blob: 8a54cd214925872a66d4d1cc36b69ec6c6047324 [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
Rik van Rielba470de2008-10-18 20:26:50 -0700208/**
Hugh Dickins408e82b2009-09-21 17:03:23 -0700209 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700210 * @vma: target vma
211 * @start: start address
212 * @end: end address
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700213 * @nonblocking:
Nick Pigginb291f002008-10-18 20:26:44 -0700214 *
Hugh Dickins408e82b2009-09-21 17:03:23 -0700215 * This takes care of making the pages present too.
Nick Pigginb291f002008-10-18 20:26:44 -0700216 *
Rik van Rielba470de2008-10-18 20:26:50 -0700217 * return 0 on success, negative error code on error.
218 *
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700219 * vma->vm_mm->mmap_sem must be held.
220 *
221 * If @nonblocking is NULL, it may be held for read or write and will
222 * be unperturbed.
223 *
224 * If @nonblocking is non-NULL, it must held for read only and may be
225 * released. If it's released, *@nonblocking will be set to 0.
Nick Pigginb291f002008-10-18 20:26:44 -0700226 */
Michel Lespinassecea10a12013-02-22 16:32:44 -0800227long __mlock_vma_pages_range(struct vm_area_struct *vma,
228 unsigned long start, unsigned long end, int *nonblocking)
Nick Pigginb291f002008-10-18 20:26:44 -0700229{
230 struct mm_struct *mm = vma->vm_mm;
Michel Lespinasse28a35712013-02-22 16:35:55 -0800231 unsigned long nr_pages = (end - start) / PAGE_SIZE;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700232 int gup_flags;
Nick Pigginb291f002008-10-18 20:26:44 -0700233
Rik van Rielba470de2008-10-18 20:26:50 -0700234 VM_BUG_ON(start & ~PAGE_MASK);
235 VM_BUG_ON(end & ~PAGE_MASK);
Sasha Levin81d1b092014-10-09 15:28:10 -0700236 VM_BUG_ON_VMA(start < vma->vm_start, vma);
237 VM_BUG_ON_VMA(end > vma->vm_end, vma);
Sasha Levin96dad672014-10-09 15:28:39 -0700238 VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
Rik van Rielba470de2008-10-18 20:26:50 -0700239
Linus Torvaldsa1fde082011-05-04 21:30:28 -0700240 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
Michel Lespinasse5ecfda02011-01-13 15:46:09 -0800241 /*
242 * We want to touch writable mappings with a write fault in order
243 * to break COW, except for shared mappings because these don't COW
244 * and we would not want to dirty them for nothing.
245 */
246 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700247 gup_flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700248
Michel Lespinassefdf4c582011-01-31 17:03:41 -0800249 /*
250 * We want mlock to succeed for regions that have any permissions
251 * other than PROT_NONE.
252 */
253 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
254 gup_flags |= FOLL_FORCE;
255
Johannes Weiner4805b022013-02-22 16:35:20 -0800256 /*
257 * We made sure addr is within a VMA, so the following will
258 * not result in a stack expansion that recurses back here.
259 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800260 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
Michel Lespinasse53a77062011-01-13 15:46:14 -0800261 NULL, NULL, nonblocking);
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700262}
263
264/*
265 * convert get_user_pages() return value to posix mlock() error
266 */
267static int __mlock_posix_error_return(long retval)
268{
269 if (retval == -EFAULT)
270 retval = -ENOMEM;
271 else if (retval == -ENOMEM)
272 retval = -EAGAIN;
273 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700274}
275
Nick Pigginb291f002008-10-18 20:26:44 -0700276/*
Vlastimil Babka56afe472013-09-11 14:22:32 -0700277 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
278 *
279 * The fast path is available only for evictable pages with single mapping.
280 * Then we can bypass the per-cpu pvec and get better performance.
281 * when mapcount > 1 we need try_to_munlock() which can fail.
282 * when !page_evictable(), we need the full redo logic of putback_lru_page to
283 * avoid leaving evictable page in unevictable list.
