blob: 79b7cf7d1bca72cee9babfb60e21a38799c8eba1 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/mempolicy.h>
15#include <linux/syscalls.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040016#include <linux/sched.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040017#include <linux/export.h>
Nick Pigginb291f002008-10-18 20:26:44 -070018#include <linux/rmap.h>
19#include <linux/mmzone.h>
20#include <linux/hugetlb.h>
21
22#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040024int can_do_mlock(void)
25{
26 if (capable(CAP_IPC_LOCK))
27 return 1;
Jiri Slaby59e99e52010-03-05 13:41:44 -080028 if (rlimit(RLIMIT_MEMLOCK) != 0)
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040029 return 1;
30 return 0;
31}
32EXPORT_SYMBOL(can_do_mlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Nick Pigginb291f002008-10-18 20:26:44 -070034/*
35 * Mlocked pages are marked with PageMlocked() flag for efficient testing
36 * in vmscan and, possibly, the fault path; and to support semi-accurate
37 * statistics.
38 *
39 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
40 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
41 * The unevictable list is an LRU sibling list to the [in]active lists.
42 * PageUnevictable is set to indicate the unevictable state.
43 *
44 * When lazy mlocking via vmscan, it is important to ensure that the
45 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
46 * may have mlocked a page that is being munlocked. So lazy mlock must take
47 * the mmap_sem for read, and verify that the vma really is locked
48 * (see mm/rmap.c).
49 */
50
51/*
52 * LRU accounting for clear_page_mlock()
53 */
Hugh Dickinse6c509f2012-10-08 16:33:19 -070054void clear_page_mlock(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -070055{
Hugh Dickinse6c509f2012-10-08 16:33:19 -070056 if (!TestClearPageMlocked(page))
Nick Pigginb291f002008-10-18 20:26:44 -070057 return;
Nick Pigginb291f002008-10-18 20:26:44 -070058
David Rientjes8449d212012-10-08 16:34:06 -070059 mod_zone_page_state(page_zone(page), NR_MLOCK,
60 -hpage_nr_pages(page));
Nick Piggin5344b7e2008-10-18 20:26:51 -070061 count_vm_event(UNEVICTABLE_PGCLEARED);
Nick Pigginb291f002008-10-18 20:26:44 -070062 if (!isolate_lru_page(page)) {
63 putback_lru_page(page);
64 } else {
65 /*
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -080066 * We lost the race. the page already moved to evictable list.
Nick Pigginb291f002008-10-18 20:26:44 -070067 */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -080068 if (PageUnevictable(page))
Nick Piggin5344b7e2008-10-18 20:26:51 -070069 count_vm_event(UNEVICTABLE_PGSTRANDED);
Nick Pigginb291f002008-10-18 20:26:44 -070070 }
71}
72
73/*
74 * Mark page as mlocked if not already.
75 * If page on LRU, isolate and putback to move to unevictable list.
76 */
77void mlock_vma_page(struct page *page)
78{
79 BUG_ON(!PageLocked(page));
80
Nick Piggin5344b7e2008-10-18 20:26:51 -070081 if (!TestSetPageMlocked(page)) {
David Rientjes8449d212012-10-08 16:34:06 -070082 mod_zone_page_state(page_zone(page), NR_MLOCK,
83 hpage_nr_pages(page));
Nick Piggin5344b7e2008-10-18 20:26:51 -070084 count_vm_event(UNEVICTABLE_PGMLOCKED);
85 if (!isolate_lru_page(page))
86 putback_lru_page(page);
87 }
Nick Pigginb291f002008-10-18 20:26:44 -070088}
89
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -080090/**
91 * munlock_vma_page - munlock a vma page
92 * @page - page to be unlocked
Nick Pigginb291f002008-10-18 20:26:44 -070093 *
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -080094 * called from munlock()/munmap() path with page supposedly on the LRU.
95 * When we munlock a page, because the vma where we found the page is being
96 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
97 * page locked so that we can leave it on the unevictable lru list and not
98 * bother vmscan with it. However, to walk the page's rmap list in
99 * try_to_munlock() we must isolate the page from the LRU. If some other
100 * task has removed the page from the LRU, we won't be able to do that.
