blob: 14645be06e301cfc4a62301e25818365e766874e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/mlock.c
3 *
4 * (C) Copyright 1995 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
6 */
7
Randy.Dunlapc59ede72006-01-11 12:17:46 -08008#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/mman.h>
10#include <linux/mm.h>
Nick Pigginb291f002008-10-18 20:26:44 -070011#include <linux/swap.h>
12#include <linux/swapops.h>
13#include <linux/pagemap.h>
Vlastimil Babka72255222013-09-11 14:22:29 -070014#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/mempolicy.h>
16#include <linux/syscalls.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040017#include <linux/sched.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040018#include <linux/export.h>
Nick Pigginb291f002008-10-18 20:26:44 -070019#include <linux/rmap.h>
20#include <linux/mmzone.h>
21#include <linux/hugetlb.h>
Vlastimil Babka72255222013-09-11 14:22:29 -070022#include <linux/memcontrol.h>
23#include <linux/mm_inline.h>
Nick Pigginb291f002008-10-18 20:26:44 -070024
25#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
Wang Xiaoqiang7f43add2016-01-15 16:57:22 -080027bool can_do_mlock(void)
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040028{
Jiri Slaby59e99e52010-03-05 13:41:44 -080029 if (rlimit(RLIMIT_MEMLOCK) != 0)
Wang Xiaoqiang7f43add2016-01-15 16:57:22 -080030 return true;
Jeff Vander Stoepa5a65792015-03-12 16:26:17 -070031 if (capable(CAP_IPC_LOCK))
Wang Xiaoqiang7f43add2016-01-15 16:57:22 -080032 return true;
33 return false;
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040034}
35EXPORT_SYMBOL(can_do_mlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Nick Pigginb291f002008-10-18 20:26:44 -070037/*
38 * Mlocked pages are marked with PageMlocked() flag for efficient testing
39 * in vmscan and, possibly, the fault path; and to support semi-accurate
40 * statistics.
41 *
42 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
43 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
44 * The unevictable list is an LRU sibling list to the [in]active lists.
45 * PageUnevictable is set to indicate the unevictable state.
46 *
47 * When lazy mlocking via vmscan, it is important to ensure that the
48 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
49 * may have mlocked a page that is being munlocked. So lazy mlock must take
50 * the mmap_sem for read, and verify that the vma really is locked
51 * (see mm/rmap.c).
52 */
53
54/*
55 * LRU accounting for clear_page_mlock()
56 */
Hugh Dickinse6c509f2012-10-08 16:33:19 -070057void clear_page_mlock(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -070058{
Hugh Dickinse6c509f2012-10-08 16:33:19 -070059 if (!TestClearPageMlocked(page))
Nick Pigginb291f002008-10-18 20:26:44 -070060 return;
Nick Pigginb291f002008-10-18 20:26:44 -070061
David Rientjes8449d212012-10-08 16:34:06 -070062 mod_zone_page_state(page_zone(page), NR_MLOCK,
63 -hpage_nr_pages(page));
Nick Piggin5344b7e2008-10-18 20:26:51 -070064 count_vm_event(UNEVICTABLE_PGCLEARED);
Nick Pigginb291f002008-10-18 20:26:44 -070065 if (!isolate_lru_page(page)) {
66 putback_lru_page(page);
67 } else {
68 /*
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -080069 * We lost the race. the page already moved to evictable list.
Nick Pigginb291f002008-10-18 20:26:44 -070070 */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -080071 if (PageUnevictable(page))
Nick Piggin5344b7e2008-10-18 20:26:51 -070072 count_vm_event(UNEVICTABLE_PGSTRANDED);
Nick Pigginb291f002008-10-18 20:26:44 -070073 }
74}
75
76/*
77 * Mark page as mlocked if not already.
78 * If page on LRU, isolate and putback to move to unevictable list.
