blob: f0b9ce572fc78ddbd0118db8aaa60d9b56086bf9 [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 */
Hugh Dickins73848b42009-12-14 17:59:22 -0800105void munlock_vma_page(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -0700106{
107 BUG_ON(!PageLocked(page));
108
Nick Piggin5344b7e2008-10-18 20:26:51 -0700109 if (TestClearPageMlocked(page)) {
David Rientjes8449d212012-10-08 16:34:06 -0700110 mod_zone_page_state(page_zone(page), NR_MLOCK,
111 -hpage_nr_pages(page));
Nick Piggin5344b7e2008-10-18 20:26:51 -0700112 if (!isolate_lru_page(page)) {
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700113 int ret = SWAP_AGAIN;
114
115 /*
116 * Optimization: if the page was mapped just once,
117 * that's our mapping and we don't need to check all the
118 * other vmas.
119 */
120 if (page_mapcount(page) > 1)
121 ret = try_to_munlock(page);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700122 /*
123 * did try_to_unlock() succeed or punt?
124 */
Hugh Dickins53f79ac2009-12-14 17:58:58 -0800125 if (ret != SWAP_MLOCK)
Nick Piggin5344b7e2008-10-18 20:26:51 -0700126 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
127
128 putback_lru_page(page);
129 } else {
130 /*
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800131 * Some other task has removed the page from the LRU.
132 * putback_lru_page() will take care of removing the
133 * page from the unevictable list, if necessary.
134 * vmscan [page_referenced()] will move the page back
135 * to the unevictable list if some other vma has it
136 * mlocked.
Nick Piggin5344b7e2008-10-18 20:26:51 -0700137 */
138 if (PageUnevictable(page))
139 count_vm_event(UNEVICTABLE_PGSTRANDED);
140 else
141 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
142 }
Nick Pigginb291f002008-10-18 20:26:44 -0700143 }
144}
145
Rik van Rielba470de2008-10-18 20:26:50 -0700146/**
Hugh Dickins408e82b2009-09-21 17:03:23 -0700147 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700148 * @vma: target vma
149 * @start: start address
150 * @end: end address
Nick Pigginb291f002008-10-18 20:26:44 -0700151 *
Hugh Dickins408e82b2009-09-21 17:03:23 -0700152 * This takes care of making the pages present too.
Nick Pigginb291f002008-10-18 20:26:44 -0700153 *
Rik van Rielba470de2008-10-18 20:26:50 -0700154 * return 0 on success, negative error code on error.
155 *
156 * vma->vm_mm->mmap_sem must be held for at least read.
Nick Pigginb291f002008-10-18 20:26:44 -0700157 */
Rik van Rielba470de2008-10-18 20:26:50 -0700158static long __mlock_vma_pages_range(struct vm_area_struct *vma,
Michel Lespinasse53a77062011-01-13 15:46:14 -0800159 unsigned long start, unsigned long end,
160 int *nonblocking)
Nick Pigginb291f002008-10-18 20:26:44 -0700161{
162 struct mm_struct *mm = vma->vm_mm;
163 unsigned long addr = start;
Nick Pigginb291f002008-10-18 20:26:44 -0700164 int nr_pages = (end - start) / PAGE_SIZE;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700165 int gup_flags;
Nick Pigginb291f002008-10-18 20:26:44 -0700166
Rik van Rielba470de2008-10-18 20:26:50 -0700167 VM_BUG_ON(start & ~PAGE_MASK);
168 VM_BUG_ON(end & ~PAGE_MASK);
169 VM_BUG_ON(start < vma->vm_start);
170 VM_BUG_ON(end > vma->vm_end);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700171 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
Rik van Rielba470de2008-10-18 20:26:50 -0700172
Linus Torvaldsa1fde082011-05-04 21:30:28 -0700173 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
Michel Lespinasse5ecfda02011-01-13 15:46:09 -0800174 /*
175 * We want to touch writable mappings with a write fault in order
176 * to break COW, except for shared mappings because these don't COW
177 * and we would not want to dirty them for nothing.
178 */
179 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700180 gup_flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700181
Michel Lespinassefdf4c582011-01-31 17:03:41 -0800182 /*
183 * We want mlock to succeed for regions that have any permissions
184 * other than PROT_NONE.
185 */
186 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
187 gup_flags |= FOLL_FORCE;
188
Michel Lespinasse53a77062011-01-13 15:46:14 -0800189 return __get_user_pages(current, mm, addr, nr_pages, gup_flags,
190 NULL, NULL, nonblocking);
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700191}
192
193/*
194 * convert get_user_pages() return value to posix mlock() error
195 */
196static int __mlock_posix_error_return(long retval)
197{
198 if (retval == -EFAULT)
199 retval = -ENOMEM;
200 else if (retval == -ENOMEM)
201 retval = -EAGAIN;
202 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700203}
204
Rik van Rielba470de2008-10-18 20:26:50 -0700205/**
206 * mlock_vma_pages_range() - mlock pages in specified vma range.
