blob: cdbed8aaa4268c94abb44a8adf6c93873d37856a [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
Kirill A. Shutemov655548b2016-11-30 15:54:05 -0800193 if (!TestClearPageMlocked(page)) {
194 /* Potentially, PTE-mapped THP: do not skip the rest PTEs */
195 nr_pages = 1;
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800196 goto unlock_out;
Kirill A. Shutemov655548b2016-11-30 15:54:05 -0800197 }
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800198
Kirill A. Shutemov655548b2016-11-30 15:54:05 -0800199 nr_pages = hpage_nr_pages(page);
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800200 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
201
202 if (__munlock_isolate_lru_page(page, true)) {
Mel Gormana52633d2016-07-28 15:45:28 -0700203 spin_unlock_irq(zone_lru_lock(zone));
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800204 __munlock_isolated_page(page);
205 goto out;
206 }
207 __munlock_isolation_failed(page);
208
209unlock_out:
Mel Gormana52633d2016-07-28 15:45:28 -0700210 spin_unlock_irq(zone_lru_lock(zone));
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800211
212out:
Vlastimil Babkac424be12014-01-02 12:58:43 -0800213 return nr_pages - 1;
Nick Pigginb291f002008-10-18 20:26:44 -0700214}
215
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700216/*
217 * convert get_user_pages() return value to posix mlock() error
218 */
219static int __mlock_posix_error_return(long retval)
220{
221 if (retval == -EFAULT)
222 retval = -ENOMEM;
223 else if (retval == -ENOMEM)
224 retval = -EAGAIN;
225 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700226}
227
Nick Pigginb291f002008-10-18 20:26:44 -0700228/*
Vlastimil Babka56afe472013-09-11 14:22:32 -0700229 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
230 *
231 * The fast path is available only for evictable pages with single mapping.
232 * Then we can bypass the per-cpu pvec and get better performance.
233 * when mapcount > 1 we need try_to_munlock() which can fail.
234 * when !page_evictable(), we need the full redo logic of putback_lru_page to
235 * avoid leaving evictable page in unevictable list.
236 *
237 * In case of success, @page is added to @pvec and @pgrescued is incremented
238 * in case that the page was previously unevictable. @page is also unlocked.
239 */
240static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
241 int *pgrescued)
242{
Sasha Levin309381fea2014-01-23 15:52:54 -0800243 VM_BUG_ON_PAGE(PageLRU(page), page);
244 VM_BUG_ON_PAGE(!PageLocked(page), page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700245
246 if (page_mapcount(page) <= 1 && page_evictable(page)) {
247 pagevec_add(pvec, page);
248 if (TestClearPageUnevictable(page))
249 (*pgrescued)++;
250 unlock_page(page);
251 return true;
252 }
253
254 return false;
255}
256
257/*
258 * Putback multiple evictable pages to the LRU
259 *
260 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
261 * the pages might have meanwhile become unevictable but that is OK.
262 */
263static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
264{
265 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
266 /*
267 *__pagevec_lru_add() calls release_pages() so we don't call
268 * put_page() explicitly
269 */
270 __pagevec_lru_add(pvec);
271 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
272}
273
274/*
Vlastimil Babka72255222013-09-11 14:22:29 -0700275 * Munlock a batch of pages from the same zone
276 *
277 * The work is split to two main phases. First phase clears the Mlocked flag
278 * and attempts to isolate the pages, all under a single zone lru lock.
279 * The second phase finishes the munlock only for pages where isolation
280 * succeeded.
281 *
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700282 * Note that the pagevec may be modified during the process.
Vlastimil Babka72255222013-09-11 14:22:29 -0700283 */
284static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
285{
286 int i;
287 int nr = pagevec_count(pvec);
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800288 int delta_munlocked;
Vlastimil Babka56afe472013-09-11 14:22:32 -0700289 struct pagevec pvec_putback;
290 int pgrescued = 0;
Vlastimil Babka72255222013-09-11 14:22:29 -0700291
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800292 pagevec_init(&pvec_putback, 0);
293
Vlastimil Babka72255222013-09-11 14:22:29 -0700294 /* Phase 1: page isolation */
Mel Gormana52633d2016-07-28 15:45:28 -0700295 spin_lock_irq(zone_lru_lock(zone));
Vlastimil Babka72255222013-09-11 14:22:29 -0700296 for (i = 0; i < nr; i++) {
297 struct page *page = pvec->pages[i];
298
299 if (TestClearPageMlocked(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700300 /*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800301 * We already have pin from follow_page_mask()
302 * so we can spare the get_page() here.