79 */
80void mlock_vma_page(struct page *page)
81{
Vlastimil Babka57e68e92014-04-07 15:37:50 -070082 /* Serialize with page migration */
Nick Pigginb291f002008-10-18 20:26:44 -070083 BUG_ON(!PageLocked(page));
84
Kirill A. Shutemove90309c2016-01-15 16:54:33 -080085 VM_BUG_ON_PAGE(PageTail(page), page);
86 VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
87
Nick Piggin5344b7e2008-10-18 20:26:51 -070088 if (!TestSetPageMlocked(page)) {
David Rientjes8449d212012-10-08 16:34:06 -070089 mod_zone_page_state(page_zone(page), NR_MLOCK,
90 hpage_nr_pages(page));
Nick Piggin5344b7e2008-10-18 20:26:51 -070091 count_vm_event(UNEVICTABLE_PGMLOCKED);
92 if (!isolate_lru_page(page))
93 putback_lru_page(page);
94 }
Nick Pigginb291f002008-10-18 20:26:44 -070095}
96
Vlastimil Babka72255222013-09-11 14:22:29 -070097/*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -080098 * Isolate a page from LRU with optional get_page() pin.
99 * Assumes lru_lock already held and page already pinned.
100 */
101static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
102{
103 if (PageLRU(page)) {
104 struct lruvec *lruvec;
105
Mel Gorman599d0c92016-07-28 15:45:31 -0700106 lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page));
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800107 if (getpage)
108 get_page(page);
109 ClearPageLRU(page);
110 del_page_from_lru_list(page, lruvec, page_lru(page));
111 return true;
112 }
113
114 return false;
115}
116
117/*
Vlastimil Babka72255222013-09-11 14:22:29 -0700118 * Finish munlock after successful page isolation
119 *
120 * Page must be locked. This is a wrapper for try_to_munlock()
121 * and putback_lru_page() with munlock accounting.
122 */
123static void __munlock_isolated_page(struct page *page)
124{
125 int ret = SWAP_AGAIN;
126
127 /*
128 * Optimization: if the page was mapped just once, that's our mapping
129 * and we don't need to check all the other vmas.
130 */
131 if (page_mapcount(page) > 1)
132 ret = try_to_munlock(page);
133
134 /* Did try_to_unlock() succeed or punt? */
135 if (ret != SWAP_MLOCK)
136 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
137
138 putback_lru_page(page);
139}
140
141/*
142 * Accounting for page isolation fail during munlock
143 *
144 * Performs accounting when page isolation fails in munlock. There is nothing
145 * else to do because it means some other task has already removed the page
146 * from the LRU. putback_lru_page() will take care of removing the page from
147 * the unevictable list, if necessary. vmscan [page_referenced()] will move
148 * the page back to the unevictable list if some other vma has it mlocked.
149 */
150static void __munlock_isolation_failed(struct page *page)
151{
152 if (PageUnevictable(page))
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800153 __count_vm_event(UNEVICTABLE_PGSTRANDED);
Vlastimil Babka72255222013-09-11 14:22:29 -0700154 else
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800155 __count_vm_event(UNEVICTABLE_PGMUNLOCKED);
Vlastimil Babka72255222013-09-11 14:22:29 -0700156}
157
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800158/**
159 * munlock_vma_page - munlock a vma page
Vlastimil Babkac424be12014-01-02 12:58:43 -0800160 * @page - page to be unlocked, either a normal page or THP page head
161 *
162 * returns the size of the page as a page mask (0 for normal page,
163 * HPAGE_PMD_NR - 1 for THP head page)
Nick Pigginb291f002008-10-18 20:26:44 -0700164 *
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800165 * called from munlock()/munmap() path with page supposedly on the LRU.
166 * When we munlock a page, because the vma where we found the page is being
167 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
168 * page locked so that we can leave it on the unevictable lru list and not
169 * bother vmscan with it. However, to walk the page's rmap list in
170 * try_to_munlock() we must isolate the page from the LRU. If some other
171 * task has removed the page from the LRU, we won't be able to do that.
172 * So we clear the PageMlocked as we might not get another chance. If we
173 * can't isolate the page, we leave it for putback_lru_page() and vmscan
174 * [page_referenced()/try_to_unmap()] to deal with.