101 * So we clear the PageMlocked as we might not get another chance. If we
102 * can't isolate the page, we leave it for putback_lru_page() and vmscan
103 * [page_referenced()/try_to_unmap()] to deal with.
Nick Pigginb291f002008-10-18 20:26:44 -0700104 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800105unsigned int munlock_vma_page(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -0700106{
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800107 unsigned int page_mask = 0;
108
Nick Pigginb291f002008-10-18 20:26:44 -0700109 BUG_ON(!PageLocked(page));
110
Nick Piggin5344b7e2008-10-18 20:26:51 -0700111 if (TestClearPageMlocked(page)) {
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800112 unsigned int nr_pages = hpage_nr_pages(page);
113 mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
114 page_mask = nr_pages - 1;
Nick Piggin5344b7e2008-10-18 20:26:51 -0700115 if (!isolate_lru_page(page)) {
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700116 int ret = SWAP_AGAIN;
117
118 /*
119 * Optimization: if the page was mapped just once,
120 * that's our mapping and we don't need to check all the
121 * other vmas.
122 */
123 if (page_mapcount(page) > 1)
124 ret = try_to_munlock(page);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700125 /*
126 * did try_to_unlock() succeed or punt?
127 */
Hugh Dickins53f79ac2009-12-14 17:58:58 -0800128 if (ret != SWAP_MLOCK)
Nick Piggin5344b7e2008-10-18 20:26:51 -0700129 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
130
131 putback_lru_page(page);
132 } else {
133 /*
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800134 * Some other task has removed the page from the LRU.
135 * putback_lru_page() will take care of removing the
136 * page from the unevictable list, if necessary.
137 * vmscan [page_referenced()] will move the page back
138 * to the unevictable list if some other vma has it
139 * mlocked.
Nick Piggin5344b7e2008-10-18 20:26:51 -0700140 */
141 if (PageUnevictable(page))
142 count_vm_event(UNEVICTABLE_PGSTRANDED);
143 else
144 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
145 }
Nick Pigginb291f002008-10-18 20:26:44 -0700146 }
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800147
148 return page_mask;
Nick Pigginb291f002008-10-18 20:26:44 -0700149}
150
Rik van Rielba470de2008-10-18 20:26:50 -0700151/**
Hugh Dickins408e82b2009-09-21 17:03:23 -0700152 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700153 * @vma: target vma
154 * @start: start address
155 * @end: end address
Nick Pigginb291f002008-10-18 20:26:44 -0700156 *
Hugh Dickins408e82b2009-09-21 17:03:23 -0700157 * This takes care of making the pages present too.
Nick Pigginb291f002008-10-18 20:26:44 -0700158 *
Rik van Rielba470de2008-10-18 20:26:50 -0700159 * return 0 on success, negative error code on error.
160 *
161 * vma->vm_mm->mmap_sem must be held for at least read.
Nick Pigginb291f002008-10-18 20:26:44 -0700162 */
Michel Lespinassecea10a12013-02-22 16:32:44 -0800163long __mlock_vma_pages_range(struct vm_area_struct *vma,
164 unsigned long start, unsigned long end, int *nonblocking)
Nick Pigginb291f002008-10-18 20:26:44 -0700165{
166 struct mm_struct *mm = vma->vm_mm;
Michel Lespinasse28a35712013-02-22 16:35:55 -0800167 unsigned long nr_pages = (end - start) / PAGE_SIZE;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700168 int gup_flags;
Nick Pigginb291f002008-10-18 20:26:44 -0700169
Rik van Rielba470de2008-10-18 20:26:50 -0700170 VM_BUG_ON(start & ~PAGE_MASK);
171 VM_BUG_ON(end & ~PAGE_MASK);
172 VM_BUG_ON(start < vma->vm_start);
173 VM_BUG_ON(end > vma->vm_end);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700174 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
Rik van Rielba470de2008-10-18 20:26:50 -0700175
Linus Torvaldsa1fde082011-05-04 21:30:28 -0700176 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
Michel Lespinasse5ecfda02011-01-13 15:46:09 -0800177 /*
178 * We want to touch writable mappings with a write fault in order
179 * to break COW, except for shared mappings because these don't COW
180 * and we would not want to dirty them for nothing.
181 */
182 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700183 gup_flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700184
Michel Lespinassefdf4c582011-01-31 17:03:41 -0800185 /*
186 * We want mlock to succeed for regions that have any permissions
187 * other than PROT_NONE.
