blob: 8f4e2dfceec1c88e796a6bafb78e7422f06bdb72 [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>
17#include <linux/module.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 */
54void __clear_page_mlock(struct page *page)
55{
56 VM_BUG_ON(!PageLocked(page));
57
58 if (!page->mapping) { /* truncated ? */
59 return;
60 }
61
Nick Piggin5344b7e2008-10-18 20:26:51 -070062 dec_zone_page_state(page, NR_MLOCK);
63 count_vm_event(UNEVICTABLE_PGCLEARED);
Nick Pigginb291f002008-10-18 20:26:44 -070064 if (!isolate_lru_page(page)) {
65 putback_lru_page(page);
66 } else {
67 /*
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -080068 * We lost the race. the page already moved to evictable list.
Nick Pigginb291f002008-10-18 20:26:44 -070069 */
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -080070 if (PageUnevictable(page))
Nick Piggin5344b7e2008-10-18 20:26:51 -070071 count_vm_event(UNEVICTABLE_PGSTRANDED);
Nick Pigginb291f002008-10-18 20:26:44 -070072 }
73}
74
75/*
76 * Mark page as mlocked if not already.
77 * If page on LRU, isolate and putback to move to unevictable list.
78 */
79void mlock_vma_page(struct page *page)
80{
81 BUG_ON(!PageLocked(page));
82
Nick Piggin5344b7e2008-10-18 20:26:51 -070083 if (!TestSetPageMlocked(page)) {
84 inc_zone_page_state(page, NR_MLOCK);
85 count_vm_event(UNEVICTABLE_PGMLOCKED);
86 if (!isolate_lru_page(page))
87 putback_lru_page(page);
88 }
Nick Pigginb291f002008-10-18 20:26:44 -070089}
90
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -080091/**
92 * munlock_vma_page - munlock a vma page
93 * @page - page to be unlocked
Nick Pigginb291f002008-10-18 20:26:44 -070094 *
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -080095 * called from munlock()/munmap() path with page supposedly on the LRU.
96 * When we munlock a page, because the vma where we found the page is being
97 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
98 * page locked so that we can leave it on the unevictable lru list and not
99 * bother vmscan with it. However, to walk the page's rmap list in
100 * try_to_munlock() we must isolate the page from the LRU. If some other
101 * task has removed the page from the LRU, we won't be able to do that.
102 * So we clear the PageMlocked as we might not get another chance. If we
103 * can't isolate the page, we leave it for putback_lru_page() and vmscan
104 * [page_referenced()/try_to_unmap()] to deal with.
Nick Pigginb291f002008-10-18 20:26:44 -0700105 */
Hugh Dickins73848b42009-12-14 17:59:22 -0800106void munlock_vma_page(struct page *page)
Nick Pigginb291f002008-10-18 20:26:44 -0700107{
108 BUG_ON(!PageLocked(page));
109
Nick Piggin5344b7e2008-10-18 20:26:51 -0700110 if (TestClearPageMlocked(page)) {
111 dec_zone_page_state(page, NR_MLOCK);
112 if (!isolate_lru_page(page)) {
113 int ret = try_to_munlock(page);
114 /*
115 * did try_to_unlock() succeed or punt?
116 */
Hugh Dickins53f79ac2009-12-14 17:58:58 -0800117 if (ret != SWAP_MLOCK)
Nick Piggin5344b7e2008-10-18 20:26:51 -0700118 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
119
120 putback_lru_page(page);
121 } else {
122 /*
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800123 * Some other task has removed the page from the LRU.
124 * putback_lru_page() will take care of removing the
125 * page from the unevictable list, if necessary.
126 * vmscan [page_referenced()] will move the page back
127 * to the unevictable list if some other vma has it
128 * mlocked.