Nick Pigginb291f002008-10-18 20:26:44 -0700175 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800176unsigned int munlock_vma_page(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -0700177{
Kirill A. Shutemov7162a1e2016-01-21 16:40:27 -0800178 int nr_pages;
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800179 struct zone *zone = page_zone(page);
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800180
Vlastimil Babka57e68e92014-04-07 15:37:50 -0700181 /* For try_to_munlock() and to serialize with page migration */
Nick Pigginb291f002008-10-18 20:26:44 -0700182 BUG_ON(!PageLocked(page));
183
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800184 VM_BUG_ON_PAGE(PageTail(page), page);
185
Vlastimil Babkac424be12014-01-02 12:58:43 -0800186 /*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800187 * Serialize with any parallel __split_huge_page_refcount() which
188 * might otherwise copy PageMlocked to part of the tail pages before
189 * we clear it in the head page. It also stabilizes hpage_nr_pages().
Vlastimil Babkac424be12014-01-02 12:58:43 -0800190 */
Mel Gormana52633d2016-07-28 15:45:28 -0700191 spin_lock_irq(zone_lru_lock(zone));
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800192
193 nr_pages = hpage_nr_pages(page);
194 if (!TestClearPageMlocked(page))
195 goto unlock_out;
196
197 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
198
199 if (__munlock_isolate_lru_page(page, true)) {
Mel Gormana52633d2016-07-28 15:45:28 -0700200 spin_unlock_irq(zone_lru_lock(zone));
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800201 __munlock_isolated_page(page);
202 goto out;
203 }
204 __munlock_isolation_failed(page);
205
206unlock_out:
Mel Gormana52633d2016-07-28 15:45:28 -0700207 spin_unlock_irq(zone_lru_lock(zone));
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800208
209out:
Vlastimil Babkac424be12014-01-02 12:58:43 -0800210 return nr_pages - 1;
Nick Pigginb291f002008-10-18 20:26:44 -0700211}
212
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700213/*
214 * convert get_user_pages() return value to posix mlock() error
215 */
216static int __mlock_posix_error_return(long retval)
217{
218 if (retval == -EFAULT)
219 retval = -ENOMEM;
220 else if (retval == -ENOMEM)
221 retval = -EAGAIN;
222 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700223}
224
Nick Pigginb291f002008-10-18 20:26:44 -0700225/*
Vlastimil Babka56afe472013-09-11 14:22:32 -0700226 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
227 *
228 * The fast path is available only for evictable pages with single mapping.
229 * Then we can bypass the per-cpu pvec and get better performance.
230 * when mapcount > 1 we need try_to_munlock() which can fail.
231 * when !page_evictable(), we need the full redo logic of putback_lru_page to
232 * avoid leaving evictable page in unevictable list.
233 *
234 * In case of success, @page is added to @pvec and @pgrescued is incremented
235 * in case that the page was previously unevictable. @page is also unlocked.
236 */
237static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
238 int *pgrescued)
239{
Sasha Levin309381fea2014-01-23 15:52:54 -0800240 VM_BUG_ON_PAGE(PageLRU(page), page);
241 VM_BUG_ON_PAGE(!PageLocked(page), page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700242
243 if (page_mapcount(page) <= 1 && page_evictable(page)) {
244 pagevec_add(pvec, page);
245 if (TestClearPageUnevictable(page))
246 (*pgrescued)++;
247 unlock_page(page);
248 return true;
249 }
250
251 return false;
252}
253
254/*
255 * Putback multiple evictable pages to the LRU
256 *
257 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
258 * the pages might have meanwhile become unevictable but that is OK.
259 */
260static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
261{
262 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
263 /*
264 *__pagevec_lru_add() calls release_pages() so we don't call
265 * put_page() explicitly
266 */
267 __pagevec_lru_add(pvec);
268 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
269}
270
271/*
Vlastimil Babka72255222013-09-11 14:22:29 -0700272 * Munlock a batch of pages from the same zone
273 *
274 * The work is split to two main phases. First phase clears the Mlocked flag
275 * and attempts to isolate the pages, all under a single zone lru lock.
276 * The second phase finishes the munlock only for pages where isolation
277 * succeeded.
278 *
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700279 * Note that the pagevec may be modified during the process.