Vlastimil Babka72255222013-09-11 14:22:29 -0700303 */
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800304 if (__munlock_isolate_lru_page(page, false))
305 continue;
306 else
307 __munlock_isolation_failed(page);
Vlastimil Babka72255222013-09-11 14:22:29 -0700308 }
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800309
310 /*
311 * We won't be munlocking this page in the next phase
312 * but we still need to release the follow_page_mask()
313 * pin. We cannot do it under lru_lock however. If it's
314 * the last pin, __page_cache_release() would deadlock.
315 */
316 pagevec_add(&pvec_putback, pvec->pages[i]);
317 pvec->pages[i] = NULL;
Vlastimil Babka72255222013-09-11 14:22:29 -0700318 }
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800319 delta_munlocked = -nr + pagevec_count(&pvec_putback);
Vlastimil Babka1ebb7cc2013-09-11 14:22:30 -0700320 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
Mel Gormana52633d2016-07-28 15:45:28 -0700321 spin_unlock_irq(zone_lru_lock(zone));
Vlastimil Babka72255222013-09-11 14:22:29 -0700322
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800323 /* Now we can release pins of pages that we are not munlocking */
324 pagevec_release(&pvec_putback);
325
Vlastimil Babka56afe472013-09-11 14:22:32 -0700326 /* Phase 2: page munlock */
Vlastimil Babka72255222013-09-11 14:22:29 -0700327 for (i = 0; i < nr; i++) {
328 struct page *page = pvec->pages[i];
329
330 if (page) {
331 lock_page(page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700332 if (!__putback_lru_fast_prepare(page, &pvec_putback,
333 &pgrescued)) {
Vlastimil Babka5b409982013-09-11 14:22:33 -0700334 /*
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() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700339 __munlock_isolated_page(page);
340 unlock_page(page);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700341 put_page(page); /* from follow_page_mask() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700342 }
Vlastimil Babka72255222013-09-11 14:22:29 -0700343 }
344 }
Vlastimil Babka56afe472013-09-11 14:22:32 -0700345
Vlastimil Babka5b409982013-09-11 14:22:33 -0700346 /*
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 */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700350 if (pagevec_count(&pvec_putback))
351 __putback_lru_fast(&pvec_putback, pgrescued);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700352}
Vlastimil Babka56afe472013-09-11 14:22:32 -0700353
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700354/*
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 */
367static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
368 struct vm_area_struct *vma, int zoneid, unsigned long start,
369 unsigned long end)
370{
371 pte_t *pte;
372 spinlock_t *ptl;
373
374 /*
375 * Initialize pte walk starting at the already pinned page where we
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700376 * are sure that there is a pte, as it was pinned under the same
377 * mmap_sem write op.
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700378 */
379 pte = get_locked_pte(vma->vm_mm, start, &ptl);
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700380 /* Make sure we do not cross the page table boundary */
381 end = pgd_addr_end(start, end);
382 end = pud_addr_end(start, end);
383 end = pmd_addr_end(start, end);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700384
385 /* The page next to the pinned page is the first we will try to get */
386 start += PAGE_SIZE;
387 while (start < end) {
388 struct page *page = NULL;
389 pte++;
390 if (pte_present(*pte))
391 page = vm_normal_page(vma, start, *pte);
392 /*
393 * Break if page could not be obtained or the page's node+zone does not
394 * match
395 */
396 if (!page || page_zone_id(page) != zoneid)
397 break;
398
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800399 /*
400 * Do not use pagevec for PTE-mapped THP,
401 * munlock_vma_pages_range() will handle them.
402 */
403 if (PageTransCompound(page))
404 break;
405
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700406 get_page(page);
407 /*
408 * Increase the address that will be returned *before* the
409 * eventual break due to pvec becoming full by adding the page
410 */
411 start += PAGE_SIZE;
412 if (pagevec_add(pvec, page) == 0)
413 break;
414 }
415 pte_unmap_unlock(pte, ptl);
416 return start;
Vlastimil Babka72255222013-09-11 14:22:29 -0700417}
418
419/*
Rik van Rielba470de2008-10-18 20:26:50 -0700420 * munlock_vma_pages_range() - munlock all pages in the vma range.'