284 *
285 * In case of success, @page is added to @pvec and @pgrescued is incremented
286 * in case that the page was previously unevictable. @page is also unlocked.
287 */
288static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
289 int *pgrescued)
290{
Sasha Levin309381fea2014-01-23 15:52:54 -0800291 VM_BUG_ON_PAGE(PageLRU(page), page);
292 VM_BUG_ON_PAGE(!PageLocked(page), page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700293
294 if (page_mapcount(page) <= 1 && page_evictable(page)) {
295 pagevec_add(pvec, page);
296 if (TestClearPageUnevictable(page))
297 (*pgrescued)++;
298 unlock_page(page);
299 return true;
300 }
301
302 return false;
303}
304
305/*
306 * Putback multiple evictable pages to the LRU
307 *
308 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
309 * the pages might have meanwhile become unevictable but that is OK.
310 */
311static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
312{
313 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
314 /*
315 *__pagevec_lru_add() calls release_pages() so we don't call
316 * put_page() explicitly
317 */
318 __pagevec_lru_add(pvec);
319 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
320}
321
322/*
Vlastimil Babka72255222013-09-11 14:22:29 -0700323 * Munlock a batch of pages from the same zone
324 *
325 * The work is split to two main phases. First phase clears the Mlocked flag
326 * and attempts to isolate the pages, all under a single zone lru lock.
327 * The second phase finishes the munlock only for pages where isolation
328 * succeeded.
329 *
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700330 * Note that the pagevec may be modified during the process.
Vlastimil Babka72255222013-09-11 14:22:29 -0700331 */
332static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
333{
334 int i;
335 int nr = pagevec_count(pvec);
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800336 int delta_munlocked;
Vlastimil Babka56afe472013-09-11 14:22:32 -0700337 struct pagevec pvec_putback;
338 int pgrescued = 0;
Vlastimil Babka72255222013-09-11 14:22:29 -0700339
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800340 pagevec_init(&pvec_putback, 0);
341
Vlastimil Babka72255222013-09-11 14:22:29 -0700342 /* Phase 1: page isolation */
343 spin_lock_irq(&zone->lru_lock);
344 for (i = 0; i < nr; i++) {
345 struct page *page = pvec->pages[i];
346
347 if (TestClearPageMlocked(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700348 /*
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800349 * We already have pin from follow_page_mask()
350 * so we can spare the get_page() here.
Vlastimil Babka72255222013-09-11 14:22:29 -0700351 */
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800352 if (__munlock_isolate_lru_page(page, false))
353 continue;
354 else
355 __munlock_isolation_failed(page);
Vlastimil Babka72255222013-09-11 14:22:29 -0700356 }
Vlastimil Babka01cc2e52014-01-23 15:52:50 -0800357
358 /*
359 * We won't be munlocking this page in the next phase
360 * but we still need to release the follow_page_mask()
361 * pin. We cannot do it under lru_lock however. If it's
362 * the last pin, __page_cache_release() would deadlock.