Nick Piggin5344b7e2008-10-18 20:26:51 -0700129 */
130 if (PageUnevictable(page))
131 count_vm_event(UNEVICTABLE_PGSTRANDED);
132 else
133 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
134 }
Nick Pigginb291f002008-10-18 20:26:44 -0700135 }
136}
137
Rik van Rielba470de2008-10-18 20:26:50 -0700138/**
Hugh Dickins408e82b2009-09-21 17:03:23 -0700139 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700140 * @vma: target vma
141 * @start: start address
142 * @end: end address
Nick Pigginb291f002008-10-18 20:26:44 -0700143 *
Hugh Dickins408e82b2009-09-21 17:03:23 -0700144 * This takes care of making the pages present too.
Nick Pigginb291f002008-10-18 20:26:44 -0700145 *
Rik van Rielba470de2008-10-18 20:26:50 -0700146 * return 0 on success, negative error code on error.
147 *
148 * vma->vm_mm->mmap_sem must be held for at least read.
Nick Pigginb291f002008-10-18 20:26:44 -0700149 */
Rik van Rielba470de2008-10-18 20:26:50 -0700150static long __mlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700151 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700152{
153 struct mm_struct *mm = vma->vm_mm;
154 unsigned long addr = start;
155 struct page *pages[16]; /* 16 gives a reasonable batch */
Nick Pigginb291f002008-10-18 20:26:44 -0700156 int nr_pages = (end - start) / PAGE_SIZE;
Helge Deller72eb8c62008-11-17 00:30:57 +0100157 int ret = 0;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700158 int gup_flags;
Nick Pigginb291f002008-10-18 20:26:44 -0700159
Rik van Rielba470de2008-10-18 20:26:50 -0700160 VM_BUG_ON(start & ~PAGE_MASK);
161 VM_BUG_ON(end & ~PAGE_MASK);
162 VM_BUG_ON(start < vma->vm_start);
163 VM_BUG_ON(end > vma->vm_end);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700164 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
Rik van Rielba470de2008-10-18 20:26:50 -0700165
Hugh Dickins58fa8792009-09-21 17:03:31 -0700166 gup_flags = FOLL_TOUCH | FOLL_GET;
Rik van Rielba470de2008-10-18 20:26:50 -0700167 if (vma->vm_flags & VM_WRITE)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700168 gup_flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700169
Nick Pigginb291f002008-10-18 20:26:44 -0700170 while (nr_pages > 0) {
171 int i;
172
173 cond_resched();
174
175 /*
176 * get_user_pages makes pages present if we are
177 * setting mlock. and this extra reference count will
178 * disable migration of this page. However, page may
179 * still be truncated out from under us.
180 */
Rik van Rielba470de2008-10-18 20:26:50 -0700181 ret = __get_user_pages(current, mm, addr,
Nick Pigginb291f002008-10-18 20:26:44 -0700182 min_t(int, nr_pages, ARRAY_SIZE(pages)),
Rik van Rielba470de2008-10-18 20:26:50 -0700183 gup_flags, pages, NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700184 /*
185 * This can happen for, e.g., VM_NONLINEAR regions before
186 * a page has been allocated and mapped at a given offset,
187 * or for addresses that map beyond end of a file.
Hugh Dickins408e82b2009-09-21 17:03:23 -0700188 * We'll mlock the pages if/when they get faulted in.
Nick Pigginb291f002008-10-18 20:26:44 -0700189 */
190 if (ret < 0)
191 break;
Nick Pigginb291f002008-10-18 20:26:44 -0700192
193 lru_add_drain(); /* push cached pages to LRU */
194
195 for (i = 0; i < ret; i++) {
196 struct page *page = pages[i];
197
Hugh Dickins6e919712009-09-21 17:03:32 -0700198 if (page->mapping) {
199 /*
200 * That preliminary check is mainly to avoid
201 * the pointless overhead of lock_page on the
202 * ZERO_PAGE: which might bounce very badly if
203 * there is contention. However, we're still
204 * dirtying its cacheline with get/put_page:
205 * we'll add another __get_user_pages flag to
206 * avoid it if that case turns out to matter.
