blob: 6fd2cf15e8687d32114f8e81516aed0a52997147 [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
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040027int can_do_mlock(void)
28{
Jiri Slaby59e99e52010-03-05 13:41:44 -080029 if (rlimit(RLIMIT_MEMLOCK) != 0)
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040030 return 1;
Jeff Vander Stoepa5a65792015-03-12 16:26:17 -070031 if (capable(CAP_IPC_LOCK))
32 return 1;
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040033 return 0;
34}
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
Nick Piggin5344b7e2008-10-18 20:26:51 -070085 if (!TestSetPageMlocked(page)) {
David Rientjes8449d212012-10-08 16:34:06 -070086 mod_zone_page_state(page_zone(page), NR_MLOCK,
87 hpage_nr_pages(page));
Nick Piggin5344b7e2008-10-18 20:26:51 -070088 count_vm_event(UNEVICTABLE_PGMLOCKED);
89 if (!isolate_lru_page(page))
90 putback_lru_page(page);
91 }
Nick Pigginb291f002008-10-18 20:26:44 -070092}
93
Vlastimil Babka72255222013-09-11 14:22:29 -070094/*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -080095 * Isolate a page from LRU with optional get_page() pin.
96 * Assumes lru_lock already held and page already pinned.
97 */
98static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
99{
100 if (PageLRU(page)) {
101 struct lruvec *lruvec;
102
103 lruvec = mem_cgroup_page_lruvec(page, page_zone(page));
104 if (getpage)
105 get_page(page);
106 ClearPageLRU(page);
107 del_page_from_lru_list(page, lruvec, page_lru(page));
108 return true;
109 }
110
111 return false;
112}
113
114/*
Vlastimil Babka72255222013-09-11 14:22:29 -0700115 * Finish munlock after successful page isolation
116 *
117 * Page must be locked. This is a wrapper for try_to_munlock()
118 * and putback_lru_page() with munlock accounting.
119 */
120static void __munlock_isolated_page(struct page *page)
121{
122 int ret = SWAP_AGAIN;
123
124 /*
125 * Optimization: if the page was mapped just once, that's our mapping
126 * and we don't need to check all the other vmas.
127 */
128 if (page_mapcount(page) > 1)
129 ret = try_to_munlock(page);
130
131 /* Did try_to_unlock() succeed or punt? */
132 if (ret != SWAP_MLOCK)
133 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
134
135 putback_lru_page(page);
136}
137
138/*
139 * Accounting for page isolation fail during munlock
140 *
141 * Performs accounting when page isolation fails in munlock. There is nothing
142 * else to do because it means some other task has already removed the page
143 * from the LRU. putback_lru_page() will take care of removing the page from
144 * the unevictable list, if necessary. vmscan [page_referenced()] will move
145 * the page back to the unevictable list if some other vma has it mlocked.
146 */
147static void __munlock_isolation_failed(struct page *page)
148{
149 if (PageUnevictable(page))
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800150 __count_vm_event(UNEVICTABLE_PGSTRANDED);
Vlastimil Babka72255222013-09-11 14:22:29 -0700151 else
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800152 __count_vm_event(UNEVICTABLE_PGMUNLOCKED);
Vlastimil Babka72255222013-09-11 14:22:29 -0700153}
154
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800155/**
156 * munlock_vma_page - munlock a vma page
Vlastimil Babkac424be12014-01-02 12:58:43 -0800157 * @page - page to be unlocked, either a normal page or THP page head
158 *
159 * returns the size of the page as a page mask (0 for normal page,
160 * HPAGE_PMD_NR - 1 for THP head page)
Nick Pigginb291f002008-10-18 20:26:44 -0700161 *
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800162 * called from munlock()/munmap() path with page supposedly on the LRU.
163 * When we munlock a page, because the vma where we found the page is being
164 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
165 * page locked so that we can leave it on the unevictable lru list and not
166 * bother vmscan with it. However, to walk the page's rmap list in
167 * try_to_munlock() we must isolate the page from the LRU. If some other
168 * task has removed the page from the LRU, we won't be able to do that.
