blob: d63802663242eb6ad13ca2ed065434f24fd7e5d1 [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{
29 if (capable(CAP_IPC_LOCK))
30 return 1;
Jiri Slaby59e99e52010-03-05 13:41:44 -080031 if (rlimit(RLIMIT_MEMLOCK) != 0)
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040032 return 1;
33 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{
82 BUG_ON(!PageLocked(page));
83
Nick Piggin5344b7e2008-10-18 20:26:51 -070084 if (!TestSetPageMlocked(page)) {
David Rientjes8449d212012-10-08 16:34:06 -070085 mod_zone_page_state(page_zone(page), NR_MLOCK,
86 hpage_nr_pages(page));
Nick Piggin5344b7e2008-10-18 20:26:51 -070087 count_vm_event(UNEVICTABLE_PGMLOCKED);
88 if (!isolate_lru_page(page))
89 putback_lru_page(page);
90 }
Nick Pigginb291f002008-10-18 20:26:44 -070091}
92
Vlastimil Babka72255222013-09-11 14:22:29 -070093/*
94 * Finish munlock after successful page isolation
95 *
96 * Page must be locked. This is a wrapper for try_to_munlock()
97 * and putback_lru_page() with munlock accounting.
98 */
99static void __munlock_isolated_page(struct page *page)
100{
101 int ret = SWAP_AGAIN;
102
103 /*
104 * Optimization: if the page was mapped just once, that's our mapping
105 * and we don't need to check all the other vmas.
106 */
107 if (page_mapcount(page) > 1)
108 ret = try_to_munlock(page);
109
110 /* Did try_to_unlock() succeed or punt? */
111 if (ret != SWAP_MLOCK)
112 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
113
114 putback_lru_page(page);
115}
116
117/*
118 * Accounting for page isolation fail during munlock
119 *
120 * Performs accounting when page isolation fails in munlock. There is nothing
121 * else to do because it means some other task has already removed the page
122 * from the LRU. putback_lru_page() will take care of removing the page from
123 * the unevictable list, if necessary. vmscan [page_referenced()] will move
124 * the page back to the unevictable list if some other vma has it mlocked.
125 */
126static void __munlock_isolation_failed(struct page *page)
127{
128 if (PageUnevictable(page))
129 count_vm_event(UNEVICTABLE_PGSTRANDED);
130 else
131 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
132}
133
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800134/**
135 * munlock_vma_page - munlock a vma page
136 * @page - page to be unlocked
Nick Pigginb291f002008-10-18 20:26:44 -0700137 *
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800138 * called from munlock()/munmap() path with page supposedly on the LRU.
139 * When we munlock a page, because the vma where we found the page is being
140 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
141 * page locked so that we can leave it on the unevictable lru list and not
142 * bother vmscan with it. However, to walk the page's rmap list in
143 * try_to_munlock() we must isolate the page from the LRU. If some other
144 * task has removed the page from the LRU, we won't be able to do that.
145 * So we clear the PageMlocked as we might not get another chance. If we
146 * can't isolate the page, we leave it for putback_lru_page() and vmscan
147 * [page_referenced()/try_to_unmap()] to deal with.
Nick Pigginb291f002008-10-18 20:26:44 -0700148 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800149unsigned int munlock_vma_page(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -0700150{
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800151 unsigned int page_mask = 0;
152
Nick Pigginb291f002008-10-18 20:26:44 -0700153 BUG_ON(!PageLocked(page));
154
Nick Piggin5344b7e2008-10-18 20:26:51 -0700155 if (TestClearPageMlocked(page)) {
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800156 unsigned int nr_pages = hpage_nr_pages(page);
157 mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
158 page_mask = nr_pages - 1;
Vlastimil Babka72255222013-09-11 14:22:29 -0700159 if (!isolate_lru_page(page))
160 __munlock_isolated_page(page);
161 else
162 __munlock_isolation_failed(page);
Nick Pigginb291f002008-10-18 20:26:44 -0700163 }
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800164
165 return page_mask;
Nick Pigginb291f002008-10-18 20:26:44 -0700166}
167
Rik van Rielba470de2008-10-18 20:26:50 -0700168/**
Hugh Dickins408e82b2009-09-21 17:03:23 -0700169 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700170 * @vma: target vma
171 * @start: start address
172 * @end: end address
Nick Pigginb291f002008-10-18 20:26:44 -0700173 *
Hugh Dickins408e82b2009-09-21 17:03:23 -0700174 * This takes care of making the pages present too.