207 */
208 lock_page(page);
209 /*
210 * Because we lock page here and migration is
211 * blocked by the elevated reference, we need
212 * only check for file-cache page truncation.
213 */
214 if (page->mapping)
215 mlock_vma_page(page);
216 unlock_page(page);
217 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700218 put_page(page); /* ref from get_user_pages() */
Nick Pigginb291f002008-10-18 20:26:44 -0700219 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700220
221 addr += ret * PAGE_SIZE;
222 nr_pages -= ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700223 ret = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700224 }
225
Hugh Dickins408e82b2009-09-21 17:03:23 -0700226 return ret; /* 0 or negative error code */
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700227}
228
229/*
230 * convert get_user_pages() return value to posix mlock() error
231 */
232static int __mlock_posix_error_return(long retval)
233{
234 if (retval == -EFAULT)
235 retval = -ENOMEM;
236 else if (retval == -ENOMEM)
237 retval = -EAGAIN;
238 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700239}
240
Rik van Rielba470de2008-10-18 20:26:50 -0700241/**
242 * mlock_vma_pages_range() - mlock pages in specified vma range.
243 * @vma - the vma containing the specfied address range
244 * @start - starting address in @vma to mlock
245 * @end - end address [+1] in @vma to mlock
246 *
247 * For mmap()/mremap()/expansion of mlocked vma.
248 *
249 * return 0 on success for "normal" vmas.
250 *
251 * return number of pages [> 0] to be removed from locked_vm on success
252 * of "special" vmas.
Nick Pigginb291f002008-10-18 20:26:44 -0700253 */
Rik van Rielba470de2008-10-18 20:26:50 -0700254long mlock_vma_pages_range(struct vm_area_struct *vma,
Nick Pigginb291f002008-10-18 20:26:44 -0700255 unsigned long start, unsigned long end)
256{
257 int nr_pages = (end - start) / PAGE_SIZE;
258 BUG_ON(!(vma->vm_flags & VM_LOCKED));
259
260 /*
261 * filter unlockable vmas
262 */
263 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
264 goto no_mlock;
265
266 if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
267 is_vm_hugetlb_page(vma) ||
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700268 vma == get_gate_vma(current))) {
Rik van Rielba470de2008-10-18 20:26:50 -0700269
Hugh Dickins408e82b2009-09-21 17:03:23 -0700270 __mlock_vma_pages_range(vma, start, end);
Hugh Dickinsd5b56232009-02-08 20:56:58 +0000271
272 /* Hide errors from mmap() and other callers */
273 return 0;
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700274 }
Nick Pigginb291f002008-10-18 20:26:44 -0700275
276 /*
277 * User mapped kernel pages or huge pages:
278 * make these pages present to populate the ptes, but
279 * fall thru' to reset VM_LOCKED--no need to unlock, and
280 * return nr_pages so these don't get counted against task's
281 * locked limit. huge pages are already counted against
282 * locked vm limit.
283 */
284 make_pages_present(start, end);
285
286no_mlock:
287 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
Rik van Rielba470de2008-10-18 20:26:50 -0700288 return nr_pages; /* error or pages NOT mlocked */
Nick Pigginb291f002008-10-18 20:26:44 -0700289}
290
Nick Pigginb291f002008-10-18 20:26:44 -0700291/*
Rik van Rielba470de2008-10-18 20:26:50 -0700292 * munlock_vma_pages_range() - munlock all pages in the vma range.'
293 * @vma - vma containing range to be munlock()ed.
294 * @start - start address in @vma of the range
295 * @end - end of range in @vma.
296 *
297 * For mremap(), munmap() and exit().
298 *
299 * Called with @vma VM_LOCKED.
300 *
301 * Returns with VM_LOCKED cleared. Callers must be prepared to
302 * deal with this.