363 */
364 pagevec_add(&pvec_putback, pvec->pages[i]);
365 pvec->pages[i] = NULL;
Vlastimil Babka72255222013-09-11 14:22:29 -0700366 }
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800367 delta_munlocked = -nr + pagevec_count(&pvec_putback);
Vlastimil Babka1ebb7cc2013-09-11 14:22:30 -0700368 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
Vlastimil Babka72255222013-09-11 14:22:29 -0700369 spin_unlock_irq(&zone->lru_lock);
370
Vlastimil Babka3b25df92014-01-02 12:58:44 -0800371 /* Now we can release pins of pages that we are not munlocking */
372 pagevec_release(&pvec_putback);
373
Vlastimil Babka56afe472013-09-11 14:22:32 -0700374 /* Phase 2: page munlock */
Vlastimil Babka72255222013-09-11 14:22:29 -0700375 for (i = 0; i < nr; i++) {
376 struct page *page = pvec->pages[i];
377
378 if (page) {
379 lock_page(page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700380 if (!__putback_lru_fast_prepare(page, &pvec_putback,
381 &pgrescued)) {
Vlastimil Babka5b409982013-09-11 14:22:33 -0700382 /*
383 * Slow path. We don't want to lose the last
384 * pin before unlock_page()
385 */
386 get_page(page); /* for putback_lru_page() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700387 __munlock_isolated_page(page);
388 unlock_page(page);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700389 put_page(page); /* from follow_page_mask() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700390 }
Vlastimil Babka72255222013-09-11 14:22:29 -0700391 }
392 }
Vlastimil Babka56afe472013-09-11 14:22:32 -0700393
Vlastimil Babka5b409982013-09-11 14:22:33 -0700394 /*
395 * Phase 3: page putback for pages that qualified for the fast path
396 * This will also call put_page() to return pin from follow_page_mask()
397 */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700398 if (pagevec_count(&pvec_putback))
399 __putback_lru_fast(&pvec_putback, pgrescued);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700400}
Vlastimil Babka56afe472013-09-11 14:22:32 -0700401
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700402/*
403 * Fill up pagevec for __munlock_pagevec using pte walk
404 *
405 * The function expects that the struct page corresponding to @start address is
406 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
407 *
408 * The rest of @pvec is filled by subsequent pages within the same pmd and same
409 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
410 * pages also get pinned.
411 *
412 * Returns the address of the next page that should be scanned. This equals
413 * @start + PAGE_SIZE when no page could be added by the pte walk.
414 */
415static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
416 struct vm_area_struct *vma, int zoneid, unsigned long start,
417 unsigned long end)
418{
419 pte_t *pte;
420 spinlock_t *ptl;
421
422 /*
423 * Initialize pte walk starting at the already pinned page where we
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700424 * are sure that there is a pte, as it was pinned under the same
425 * mmap_sem write op.
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700426 */
427 pte = get_locked_pte(vma->vm_mm, start, &ptl);
Vlastimil Babkaeadb41a2013-09-30 13:45:18 -0700428 /* Make sure we do not cross the page table boundary */
429 end = pgd_addr_end(start, end);
430 end = pud_addr_end(start, end);
431 end = pmd_addr_end(start, end);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700432
433 /* The page next to the pinned page is the first we will try to get */
434 start += PAGE_SIZE;
435 while (start < end) {
436 struct page *page = NULL;
437 pte++;
438 if (pte_present(*pte))
439 page = vm_normal_page(vma, start, *pte);
440 /*
441 * Break if page could not be obtained or the page's node+zone does not
442 * match
443 */
444 if (!page || page_zone_id(page) != zoneid)
445 break;
446
447 get_page(page);
448 /*
449 * Increase the address that will be returned *before* the
450 * eventual break due to pvec becoming full by adding the page
451 */
452 start += PAGE_SIZE;
453 if (pagevec_add(pvec, page) == 0)
454 break;
455 }
456 pte_unmap_unlock(pte, ptl);
457 return start;
Vlastimil Babka72255222013-09-11 14:22:29 -0700458}
459
460/*
Rik van Rielba470de2008-10-18 20:26:50 -0700461 * munlock_vma_pages_range() - munlock all pages in the vma range.'
462 * @vma - vma containing range to be munlock()ed.
463 * @start - start address in @vma of the range
464 * @end - end of range in @vma.
465 *
466 * For mremap(), munmap() and exit().
467 *
468 * Called with @vma VM_LOCKED.
469 *
470 * Returns with VM_LOCKED cleared. Callers must be prepared to
471 * deal with this.
472 *
473 * We don't save and restore VM_LOCKED here because pages are
474 * still on lru. In unmap path, pages might be scanned by reclaim
475 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
476 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700477 */
Rik van Rielba470de2008-10-18 20:26:50 -0700478void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700479 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700480{
481 vma->vm_flags &= ~VM_LOCKED;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700482
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800483 while (start < end) {
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700484 struct page *page = NULL;
Vlastimil Babkac424be12014-01-02 12:58:43 -0800485 unsigned int page_mask;
486 unsigned long page_increm;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700487 struct pagevec pvec;
488 struct zone *zone;
489 int zoneid;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800490
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700491 pagevec_init(&pvec, 0);
Hugh Dickins6e919712009-09-21 17:03:32 -0700492 /*
493 * Although FOLL_DUMP is intended for get_dump_page(),
494 * it just so happens that its special treatment of the
495 * ZERO_PAGE (returning an error instead of doing get_page)
496 * suits munlock very well (and if somehow an abnormal page
497 * has sneaked into the range, we won't oops here: great).