207 * @vma - the vma containing the specfied address range
208 * @start - starting address in @vma to mlock
209 * @end - end address [+1] in @vma to mlock
210 *
211 * For mmap()/mremap()/expansion of mlocked vma.
212 *
213 * return 0 on success for "normal" vmas.
214 *
215 * return number of pages [> 0] to be removed from locked_vm on success
216 * of "special" vmas.
Nick Pigginb291f002008-10-18 20:26:44 -0700217 */
Rik van Rielba470de2008-10-18 20:26:50 -0700218long mlock_vma_pages_range(struct vm_area_struct *vma,
Nick Pigginb291f002008-10-18 20:26:44 -0700219 unsigned long start, unsigned long end)
220{
221 int nr_pages = (end - start) / PAGE_SIZE;
222 BUG_ON(!(vma->vm_flags & VM_LOCKED));
223
224 /*
225 * filter unlockable vmas
226 */
227 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
228 goto no_mlock;
229
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -0700230 if (!((vma->vm_flags & VM_DONTEXPAND) ||
Nick Pigginb291f002008-10-18 20:26:44 -0700231 is_vm_hugetlb_page(vma) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400232 vma == get_gate_vma(current->mm))) {
Rik van Rielba470de2008-10-18 20:26:50 -0700233
Michel Lespinasse53a77062011-01-13 15:46:14 -0800234 __mlock_vma_pages_range(vma, start, end, NULL);
Hugh Dickinsd5b56232009-02-08 20:56:58 +0000235
236 /* Hide errors from mmap() and other callers */
237 return 0;
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700238 }
Nick Pigginb291f002008-10-18 20:26:44 -0700239
240 /*
241 * User mapped kernel pages or huge pages:
242 * make these pages present to populate the ptes, but
243 * fall thru' to reset VM_LOCKED--no need to unlock, and
244 * return nr_pages so these don't get counted against task's
245 * locked limit. huge pages are already counted against
246 * locked vm limit.
247 */
248 make_pages_present(start, end);
249
250no_mlock:
251 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
Rik van Rielba470de2008-10-18 20:26:50 -0700252 return nr_pages; /* error or pages NOT mlocked */
Nick Pigginb291f002008-10-18 20:26:44 -0700253}
254
Nick Pigginb291f002008-10-18 20:26:44 -0700255/*
Rik van Rielba470de2008-10-18 20:26:50 -0700256 * munlock_vma_pages_range() - munlock all pages in the vma range.'
257 * @vma - vma containing range to be munlock()ed.
258 * @start - start address in @vma of the range
259 * @end - end of range in @vma.
260 *
261 * For mremap(), munmap() and exit().
262 *
263 * Called with @vma VM_LOCKED.
264 *
265 * Returns with VM_LOCKED cleared. Callers must be prepared to
266 * deal with this.
267 *
268 * We don't save and restore VM_LOCKED here because pages are
269 * still on lru. In unmap path, pages might be scanned by reclaim
270 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
271 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700272 */
Rik van Rielba470de2008-10-18 20:26:50 -0700273void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700274 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700275{
Hugh Dickins408e82b2009-09-21 17:03:23 -0700276 unsigned long addr;
277
278 lru_add_drain();
Nick Pigginb291f002008-10-18 20:26:44 -0700279 vma->vm_flags &= ~VM_LOCKED;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700280
281 for (addr = start; addr < end; addr += PAGE_SIZE) {
Hugh Dickins6e919712009-09-21 17:03:32 -0700282 struct page *page;
283 /*
284 * Although FOLL_DUMP is intended for get_dump_page(),
285 * it just so happens that its special treatment of the
286 * ZERO_PAGE (returning an error instead of doing get_page)
287 * suits munlock very well (and if somehow an abnormal page
288 * has sneaked into the range, we won't oops here: great).
289 */
290 page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP);
291 if (page && !IS_ERR(page)) {
Hugh Dickins408e82b2009-09-21 17:03:23 -0700292 lock_page(page);
Hugh Dickinse6c509f2012-10-08 16:33:19 -0700293 munlock_vma_page(page);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700294 unlock_page(page);
295 put_page(page);
296 }
297 cond_resched();
298 }
Nick Pigginb291f002008-10-18 20:26:44 -0700299}
300
301/*
302 * mlock_fixup - handle mlock[all]/munlock[all] requests.
303 *
304 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
305 * munlock is a no-op. However, for some special vmas, we go ahead and
306 * populate the ptes via make_pages_present().
307 *
308 * For vmas that pass the filters, merge/split as appropriate.
