blob: 22041aa9f5c1d7793b9cc8094153c2c09975e211 [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;
28 if (current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur != 0)
29 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
91/*
92 * called from munlock()/munmap() path with page supposedly on the LRU.
93 *
94 * Note: unlike mlock_vma_page(), we can't just clear the PageMlocked
95 * [in try_to_munlock()] and then attempt to isolate the page. We must
96 * isolate the page to keep others from messing with its unevictable
97 * and mlocked state while trying to munlock. However, we pre-clear the
98 * mlocked state anyway as we might lose the isolation race and we might
99 * not get another chance to clear PageMlocked. If we successfully
100 * isolate the page and try_to_munlock() detects other VM_LOCKED vmas
101 * mapping the page, it will restore the PageMlocked state, unless the page
102 * is mapped in a non-linear vma. So, we go ahead and SetPageMlocked(),
103 * perhaps redundantly.
104 * If we lose the isolation race, and the page is mapped by other VM_LOCKED
105 * vmas, we'll detect this in vmscan--via try_to_munlock() or try_to_unmap()
106 * either of which will restore the PageMlocked state by calling
107 * mlock_vma_page() above, if it can grab the vma's mmap sem.
108 */
109static void munlock_vma_page(struct page *page)
110{
111 BUG_ON(!PageLocked(page));
112
Nick Piggin5344b7e2008-10-18 20:26:51 -0700113 if (TestClearPageMlocked(page)) {
114 dec_zone_page_state(page, NR_MLOCK);
115 if (!isolate_lru_page(page)) {
116 int ret = try_to_munlock(page);
117 /*
118 * did try_to_unlock() succeed or punt?
119 */
120 if (ret == SWAP_SUCCESS || ret == SWAP_AGAIN)
121 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
122
123 putback_lru_page(page);
124 } else {
125 /*
126 * We lost the race. let try_to_unmap() deal
127 * with it. At least we get the page state and
128 * mlock stats right. However, page is still on
129 * the noreclaim list. We'll fix that up when
130 * the page is eventually freed or we scan the
131 * noreclaim list.
132 */
133 if (PageUnevictable(page))
134 count_vm_event(UNEVICTABLE_PGSTRANDED);
135 else
136 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
137 }
Nick Pigginb291f002008-10-18 20:26:44 -0700138 }
139}
140
Rik van Rielba470de2008-10-18 20:26:50 -0700141/**
Hugh Dickins408e82b2009-09-21 17:03:23 -0700142 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700143 * @vma: target vma
144 * @start: start address
145 * @end: end address
Nick Pigginb291f002008-10-18 20:26:44 -0700146 *
Hugh Dickins408e82b2009-09-21 17:03:23 -0700147 * This takes care of making the pages present too.
Nick Pigginb291f002008-10-18 20:26:44 -0700148 *
Rik van Rielba470de2008-10-18 20:26:50 -0700149 * return 0 on success, negative error code on error.
150 *
151 * vma->vm_mm->mmap_sem must be held for at least read.
Nick Pigginb291f002008-10-18 20:26:44 -0700152 */
Rik van Rielba470de2008-10-18 20:26:50 -0700153static long __mlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700154 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700155{
156 struct mm_struct *mm = vma->vm_mm;
157 unsigned long addr = start;
158 struct page *pages[16]; /* 16 gives a reasonable batch */
Nick Pigginb291f002008-10-18 20:26:44 -0700159 int nr_pages = (end - start) / PAGE_SIZE;
Helge Deller72eb8c62008-11-17 00:30:57 +0100160 int ret = 0;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700161 int gup_flags;
Nick Pigginb291f002008-10-18 20:26:44 -0700162
Rik van Rielba470de2008-10-18 20:26:50 -0700163 VM_BUG_ON(start & ~PAGE_MASK);
164 VM_BUG_ON(end & ~PAGE_MASK);
165 VM_BUG_ON(start < vma->vm_start);
166 VM_BUG_ON(end > vma->vm_end);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700167 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
Rik van Rielba470de2008-10-18 20:26:50 -0700168
Hugh Dickins58fa8792009-09-21 17:03:31 -0700169 gup_flags = FOLL_TOUCH | FOLL_GET;
Rik van Rielba470de2008-10-18 20:26:50 -0700170 if (vma->vm_flags & VM_WRITE)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700171 gup_flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700172
Nick Pigginb291f002008-10-18 20:26:44 -0700173 while (nr_pages > 0) {
174 int i;
175
176 cond_resched();
177
178 /*
179 * get_user_pages makes pages present if we are
180 * setting mlock. and this extra reference count will
181 * disable migration of this page. However, page may
182 * still be truncated out from under us.