498 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800499 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700500 &page_mask);
501
Hugh Dickins6e919712009-09-21 17:03:32 -0700502 if (page && !IS_ERR(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700503 if (PageTransHuge(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700504 lock_page(page);
505 /*
506 * Any THP page found by follow_page_mask() may
507 * have gotten split before reaching
508 * munlock_vma_page(), so we need to recompute
509 * the page_mask here.
510 */
511 page_mask = munlock_vma_page(page);
512 unlock_page(page);
513 put_page(page); /* follow_page_mask() */
514 } else {
515 /*
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700516 * Non-huge pages are handled in batches via
517 * pagevec. The pin from follow_page_mask()
518 * prevents them from collapsing by THP.
Vlastimil Babka72255222013-09-11 14:22:29 -0700519 */
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700520 pagevec_add(&pvec, page);
521 zone = page_zone(page);
522 zoneid = page_zone_id(page);
523
524 /*
525 * Try to fill the rest of pagevec using fast
526 * pte walk. This will also update start to
527 * the next page to process. Then munlock the
528 * pagevec.
529 */
530 start = __munlock_pagevec_fill(&pvec, vma,
531 zoneid, start, end);
532 __munlock_pagevec(&pvec, zone);
533 goto next;
Vlastimil Babka72255222013-09-11 14:22:29 -0700534 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700535 }
Vlastimil Babkac424be12014-01-02 12:58:43 -0800536 /* It's a bug to munlock in the middle of a THP page */
537 VM_BUG_ON((start >> PAGE_SHIFT) & page_mask);
538 page_increm = 1 + page_mask;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800539 start += page_increm * PAGE_SIZE;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700540next:
Hugh Dickins408e82b2009-09-21 17:03:23 -0700541 cond_resched();
542 }
Nick Pigginb291f002008-10-18 20:26:44 -0700543}
544
545/*
546 * mlock_fixup - handle mlock[all]/munlock[all] requests.
547 *
548 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
549 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800550 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700551 *
552 * For vmas that pass the filters, merge/split as appropriate.
553 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900555 unsigned long start, unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Nick Pigginb291f002008-10-18 20:26:44 -0700557 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700559 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 int ret = 0;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900561 int lock = !!(newflags & VM_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Michel Lespinassefed067d2011-01-13 15:46:10 -0800563 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400564 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
Nick Pigginb291f002008-10-18 20:26:44 -0700565 goto out; /* don't set VM_LOCKED, don't count */
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
568 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
569 vma->vm_file, pgoff, vma_policy(vma));
570 if (*prev) {
571 vma = *prev;
572 goto success;
573 }
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (start != vma->vm_start) {
576 ret = split_vma(mm, vma, start, 1);
577 if (ret)
578 goto out;
579 }
580
581 if (end != vma->vm_end) {
582 ret = split_vma(mm, vma, end, 0);
583 if (ret)
584 goto out;
585 }
586
587success:
588 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700589 * Keep track of amount of locked VM.
590 */
591 nr_pages = (end - start) >> PAGE_SHIFT;
592 if (!lock)
593 nr_pages = -nr_pages;
594 mm->locked_vm += nr_pages;
595
596 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 * vm_flags is protected by the mmap_sem held in write mode.