303 *
304 * We don't save and restore VM_LOCKED here because pages are
305 * still on lru. In unmap path, pages might be scanned by reclaim
306 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
307 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700308 */
Rik van Rielba470de2008-10-18 20:26:50 -0700309void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700310 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700311{
Hugh Dickins408e82b2009-09-21 17:03:23 -0700312 unsigned long addr;
313
314 lru_add_drain();
Nick Pigginb291f002008-10-18 20:26:44 -0700315 vma->vm_flags &= ~VM_LOCKED;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700316
317 for (addr = start; addr < end; addr += PAGE_SIZE) {
Hugh Dickins6e919712009-09-21 17:03:32 -0700318 struct page *page;
319 /*
320 * Although FOLL_DUMP is intended for get_dump_page(),
321 * it just so happens that its special treatment of the
322 * ZERO_PAGE (returning an error instead of doing get_page)
323 * suits munlock very well (and if somehow an abnormal page
324 * has sneaked into the range, we won't oops here: great).
325 */
326 page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP);
327 if (page && !IS_ERR(page)) {
Hugh Dickins408e82b2009-09-21 17:03:23 -0700328 lock_page(page);
Hugh Dickins6e919712009-09-21 17:03:32 -0700329 /*
330 * Like in __mlock_vma_pages_range(),
331 * because we lock page here and migration is
332 * blocked by the elevated reference, we need
333 * only check for file-cache page truncation.
334 */
Hugh Dickins408e82b2009-09-21 17:03:23 -0700335 if (page->mapping)
336 munlock_vma_page(page);
337 unlock_page(page);
338 put_page(page);
339 }
340 cond_resched();
341 }
Nick Pigginb291f002008-10-18 20:26:44 -0700342}
343
344/*
345 * mlock_fixup - handle mlock[all]/munlock[all] requests.
346 *
347 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
348 * munlock is a no-op. However, for some special vmas, we go ahead and
349 * populate the ptes via make_pages_present().
350 *
351 * For vmas that pass the filters, merge/split as appropriate.
352 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
354 unsigned long start, unsigned long end, unsigned int newflags)
355{
Nick Pigginb291f002008-10-18 20:26:44 -0700356 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700358 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 int ret = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700360 int lock = newflags & VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
Nick Pigginb291f002008-10-18 20:26:44 -0700362 if (newflags == vma->vm_flags ||
363 (vma->vm_flags & (VM_IO | VM_PFNMAP)))
364 goto out; /* don't set VM_LOCKED, don't count */
365
366 if ((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
367 is_vm_hugetlb_page(vma) ||
368 vma == get_gate_vma(current)) {
369 if (lock)
370 make_pages_present(start, end);
371 goto out; /* don't set VM_LOCKED, don't count */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 }
373
374 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
375 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
376 vma->vm_file, pgoff, vma_policy(vma));
377 if (*prev) {
378 vma = *prev;
379 goto success;
380 }
381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (start != vma->vm_start) {
383 ret = split_vma(mm, vma, start, 1);
384 if (ret)
385 goto out;
386 }
387
388 if (end != vma->vm_end) {
389 ret = split_vma(mm, vma, end, 0);
390 if (ret)
391 goto out;
392 }
393
394success:
395 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700396 * Keep track of amount of locked VM.
397 */
398 nr_pages = (end - start) >> PAGE_SHIFT;
399 if (!lock)
400 nr_pages = -nr_pages;
401 mm->locked_vm += nr_pages;
402
403 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 * vm_flags is protected by the mmap_sem held in write mode.