169 * So we clear the PageMlocked as we might not get another chance. If we
170 * can't isolate the page, we leave it for putback_lru_page() and vmscan
171 * [page_referenced()/try_to_unmap()] to deal with.
Nick Pigginb291f002008-10-18 20:26:44 -0700172 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800173unsigned int munlock_vma_page(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -0700174{
Vlastimil Babkac424be12014-01-02 12:58:43 -0800175 unsigned int nr_pages;
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800176 struct zone *zone = page_zone(page);
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800177
Vlastimil Babka57e68e92014-04-07 15:37:50 -0700178 /* For try_to_munlock() and to serialize with page migration */
Nick Pigginb291f002008-10-18 20:26:44 -0700179 BUG_ON(!PageLocked(page));
180
Vlastimil Babkac424be12014-01-02 12:58:43 -0800181 /*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800182 * Serialize with any parallel __split_huge_page_refcount() which
183 * might otherwise copy PageMlocked to part of the tail pages before
184 * we clear it in the head page. It also stabilizes hpage_nr_pages().
Vlastimil Babkac424be12014-01-02 12:58:43 -0800185 */
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800186 spin_lock_irq(&zone->lru_lock);
187
188 nr_pages = hpage_nr_pages(page);
189 if (!TestClearPageMlocked(page))
190 goto unlock_out;
191
192 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
193
194 if (__munlock_isolate_lru_page(page, true)) {
195 spin_unlock_irq(&zone->lru_lock);
196 __munlock_isolated_page(page);
197 goto out;
198 }
199 __munlock_isolation_failed(page);
200
201unlock_out:
202 spin_unlock_irq(&zone->lru_lock);
203
204out:
Vlastimil Babkac424be12014-01-02 12:58:43 -0800205 return nr_pages - 1;
Nick Pigginb291f002008-10-18 20:26:44 -0700206}
207
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700208/*
209 * convert get_user_pages() return value to posix mlock() error
210 */
211static int __mlock_posix_error_return(long retval)
212{
213 if (retval == -EFAULT)
214 retval = -ENOMEM;
215 else if (retval == -ENOMEM)
216 retval = -EAGAIN;
217 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700218}
219
Nick Pigginb291f002008-10-18 20:26:44 -0700220/*
Vlastimil Babka56afe472013-09-11 14:22:32 -0700221 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
222 *
223 * The fast path is available only for evictable pages with single mapping.
224 * Then we can bypass the per-cpu pvec and get better performance.
225 * when mapcount > 1 we need try_to_munlock() which can fail.
226 * when !page_evictable(), we need the full redo logic of putback_lru_page to
227 * avoid leaving evictable page in unevictable list.
228 *
229 * In case of success, @page is added to @pvec and @pgrescued is incremented
230 * in case that the page was previously unevictable. @page is also unlocked.
231 */
232static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
233 int *pgrescued)
234{
Sasha Levin309381fea2014-01-23 15:52:54 -0800235 VM_BUG_ON_PAGE(PageLRU(page), page);
236 VM_BUG_ON_PAGE(!PageLocked(page), page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700237
238 if (page_mapcount(page) <= 1 && page_evictable(page)) {
239 pagevec_add(pvec, page);
240 if (TestClearPageUnevictable(page))
241 (*pgrescued)++;
242 unlock_page(page);
243 return true;
244 }
245
246 return false;
247}
248
249/*
250 * Putback multiple evictable pages to the LRU
251 *
252 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
253 * the pages might have meanwhile become unevictable but that is OK.
254 */
255static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
256{
257 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
258 /*
259 *__pagevec_lru_add() calls release_pages() so we don't call
260 * put_page() explicitly
261 */
262 __pagevec_lru_add(pvec);
263 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
264}
265
266/*
Vlastimil Babka72255222013-09-11 14:22:29 -0700267 * Munlock a batch of pages from the same zone
268 *
269 * The work is split to two main phases. First phase clears the Mlocked flag
270 * and attempts to isolate the pages, all under a single zone lru lock.