309 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900311 unsigned long start, unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Nick Pigginb291f002008-10-18 20:26:44 -0700313 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700315 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 int ret = 0;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900317 int lock = !!(newflags & VM_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Michel Lespinassefed067d2011-01-13 15:46:10 -0800319 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
Stephen Wilson31db58b2011-03-13 15:49:15 -0400320 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
Nick Pigginb291f002008-10-18 20:26:44 -0700321 goto out; /* don't set VM_LOCKED, don't count */
322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
324 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
325 vma->vm_file, pgoff, vma_policy(vma));
326 if (*prev) {
327 vma = *prev;
328 goto success;
329 }
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 if (start != vma->vm_start) {
332 ret = split_vma(mm, vma, start, 1);
333 if (ret)
334 goto out;
335 }
336
337 if (end != vma->vm_end) {
338 ret = split_vma(mm, vma, end, 0);
339 if (ret)
340 goto out;
341 }
342
343success:
344 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700345 * Keep track of amount of locked VM.
346 */
347 nr_pages = (end - start) >> PAGE_SHIFT;
348 if (!lock)
349 nr_pages = -nr_pages;
350 mm->locked_vm += nr_pages;
351
352 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 * vm_flags is protected by the mmap_sem held in write mode.
354 * It's okay if try_to_unmap_one unmaps a page just after we
Nick Pigginb291f002008-10-18 20:26:44 -0700355 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
Michel Lespinassefed067d2011-01-13 15:46:10 -0800358 if (lock)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700359 vma->vm_flags = newflags;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800360 else
Hugh Dickins408e82b2009-09-21 17:03:23 -0700361 munlock_vma_pages_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363out:
Nick Pigginb291f002008-10-18 20:26:44 -0700364 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 return ret;
366}
367
368static int do_mlock(unsigned long start, size_t len, int on)
369{
370 unsigned long nstart, end, tmp;
371 struct vm_area_struct * vma, * prev;
372 int error;
373
Michel Lespinassefed067d2011-01-13 15:46:10 -0800374 VM_BUG_ON(start & ~PAGE_MASK);
375 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 end = start + len;
377 if (end < start)
378 return -EINVAL;
379 if (end == start)
380 return 0;
Linus Torvalds097d5912012-03-06 18:23:36 -0800381 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (!vma || vma->vm_start > start)
383 return -ENOMEM;
384
Linus Torvalds097d5912012-03-06 18:23:36 -0800385 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 if (start > vma->vm_start)
387 prev = vma;
388
389 for (nstart = start ; ; ) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900390 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
393
394 newflags = vma->vm_flags | VM_LOCKED;
395 if (!on)
396 newflags &= ~VM_LOCKED;
397
398 tmp = vma->vm_end;
399 if (tmp > end)
400 tmp = end;
401 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
402 if (error)
403 break;
404 nstart = tmp;
405 if (nstart < prev->vm_end)
406 nstart = prev->vm_end;
407 if (nstart >= end)
408 break;
409
410 vma = prev->vm_next;
411 if (!vma || vma->vm_start != nstart) {
412 error = -ENOMEM;
413 break;
414 }
415 }
416 return error;
417}
418
Michel Lespinassefed067d2011-01-13 15:46:10 -0800419static int do_mlock_pages(unsigned long start, size_t len, int ignore_errors)
420{
421 struct mm_struct *mm = current->mm;
422 unsigned long end, nstart, nend;
423 struct vm_area_struct *vma = NULL;
Michel Lespinasse53a77062011-01-13 15:46:14 -0800424 int locked = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800425 int ret = 0;
426
427 VM_BUG_ON(start & ~PAGE_MASK);
428 VM_BUG_ON(len != PAGE_ALIGN(len));
429 end = start + len;
430
Michel Lespinassefed067d2011-01-13 15:46:10 -0800431 for (nstart = start; nstart < end; nstart = nend) {
432 /*
433 * We want to fault in pages for [nstart; end) address range.
434 * Find first corresponding VMA.
435 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800436 if (!locked) {
437 locked = 1;
438 down_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800439 vma = find_vma(mm, nstart);
Michel Lespinasse53a77062011-01-13 15:46:14 -0800440 } else if (nstart >= vma->vm_end)
Michel Lespinassefed067d2011-01-13 15:46:10 -0800441 vma = vma->vm_next;
442 if (!vma || vma->vm_start >= end)
443 break;
444 /*
445 * Set [nstart; nend) to intersection of desired address
446 * range with the first VMA. Also, skip undesirable VMA types.
447 */
448 nend = min(end, vma->vm_end);
449 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
450 continue;
451 if (nstart < vma->vm_start)
452 nstart = vma->vm_start;
453 /*
Michel Lespinasse53a77062011-01-13 15:46:14 -0800454 * Now fault in a range of pages. __mlock_vma_pages_range()
455 * double checks the vma flags, so that it won't mlock pages
456 * if the vma was already munlocked.