421 * @vma - vma containing range to be munlock()ed.
422 * @start - start address in @vma of the range
423 * @end - end of range in @vma.
424 *
425 * For mremap(), munmap() and exit().
426 *
427 * Called with @vma VM_LOCKED.
428 *
429 * Returns with VM_LOCKED cleared. Callers must be prepared to
430 * deal with this.
431 *
432 * We don't save and restore VM_LOCKED here because pages are
433 * still on lru. In unmap path, pages might be scanned by reclaim
434 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
435 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700436 */
Rik van Rielba470de2008-10-18 20:26:50 -0700437void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700438 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700439{
Eric B Munsonde60f5f2015-11-05 18:51:36 -0800440 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700441
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800442 while (start < end) {
Alexey Klimovab7a5af2016-01-14 15:18:24 -0800443 struct page *page;
Vlastimil Babkac424be12014-01-02 12:58:43 -0800444 unsigned int page_mask;
445 unsigned long page_increm;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700446 struct pagevec pvec;
447 struct zone *zone;
448 int zoneid;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800449
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700450 pagevec_init(&pvec, 0);
Hugh Dickins6e919712009-09-21 17:03:32 -0700451 /*
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 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800458 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700459 &page_mask);
460
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800461 if (page && !IS_ERR(page)) {
462 if (PageTransTail(page)) {
463 VM_BUG_ON_PAGE(PageMlocked(page), page);
464 put_page(page); /* follow_page_mask() */
465 } else if (PageTransHuge(page)) {
466 lock_page(page);
467 /*
468 * Any THP page found by follow_page_mask() may
469 * have gotten split before reaching
470 * munlock_vma_page(), so we need to recompute
471 * the page_mask here.
472 */
473 page_mask = munlock_vma_page(page);
474 unlock_page(page);
475 put_page(page); /* follow_page_mask() */
476 } else {
477 /*
478 * Non-huge pages are handled in batches via
479 * pagevec. The pin from follow_page_mask()
480 * prevents them from collapsing by THP.
481 */
482 pagevec_add(&pvec, page);
483 zone = page_zone(page);
484 zoneid = page_zone_id(page);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700485
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800486 /*
487 * Try to fill the rest of pagevec using fast
488 * pte walk. This will also update start to
489 * the next page to process. Then munlock the
490 * pagevec.
491 */
492 start = __munlock_pagevec_fill(&pvec, vma,
493 zoneid, start, end);
494 __munlock_pagevec(&pvec, zone);
495 goto next;
496 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700497 }
Vlastimil Babkac424be12014-01-02 12:58:43 -0800498 page_increm = 1 + page_mask;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800499 start += page_increm * PAGE_SIZE;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700500next:
Hugh Dickins408e82b2009-09-21 17:03:23 -0700501 cond_resched();
502 }
Nick Pigginb291f002008-10-18 20:26:44 -0700503}
504
505/*
506 * mlock_fixup - handle mlock[all]/munlock[all] requests.
507 *
508 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
509 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800510 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700511 *
512 * For vmas that pass the filters, merge/split as appropriate.
513 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900515 unsigned long start, unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Nick Pigginb291f002008-10-18 20:26:44 -0700517 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700519 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 int ret = 0;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900521 int lock = !!(newflags & VM_LOCKED);
Simon Guob155b4f2016-10-07 16:59:40 -0700522 vm_flags_t old_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Michel Lespinassefed067d2011-01-13 15:46:10 -0800524 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400525 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800526 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
527 goto out;
Nick Pigginb291f002008-10-18 20:26:44 -0700528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
530 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700531 vma->vm_file, pgoff, vma_policy(vma),
532 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 if (*prev) {
534 vma = *prev;
535 goto success;
536 }
537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 if (start != vma->vm_start) {
539 ret = split_vma(mm, vma, start, 1);
540 if (ret)
541 goto out;
542 }
543
544 if (end != vma->vm_end) {
545 ret = split_vma(mm, vma, end, 0);
546 if (ret)
547 goto out;
548 }
549
550success:
551 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700552 * Keep track of amount of locked VM.