188 */
189 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
190 gup_flags |= FOLL_FORCE;
191
Johannes Weiner4805b022013-02-22 16:35:20 -0800192 /*
193 * We made sure addr is within a VMA, so the following will
194 * not result in a stack expansion that recurses back here.
195 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800196 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
Michel Lespinasse53a77062011-01-13 15:46:14 -0800197 NULL, NULL, nonblocking);
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700198}
199
200/*
201 * convert get_user_pages() return value to posix mlock() error
202 */
203static int __mlock_posix_error_return(long retval)
204{
205 if (retval == -EFAULT)
206 retval = -ENOMEM;
207 else if (retval == -ENOMEM)
208 retval = -EAGAIN;
209 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700210}
211
Nick Pigginb291f002008-10-18 20:26:44 -0700212/*
Rik van Rielba470de2008-10-18 20:26:50 -0700213 * munlock_vma_pages_range() - munlock all pages in the vma range.'
214 * @vma - vma containing range to be munlock()ed.
215 * @start - start address in @vma of the range
216 * @end - end of range in @vma.
217 *
218 * For mremap(), munmap() and exit().
219 *
220 * Called with @vma VM_LOCKED.
221 *
222 * Returns with VM_LOCKED cleared. Callers must be prepared to
223 * deal with this.
224 *
225 * We don't save and restore VM_LOCKED here because pages are
226 * still on lru. In unmap path, pages might be scanned by reclaim
227 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
228 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700229 */
Rik van Rielba470de2008-10-18 20:26:50 -0700230void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700231 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700232{
233 vma->vm_flags &= ~VM_LOCKED;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700234
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800235 while (start < end) {
Hugh Dickins6e919712009-09-21 17:03:32 -0700236 struct page *page;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800237 unsigned int page_mask, page_increm;
238
Hugh Dickins6e919712009-09-21 17:03:32 -0700239 /*
240 * Although FOLL_DUMP is intended for get_dump_page(),
241 * it just so happens that its special treatment of the
242 * ZERO_PAGE (returning an error instead of doing get_page)
243 * suits munlock very well (and if somehow an abnormal page
244 * has sneaked into the range, we won't oops here: great).
245 */
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800246 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
247 &page_mask);
Hugh Dickins6e919712009-09-21 17:03:32 -0700248 if (page && !IS_ERR(page)) {
Hugh Dickins408e82b2009-09-21 17:03:23 -0700249 lock_page(page);
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800250 lru_add_drain();
251 /*
252 * Any THP page found by follow_page_mask() may have
253 * gotten split before reaching munlock_vma_page(),
254 * so we need to recompute the page_mask here.
255 */
256 page_mask = munlock_vma_page(page);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700257 unlock_page(page);
258 put_page(page);
259 }
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800260 page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);
261 start += page_increm * PAGE_SIZE;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700262 cond_resched();
263 }
Nick Pigginb291f002008-10-18 20:26:44 -0700264}
265
266/*
267 * mlock_fixup - handle mlock[all]/munlock[all] requests.
268 *
269 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
270 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800271 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700272 *
273 * For vmas that pass the filters, merge/split as appropriate.
274 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900276 unsigned long start, unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
Nick Pigginb291f002008-10-18 20:26:44 -0700278 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700280 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 int ret = 0;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900282 int lock = !!(newflags & VM_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
Michel Lespinassefed067d2011-01-13 15:46:10 -0800284 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400285 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
Nick Pigginb291f002008-10-18 20:26:44 -0700286 goto out; /* don't set VM_LOCKED, don't count */
287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
289 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
290 vma->vm_file, pgoff, vma_policy(vma));
291 if (*prev) {
292 vma = *prev;
293 goto success;
294 }
295
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 if (start != vma->vm_start) {
297 ret = split_vma(mm, vma, start, 1);
298 if (ret)
299 goto out;
300 }
301
302 if (end != vma->vm_end) {
303 ret = split_vma(mm, vma, end, 0);
304 if (ret)
305 goto out;
306 }
307
308success:
309 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700310 * Keep track of amount of locked VM.