271 * The second phase finishes the munlock only for pages where isolation
272 * succeeded.
273 *
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700274 * Note that the pagevec may be modified during the process.
Vlastimil Babka72255222013-09-11 14:22:29 -0700275 */
276static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
277{
278 int i;
279 int nr = pagevec_count(pvec);
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800280 int delta_munlocked;
Vlastimil Babka56afe472013-09-11 14:22:32 -0700281 struct pagevec pvec_putback;
282 int pgrescued = 0;
Vlastimil Babka72255222013-09-11 14:22:29 -0700283
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800284 pagevec_init(&pvec_putback, 0);
285
Vlastimil Babka72255222013-09-11 14:22:29 -0700286 /* Phase 1: page isolation */
287 spin_lock_irq(&zone->lru_lock);
288 for (i = 0; i < nr; i++) {
289 struct page *page = pvec->pages[i];
290
291 if (TestClearPageMlocked(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700292 /*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800293 * We already have pin from follow_page_mask()
294 * so we can spare the get_page() here.
Vlastimil Babka72255222013-09-11 14:22:29 -0700295 */
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800296 if (__munlock_isolate_lru_page(page, false))
297 continue;
298 else
299 __munlock_isolation_failed(page);
Vlastimil Babka72255222013-09-11 14:22:29 -0700300 }
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800301
302 /*
303 * We won't be munlocking this page in the next phase
304 * but we still need to release the follow_page_mask()
305 * pin. We cannot do it under lru_lock however. If it's
306 * the last pin, __page_cache_release() would deadlock.
307 */
308 pagevec_add(&pvec_putback, pvec->pages[i]);
309 pvec->pages[i] = NULL;
Vlastimil Babka72255222013-09-11 14:22:29 -0700310 }
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800311 delta_munlocked = -nr + pagevec_count(&pvec_putback);
Vlastimil Babka1ebb7cc2013-09-11 14:22:30 -0700312 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
Vlastimil Babka72255222013-09-11 14:22:29 -0700313 spin_unlock_irq(&zone->lru_lock);
314
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800315 /* Now we can release pins of pages that we are not munlocking */
316 pagevec_release(&pvec_putback);
317
Vlastimil Babka56afe472013-09-11 14:22:32 -0700318 /* Phase 2: page munlock */
Vlastimil Babka72255222013-09-11 14:22:29 -0700319 for (i = 0; i < nr; i++) {
320 struct page *page = pvec->pages[i];
321
322 if (page) {
323 lock_page(page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700324 if (!__putback_lru_fast_prepare(page, &pvec_putback,
325 &pgrescued)) {
Vlastimil Babka5b409982013-09-11 14:22:33 -0700326 /*
327 * Slow path. We don't want to lose the last
328 * pin before unlock_page()
329 */
330 get_page(page); /* for putback_lru_page() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700331 __munlock_isolated_page(page);
332 unlock_page(page);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700333 put_page(page); /* from follow_page_mask() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700334 }
Vlastimil Babka72255222013-09-11 14:22:29 -0700335 }
336 }
Vlastimil Babka56afe472013-09-11 14:22:32 -0700337
Vlastimil Babka5b409982013-09-11 14:22:33 -0700338 /*
339 * Phase 3: page putback for pages that qualified for the fast path
340 * This will also call put_page() to return pin from follow_page_mask()
341 */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700342 if (pagevec_count(&pvec_putback))
343 __putback_lru_fast(&pvec_putback, pgrescued);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700344}
Vlastimil Babka56afe472013-09-11 14:22:32 -0700345
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700346/*
347 * Fill up pagevec for __munlock_pagevec using pte walk
348 *
349 * The function expects that the struct page corresponding to @start address is
350 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
351 *
352 * The rest of @pvec is filled by subsequent pages within the same pmd and same
353 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
354 * pages also get pinned.
355 *
356 * Returns the address of the next page that should be scanned. This equals
357 * @start + PAGE_SIZE when no page could be added by the pte walk.