Nick Pigginb291f002008-10-18 20:26:44 -0700175 *
Rik van Rielba470de2008-10-18 20:26:50 -0700176 * return 0 on success, negative error code on error.
177 *
178 * vma->vm_mm->mmap_sem must be held for at least read.
Nick Pigginb291f002008-10-18 20:26:44 -0700179 */
Michel Lespinassecea10a12013-02-22 16:32:44 -0800180long __mlock_vma_pages_range(struct vm_area_struct *vma,
181 unsigned long start, unsigned long end, int *nonblocking)
Nick Pigginb291f002008-10-18 20:26:44 -0700182{
183 struct mm_struct *mm = vma->vm_mm;
Michel Lespinasse28a35712013-02-22 16:35:55 -0800184 unsigned long nr_pages = (end - start) / PAGE_SIZE;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700185 int gup_flags;
Nick Pigginb291f002008-10-18 20:26:44 -0700186
Rik van Rielba470de2008-10-18 20:26:50 -0700187 VM_BUG_ON(start & ~PAGE_MASK);
188 VM_BUG_ON(end & ~PAGE_MASK);
189 VM_BUG_ON(start < vma->vm_start);
190 VM_BUG_ON(end > vma->vm_end);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700191 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
Rik van Rielba470de2008-10-18 20:26:50 -0700192
Linus Torvaldsa1fde082011-05-04 21:30:28 -0700193 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
Michel Lespinasse5ecfda02011-01-13 15:46:09 -0800194 /*
195 * We want to touch writable mappings with a write fault in order
196 * to break COW, except for shared mappings because these don't COW
197 * and we would not want to dirty them for nothing.
198 */
199 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700200 gup_flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700201
Michel Lespinassefdf4c582011-01-31 17:03:41 -0800202 /*
203 * We want mlock to succeed for regions that have any permissions
204 * other than PROT_NONE.
205 */
206 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
207 gup_flags |= FOLL_FORCE;
208
Johannes Weiner4805b022013-02-22 16:35:20 -0800209 /*
210 * We made sure addr is within a VMA, so the following will
211 * not result in a stack expansion that recurses back here.
212 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800213 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
Michel Lespinasse53a77062011-01-13 15:46:14 -0800214 NULL, NULL, nonblocking);
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700215}
216
217/*
218 * convert get_user_pages() return value to posix mlock() error
219 */
220static int __mlock_posix_error_return(long retval)
221{
222 if (retval == -EFAULT)
223 retval = -ENOMEM;
224 else if (retval == -ENOMEM)
225 retval = -EAGAIN;
226 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700227}
228
Nick Pigginb291f002008-10-18 20:26:44 -0700229/*
Vlastimil Babka56afe472013-09-11 14:22:32 -0700230 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
231 *
232 * The fast path is available only for evictable pages with single mapping.
233 * Then we can bypass the per-cpu pvec and get better performance.
234 * when mapcount > 1 we need try_to_munlock() which can fail.
235 * when !page_evictable(), we need the full redo logic of putback_lru_page to
236 * avoid leaving evictable page in unevictable list.
237 *
238 * In case of success, @page is added to @pvec and @pgrescued is incremented
239 * in case that the page was previously unevictable. @page is also unlocked.