311 */
312 nr_pages = (end - start) >> PAGE_SHIFT;
313 if (!lock)
314 nr_pages = -nr_pages;
315 mm->locked_vm += nr_pages;
316
317 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * vm_flags is protected by the mmap_sem held in write mode.
319 * It's okay if try_to_unmap_one unmaps a page just after we
Nick Pigginb291f002008-10-18 20:26:44 -0700320 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Michel Lespinassefed067d2011-01-13 15:46:10 -0800323 if (lock)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700324 vma->vm_flags = newflags;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800325 else
Hugh Dickins408e82b2009-09-21 17:03:23 -0700326 munlock_vma_pages_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328out:
Nick Pigginb291f002008-10-18 20:26:44 -0700329 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 return ret;
331}
332
333static int do_mlock(unsigned long start, size_t len, int on)
334{
335 unsigned long nstart, end, tmp;
336 struct vm_area_struct * vma, * prev;
337 int error;
338
Michel Lespinassefed067d2011-01-13 15:46:10 -0800339 VM_BUG_ON(start & ~PAGE_MASK);
340 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 end = start + len;
342 if (end < start)
343 return -EINVAL;
344 if (end == start)
345 return 0;
Linus Torvalds097d5912012-03-06 18:23:36 -0800346 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 if (!vma || vma->vm_start > start)
348 return -ENOMEM;
349
Linus Torvalds097d5912012-03-06 18:23:36 -0800350 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (start > vma->vm_start)
352 prev = vma;
353
354 for (nstart = start ; ; ) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900355 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
357 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
358
Michel Lespinasse18693052013-02-22 16:32:46 -0800359 newflags = vma->vm_flags & ~VM_LOCKED;
360 if (on)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700361 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 tmp = vma->vm_end;
364 if (tmp > end)
365 tmp = end;
366 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
367 if (error)
368 break;
369 nstart = tmp;
370 if (nstart < prev->vm_end)
371 nstart = prev->vm_end;
372 if (nstart >= end)
373 break;
374
375 vma = prev->vm_next;
376 if (!vma || vma->vm_start != nstart) {
377 error = -ENOMEM;
378 break;
379 }
380 }
381 return error;
382}
383
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800384/*
385 * __mm_populate - populate and/or mlock pages within a range of address space.
386 *
387 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
388 * flags. VMAs must be already marked with the desired vm_flags, and
389 * mmap_sem must not be held.
390 */
391int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
Michel Lespinassefed067d2011-01-13 15:46:10 -0800392{
393 struct mm_struct *mm = current->mm;
394 unsigned long end, nstart, nend;
395 struct vm_area_struct *vma = NULL;
Michel Lespinasse53a77062011-01-13 15:46:14 -0800396 int locked = 0;
Michel Lespinasse28a35712013-02-22 16:35:55 -0800397 long ret = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800398
399 VM_BUG_ON(start & ~PAGE_MASK);
400 VM_BUG_ON(len != PAGE_ALIGN(len));
401 end = start + len;
402
Michel Lespinassefed067d2011-01-13 15:46:10 -0800403 for (nstart = start; nstart < end; nstart = nend) {
404 /*
405 * We want to fault in pages for [nstart; end) address range.
406 * Find first corresponding VMA.
407 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800408 if (!locked) {
409 locked = 1;
410 down_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800411 vma = find_vma(mm, nstart);
Michel Lespinasse53a77062011-01-13 15:46:14 -0800412 } else if (nstart >= vma->vm_end)
Michel Lespinassefed067d2011-01-13 15:46:10 -0800413 vma = vma->vm_next;
414 if (!vma || vma->vm_start >= end)
415 break;
416 /*
417 * Set [nstart; nend) to intersection of desired address
418 * range with the first VMA. Also, skip undesirable VMA types.
419 */
420 nend = min(end, vma->vm_end);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700421 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
Michel Lespinassefed067d2011-01-13 15:46:10 -0800422 continue;
423 if (nstart < vma->vm_start)
424 nstart = vma->vm_start;
425 /*
Michel Lespinasse53a77062011-01-13 15:46:14 -0800426 * Now fault in a range of pages. __mlock_vma_pages_range()
427 * double checks the vma flags, so that it won't mlock pages
428 * if the vma was already munlocked.