598 * It's okay if try_to_unmap_one unmaps a page just after we
Nick Pigginb291f002008-10-18 20:26:44 -0700599 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Michel Lespinassefed067d2011-01-13 15:46:10 -0800602 if (lock)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700603 vma->vm_flags = newflags;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800604 else
Hugh Dickins408e82b2009-09-21 17:03:23 -0700605 munlock_vma_pages_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607out:
Nick Pigginb291f002008-10-18 20:26:44 -0700608 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 return ret;
610}
611
612static int do_mlock(unsigned long start, size_t len, int on)
613{
614 unsigned long nstart, end, tmp;
615 struct vm_area_struct * vma, * prev;
616 int error;
617
Michel Lespinassefed067d2011-01-13 15:46:10 -0800618 VM_BUG_ON(start & ~PAGE_MASK);
619 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 end = start + len;
621 if (end < start)
622 return -EINVAL;
623 if (end == start)
624 return 0;
Linus Torvalds097d5912012-03-06 18:23:36 -0800625 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 if (!vma || vma->vm_start > start)
627 return -ENOMEM;
628
Linus Torvalds097d5912012-03-06 18:23:36 -0800629 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (start > vma->vm_start)
631 prev = vma;
632
633 for (nstart = start ; ; ) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900634 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
636 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
637
Michel Lespinasse18693052013-02-22 16:32:46 -0800638 newflags = vma->vm_flags & ~VM_LOCKED;
639 if (on)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700640 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642 tmp = vma->vm_end;
643 if (tmp > end)
644 tmp = end;
645 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
646 if (error)
647 break;
648 nstart = tmp;
649 if (nstart < prev->vm_end)
650 nstart = prev->vm_end;
651 if (nstart >= end)
652 break;
653
654 vma = prev->vm_next;
655 if (!vma || vma->vm_start != nstart) {
656 error = -ENOMEM;
657 break;
658 }
659 }
660 return error;
661}
662
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800663/*
664 * __mm_populate - populate and/or mlock pages within a range of address space.
665 *
666 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
667 * flags. VMAs must be already marked with the desired vm_flags, and
668 * mmap_sem must not be held.
669 */
670int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
Michel Lespinassefed067d2011-01-13 15:46:10 -0800671{
672 struct mm_struct *mm = current->mm;
673 unsigned long end, nstart, nend;
674 struct vm_area_struct *vma = NULL;
Michel Lespinasse53a77062011-01-13 15:46:14 -0800675 int locked = 0;
Michel Lespinasse28a35712013-02-22 16:35:55 -0800676 long ret = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800677
678 VM_BUG_ON(start & ~PAGE_MASK);
679 VM_BUG_ON(len != PAGE_ALIGN(len));
680 end = start + len;
681
Michel Lespinassefed067d2011-01-13 15:46:10 -0800682 for (nstart = start; nstart < end; nstart = nend) {
683 /*
684 * We want to fault in pages for [nstart; end) address range.
685 * Find first corresponding VMA.
686 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800687 if (!locked) {
688 locked = 1;
689 down_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800690 vma = find_vma(mm, nstart);
Michel Lespinasse53a77062011-01-13 15:46:14 -0800691 } else if (nstart >= vma->vm_end)
Michel Lespinassefed067d2011-01-13 15:46:10 -0800692 vma = vma->vm_next;
693 if (!vma || vma->vm_start >= end)
694 break;
695 /*
696 * Set [nstart; nend) to intersection of desired address
697 * range with the first VMA. Also, skip undesirable VMA types.
698 */
699 nend = min(end, vma->vm_end);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700700 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
Michel Lespinassefed067d2011-01-13 15:46:10 -0800701 continue;
702 if (nstart < vma->vm_start)
703 nstart = vma->vm_start;
704 /*
Michel Lespinasse53a77062011-01-13 15:46:14 -0800705 * Now fault in a range of pages. __mlock_vma_pages_range()
706 * double checks the vma flags, so that it won't mlock pages
707 * if the vma was already munlocked.