Vlastimil Babka72255222013-09-11 14:22:29 -0700280 */
281static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
282{
283 int i;
284 int nr = pagevec_count(pvec);
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800285 int delta_munlocked;
Vlastimil Babka56afe472013-09-11 14:22:32 -0700286 struct pagevec pvec_putback;
287 int pgrescued = 0;
Vlastimil Babka72255222013-09-11 14:22:29 -0700288
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800289 pagevec_init(&pvec_putback, 0);
290
Vlastimil Babka72255222013-09-11 14:22:29 -0700291 /* Phase 1: page isolation */
Mel Gormana52633d2016-07-28 15:45:28 -0700292 spin_lock_irq(zone_lru_lock(zone));
Vlastimil Babka72255222013-09-11 14:22:29 -0700293 for (i = 0; i < nr; i++) {
294 struct page *page = pvec->pages[i];
295
296 if (TestClearPageMlocked(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700297 /*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800298 * We already have pin from follow_page_mask()
299 * so we can spare the get_page() here.
Vlastimil Babka72255222013-09-11 14:22:29 -0700300 */
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800301 if (__munlock_isolate_lru_page(page, false))
302 continue;
303 else
304 __munlock_isolation_failed(page);
Vlastimil Babka72255222013-09-11 14:22:29 -0700305 }
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800306
307 /*
308 * We won't be munlocking this page in the next phase
309 * but we still need to release the follow_page_mask()
310 * pin. We cannot do it under lru_lock however. If it's
311 * the last pin, __page_cache_release() would deadlock.
312 */
313 pagevec_add(&pvec_putback, pvec->pages[i]);
314 pvec->pages[i] = NULL;
Vlastimil Babka72255222013-09-11 14:22:29 -0700315 }
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800316 delta_munlocked = -nr + pagevec_count(&pvec_putback);
Vlastimil Babka1ebb7cc2013-09-11 14:22:30 -0700317 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
Mel Gormana52633d2016-07-28 15:45:28 -0700318 spin_unlock_irq(zone_lru_lock(zone));
Vlastimil Babka72255222013-09-11 14:22:29 -0700319
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800320 /* Now we can release pins of pages that we are not munlocking */
321 pagevec_release(&pvec_putback);
322
Vlastimil Babka56afe472013-09-11 14:22:32 -0700323 /* Phase 2: page munlock */
Vlastimil Babka72255222013-09-11 14:22:29 -0700324 for (i = 0; i < nr; i++) {
325 struct page *page = pvec->pages[i];
326
327 if (page) {
328 lock_page(page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700329 if (!__putback_lru_fast_prepare(page, &pvec_putback,
330 &pgrescued)) {
Vlastimil Babka5b409982013-09-11 14:22:33 -0700331 /*
332 * Slow path. We don't want to lose the last
333 * pin before unlock_page()
334 */
335 get_page(page); /* for putback_lru_page() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700336 __munlock_isolated_page(page);
337 unlock_page(page);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700338 put_page(page); /* from follow_page_mask() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700339 }
Vlastimil Babka72255222013-09-11 14:22:29 -0700340 }
341 }
Vlastimil Babka56afe472013-09-11 14:22:32 -0700342
Vlastimil Babka5b409982013-09-11 14:22:33 -0700343 /*
344 * Phase 3: page putback for pages that qualified for the fast path
345 * This will also call put_page() to return pin from follow_page_mask()
346 */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700347 if (pagevec_count(&pvec_putback))
348 __putback_lru_fast(&pvec_putback, pgrescued);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700349}
Vlastimil Babka56afe472013-09-11 14:22:32 -0700350
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700351/*
352 * Fill up pagevec for __munlock_pagevec using pte walk
353 *
354 * The function expects that the struct page corresponding to @start address is
355 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
356 *
357 * The rest of @pvec is filled by subsequent pages within the same pmd and same
358 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
359 * pages also get pinned.
360 *
361 * Returns the address of the next page that should be scanned. This equals
362 * @start + PAGE_SIZE when no page could be added by the pte walk.
363 */
364static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
365 struct vm_area_struct *vma, int zoneid, unsigned long start,
366 unsigned long end)
367{
368 pte_t *pte;
369 spinlock_t *ptl;
370
371 /*
372 * Initialize pte walk starting at the already pinned page where we
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700373 * are sure that there is a pte, as it was pinned under the same
374 * mmap_sem write op.