553 */
554 nr_pages = (end - start) >> PAGE_SHIFT;
555 if (!lock)
556 nr_pages = -nr_pages;
Simon Guob155b4f2016-10-07 16:59:40 -0700557 else if (old_flags & VM_LOCKED)
558 nr_pages = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700559 mm->locked_vm += nr_pages;
560
561 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 * vm_flags is protected by the mmap_sem held in write mode.
563 * It's okay if try_to_unmap_one unmaps a page just after we
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700564 * set VM_LOCKED, populate_vma_page_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Michel Lespinassefed067d2011-01-13 15:46:10 -0800567 if (lock)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700568 vma->vm_flags = newflags;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800569 else
Hugh Dickins408e82b2009-09-21 17:03:23 -0700570 munlock_vma_pages_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572out:
Nick Pigginb291f002008-10-18 20:26:44 -0700573 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 return ret;
575}
576
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800577static int apply_vma_lock_flags(unsigned long start, size_t len,
578 vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
580 unsigned long nstart, end, tmp;
581 struct vm_area_struct * vma, * prev;
582 int error;
583
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800584 VM_BUG_ON(offset_in_page(start));
Michel Lespinassefed067d2011-01-13 15:46:10 -0800585 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 end = start + len;
587 if (end < start)
588 return -EINVAL;
589 if (end == start)
590 return 0;
Linus Torvalds097d5912012-03-06 18:23:36 -0800591 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (!vma || vma->vm_start > start)
593 return -ENOMEM;
594
Linus Torvalds097d5912012-03-06 18:23:36 -0800595 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 if (start > vma->vm_start)
597 prev = vma;
598
599 for (nstart = start ; ; ) {
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800600 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800601
602 newflags |= flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 tmp = vma->vm_end;
606 if (tmp > end)
607 tmp = end;
608 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
609 if (error)
610 break;
611 nstart = tmp;
612 if (nstart < prev->vm_end)
613 nstart = prev->vm_end;
614 if (nstart >= end)
615 break;
616
617 vma = prev->vm_next;
618 if (!vma || vma->vm_start != nstart) {
619 error = -ENOMEM;
620 break;
621 }
622 }
623 return error;
624}
625
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700626/*
627 * Go through vma areas and sum size of mlocked
628 * vma pages, as return value.
629 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
630 * is also counted.
631 * Return value: previously mlocked page counts
632 */
633static int count_mm_mlocked_page_nr(struct mm_struct *mm,
634 unsigned long start, size_t len)
635{
636 struct vm_area_struct *vma;
637 int count = 0;
638
639 if (mm == NULL)
640 mm = current->mm;
641
642 vma = find_vma(mm, start);
643 if (vma == NULL)
644 vma = mm->mmap;
645
646 for (; vma ; vma = vma->vm_next) {
647 if (start >= vma->vm_end)
648 continue;
649 if (start + len <= vma->vm_start)
650 break;
651 if (vma->vm_flags & VM_LOCKED) {
652 if (start > vma->vm_start)
653 count -= (start - vma->vm_start);
654 if (start + len < vma->vm_end) {
655 count += start + len - vma->vm_start;
656 break;
657 }
658 count += vma->vm_end - vma->vm_start;
659 }
660 }
661
662 return count >> PAGE_SHIFT;
663}
664
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700665static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 unsigned long locked;
668 unsigned long lock_limit;
669 int error = -ENOMEM;
670
671 if (!can_do_mlock())
672 return -EPERM;
673
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800674 lru_add_drain_all(); /* flush pagevec */
675
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800676 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 start &= PAGE_MASK;
678
Jiri Slaby59e99e52010-03-05 13:41:44 -0800679 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 lock_limit >>= PAGE_SHIFT;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800681 locked = len >> PAGE_SHIFT;
682
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700683 if (down_write_killable(&current->mm->mmap_sem))
684 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800685
686 locked += current->mm->locked_vm;
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700687 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
688 /*
689 * It is possible that the regions requested intersect with
690 * previously mlocked areas, that part area in "mm->locked_vm"
691 * should not be counted to new mlock increment count. So check
692 * and adjust locked count if necessary.