358 */
359static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
360 struct vm_area_struct *vma, int zoneid, unsigned long start,
361 unsigned long end)
362{
363 pte_t *pte;
364 spinlock_t *ptl;
365
366 /*
367 * Initialize pte walk starting at the already pinned page where we
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700368 * are sure that there is a pte, as it was pinned under the same
369 * mmap_sem write op.
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700370 */
371 pte = get_locked_pte(vma->vm_mm, start, &ptl);
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700372 /* Make sure we do not cross the page table boundary */
373 end = pgd_addr_end(start, end);
374 end = pud_addr_end(start, end);
375 end = pmd_addr_end(start, end);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700376
377 /* The page next to the pinned page is the first we will try to get */
378 start += PAGE_SIZE;
379 while (start < end) {
380 struct page *page = NULL;
381 pte++;
382 if (pte_present(*pte))
383 page = vm_normal_page(vma, start, *pte);
384 /*
385 * Break if page could not be obtained or the page's node+zone does not
386 * match
387 */
388 if (!page || page_zone_id(page) != zoneid)
389 break;
390
391 get_page(page);
392 /*
393 * Increase the address that will be returned *before* the
394 * eventual break due to pvec becoming full by adding the page
395 */
396 start += PAGE_SIZE;
397 if (pagevec_add(pvec, page) == 0)
398 break;
399 }
400 pte_unmap_unlock(pte, ptl);
401 return start;
Vlastimil Babka72255222013-09-11 14:22:29 -0700402}
403
404/*
Rik van Rielba470de2008-10-18 20:26:50 -0700405 * munlock_vma_pages_range() - munlock all pages in the vma range.'
406 * @vma - vma containing range to be munlock()ed.
407 * @start - start address in @vma of the range
408 * @end - end of range in @vma.
409 *
410 * For mremap(), munmap() and exit().
411 *
412 * Called with @vma VM_LOCKED.
413 *
414 * Returns with VM_LOCKED cleared. Callers must be prepared to
415 * deal with this.
416 *
417 * We don't save and restore VM_LOCKED here because pages are
418 * still on lru. In unmap path, pages might be scanned by reclaim
419 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
420 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700421 */
Rik van Rielba470de2008-10-18 20:26:50 -0700422void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700423 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700424{
425 vma->vm_flags &= ~VM_LOCKED;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700426
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800427 while (start < end) {
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700428 struct page *page = NULL;
Vlastimil Babkac424be12014-01-02 12:58:43 -0800429 unsigned int page_mask;
430 unsigned long page_increm;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700431 struct pagevec pvec;
432 struct zone *zone;
433 int zoneid;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800434
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700435 pagevec_init(&pvec, 0);
Hugh Dickins6e919712009-09-21 17:03:32 -0700436 /*
437 * Although FOLL_DUMP is intended for get_dump_page(),
438 * it just so happens that its special treatment of the
439 * ZERO_PAGE (returning an error instead of doing get_page)
440 * suits munlock very well (and if somehow an abnormal page
441 * has sneaked into the range, we won't oops here: great).
442 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800443 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700444 &page_mask);
445
Hugh Dickins6e919712009-09-21 17:03:32 -0700446 if (page && !IS_ERR(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700447 if (PageTransHuge(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700448 lock_page(page);
449 /*
450 * Any THP page found by follow_page_mask() may
451 * have gotten split before reaching
452 * munlock_vma_page(), so we need to recompute
453 * the page_mask here.
454 */
455 page_mask = munlock_vma_page(page);
456 unlock_page(page);
457 put_page(page); /* follow_page_mask() */
458 } else {
459 /*
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700460 * Non-huge pages are handled in batches via
461 * pagevec. The pin from follow_page_mask()
462 * prevents them from collapsing by THP.
Vlastimil Babka72255222013-09-11 14:22:29 -0700463 */
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700464 pagevec_add(&pvec, page);
465 zone = page_zone(page);
466 zoneid = page_zone_id(page);
467
468 /*
469 * Try to fill the rest of pagevec using fast
470 * pte walk. This will also update start to
471 * the next page to process. Then munlock the
472 * pagevec.