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700375 */
376 pte = get_locked_pte(vma->vm_mm, start, &ptl);
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700377 /* Make sure we do not cross the page table boundary */
378 end = pgd_addr_end(start, end);
379 end = pud_addr_end(start, end);
380 end = pmd_addr_end(start, end);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700381
382 /* The page next to the pinned page is the first we will try to get */
383 start += PAGE_SIZE;
384 while (start < end) {
385 struct page *page = NULL;
386 pte++;
387 if (pte_present(*pte))
388 page = vm_normal_page(vma, start, *pte);
389 /*
390 * Break if page could not be obtained or the page's node+zone does not
391 * match
392 */
393 if (!page || page_zone_id(page) != zoneid)
394 break;
395
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800396 /*
397 * Do not use pagevec for PTE-mapped THP,
398 * munlock_vma_pages_range() will handle them.
399 */
400 if (PageTransCompound(page))
401 break;
402
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700403 get_page(page);
404 /*
405 * Increase the address that will be returned *before* the
406 * eventual break due to pvec becoming full by adding the page
407 */
408 start += PAGE_SIZE;
409 if (pagevec_add(pvec, page) == 0)
410 break;
411 }
412 pte_unmap_unlock(pte, ptl);
413 return start;
Vlastimil Babka72255222013-09-11 14:22:29 -0700414}
415
416/*
Rik van Rielba470de2008-10-18 20:26:50 -0700417 * munlock_vma_pages_range() - munlock all pages in the vma range.'
418 * @vma - vma containing range to be munlock()ed.
419 * @start - start address in @vma of the range
420 * @end - end of range in @vma.
421 *
422 * For mremap(), munmap() and exit().
423 *
424 * Called with @vma VM_LOCKED.
425 *
426 * Returns with VM_LOCKED cleared. Callers must be prepared to
427 * deal with this.
428 *
429 * We don't save and restore VM_LOCKED here because pages are
430 * still on lru. In unmap path, pages might be scanned by reclaim
431 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
432 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700433 */
Rik van Rielba470de2008-10-18 20:26:50 -0700434void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700435 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700436{
Eric B Munsonde60f5f2015-11-05 18:51:36 -0800437 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700438
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800439 while (start < end) {
Alexey Klimovab7a5af2016-01-14 15:18:24 -0800440 struct page *page;
Vlastimil Babkac424be12014-01-02 12:58:43 -0800441 unsigned int page_mask;
442 unsigned long page_increm;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700443 struct pagevec pvec;
444 struct zone *zone;
445 int zoneid;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800446
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700447 pagevec_init(&pvec, 0);
Hugh Dickins6e919712009-09-21 17:03:32 -0700448 /*
449 * Although FOLL_DUMP is intended for get_dump_page(),
450 * it just so happens that its special treatment of the
451 * ZERO_PAGE (returning an error instead of doing get_page)
452 * suits munlock very well (and if somehow an abnormal page
453 * has sneaked into the range, we won't oops here: great).
454 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800455 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700456 &page_mask);
457
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800458 if (page && !IS_ERR(page)) {
459 if (PageTransTail(page)) {
460 VM_BUG_ON_PAGE(PageMlocked(page), page);
461 put_page(page); /* follow_page_mask() */
462 } else if (PageTransHuge(page)) {
463 lock_page(page);
464 /*
465 * Any THP page found by follow_page_mask() may
466 * have gotten split before reaching
467 * munlock_vma_page(), so we need to recompute
468 * the page_mask here.
469 */
470 page_mask = munlock_vma_page(page);
471 unlock_page(page);
472 put_page(page); /* follow_page_mask() */
473 } else {
474 /*
475 * Non-huge pages are handled in batches via
476 * pagevec. The pin from follow_page_mask()
477 * prevents them from collapsing by THP.
478 */
479 pagevec_add(&pvec, page);
480 zone = page_zone(page);
481 zoneid = page_zone_id(page);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700482
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800483 /*
484 * Try to fill the rest of pagevec using fast
485 * pte walk. This will also update start to
486 * the next page to process. Then munlock the
487 * pagevec.