405 * It's okay if try_to_unmap_one unmaps a page just after we
Nick Pigginb291f002008-10-18 20:26:44 -0700406 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Nick Pigginb291f002008-10-18 20:26:44 -0700409 if (lock) {
Hugh Dickins408e82b2009-09-21 17:03:23 -0700410 vma->vm_flags = newflags;
411 ret = __mlock_vma_pages_range(vma, start, end);
412 if (ret < 0)
413 ret = __mlock_posix_error_return(ret);
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700414 } else {
Hugh Dickins408e82b2009-09-21 17:03:23 -0700415 munlock_vma_pages_range(vma, start, end);
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418out:
Nick Pigginb291f002008-10-18 20:26:44 -0700419 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 return ret;
421}
422
423static int do_mlock(unsigned long start, size_t len, int on)
424{
425 unsigned long nstart, end, tmp;
426 struct vm_area_struct * vma, * prev;
427 int error;
428
429 len = PAGE_ALIGN(len);
430 end = start + len;
431 if (end < start)
432 return -EINVAL;
433 if (end == start)
434 return 0;
435 vma = find_vma_prev(current->mm, start, &prev);
436 if (!vma || vma->vm_start > start)
437 return -ENOMEM;
438
439 if (start > vma->vm_start)
440 prev = vma;
441
442 for (nstart = start ; ; ) {
443 unsigned int newflags;
444
445 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
446
447 newflags = vma->vm_flags | VM_LOCKED;
448 if (!on)
449 newflags &= ~VM_LOCKED;
450
451 tmp = vma->vm_end;
452 if (tmp > end)
453 tmp = end;
454 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
455 if (error)
456 break;
457 nstart = tmp;
458 if (nstart < prev->vm_end)
459 nstart = prev->vm_end;
460 if (nstart >= end)
461 break;
462
463 vma = prev->vm_next;
464 if (!vma || vma->vm_start != nstart) {
465 error = -ENOMEM;
466 break;
467 }
468 }
469 return error;
470}
471
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100472SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473{
474 unsigned long locked;
475 unsigned long lock_limit;
476 int error = -ENOMEM;
477
478 if (!can_do_mlock())
479 return -EPERM;
480
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800481 lru_add_drain_all(); /* flush pagevec */
482
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 down_write(&current->mm->mmap_sem);
484 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
485 start &= PAGE_MASK;
486
487 locked = len >> PAGE_SHIFT;
488 locked += current->mm->locked_vm;
489
Jiri Slaby59e99e52010-03-05 13:41:44 -0800490 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 lock_limit >>= PAGE_SHIFT;
492
493 /* check against resource limits */
494 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
495 error = do_mlock(start, len, 1);
496 up_write(&current->mm->mmap_sem);
497 return error;
498}
499
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100500SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
502 int ret;
503
504 down_write(&current->mm->mmap_sem);
505 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
506 start &= PAGE_MASK;
507 ret = do_mlock(start, len, 0);
508 up_write(&current->mm->mmap_sem);
509 return ret;
510}
511
512static int do_mlockall(int flags)
513{
514 struct vm_area_struct * vma, * prev = NULL;
515 unsigned int def_flags = 0;
516
517 if (flags & MCL_FUTURE)
518 def_flags = VM_LOCKED;
519 current->mm->def_flags = def_flags;
520 if (flags == MCL_FUTURE)
521 goto out;
522
523 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
524 unsigned int newflags;
525
526 newflags = vma->vm_flags | VM_LOCKED;
527 if (!(flags & MCL_CURRENT))
528 newflags &= ~VM_LOCKED;
529
530 /* Ignore errors */
531 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
532 }
533out:
534 return 0;
535}
536
Heiko Carstens3480b252009-01-14 14:14:16 +0100537SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
539 unsigned long lock_limit;
540 int ret = -EINVAL;
541
542 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
543 goto out;
544
545 ret = -EPERM;
546 if (!can_do_mlock())
547 goto out;
548
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800549 lru_add_drain_all(); /* flush pagevec */
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 down_write(&current->mm->mmap_sem);
552
Jiri Slaby59e99e52010-03-05 13:41:44 -0800553 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 lock_limit >>= PAGE_SHIFT;
555
556 ret = -ENOMEM;
557 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
558 capable(CAP_IPC_LOCK))
559 ret = do_mlockall(flags);
560 up_write(&current->mm->mmap_sem);
561out:
562 return ret;
563}
564
Heiko Carstens3480b252009-01-14 14:14:16 +0100565SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
567 int ret;
568
569 down_write(&current->mm->mmap_sem);
570 ret = do_mlockall(0);
571 up_write(&current->mm->mmap_sem);
572 return ret;
573}
574
575/*
576 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
577 * shm segments) get accounted against the user_struct instead.