240 */
241static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
242 int *pgrescued)
243{
244 VM_BUG_ON(PageLRU(page));
245 VM_BUG_ON(!PageLocked(page));
246
247 if (page_mapcount(page) <= 1 && page_evictable(page)) {
248 pagevec_add(pvec, page);
249 if (TestClearPageUnevictable(page))
250 (*pgrescued)++;
251 unlock_page(page);
252 return true;
253 }
254
255 return false;
256}
257
258/*
259 * Putback multiple evictable pages to the LRU
260 *
261 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
262 * the pages might have meanwhile become unevictable but that is OK.
263 */
264static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
265{
266 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
267 /*
268 *__pagevec_lru_add() calls release_pages() so we don't call
269 * put_page() explicitly
270 */
271 __pagevec_lru_add(pvec);
272 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
273}
274
275/*
Vlastimil Babka72255222013-09-11 14:22:29 -0700276 * Munlock a batch of pages from the same zone
277 *
278 * The work is split to two main phases. First phase clears the Mlocked flag
279 * and attempts to isolate the pages, all under a single zone lru lock.
280 * The second phase finishes the munlock only for pages where isolation
281 * succeeded.
282 *
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700283 * Note that the pagevec may be modified during the process.
Vlastimil Babka72255222013-09-11 14:22:29 -0700284 */
285static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
286{
287 int i;
288 int nr = pagevec_count(pvec);
Vlastimil Babka1ebb7cc2013-09-11 14:22:30 -0700289 int delta_munlocked = -nr;
Vlastimil Babka56afe472013-09-11 14:22:32 -0700290 struct pagevec pvec_putback;
291 int pgrescued = 0;
Vlastimil Babka72255222013-09-11 14:22:29 -0700292
293 /* Phase 1: page isolation */
294 spin_lock_irq(&zone->lru_lock);
295 for (i = 0; i < nr; i++) {
296 struct page *page = pvec->pages[i];
297
298 if (TestClearPageMlocked(page)) {
299 struct lruvec *lruvec;
300 int lru;
301
Vlastimil Babka72255222013-09-11 14:22:29 -0700302 if (PageLRU(page)) {
303 lruvec = mem_cgroup_page_lruvec(page, zone);
304 lru = page_lru(page);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700305 /*
306 * We already have pin from follow_page_mask()
307 * so we can spare the get_page() here.
308 */
Vlastimil Babka72255222013-09-11 14:22:29 -0700309 ClearPageLRU(page);
310 del_page_from_lru_list(page, lruvec, lru);
311 } else {
312 __munlock_isolation_failed(page);
313 goto skip_munlock;
314 }
315
316 } else {
317skip_munlock:
318 /*
319 * We won't be munlocking this page in the next phase
320 * but we still need to release the follow_page_mask()
321 * pin.
