKED. | |
430 | * | |
431 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
432 | * deal with this. | |
433 | * | |
434 | * We don't save and restore VM_LOCKED here because pages are | |
435 | * still on lru. In unmap path, pages might be scanned by reclaim | |
436 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
437 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 438 | */ |
ba470de4 | 439 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 440 | unsigned long start, unsigned long end) |
b291f000 | 441 | { |
de60f5f1 | 442 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
408e82b7 | 443 | |
ff6a6da6 | 444 | while (start < end) { |
ab7a5af7 | 445 | struct page *page; |
6ebb4a1b | 446 | unsigned int page_mask = 0; |
c424be1c | 447 | unsigned long page_increm; |
7a8010cd VB |
448 | struct pagevec pvec; |
449 | struct zone *zone; | |
450 | int zoneid; | |
ff6a6da6 | 451 | |
7a8010cd | 452 | pagevec_init(&pvec, 0); |
6e919717 HD |
453 | /* |
454 | * Although FOLL_DUMP is intended for get_dump_page(), | |
455 | * it just so happens that its special treatment of the | |
456 | * ZERO_PAGE (returning an error instead of doing get_page) | |
457 | * suits munlock very well (and if somehow an abnormal page | |
458 | * has sneaked into the range, we won't oops here: great). | |
459 | */ | |
6ebb4a1b | 460 | page = follow_page(vma, start, FOLL_GET | FOLL_DUMP); |
7a8010cd | 461 | |
e90309c9 KS |
462 | if (page && !IS_ERR(page)) { |
463 | if (PageTransTail(page)) { | |
464 | VM_BUG_ON_PAGE(PageMlocked(page), page); | |
465 | put_page(page); /* follow_page_mask() */ | |
466 | } else if (PageTransHuge(page)) { | |
467 | lock_page(page); | |
468 | /* | |
469 | * Any THP page found by follow_page_mask() may | |
470 | * have gotten split before reaching | |
6ebb4a1b KS |
471 | * munlock_vma_page(), so we need to compute |
472 | * the page_mask here instead. | |
e90309c9 KS |
473 | */ |
474 | page_mask = munlock_vma_page(page); | |
475 | unlock_page(page); | |
476 | put_page(page); /* follow_page_mask() */ | |
477 | } else { | |
478 | /* | |
479 | * Non-huge pages are handled in batches via | |
480 | * pagevec. The pin from follow_page_mask() | |
481 | * prevents them from collapsing by THP. | |
482 | */ | |
483 | pagevec_add(&pvec, page); | |
484 | zone = page_zone(page); | |
485 | zoneid = page_zone_id(page); | |
7a8010cd | 486 | |
e90309c9 KS |
487 | /* |
488 | * Try to fill the rest of pagevec using fast | |
489 | * pte walk. This will also update start to | |
490 | * the next page to process. Then munlock the | |
491 | * pagevec. | |
492 | */ | |
493 | start = __munlock_pagevec_fill(&pvec, vma, | |
494 | zoneid, start, end); | |
495 | __munlock_pagevec(&pvec, zone); | |
496 | goto next; | |
497 | } | |
408e82b7 | 498 | } |
c424be1c | 499 | page_increm = 1 + page_mask; |
ff6a6da6 | 500 | start += page_increm * PAGE_SIZE; |
7a8010cd | 501 | next: |
408e82b7 HD |
502 | cond_resched(); |
503 | } | |
b291f000 NP |
504 | } |
505 | ||
506 | /* | |
507 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
508 | * | |
509 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
510 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 511 | * populate the ptes. |
b291f000 NP |
512 | * |
513 | * For vmas that pass the filters, merge/split as appropriate. | |
514 | */ | |
1da177e4 | 515 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 516 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 517 | { |
b291f000 | 518 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 519 | pgoff_t pgoff; |
b291f000 | 520 | int nr_pages; |
1da177e4 | 521 | int ret = 0; |
ca16d140 | 522 | int lock = !!(newflags & VM_LOCKED); |
b155b4fd | 523 | vm_flags_t old_flags = vma->vm_flags; |
1da177e4 | 524 | |
fed067da | 525 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
31db58b3 | 526 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
b0f205c2 EM |