183 */
Rik van Rielba470de2008-10-18 20:26:50 -0700184 ret = __get_user_pages(current, mm, addr,
Nick Pigginb291f002008-10-18 20:26:44 -0700185 min_t(int, nr_pages, ARRAY_SIZE(pages)),
Rik van Rielba470de2008-10-18 20:26:50 -0700186 gup_flags, pages, NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700187 /*
188 * This can happen for, e.g., VM_NONLINEAR regions before
189 * a page has been allocated and mapped at a given offset,
190 * or for addresses that map beyond end of a file.
Hugh Dickins408e82b2009-09-21 17:03:23 -0700191 * We'll mlock the pages if/when they get faulted in.
Nick Pigginb291f002008-10-18 20:26:44 -0700192 */
193 if (ret < 0)
194 break;
Nick Pigginb291f002008-10-18 20:26:44 -0700195
196 lru_add_drain(); /* push cached pages to LRU */
197
198 for (i = 0; i < ret; i++) {
199 struct page *page = pages[i];
200
201 lock_page(page);
202 /*
203 * Because we lock page here and migration is blocked
204 * by the elevated reference, we need only check for
Hugh Dickins408e82b2009-09-21 17:03:23 -0700205 * file-cache page truncation. This page->mapping
206 * check also neatly skips over the ZERO_PAGE(),
207 * though if that's common we'd prefer not to lock it.
Nick Pigginb291f002008-10-18 20:26:44 -0700208 */
Hugh Dickins408e82b2009-09-21 17:03:23 -0700209 if (page->mapping)
210 mlock_vma_page(page);
Nick Pigginb291f002008-10-18 20:26:44 -0700211 unlock_page(page);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700212 put_page(page); /* ref from get_user_pages() */
Nick Pigginb291f002008-10-18 20:26:44 -0700213 }
Hugh Dickins408e82b2009-09-21 17:03:23 -0700214
215 addr += ret * PAGE_SIZE;
216 nr_pages -= ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700217 ret = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700218 }
219
Hugh Dickins408e82b2009-09-21 17:03:23 -0700220 return ret; /* 0 or negative error code */
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700221}
222
223/*
224 * convert get_user_pages() return value to posix mlock() error
225 */
226static int __mlock_posix_error_return(long retval)
227{
228 if (retval == -EFAULT)
229 retval = -ENOMEM;
230 else if (retval == -ENOMEM)
231 retval = -EAGAIN;
232 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700233}
234
Rik van Rielba470de2008-10-18 20:26:50 -0700235/**
236 * mlock_vma_pages_range() - mlock pages in specified vma range.
237 * @vma - the vma containing the specfied address range
238 * @start - starting address in @vma to mlock
239 * @end - end address [+1] in @vma to mlock
240 *
241 * For mmap()/mremap()/expansion of mlocked vma.
242 *
243 * return 0 on success for "normal" vmas.
244 *
245 * return number of pages [> 0] to be removed from locked_vm on success
246 * of "special" vmas.
Nick Pigginb291f002008-10-18 20:26:44 -0700247 */
Rik van Rielba470de2008-10-18 20:26:50 -0700248long mlock_vma_pages_range(struct vm_area_struct *vma,
Nick Pigginb291f002008-10-18 20:26:44 -0700249 unsigned long start, unsigned long end)
250{
251 int nr_pages = (end - start) / PAGE_SIZE;
252 BUG_ON(!(vma->vm_flags & VM_LOCKED));
253
254 /*
255 * filter unlockable vmas
256 */
257 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
258 goto no_mlock;
259
260 if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
261 is_vm_hugetlb_page(vma) ||
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700262 vma == get_gate_vma(current))) {
Rik van Rielba470de2008-10-18 20:26:50 -0700263
Hugh Dickins408e82b2009-09-21 17:03:23 -0700264 __mlock_vma_pages_range(vma, start, end);
Hugh Dickinsd5b56232009-02-08 20:56:58 +0000265
266 /* Hide errors from mmap() and other callers */
267 return 0;
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700268 }
Nick Pigginb291f002008-10-18 20:26:44 -0700269
270 /*
271 * User mapped kernel pages or huge pages:
272 * make these pages present to populate the ptes, but
273 * fall thru' to reset VM_LOCKED--no need to unlock, and
274 * return nr_pages so these don't get counted against task's
275 * locked limit. huge pages are already counted against
276 * locked vm limit.