322 */
323 pvec->pages[i] = NULL;
324 put_page(page);
Vlastimil Babka1ebb7cc2013-09-11 14:22:30 -0700325 delta_munlocked++;
Vlastimil Babka72255222013-09-11 14:22:29 -0700326 }
327 }
Vlastimil Babka1ebb7cc2013-09-11 14:22:30 -0700328 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
Vlastimil Babka72255222013-09-11 14:22:29 -0700329 spin_unlock_irq(&zone->lru_lock);
330
Vlastimil Babka56afe472013-09-11 14:22:32 -0700331 /* Phase 2: page munlock */
332 pagevec_init(&pvec_putback, 0);
Vlastimil Babka72255222013-09-11 14:22:29 -0700333 for (i = 0; i < nr; i++) {
334 struct page *page = pvec->pages[i];
335
336 if (page) {
337 lock_page(page);
Vlastimil Babka56afe472013-09-11 14:22:32 -0700338 if (!__putback_lru_fast_prepare(page, &pvec_putback,
339 &pgrescued)) {
Vlastimil Babka5b409982013-09-11 14:22:33 -0700340 /*
341 * Slow path. We don't want to lose the last
342 * pin before unlock_page()
343 */
344 get_page(page); /* for putback_lru_page() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700345 __munlock_isolated_page(page);
346 unlock_page(page);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700347 put_page(page); /* from follow_page_mask() */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700348 }
Vlastimil Babka72255222013-09-11 14:22:29 -0700349 }
350 }
Vlastimil Babka56afe472013-09-11 14:22:32 -0700351
Vlastimil Babka5b409982013-09-11 14:22:33 -0700352 /*
353 * Phase 3: page putback for pages that qualified for the fast path
354 * This will also call put_page() to return pin from follow_page_mask()
355 */
Vlastimil Babka56afe472013-09-11 14:22:32 -0700356 if (pagevec_count(&pvec_putback))
357 __putback_lru_fast(&pvec_putback, pgrescued);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700358}
Vlastimil Babka56afe472013-09-11 14:22:32 -0700359
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700360/*
361 * Fill up pagevec for __munlock_pagevec using pte walk
362 *
363 * The function expects that the struct page corresponding to @start address is
364 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
365 *
366 * The rest of @pvec is filled by subsequent pages within the same pmd and same
367 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
368 * pages also get pinned.
369 *
370 * Returns the address of the next page that should be scanned. This equals
371 * @start + PAGE_SIZE when no page could be added by the pte walk.
372 */
373static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
374 struct vm_area_struct *vma, int zoneid, unsigned long start,
375 unsigned long end)
376{
377 pte_t *pte;
378 spinlock_t *ptl;
379
380 /*
381 * Initialize pte walk starting at the already pinned page where we
382 * are sure that there is a pte.
383 */
384 pte = get_locked_pte(vma->vm_mm, start, &ptl);
385 end = min(end, pmd_addr_end(start, end));
386
387 /* The page next to the pinned page is the first we will try to get */
388 start += PAGE_SIZE;
389 while (start < end) {
390 struct page *page = NULL;
391 pte++;
392 if (pte_present(*pte))
393 page = vm_normal_page(vma, start, *pte);
394 /*
395 * Break if page could not be obtained or the page's node+zone does not
396 * match
397 */
398 if (!page || page_zone_id(page) != zoneid)
399 break;
400
401 get_page(page);
402 /*
403 * Increase the address that will be returned *before* the
404 * eventual break due to pvec becoming full by adding the page
405 */
406 start += PAGE_SIZE;
407 if (pagevec_add(pvec, page) == 0)
408 break;
409 }
410 pte_unmap_unlock(pte, ptl);
411 return start;
Vlastimil Babka72255222013-09-11 14:22:29 -0700412}
413
414/*
Rik van Rielba470de2008-10-18 20:26:50 -0700415 * munlock_vma_pages_range() - munlock all pages in the vma range.'
416 * @vma - vma containing range to be munlock()ed.
417 * @start - start address in @vma of the range
418 * @end - end of range in @vma.
419 *
420 * For mremap(), munmap() and exit().
421 *
422 * Called with @vma VM_LOCKED.
423 *
424 * Returns with VM_LOCKED cleared. Callers must be prepared to
425 * deal with this.
426 *
427 * We don't save and restore VM_LOCKED here because pages are
428 * still on lru. In unmap path, pages might be scanned by reclaim
429 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
430 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700431 */
Rik van Rielba470de2008-10-18 20:26:50 -0700432void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700433 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700434{
435 vma->vm_flags &= ~VM_LOCKED;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700436
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800437 while (start < end) {
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700438 struct page *page = NULL;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800439 unsigned int page_mask, page_increm;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700440 struct pagevec pvec;
441 struct zone *zone;
442 int zoneid;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800443
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700444 pagevec_init(&pvec, 0);
Hugh Dickins6e919712009-09-21 17:03:32 -0700445 /*
446 * Although FOLL_DUMP is intended for get_dump_page(),
447 * it just so happens that its special treatment of the
448 * ZERO_PAGE (returning an error instead of doing get_page)
449 * suits munlock very well (and if somehow an abnormal page
450 * has sneaked into the range, we won't oops here: great).