Michel Lespinassefed067d2011-01-13 15:46:10 -0800457 */
Michel Lespinasse53a77062011-01-13 15:46:14 -0800458 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
459 if (ret < 0) {
460 if (ignore_errors) {
461 ret = 0;
462 continue; /* continue at next VMA */
463 }
Michel Lespinasse5fdb2002011-01-13 15:46:12 -0800464 ret = __mlock_posix_error_return(ret);
465 break;
466 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800467 nend = nstart + ret * PAGE_SIZE;
468 ret = 0;
Michel Lespinassefed067d2011-01-13 15:46:10 -0800469 }
Michel Lespinasse53a77062011-01-13 15:46:14 -0800470 if (locked)
471 up_read(&mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800472 return ret; /* 0 or negative error code */
473}
474
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100475SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 unsigned long locked;
478 unsigned long lock_limit;
479 int error = -ENOMEM;
480
481 if (!can_do_mlock())
482 return -EPERM;
483
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800484 lru_add_drain_all(); /* flush pagevec */
485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 down_write(&current->mm->mmap_sem);
487 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
488 start &= PAGE_MASK;
489
490 locked = len >> PAGE_SHIFT;
491 locked += current->mm->locked_vm;
492
Jiri Slaby59e99e52010-03-05 13:41:44 -0800493 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 lock_limit >>= PAGE_SHIFT;
495
496 /* check against resource limits */
497 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
498 error = do_mlock(start, len, 1);
499 up_write(&current->mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800500 if (!error)
501 error = do_mlock_pages(start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return error;
503}
504
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100505SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
507 int ret;
508
509 down_write(&current->mm->mmap_sem);
510 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
511 start &= PAGE_MASK;
512 ret = do_mlock(start, len, 0);
513 up_write(&current->mm->mmap_sem);
514 return ret;
515}
516
517static int do_mlockall(int flags)
518{
519 struct vm_area_struct * vma, * prev = NULL;
520 unsigned int def_flags = 0;
521
522 if (flags & MCL_FUTURE)
523 def_flags = VM_LOCKED;
524 current->mm->def_flags = def_flags;
525 if (flags == MCL_FUTURE)
526 goto out;
527
528 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900529 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531 newflags = vma->vm_flags | VM_LOCKED;
532 if (!(flags & MCL_CURRENT))
533 newflags &= ~VM_LOCKED;
534
535 /* Ignore errors */
536 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
537 }
538out:
539 return 0;
540}
541
Heiko Carstens3480b252009-01-14 14:14:16 +0100542SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
544 unsigned long lock_limit;
545 int ret = -EINVAL;
546
547 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
548 goto out;
549
550 ret = -EPERM;
551 if (!can_do_mlock())
552 goto out;
553
Christoph Lameterdf9d6982011-10-31 17:09:35 -0700554 if (flags & MCL_CURRENT)
555 lru_add_drain_all(); /* flush pagevec */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 down_write(&current->mm->mmap_sem);
558
Jiri Slaby59e99e52010-03-05 13:41:44 -0800559 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 lock_limit >>= PAGE_SHIFT;
561
562 ret = -ENOMEM;
563 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
564 capable(CAP_IPC_LOCK))
565 ret = do_mlockall(flags);
566 up_write(&current->mm->mmap_sem);
Michel Lespinassefed067d2011-01-13 15:46:10 -0800567 if (!ret && (flags & MCL_CURRENT)) {
568 /* Ignore errors */
569 do_mlock_pages(0, TASK_SIZE, 1);
570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571out:
572 return ret;
573}
574
Heiko Carstens3480b252009-01-14 14:14:16 +0100575SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
577 int ret;
578
579 down_write(&current->mm->mmap_sem);
580 ret = do_mlockall(0);
581 up_write(&current->mm->mmap_sem);
582 return ret;
583}
584
585/*
586 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
587 * shm segments) get accounted against the user_struct instead.
588 */
589static DEFINE_SPINLOCK(shmlock_user_lock);
590
591int user_shm_lock(size_t size, struct user_struct *user)
592{
593 unsigned long lock_limit, locked;
594 int allowed = 0;
595
596 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800597 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700598 if (lock_limit == RLIM_INFINITY)
599 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 lock_limit >>= PAGE_SHIFT;
601 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700602 if (!allowed &&
603 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 goto out;
605 get_uid(user);
606 user->locked_shm += locked;
607 allowed = 1;
608out:
609 spin_unlock(&shmlock_user_lock);
610 return allowed;
611}
612
613void user_shm_unlock(size_t size, struct user_struct *user)
614{
615 spin_lock(&shmlock_user_lock);
616 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
617 spin_unlock(&shmlock_user_lock);
618 free_uid(user);
619}