693 */
694 locked -= count_mm_mlocked_page_nr(current->mm,
695 start, len);
696 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 /* check against resource limits */
699 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800700 error = apply_vma_lock_flags(start, len, flags);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 up_write(&current->mm->mmap_sem);
Kirill A. Shutemovc5612592015-04-14 15:44:42 -0700703 if (error)
704 return error;
705
706 error = __mm_populate(start, len, 0);
707 if (error)
708 return __mlock_posix_error_return(error);
709 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800712SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
713{
714 return do_mlock(start, len, VM_LOCKED);
715}
716
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800717SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
718{
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800719 vm_flags_t vm_flags = VM_LOCKED;
720
721 if (flags & ~MLOCK_ONFAULT)
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800722 return -EINVAL;
723
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800724 if (flags & MLOCK_ONFAULT)
725 vm_flags |= VM_LOCKONFAULT;
726
727 return do_mlock(start, len, vm_flags);
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800728}
729
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100730SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
732 int ret;
733
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800734 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 start &= PAGE_MASK;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800736
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700737 if (down_write_killable(&current->mm->mmap_sem))
738 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800739 ret = apply_vma_lock_flags(start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 up_write(&current->mm->mmap_sem);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 return ret;
743}
744
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800745/*
746 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
747 * and translate into the appropriate modifications to mm->def_flags and/or the
748 * flags for all current VMAs.
749 *
750 * There are a couple of subtleties with this. If mlockall() is called multiple
751 * times with different flags, the values do not necessarily stack. If mlockall
752 * is called once including the MCL_FUTURE flag and then a second time without
753 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
754 */
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800755static int apply_mlockall_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
757 struct vm_area_struct * vma, * prev = NULL;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800758 vm_flags_t to_add = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800760 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
761 if (flags & MCL_FUTURE) {
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700762 current->mm->def_flags |= VM_LOCKED;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800763
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800764 if (flags & MCL_ONFAULT)
765 current->mm->def_flags |= VM_LOCKONFAULT;
766
767 if (!(flags & MCL_CURRENT))
768 goto out;
769 }
770
771 if (flags & MCL_CURRENT) {
772 to_add |= VM_LOCKED;
773 if (flags & MCL_ONFAULT)
774 to_add |= VM_LOCKONFAULT;
775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900778 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800780 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
781 newflags |= to_add;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 /* Ignore errors */
784 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700785 cond_resched_rcu_qs();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
787out:
788 return 0;
789}
790
Heiko Carstens3480b252009-01-14 14:14:16 +0100791SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
793 unsigned long lock_limit;
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800794 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800796 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800797 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 if (!can_do_mlock())
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800800 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Christoph Lameterdf9d6982011-10-31 17:09:35 -0700802 if (flags & MCL_CURRENT)
803 lru_add_drain_all(); /* flush pagevec */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800804
Jiri Slaby59e99e52010-03-05 13:41:44 -0800805 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 lock_limit >>= PAGE_SHIFT;
807
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700808 if (down_write_killable(&current->mm->mmap_sem))
809 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800810
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700811 ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
813 capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800814 ret = apply_mlockall_flags(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800816 if (!ret && (flags & MCL_CURRENT))
817 mm_populate(0, TASK_SIZE);
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800818
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return ret;
820}
821
Heiko Carstens3480b252009-01-14 14:14:16 +0100822SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
824 int ret;
825
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700826 if (down_write_killable(&current->mm->mmap_sem))
827 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800828 ret = apply_mlockall_flags(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 up_write(&current->mm->mmap_sem);
830 return ret;
831}
832
833/*
834 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
835 * shm segments) get accounted against the user_struct instead.
836 */
837static DEFINE_SPINLOCK(shmlock_user_lock);
838
839int user_shm_lock(size_t size, struct user_struct *user)
840{
841 unsigned long lock_limit, locked;
842 int allowed = 0;
843
844 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800845 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700846 if (lock_limit == RLIM_INFINITY)
847 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 lock_limit >>= PAGE_SHIFT;
849 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700850 if (!allowed &&
851 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 goto out;
853 get_uid(user);
854 user->locked_shm += locked;
855 allowed = 1;
856out:
857 spin_unlock(&shmlock_user_lock);
858 return allowed;
859}
860
861void user_shm_unlock(size_t size, struct user_struct *user)
862{
863 spin_lock(&shmlock_user_lock);
864 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
865 spin_unlock(&shmlock_user_lock);
866 free_uid(user);
867}