451 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800452 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700453 &page_mask);
454
Hugh Dickins6e919712009-09-21 17:03:32 -0700455 if (page && !IS_ERR(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700456 if (PageTransHuge(page)) {
Vlastimil Babka72255222013-09-11 14:22:29 -0700457 lock_page(page);
458 /*
459 * Any THP page found by follow_page_mask() may
460 * have gotten split before reaching
461 * munlock_vma_page(), so we need to recompute
462 * the page_mask here.
463 */
464 page_mask = munlock_vma_page(page);
465 unlock_page(page);
466 put_page(page); /* follow_page_mask() */
467 } else {
468 /*
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700469 * Non-huge pages are handled in batches via
470 * pagevec. The pin from follow_page_mask()
471 * prevents them from collapsing by THP.
Vlastimil Babka72255222013-09-11 14:22:29 -0700472 */
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700473 pagevec_add(&pvec, page);
474 zone = page_zone(page);
475 zoneid = page_zone_id(page);
476
477 /*
478 * Try to fill the rest of pagevec using fast
479 * pte walk. This will also update start to
480 * the next page to process. Then munlock the
481 * pagevec.
482 */
483 start = __munlock_pagevec_fill(&pvec, vma,
484 zoneid, start, end);
485 __munlock_pagevec(&pvec, zone);
486 goto next;
Vlastimil Babka72255222013-09-11 14:22:29 -0700487 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700488 }
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800489 page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);
490 start += page_increm * PAGE_SIZE;
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700491next:
Hugh Dickins408e82b2009-09-21 17:03:23 -0700492 cond_resched();
493 }
Nick Pigginb291f002008-10-18 20:26:44 -0700494}
495
496/*
497 * mlock_fixup - handle mlock[all]/munlock[all] requests.
498 *
499 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
500 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800501 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700502 *
503 * For vmas that pass the filters, merge/split as appropriate.
504 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900506 unsigned long start, unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
Nick Pigginb291f002008-10-18 20:26:44 -0700508 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700510 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 int ret = 0;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900512 int lock = !!(newflags & VM_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Michel Lespinassefed067d2011-01-13 15:46:10 -0800514 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400515 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
Nick Pigginb291f002008-10-18 20:26:44 -0700516 goto out; /* don't set VM_LOCKED, don't count */
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
519 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
520 vma->vm_file, pgoff, vma_policy(vma));
521 if (*prev) {
522 vma = *prev;
523 goto success;
524 }
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (start != vma->vm_start) {
527 ret = split_vma(mm, vma, start, 1);
528 if (ret)
529 goto out;
530 }
531
532 if (end != vma->vm_end) {
533 ret = split_vma(mm, vma, end, 0);
534 if (ret)
535 goto out;
536 }
537
538success:
539 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700540 * Keep track of amount of locked VM.
541 */
542 nr_pages = (end - start) >> PAGE_SHIFT;
543 if (!lock)
544 nr_pages = -nr_pages;
545 mm->locked_vm += nr_pages;
546
547 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 * vm_flags is protected by the mmap_sem held in write mode.