488 */
489 start = __munlock_pagevec_fill(&pvec, vma,
490 zoneid, start, end);
491 __munlock_pagevec(&pvec, zone);
492 goto next;
493 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700494 }
Vlastimil Babkac424be12014-01-02 12:58:43 -0800495 page_increm = 1 + page_mask;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800496 start += page_increm * PAGE_SIZE;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700497next:
Hugh Dickins408e82b2009-09-21 17:03:23 -0700498 cond_resched();
499 }
Nick Pigginb291f002008-10-18 20:26:44 -0700500}
501
502/*
503 * mlock_fixup - handle mlock[all]/munlock[all] requests.
504 *
505 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
506 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800507 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700508 *
509 * For vmas that pass the filters, merge/split as appropriate.
510 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900512 unsigned long start, unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
Nick Pigginb291f002008-10-18 20:26:44 -0700514 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700516 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 int ret = 0;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900518 int lock = !!(newflags & VM_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Michel Lespinassefed067d2011-01-13 15:46:10 -0800520 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400521 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800522 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
523 goto out;
Nick Pigginb291f002008-10-18 20:26:44 -0700524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
526 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700527 vma->vm_file, pgoff, vma_policy(vma),
528 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 if (*prev) {
530 vma = *prev;
531 goto success;
532 }
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (start != vma->vm_start) {
535 ret = split_vma(mm, vma, start, 1);
536 if (ret)
537 goto out;
538 }
539
540 if (end != vma->vm_end) {
541 ret = split_vma(mm, vma, end, 0);
542 if (ret)
543 goto out;
544 }
545
546success:
547 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700548 * Keep track of amount of locked VM.
549 */
550 nr_pages = (end - start) >> PAGE_SHIFT;
551 if (!lock)
552 nr_pages = -nr_pages;
553 mm->locked_vm += nr_pages;
554
555 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 * vm_flags is protected by the mmap_sem held in write mode.
557 * It's okay if try_to_unmap_one unmaps a page just after we
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700558 * set VM_LOCKED, populate_vma_page_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Michel Lespinassefed067d2011-01-13 15:46:10 -0800561 if (lock)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700562 vma->vm_flags = newflags;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800563 else
Hugh Dickins408e82b2009-09-21 17:03:23 -0700564 munlock_vma_pages_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566out:
Nick Pigginb291f002008-10-18 20:26:44 -0700567 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 return ret;
569}
570
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800571static int apply_vma_lock_flags(unsigned long start, size_t len,
572 vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
574 unsigned long nstart, end, tmp;
575 struct vm_area_struct * vma, * prev;
576 int error;
577
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800578 VM_BUG_ON(offset_in_page(start));
Michel Lespinassefed067d2011-01-13 15:46:10 -0800579 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 end = start + len;
581 if (end < start)
582 return -EINVAL;
583 if (end == start)
584 return 0;
Linus Torvalds097d5912012-03-06 18:23:36 -0800585 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (!vma || vma->vm_start > start)
587 return -ENOMEM;
588
Linus Torvalds097d5912012-03-06 18:23:36 -0800589 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 if (start > vma->vm_start)
591 prev = vma;
592
593 for (nstart = start ; ; ) {
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800594 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800595
596 newflags |= flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 tmp = vma->vm_end;
600 if (tmp > end)
601 tmp = end;
602 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
603 if (error)
604 break;
605 nstart = tmp;
606 if (nstart < prev->vm_end)
607 nstart = prev->vm_end;
608 if (nstart >= end)
609 break;
610
611 vma = prev->vm_next;
612 if (!vma || vma->vm_start != nstart) {
613 error = -ENOMEM;
614 break;
615 }
616 }
617 return error;
618}
619
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700620static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
622 unsigned long locked;
623 unsigned long lock_limit;
624 int error = -ENOMEM;
625
626 if (!can_do_mlock())
627 return -EPERM;
628
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800629 lru_add_drain_all(); /* flush pagevec */
630
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800631 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 start &= PAGE_MASK;
633
Jiri Slaby59e99e52010-03-05 13:41:44 -0800634 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 lock_limit >>= PAGE_SHIFT;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800636 locked = len >> PAGE_SHIFT;
637
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700638 if (down_write_killable(&current->mm->mmap_sem))
639 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800640
641 locked += current->mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 /* check against resource limits */
644 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800645 error = apply_vma_lock_flags(start, len, flags);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 up_write(&current->mm->mmap_sem);
Kirill A. Shutemovc5612592015-04-14 15:44:42 -0700648 if (error)
649 return error;
650
651 error = __mm_populate(start, len, 0);
652 if (error)
653 return __mlock_posix_error_return(error);
654 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655}
656
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800657SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
658{
659 return do_mlock(start, len, VM_LOCKED);
660}
661
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800662SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
663{
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800664 vm_flags_t vm_flags = VM_LOCKED;
665
666 if (flags & ~MLOCK_ONFAULT)
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800667 return -EINVAL;
668
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800669 if (flags & MLOCK_ONFAULT)
670 vm_flags |= VM_LOCKONFAULT;
671
672 return do_mlock(start, len, vm_flags);
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800673}
674
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100675SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
677 int ret;
678
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800679 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 start &= PAGE_MASK;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800681
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700682 if (down_write_killable(&current->mm->mmap_sem))
683 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800684 ret = apply_vma_lock_flags(start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 up_write(&current->mm->mmap_sem);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 return ret;
688}
689
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800690/*
691 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
692 * and translate into the appropriate modifications to mm->def_flags and/or the
693 * flags for all current VMAs.