527 | /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */ |
528 | goto out; | |
b291f000 | 529 | |
1da177e4 LT |
530 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
531 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
19a809af AA |
532 | vma->vm_file, pgoff, vma_policy(vma), |
533 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
534 | if (*prev) { |
535 | vma = *prev; | |
536 | goto success; | |
537 | } | |
538 | ||
1da177e4 LT |
539 | if (start != vma->vm_start) { |
540 | ret = split_vma(mm, vma, start, 1); | |
541 | if (ret) | |
542 | goto out; | |
543 | } | |
544 | ||
545 | if (end != vma->vm_end) { | |
546 | ret = split_vma(mm, vma, end, 0); | |
547 | if (ret) | |
548 | goto out; | |
549 | } | |
550 | ||
551 | success: | |
b291f000 NP |
552 | /* |
553 | * Keep track of amount of locked VM. | |
554 | */ | |
555 | nr_pages = (end - start) >> PAGE_SHIFT; | |
556 | if (!lock) | |
557 | nr_pages = -nr_pages; | |
b155b4fd SG |
558 | else if (old_flags & VM_LOCKED) |
559 | nr_pages = 0; | |
b291f000 NP |
560 | mm->locked_vm += nr_pages; |
561 | ||
1da177e4 LT |
562 | /* |
563 | * vm_flags is protected by the mmap_sem held in write mode. | |
564 | * It's okay if try_to_unmap_one unmaps a page just after we | |
fc05f566 | 565 | * set VM_LOCKED, populate_vma_page_range will bring it back. |
1da177e4 | 566 | */ |
1da177e4 | 567 | |
fed067da | 568 | if (lock) |
408e82b7 | 569 | vma->vm_flags = newflags; |
fed067da | 570 | else |
408e82b7 | 571 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 572 | |
1da177e4 | 573 | out: |
b291f000 | 574 | *prev = vma; |
1da177e4 LT |
575 | return ret; |
576 | } | |
577 | ||
1aab92ec EM |
578 | static int apply_vma_lock_flags(unsigned long start, size_t len, |
579 | vm_flags_t flags) | |
1da177e4 LT |
580 | { |
581 | unsigned long nstart, end, tmp; | |
582 | struct vm_area_struct * vma, * prev; | |
583 | int error; | |
584 | ||
8fd9e488 | 585 | VM_BUG_ON(offset_in_page(start)); |
fed067da | 586 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
1da177e4 LT |
587 | end = start + len; |
588 | if (end < start) | |
589 | return -EINVAL; | |
590 | if (end == start) | |
591 | return 0; | |
097d5910 | 592 | vma = find_vma(current->mm, start); |
1da177e4 LT |
593 | if (!vma || vma->vm_start > start) |
594 | return -ENOMEM; | |
595 | ||
097d5910 | 596 | prev = vma->vm_prev; |
1da177e4 LT |
597 | if (start > vma->vm_start) |
598 | prev = vma; | |
599 | ||
600 | for (nstart = start ; ; ) { | |
b0f205c2 | 601 | vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
1da177e4 | 602 | |
1aab92ec | 603 | newflags |= flags; |
1da177e4 | 604 | |
1aab92ec | 605 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 LT |
606 | tmp = vma->vm_end; |
607 | if (tmp > end) | |
608 | tmp = end; | |
609 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
610 | if (error) | |
611 | break; | |
612 | nstart = tmp; | |
613 | if (nstart < prev->vm_end) | |
614 | nstart = prev->vm_end; | |
615 | if (nstart >= end) | |
616 | break; | |
617 | ||
618 | vma = prev->vm_next; | |
619 | if (!vma || vma->vm_start != nstart) { | |
620 | error = -ENOMEM; | |
621 | break; | |
622 | } | |
623 | } | |
624 | return error; | |
625 | } | |
626 | ||
0cf2f6f6 SG |
627 | /* |
628 | * Go through vma areas and sum size of mlocked | |
629 | * vma pages, as return value. | |
630 | * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT) | |
631 | * is also counted. | |
632 | * Return value: previously mlocked page counts | |
633 | */ | |
634 | static int count_mm_mlocked_page_nr(struct mm_struct *mm, | |
635 | unsigned long start, size_t len) | |
636 | { | |
637 | struct vm_area_struct *vma; | |
638 | int count = 0; | |
639 | ||
640 | if (mm == NULL) | |
641 | mm = current->mm; | |
642 | ||
643 | vma = find_vma(mm, start); | |
644 | if (vma == NULL) | |
645 | vma = mm->mmap; | |
646 | ||
647 | for (; vma ; vma = vma->vm_next) { | |