473 */
474 start = __munlock_pagevec_fill(&pvec, vma,
475 zoneid, start, end);
476 __munlock_pagevec(&pvec, zone);
477 goto next;
Vlastimil Babka72255222013-09-11 14:22:29 -0700478 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700479 }
Vlastimil Babkac424be12014-01-02 12:58:43 -0800480 /* It's a bug to munlock in the middle of a THP page */
481 VM_BUG_ON((start >> PAGE_SHIFT) & page_mask);
482 page_increm = 1 + page_mask;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800483 start += page_increm * PAGE_SIZE;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700484next:
Hugh Dickins408e82b2009-09-21 17:03:23 -0700485 cond_resched();
486 }
Nick Pigginb291f002008-10-18 20:26:44 -0700487}
488
489/*
490 * mlock_fixup - handle mlock[all]/munlock[all] requests.
491 *
492 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
493 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800494 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700495 *
496 * For vmas that pass the filters, merge/split as appropriate.
497 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900499 unsigned long start, unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Nick Pigginb291f002008-10-18 20:26:44 -0700501 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700503 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 int ret = 0;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900505 int lock = !!(newflags & VM_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Michel Lespinassefed067d2011-01-13 15:46:10 -0800507 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400508 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
Nick Pigginb291f002008-10-18 20:26:44 -0700509 goto out; /* don't set VM_LOCKED, don't count */
510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
512 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
513 vma->vm_file, pgoff, vma_policy(vma));
514 if (*prev) {
515 vma = *prev;
516 goto success;
517 }
518
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 if (start != vma->vm_start) {
520 ret = split_vma(mm, vma, start, 1);
521 if (ret)
522 goto out;
523 }
524
525 if (end != vma->vm_end) {
526 ret = split_vma(mm, vma, end, 0);
527 if (ret)
528 goto out;
529 }
530
531success:
532 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700533 * Keep track of amount of locked VM.
534 */
535 nr_pages = (end - start) >> PAGE_SHIFT;
536 if (!lock)
537 nr_pages = -nr_pages;
538 mm->locked_vm += nr_pages;
539
540 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 * vm_flags is protected by the mmap_sem held in write mode.
542 * It's okay if try_to_unmap_one unmaps a page just after we
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700543 * set VM_LOCKED, populate_vma_page_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Michel Lespinassefed067d2011-01-13 15:46:10 -0800546 if (lock)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700547 vma->vm_flags = newflags;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800548 else
Hugh Dickins408e82b2009-09-21 17:03:23 -0700549 munlock_vma_pages_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551out:
Nick Pigginb291f002008-10-18 20:26:44 -0700552 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 return ret;
554}
555
556static int do_mlock(unsigned long start, size_t len, int on)
557{
558 unsigned long nstart, end, tmp;
559 struct vm_area_struct * vma, * prev;
560 int error;
561
Michel Lespinassefed067d2011-01-13 15:46:10 -0800562 VM_BUG_ON(start & ~PAGE_MASK);
563 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 end = start + len;
565 if (end < start)
566 return -EINVAL;
567 if (end == start)
568 return 0;
Linus Torvalds097d5912012-03-06 18:23:36 -0800569 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 if (!vma || vma->vm_start > start)
571 return -ENOMEM;
572
Linus Torvalds097d5912012-03-06 18:23:36 -0800573 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (start > vma->vm_start)
575 prev = vma;
576
577 for (nstart = start ; ; ) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900578 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
581
Michel Lespinasse18693052013-02-22 16:32:46 -0800582 newflags = vma->vm_flags & ~VM_LOCKED;
583 if (on)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700584 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586 tmp = vma->vm_end;
587 if (tmp > end)
588 tmp = end;
589 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
590 if (error)
591 break;
592 nstart = tmp;
593 if (nstart < prev->vm_end)
594 nstart = prev->vm_end;
595 if (nstart >= end)
596 break;
597
598 vma = prev->vm_next;
599 if (!vma || vma->vm_start != nstart) {
600 error = -ENOMEM;
601 break;
602 }
603 }
604 return error;
605}
606
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100607SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
609 unsigned long locked;
610 unsigned long lock_limit;
611 int error = -ENOMEM;