549 * It's okay if try_to_unmap_one unmaps a page just after we
Nick Pigginb291f002008-10-18 20:26:44 -0700550 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Michel Lespinassefed067d2011-01-13 15:46:10 -0800553 if (lock)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700554 vma->vm_flags = newflags;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800555 else
Hugh Dickins408e82b2009-09-21 17:03:23 -0700556 munlock_vma_pages_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558out:
Nick Pigginb291f002008-10-18 20:26:44 -0700559 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 return ret;
561}
562
563static int do_mlock(unsigned long start, size_t len, int on)
564{
565 unsigned long nstart, end, tmp;
566 struct vm_area_struct * vma, * prev;
567 int error;
568
Michel Lespinassefed067d2011-01-13 15:46:10 -0800569 VM_BUG_ON(start & ~PAGE_MASK);
570 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 end = start + len;
572 if (end < start)
573 return -EINVAL;
574 if (end == start)
575 return 0;
Linus Torvalds097d5912012-03-06 18:23:36 -0800576 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 if (!vma || vma->vm_start > start)
578 return -ENOMEM;
579
Linus Torvalds097d5912012-03-06 18:23:36 -0800580 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (start > vma->vm_start)
582 prev = vma;
583
584 for (nstart = start ; ; ) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900585 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
588
Michel Lespinasse18693052013-02-22 16:32:46 -0800589 newflags = vma->vm_flags & ~VM_LOCKED;
590 if (on)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700591 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
593 tmp = vma->vm_end;
594 if (tmp > end)
595 tmp = end;
596 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
597 if (error)
598 break;
599 nstart = tmp;
600 if (nstart < prev->vm_end)
601 nstart = prev->vm_end;
602 if (nstart >= end)
603 break;
604
605 vma = prev->vm_next;
606 if (!vma || vma->vm_start != nstart) {
607 error = -ENOMEM;
608 break;
609 }
610 }
611 return error;
612}
613
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800614/*
615 * __mm_populate - populate and/or mlock pages within a range of address space.
616 *
617 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
618 * flags. VMAs must be already marked with the desired vm_flags, and
619 * mmap_sem must not be held.
620 */
621int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
Michel Lespinassefed067d2011-01-13 15:46:10 -0800622{
623 struct mm_struct *mm = current->mm;
624 unsigned long end, nstart, nend;
625 struct vm_area_struct *vma = NULL;
Michel Lespinasse53a77062011-01-13 15:46:14 -0800626 int locked = 0;
Michel Lespinasse28a35712013-02-22 16:35:55 -0800627 long ret = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800628
629 VM_BUG_ON(start & ~PAGE_MASK);
630 VM_BUG_ON(len != PAGE_ALIGN(len));
631 end = start + len;
632
Michel Lespinassefed067d2011-01-13 15:46:10 -0800633 for (nstart = start; nstart < end; nstart = nend) {
634 /*
635 * We want to fault in pages for [nstart; end) address range.
636 * Find first corresponding VMA.
637 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800638 if (!locked) {
639 locked = 1;
640 down_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800641 vma = find_vma(mm, nstart);
Michel Lespinasse53a77062011-01-13 15:46:14 -0800642 } else if (nstart >= vma->vm_end)
Michel Lespinassefed067d2011-01-13 15:46:10 -0800643 vma = vma->vm_next;
644 if (!vma || vma->vm_start >= end)
645 break;
646 /*
647 * Set [nstart; nend) to intersection of desired address
648 * range with the first VMA. Also, skip undesirable VMA types.
649 */
650 nend = min(end, vma->vm_end);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700651 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
Michel Lespinassefed067d2011-01-13 15:46:10 -0800652 continue;
653 if (nstart < vma->vm_start)
654 nstart = vma->vm_start;
655 /*
Michel Lespinasse53a77062011-01-13 15:46:14 -0800656 * Now fault in a range of pages. __mlock_vma_pages_range()
657 * double checks the vma flags, so that it won't mlock pages
658 * if the vma was already munlocked.