648 | if (start >= vma->vm_end) | |
649 | continue; | |
650 | if (start + len <= vma->vm_start) | |
651 | break; | |
652 | if (vma->vm_flags & VM_LOCKED) { | |
653 | if (start > vma->vm_start) | |
654 | count -= (start - vma->vm_start); | |
655 | if (start + len < vma->vm_end) { | |
656 | count += start + len - vma->vm_start; | |
657 | break; | |
658 | } | |
659 | count += vma->vm_end - vma->vm_start; | |
660 | } | |
661 | } | |
662 | ||
663 | return count >> PAGE_SHIFT; | |
664 | } | |
665 | ||
dc0ef0df | 666 | static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags) |
1da177e4 LT |
667 | { |
668 | unsigned long locked; | |
669 | unsigned long lock_limit; | |
670 | int error = -ENOMEM; | |
671 | ||
672 | if (!can_do_mlock()) | |
673 | return -EPERM; | |
674 | ||
8891d6da KM |
675 | lru_add_drain_all(); /* flush pagevec */ |
676 | ||
8fd9e488 | 677 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 LT |
678 | start &= PAGE_MASK; |
679 | ||
59e99e5b | 680 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 | 681 | lock_limit >>= PAGE_SHIFT; |
1f1cd705 DB |
682 | locked = len >> PAGE_SHIFT; |
683 | ||
dc0ef0df MH |
684 | if (down_write_killable(¤t->mm->mmap_sem)) |
685 | return -EINTR; | |
1f1cd705 DB |
686 | |
687 | locked += current->mm->locked_vm; | |
0cf2f6f6 SG |
688 | if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) { |
689 | /* | |
690 | * It is possible that the regions requested intersect with | |
691 | * previously mlocked areas, that part area in "mm->locked_vm" | |
692 | * should not be counted to new mlock increment count. So check | |
693 | * and adjust locked count if necessary. | |
694 | */ | |
695 | locked -= count_mm_mlocked_page_nr(current->mm, | |
696 | start, len); | |
697 | } | |
1da177e4 LT |
698 | |
699 | /* check against resource limits */ | |
700 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
1aab92ec | 701 | error = apply_vma_lock_flags(start, len, flags); |
1f1cd705 | 702 | |
1da177e4 | 703 | up_write(¤t->mm->mmap_sem); |
c561259c KS |
704 | if (error) |
705 | return error; | |
706 | ||
707 | error = __mm_populate(start, len, 0); | |
708 | if (error) | |
709 | return __mlock_posix_error_return(error); | |
710 | return 0; | |
1da177e4 LT |
711 | } |
712 | ||
1aab92ec EM |
713 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
714 | { | |
715 | return do_mlock(start, len, VM_LOCKED); | |
716 | } | |
717 | ||
a8ca5d0e EM |
718 | SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags) |
719 | { | |
b0f205c2 EM |
720 | vm_flags_t vm_flags = VM_LOCKED; |
721 | ||
722 | if (flags & ~MLOCK_ONFAULT) | |
a8ca5d0e EM |
723 | return -EINVAL; |
724 | ||
b0f205c2 EM |
725 | if (flags & MLOCK_ONFAULT) |
726 | vm_flags |= VM_LOCKONFAULT; | |
727 | ||
728 | return do_mlock(start, len, vm_flags); | |
a8ca5d0e EM |
729 | } |
730 | ||
6a6160a7 | 731 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
732 | { |
733 | int ret; | |
734 | ||
8fd9e488 | 735 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
1da177e4 | 736 | start &= PAGE_MASK; |
1f1cd705 | 737 | |
dc0ef0df MH |
738 | if (down_write_killable(¤t->mm->mmap_sem)) |
739 | return -EINTR; | |
1aab92ec | 740 | ret = apply_vma_lock_flags(start, len, 0); |
1da177e4 | 741 | up_write(¤t->mm->mmap_sem); |
1f1cd705 | 742 | |
1da177e4 LT |
743 | return ret; |
744 | } | |
745 | ||
b0f205c2 EM |
746 | /* |
747 | * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall) | |
748 | * and translate into the appropriate modifications to mm->def_flags and/or the | |
749 | * flags for all current VMAs. | |
750 | * | |
751 | * There are a couple of subtleties with this. If mlockall() is called multiple | |
752 | * times with different flags, the values do not necessarily stack. If mlockall | |
753 | * is called once including the MCL_FUTURE flag and then a second time without | |
754 | * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags. | |