694 *
695 * There are a couple of subtleties with this. If mlockall() is called multiple
696 * times with different flags, the values do not necessarily stack. If mlockall
697 * is called once including the MCL_FUTURE flag and then a second time without
698 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
699 */
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800700static int apply_mlockall_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 struct vm_area_struct * vma, * prev = NULL;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800703 vm_flags_t to_add = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800705 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
706 if (flags & MCL_FUTURE) {
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700707 current->mm->def_flags |= VM_LOCKED;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800708
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800709 if (flags & MCL_ONFAULT)
710 current->mm->def_flags |= VM_LOCKONFAULT;
711
712 if (!(flags & MCL_CURRENT))
713 goto out;
714 }
715
716 if (flags & MCL_CURRENT) {
717 to_add |= VM_LOCKED;
718 if (flags & MCL_ONFAULT)
719 to_add |= VM_LOCKONFAULT;
720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900723 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800725 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
726 newflags |= to_add;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 /* Ignore errors */
729 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700730 cond_resched_rcu_qs();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732out:
733 return 0;
734}
735
Heiko Carstens3480b252009-01-14 14:14:16 +0100736SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737{
738 unsigned long lock_limit;
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800739 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800741 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800742 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (!can_do_mlock())
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800745 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Christoph Lameterdf9d6982011-10-31 17:09:35 -0700747 if (flags & MCL_CURRENT)
748 lru_add_drain_all(); /* flush pagevec */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800749
Jiri Slaby59e99e52010-03-05 13:41:44 -0800750 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 lock_limit >>= PAGE_SHIFT;
752
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700753 if (down_write_killable(&current->mm->mmap_sem))
754 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800755
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700756 ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
758 capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800759 ret = apply_mlockall_flags(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800761 if (!ret && (flags & MCL_CURRENT))
762 mm_populate(0, TASK_SIZE);
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 return ret;
765}
766
Heiko Carstens3480b252009-01-14 14:14:16 +0100767SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
769 int ret;
770
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700771 if (down_write_killable(&current->mm->mmap_sem))
772 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800773 ret = apply_mlockall_flags(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 up_write(&current->mm->mmap_sem);
775 return ret;
776}
777
778/*
779 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
780 * shm segments) get accounted against the user_struct instead.
781 */
782static DEFINE_SPINLOCK(shmlock_user_lock);
783
784int user_shm_lock(size_t size, struct user_struct *user)
785{
786 unsigned long lock_limit, locked;
787 int allowed = 0;
788
789 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800790 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700791 if (lock_limit == RLIM_INFINITY)
792 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 lock_limit >>= PAGE_SHIFT;
794 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700795 if (!allowed &&
796 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 goto out;
798 get_uid(user);
799 user->locked_shm += locked;
800 allowed = 1;
801out:
802 spin_unlock(&shmlock_user_lock);
803 return allowed;
804}
805
806void user_shm_unlock(size_t size, struct user_struct *user)
807{
808 spin_lock(&shmlock_user_lock);
809 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
810 spin_unlock(&shmlock_user_lock);
811 free_uid(user);
812}