578 */
579static DEFINE_SPINLOCK(shmlock_user_lock);
580
581int user_shm_lock(size_t size, struct user_struct *user)
582{
583 unsigned long lock_limit, locked;
584 int allowed = 0;
585
586 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800587 lock_limit = rlimit(RLIMIT_MEMLOCK);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700588 if (lock_limit == RLIM_INFINITY)
589 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 lock_limit >>= PAGE_SHIFT;
591 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700592 if (!allowed &&
593 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 goto out;
595 get_uid(user);
596 user->locked_shm += locked;
597 allowed = 1;
598out:
599 spin_unlock(&shmlock_user_lock);
600 return allowed;
601}
602
603void user_shm_unlock(size_t size, struct user_struct *user)
604{
605 spin_lock(&shmlock_user_lock);
606 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
607 spin_unlock(&shmlock_user_lock);
608 free_uid(user);
609}
Markus Metzgerc5dee612008-12-19 15:17:02 +0100610
Markus Metzger1cb81b12009-04-24 09:51:43 +0200611int account_locked_memory(struct mm_struct *mm, struct rlimit *rlim,
612 size_t size)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100613{
Markus Metzger1cb81b12009-04-24 09:51:43 +0200614 unsigned long lim, vm, pgsz;
615 int error = -ENOMEM;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100616
617 pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
618
Markus Metzger1cb81b12009-04-24 09:51:43 +0200619 down_write(&mm->mmap_sem);
Markus Metzgerc5dee612008-12-19 15:17:02 +0100620
Jiri Slaby59e99e52010-03-05 13:41:44 -0800621 lim = ACCESS_ONCE(rlim[RLIMIT_AS].rlim_cur) >> PAGE_SHIFT;
Markus Metzger1cb81b12009-04-24 09:51:43 +0200622 vm = mm->total_vm + pgsz;
623 if (lim < vm)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100624 goto out;
625
Jiri Slaby59e99e52010-03-05 13:41:44 -0800626 lim = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur) >> PAGE_SHIFT;
Markus Metzger1cb81b12009-04-24 09:51:43 +0200627 vm = mm->locked_vm + pgsz;
628 if (lim < vm)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100629 goto out;
630
Markus Metzger1cb81b12009-04-24 09:51:43 +0200631 mm->total_vm += pgsz;
632 mm->locked_vm += pgsz;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100633
Markus Metzger1cb81b12009-04-24 09:51:43 +0200634 error = 0;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100635 out:
Markus Metzger1cb81b12009-04-24 09:51:43 +0200636 up_write(&mm->mmap_sem);
637 return error;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100638}
639
Markus Metzger1cb81b12009-04-24 09:51:43 +0200640void refund_locked_memory(struct mm_struct *mm, size_t size)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100641{
642 unsigned long pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
643
Markus Metzgere2b371f2009-04-03 16:43:35 +0200644 down_write(&mm->mmap_sem);
Markus Metzgerc5dee612008-12-19 15:17:02 +0100645
Markus Metzgere2b371f2009-04-03 16:43:35 +0200646 mm->total_vm -= pgsz;
647 mm->locked_vm -= pgsz;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100648
Markus Metzgere2b371f2009-04-03 16:43:35 +0200649 up_write(&mm->mmap_sem);
Markus Metzger9f339e72009-02-11 15:10:27 +0100650}