Michel Lespinassefed067d2011-01-13 15:46:10 -0800429 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800430 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
431 if (ret < 0) {
432 if (ignore_errors) {
433 ret = 0;
434 continue; /* continue at next VMA */
435 }
Michel Lespinasse5fdb2002011-01-13 15:46:12 -0800436 ret = __mlock_posix_error_return(ret);
437 break;
438 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800439 nend = nstart + ret * PAGE_SIZE;
440 ret = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800441 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800442 if (locked)
443 up_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800444 return ret; /* 0 or negative error code */
445}
446
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100447SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
449 unsigned long locked;
450 unsigned long lock_limit;
451 int error = -ENOMEM;
452
453 if (!can_do_mlock())
454 return -EPERM;
455
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800456 lru_add_drain_all(); /* flush pagevec */
457
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 down_write(&current->mm->mmap_sem);
459 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
460 start &= PAGE_MASK;
461
462 locked = len >> PAGE_SHIFT;
463 locked += current->mm->locked_vm;
464
Jiri Slaby59e99e52010-03-05 13:41:44 -0800465 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 lock_limit >>= PAGE_SHIFT;
467
468 /* check against resource limits */
469 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
470 error = do_mlock(start, len, 1);
471 up_write(&current->mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800472 if (!error)
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800473 error = __mm_populate(start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 return error;
475}
476
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100477SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
479 int ret;
480
481 down_write(&current->mm->mmap_sem);
482 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
483 start &= PAGE_MASK;
484 ret = do_mlock(start, len, 0);
485 up_write(&current->mm->mmap_sem);
486 return ret;
487}
488
489static int do_mlockall(int flags)
490{
491 struct vm_area_struct * vma, * prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493 if (flags & MCL_FUTURE)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700494 current->mm->def_flags |= VM_LOCKED;
Gerald Schaefer9977f0f2013-02-12 13:46:20 -0800495 else
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700496 current->mm->def_flags &= ~VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 if (flags == MCL_FUTURE)
498 goto out;
499
500 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900501 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Michel Lespinasse18693052013-02-22 16:32:46 -0800503 newflags = vma->vm_flags & ~VM_LOCKED;
504 if (flags & MCL_CURRENT)
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700505 newflags |= VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
507 /* Ignore errors */
508 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
509 }
510out:
511 return 0;
512}
513
Heiko Carstens3480b252009-01-14 14:14:16 +0100514SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515{
516 unsigned long lock_limit;
517 int ret = -EINVAL;
518
519 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
520 goto out;
521
522 ret = -EPERM;
523 if (!can_do_mlock())
524 goto out;
525
Christoph Lameterdf9d6982011-10-31 17:09:35 -0700526 if (flags & MCL_CURRENT)
527 lru_add_drain_all(); /* flush pagevec */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 down_write(&current->mm->mmap_sem);
530
Jiri Slaby59e99e52010-03-05 13:41:44 -0800531 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 lock_limit >>= PAGE_SHIFT;
533
534 ret = -ENOMEM;
535 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
536 capable(CAP_IPC_LOCK))
537 ret = do_mlockall(flags);
538 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800539 if (!ret && (flags & MCL_CURRENT))
540 mm_populate(0, TASK_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541out:
542 return ret;
543}
544
Heiko Carstens3480b252009-01-14 14:14:16 +0100545SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 int ret;
548
549 down_write(&current->mm->mmap_sem);
550 ret = do_mlockall(0);
551 up_write(&current->mm->mmap_sem);
552 return ret;
553}
554
555/*
556 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
557 * shm segments) get accounted against the user_struct instead.
558 */
559static DEFINE_SPINLOCK(shmlock_user_lock);
560
561int user_shm_lock(size_t size, struct user_struct *user)
562{
563 unsigned long lock_limit, locked;
564 int allowed = 0;
565
566 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800567 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700568 if (lock_limit == RLIM_INFINITY)
569 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 lock_limit >>= PAGE_SHIFT;
571 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700572 if (!allowed &&
573 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 goto out;
575 get_uid(user);
576 user->locked_shm += locked;
577 allowed = 1;
578out:
579 spin_unlock(&shmlock_user_lock);
580 return allowed;
581}
582
583void user_shm_unlock(size_t size, struct user_struct *user)
584{
585 spin_lock(&shmlock_user_lock);
586 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
587 spin_unlock(&shmlock_user_lock);
588 free_uid(user);
589}