755 | */ | |
1aab92ec | 756 | static int apply_mlockall_flags(int flags) |
1da177e4 LT |
757 | { |
758 | struct vm_area_struct * vma, * prev = NULL; | |
b0f205c2 | 759 | vm_flags_t to_add = 0; |
1da177e4 | 760 | |
b0f205c2 EM |
761 | current->mm->def_flags &= VM_LOCKED_CLEAR_MASK; |
762 | if (flags & MCL_FUTURE) { | |
09a9f1d2 | 763 | current->mm->def_flags |= VM_LOCKED; |
1aab92ec | 764 | |
b0f205c2 EM |
765 | if (flags & MCL_ONFAULT) |
766 | current->mm->def_flags |= VM_LOCKONFAULT; | |
767 | ||
768 | if (!(flags & MCL_CURRENT)) | |
769 | goto out; | |
770 | } | |
771 | ||
772 | if (flags & MCL_CURRENT) { | |
773 | to_add |= VM_LOCKED; | |
774 | if (flags & MCL_ONFAULT) | |
775 | to_add |= VM_LOCKONFAULT; | |
776 | } | |
1da177e4 LT |
777 | |
778 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 779 | vm_flags_t newflags; |
1da177e4 | 780 | |
b0f205c2 EM |
781 | newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
782 | newflags |= to_add; | |
1da177e4 LT |
783 | |
784 | /* Ignore errors */ | |
785 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
bde6c3aa | 786 | cond_resched_rcu_qs(); |
1da177e4 LT |
787 | } |
788 | out: | |
789 | return 0; | |
790 | } | |
791 | ||
3480b257 | 792 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
793 | { |
794 | unsigned long lock_limit; | |
86d2adcc | 795 | int ret; |
1da177e4 | 796 | |
b0f205c2 | 797 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT))) |
86d2adcc | 798 | return -EINVAL; |
1da177e4 | 799 | |
1da177e4 | 800 | if (!can_do_mlock()) |
86d2adcc | 801 | return -EPERM; |
1da177e4 | 802 | |
df9d6985 CL |
803 | if (flags & MCL_CURRENT) |
804 | lru_add_drain_all(); /* flush pagevec */ | |
8891d6da | 805 | |
59e99e5b | 806 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
807 | lock_limit >>= PAGE_SHIFT; |
808 | ||
dc0ef0df MH |
809 | if (down_write_killable(¤t->mm->mmap_sem)) |
810 | return -EINTR; | |
1f1cd705 | 811 | |
dc0ef0df | 812 | ret = -ENOMEM; |
1da177e4 LT |
813 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
814 | capable(CAP_IPC_LOCK)) | |
1aab92ec | 815 | ret = apply_mlockall_flags(flags); |
1da177e4 | 816 | up_write(¤t->mm->mmap_sem); |
bebeb3d6 ML |
817 | if (!ret && (flags & MCL_CURRENT)) |
818 | mm_populate(0, TASK_SIZE); | |
86d2adcc | 819 | |
1da177e4 LT |
820 | return ret; |
821 | } | |
822 | ||
3480b257 | 823 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
824 | { |
825 | int ret; | |
826 | ||
dc0ef0df MH |
827 | if (down_write_killable(¤t->mm->mmap_sem)) |
828 | return -EINTR; | |
1aab92ec | 829 | ret = apply_mlockall_flags(0); |
1da177e4 LT |
830 | up_write(¤t->mm->mmap_sem); |
831 | return ret; | |
832 | } | |
833 | ||
834 | /* | |
835 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
836 | * shm segments) get accounted against the user_struct instead. | |
837 | */ | |
838 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
839 | ||
840 | int user_shm_lock(size_t size, struct user_struct *user) | |
841 | { | |
842 | unsigned long lock_limit, locked; | |
843 | int allowed = 0; | |
844 | ||
845 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 846 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
847 | if (lock_limit == RLIM_INFINITY) |
848 | allowed = 1; | |
1da177e4 LT |
849 | lock_limit >>= PAGE_SHIFT; |
850 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
851 | if (!allowed && |
852 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
853 | goto out; |
854 | get_uid(user); | |
855 | user->locked_shm += locked; | |
856 | allowed = 1; | |
857 | out: | |
858 | spin_unlock(&shmlock_user_lock); | |
859 | return allowed; | |
860 | } | |
861 | ||
862 | void user_shm_unlock(size_t size, struct user_struct *user) | |
863 | { | |
864 | spin_lock(&shmlock_user_lock); | |
865 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
866 | spin_unlock(&shmlock_user_lock); | |
867 | free_uid(user); | |
868 | } |