Michel Lespinassefed067d2011-01-13 15:46:10 -0800708 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800709 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
710 if (ret < 0) {
711 if (ignore_errors) {
712 ret = 0;
713 continue; /* continue at next VMA */
714 }
Michel Lespinasse5fdb2002011-01-13 15:46:12 -0800715 ret = __mlock_posix_error_return(ret);
716 break;
717 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800718 nend = nstart + ret * PAGE_SIZE;
719 ret = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800720 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800721 if (locked)
722 up_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800723 return ret; /* 0 or negative error code */
724}
725
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100726SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
728 unsigned long locked;
729 unsigned long lock_limit;
730 int error = -ENOMEM;
731
732 if (!can_do_mlock())
733 return -EPERM;
734
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800735 lru_add_drain_all(); /* flush pagevec */
736
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
738 start &= PAGE_MASK;
739
Jiri Slaby59e99e52010-03-05 13:41:44 -0800740 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 lock_limit >>= PAGE_SHIFT;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800742 locked = len >> PAGE_SHIFT;
743
744 down_write(&current->mm->mmap_sem);
745
746 locked += current->mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748 /* check against resource limits */
749 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
750 error = do_mlock(start, len, 1);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 up_write(&current->mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800753 if (!error)
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800754 error = __mm_populate(start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 return error;
756}
757
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100758SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
760 int ret;
761
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
763 start &= PAGE_MASK;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800764
765 down_write(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 ret = do_mlock(start, len, 0);
767 up_write(&current->mm->mmap_sem);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800768
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 return ret;
770}
771
772static int do_mlockall(int flags)
773{
774 struct vm_area_struct * vma, * prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
776 if (flags & MCL_FUTURE)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700777 current->mm->def_flags |= VM_LOCKED;
Gerald Schaefer9977f0f2013-02-12 13:46:20 -0800778 else
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700779 current->mm->def_flags &= ~VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 if (flags == MCL_FUTURE)
781 goto out;
782
783 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900784 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Michel Lespinasse18693052013-02-22 16:32:46 -0800786 newflags = vma->vm_flags & ~VM_LOCKED;
787 if (flags & MCL_CURRENT)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700788 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 /* Ignore errors */
791 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700792 cond_resched_rcu_qs();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 }
794out:
795 return 0;
796}
797
Heiko Carstens3480b252009-01-14 14:14:16 +0100798SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
800 unsigned long lock_limit;
801 int ret = -EINVAL;
802
803 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
804 goto out;
805
806 ret = -EPERM;
807 if (!can_do_mlock())
808 goto out;
809
Christoph Lameterdf9d6982011-10-31 17:09:35 -0700810 if (flags & MCL_CURRENT)
811 lru_add_drain_all(); /* flush pagevec */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800812
Jiri Slaby59e99e52010-03-05 13:41:44 -0800813 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 lock_limit >>= PAGE_SHIFT;
815
816 ret = -ENOMEM;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800817 down_write(&current->mm->mmap_sem);
818
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
820 capable(CAP_IPC_LOCK))
821 ret = do_mlockall(flags);
822 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800823 if (!ret && (flags & MCL_CURRENT))
824 mm_populate(0, TASK_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825out:
826 return ret;
827}
828
Heiko Carstens3480b252009-01-14 14:14:16 +0100829SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
831 int ret;
832
833 down_write(&current->mm->mmap_sem);
834 ret = do_mlockall(0);
835 up_write(&current->mm->mmap_sem);
836 return ret;
837}
838
839/*
840 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
841 * shm segments) get accounted against the user_struct instead.
842 */
843static DEFINE_SPINLOCK(shmlock_user_lock);
844
845int user_shm_lock(size_t size, struct user_struct *user)
846{
847 unsigned long lock_limit, locked;
848 int allowed = 0;
849
850 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800851 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700852 if (lock_limit == RLIM_INFINITY)
853 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 lock_limit >>= PAGE_SHIFT;
855 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700856 if (!allowed &&
857 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 goto out;
859 get_uid(user);
860 user->locked_shm += locked;
861 allowed = 1;
862out:
863 spin_unlock(&shmlock_user_lock);
864 return allowed;
865}
866
867void user_shm_unlock(size_t size, struct user_struct *user)
868{
869 spin_lock(&shmlock_user_lock);
870 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
871 spin_unlock(&shmlock_user_lock);
872 free_uid(user);
873}