612
613 if (!can_do_mlock())
614 return -EPERM;
615
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800616 lru_add_drain_all(); /* flush pagevec */
617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
619 start &= PAGE_MASK;
620
Jiri Slaby59e99e52010-03-05 13:41:44 -0800621 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 lock_limit >>= PAGE_SHIFT;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800623 locked = len >> PAGE_SHIFT;
624
625 down_write(&current->mm->mmap_sem);
626
627 locked += current->mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 /* check against resource limits */
630 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
631 error = do_mlock(start, len, 1);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 up_write(&current->mm->mmap_sem);
Kirill A. Shutemovc5612592015-04-14 15:44:42 -0700634 if (error)
635 return error;
636
637 error = __mm_populate(start, len, 0);
638 if (error)
639 return __mlock_posix_error_return(error);
640 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
642
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100643SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
645 int ret;
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
648 start &= PAGE_MASK;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800649
650 down_write(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 ret = do_mlock(start, len, 0);
652 up_write(&current->mm->mmap_sem);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 return ret;
655}
656
657static int do_mlockall(int flags)
658{
659 struct vm_area_struct * vma, * prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661 if (flags & MCL_FUTURE)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700662 current->mm->def_flags |= VM_LOCKED;
Gerald Schaefer9977f0f2013-02-12 13:46:20 -0800663 else
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700664 current->mm->def_flags &= ~VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 if (flags == MCL_FUTURE)
666 goto out;
667
668 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900669 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Michel Lespinasse18693052013-02-22 16:32:46 -0800671 newflags = vma->vm_flags & ~VM_LOCKED;
672 if (flags & MCL_CURRENT)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700673 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 /* Ignore errors */
676 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700677 cond_resched_rcu_qs();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679out:
680 return 0;
681}
682
Heiko Carstens3480b252009-01-14 14:14:16 +0100683SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
685 unsigned long lock_limit;
686 int ret = -EINVAL;
687
688 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
689 goto out;
690
691 ret = -EPERM;
692 if (!can_do_mlock())
693 goto out;
694
Christoph Lameterdf9d6982011-10-31 17:09:35 -0700695 if (flags & MCL_CURRENT)
696 lru_add_drain_all(); /* flush pagevec */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800697
Jiri Slaby59e99e52010-03-05 13:41:44 -0800698 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 lock_limit >>= PAGE_SHIFT;
700
701 ret = -ENOMEM;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800702 down_write(&current->mm->mmap_sem);
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
705 capable(CAP_IPC_LOCK))
706 ret = do_mlockall(flags);
707 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800708 if (!ret && (flags & MCL_CURRENT))
709 mm_populate(0, TASK_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710out:
711 return ret;
712}
713
Heiko Carstens3480b252009-01-14 14:14:16 +0100714SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
716 int ret;
717
718 down_write(&current->mm->mmap_sem);
719 ret = do_mlockall(0);
720 up_write(&current->mm->mmap_sem);
721 return ret;
722}
723
724/*
725 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
726 * shm segments) get accounted against the user_struct instead.
727 */
728static DEFINE_SPINLOCK(shmlock_user_lock);
729
730int user_shm_lock(size_t size, struct user_struct *user)
731{
732 unsigned long lock_limit, locked;
733 int allowed = 0;
734
735 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800736 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700737 if (lock_limit == RLIM_INFINITY)
738 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 lock_limit >>= PAGE_SHIFT;
740 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700741 if (!allowed &&
742 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 goto out;
744 get_uid(user);
745 user->locked_shm += locked;
746 allowed = 1;
747out:
748 spin_unlock(&shmlock_user_lock);
749 return allowed;
750}
751
752void user_shm_unlock(size_t size, struct user_struct *user)
753{
754 spin_lock(&shmlock_user_lock);
755 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
756 spin_unlock(&shmlock_user_lock);
757 free_uid(user);
758}