Michel Lespinassefed067d2011-01-13 15:46:10 -0800659 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800660 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
661 if (ret < 0) {
662 if (ignore_errors) {
663 ret = 0;
664 continue; /* continue at next VMA */
665 }
Michel Lespinasse5fdb2002011-01-13 15:46:12 -0800666 ret = __mlock_posix_error_return(ret);
667 break;
668 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800669 nend = nstart + ret * PAGE_SIZE;
670 ret = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800671 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800672 if (locked)
673 up_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800674 return ret; /* 0 or negative error code */
675}
676
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100677SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 unsigned long locked;
680 unsigned long lock_limit;
681 int error = -ENOMEM;
682
683 if (!can_do_mlock())
684 return -EPERM;
685
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800686 lru_add_drain_all(); /* flush pagevec */
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 down_write(&current->mm->mmap_sem);
689 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
690 start &= PAGE_MASK;
691
692 locked = len >> PAGE_SHIFT;
693 locked += current->mm->locked_vm;
694
Jiri Slaby59e99e52010-03-05 13:41:44 -0800695 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 lock_limit >>= PAGE_SHIFT;
697
698 /* check against resource limits */
699 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
700 error = do_mlock(start, len, 1);
701 up_write(&current->mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800702 if (!error)
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800703 error = __mm_populate(start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 return error;
705}
706
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100707SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
709 int ret;
710
711 down_write(&current->mm->mmap_sem);
712 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
713 start &= PAGE_MASK;
714 ret = do_mlock(start, len, 0);
715 up_write(&current->mm->mmap_sem);
716 return ret;
717}
718
719static int do_mlockall(int flags)
720{
721 struct vm_area_struct * vma, * prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 if (flags & MCL_FUTURE)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700724 current->mm->def_flags |= VM_LOCKED;
Gerald Schaefer9977f0f2013-02-12 13:46:20 -0800725 else
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700726 current->mm->def_flags &= ~VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 if (flags == MCL_FUTURE)
728 goto out;
729
730 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900731 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Michel Lespinasse18693052013-02-22 16:32:46 -0800733 newflags = vma->vm_flags & ~VM_LOCKED;
734 if (flags & MCL_CURRENT)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700735 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
737 /* Ignore errors */
738 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
739 }
740out:
741 return 0;
742}
743
Heiko Carstens3480b252009-01-14 14:14:16 +0100744SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
746 unsigned long lock_limit;
747 int ret = -EINVAL;
748
749 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
750 goto out;
751
752 ret = -EPERM;
753 if (!can_do_mlock())
754 goto out;
755
Christoph Lameterdf9d6982011-10-31 17:09:35 -0700756 if (flags & MCL_CURRENT)
757 lru_add_drain_all(); /* flush pagevec */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 down_write(&current->mm->mmap_sem);
760
Jiri Slaby59e99e52010-03-05 13:41:44 -0800761 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 lock_limit >>= PAGE_SHIFT;
763
764 ret = -ENOMEM;
765 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
766 capable(CAP_IPC_LOCK))
767 ret = do_mlockall(flags);
768 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800769 if (!ret && (flags & MCL_CURRENT))
770 mm_populate(0, TASK_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771out:
772 return ret;
773}
774
Heiko Carstens3480b252009-01-14 14:14:16 +0100775SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
777 int ret;
778
779 down_write(&current->mm->mmap_sem);
780 ret = do_mlockall(0);
781 up_write(&current->mm->mmap_sem);
782 return ret;
783}
784
785/*
786 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
787 * shm segments) get accounted against the user_struct instead.
788 */
789static DEFINE_SPINLOCK(shmlock_user_lock);
790
791int user_shm_lock(size_t size, struct user_struct *user)
792{
793 unsigned long lock_limit, locked;
794 int allowed = 0;
795
796 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800797 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700798 if (lock_limit == RLIM_INFINITY)
799 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 lock_limit >>= PAGE_SHIFT;
801 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700802 if (!allowed &&
803 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 goto out;
805 get_uid(user);
806 user->locked_shm += locked;
807 allowed = 1;
808out:
809 spin_unlock(&shmlock_user_lock);
810 return allowed;
811}
812
813void user_shm_unlock(size_t size, struct user_struct *user)
814{
815 spin_lock(&shmlock_user_lock);
816 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
817 spin_unlock(&shmlock_user_lock);
818 free_uid(user);
819}