277 */
278 make_pages_present(start, end);
279
280no_mlock:
281 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
Rik van Rielba470de2008-10-18 20:26:50 -0700282 return nr_pages; /* error or pages NOT mlocked */
Nick Pigginb291f002008-10-18 20:26:44 -0700283}
284
Nick Pigginb291f002008-10-18 20:26:44 -0700285/*
Rik van Rielba470de2008-10-18 20:26:50 -0700286 * munlock_vma_pages_range() - munlock all pages in the vma range.'
287 * @vma - vma containing range to be munlock()ed.
288 * @start - start address in @vma of the range
289 * @end - end of range in @vma.
290 *
291 * For mremap(), munmap() and exit().
292 *
293 * Called with @vma VM_LOCKED.
294 *
295 * Returns with VM_LOCKED cleared. Callers must be prepared to
296 * deal with this.
297 *
298 * We don't save and restore VM_LOCKED here because pages are
299 * still on lru. In unmap path, pages might be scanned by reclaim
300 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
301 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700302 */
Rik van Rielba470de2008-10-18 20:26:50 -0700303void munlock_vma_pages_range(struct vm_area_struct *vma,
Hugh Dickins408e82b2009-09-21 17:03:23 -0700304 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700305{
Hugh Dickins408e82b2009-09-21 17:03:23 -0700306 unsigned long addr;
307
308 lru_add_drain();
Nick Pigginb291f002008-10-18 20:26:44 -0700309 vma->vm_flags &= ~VM_LOCKED;
Hugh Dickins408e82b2009-09-21 17:03:23 -0700310
311 for (addr = start; addr < end; addr += PAGE_SIZE) {
312 struct page *page = follow_page(vma, addr, FOLL_GET);
313 if (page) {
314 lock_page(page);
315 if (page->mapping)
316 munlock_vma_page(page);
317 unlock_page(page);
318 put_page(page);
319 }
320 cond_resched();
321 }
Nick Pigginb291f002008-10-18 20:26:44 -0700322}
323
324/*
325 * mlock_fixup - handle mlock[all]/munlock[all] requests.
326 *
327 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
328 * munlock is a no-op. However, for some special vmas, we go ahead and
329 * populate the ptes via make_pages_present().
330 *
331 * For vmas that pass the filters, merge/split as appropriate.
332 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
334 unsigned long start, unsigned long end, unsigned int newflags)
335{
Nick Pigginb291f002008-10-18 20:26:44 -0700336 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700338 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 int ret = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700340 int lock = newflags & VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Nick Pigginb291f002008-10-18 20:26:44 -0700342 if (newflags == vma->vm_flags ||
343 (vma->vm_flags & (VM_IO | VM_PFNMAP)))
344 goto out; /* don't set VM_LOCKED, don't count */
345
346 if ((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
347 is_vm_hugetlb_page(vma) ||
348 vma == get_gate_vma(current)) {
349 if (lock)
350 make_pages_present(start, end);
351 goto out; /* don't set VM_LOCKED, don't count */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 }
353
354 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
355 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
356 vma->vm_file, pgoff, vma_policy(vma));
357 if (*prev) {
358 vma = *prev;
359 goto success;
360 }
361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 if (start != vma->vm_start) {
363 ret = split_vma(mm, vma, start, 1);
364 if (ret)
365 goto out;
366 }
367
368 if (end != vma->vm_end) {
369 ret = split_vma(mm, vma, end, 0);
370 if (ret)
371 goto out;
372 }
373
374success:
375 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700376 * Keep track of amount of locked VM.
377 */
378 nr_pages = (end - start) >> PAGE_SHIFT;
379 if (!lock)
380 nr_pages = -nr_pages;
381 mm->locked_vm += nr_pages;
382
383 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 * vm_flags is protected by the mmap_sem held in write mode.
385 * It's okay if try_to_unmap_one unmaps a page just after we
Nick Pigginb291f002008-10-18 20:26:44 -0700386 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Nick Pigginb291f002008-10-18 20:26:44 -0700389 if (lock) {
Hugh Dickins408e82b2009-09-21 17:03:23 -0700390 vma->vm_flags = newflags;
391 ret = __mlock_vma_pages_range(vma, start, end);
392 if (ret < 0)
393 ret = __mlock_posix_error_return(ret);
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700394 } else {
Hugh Dickins408e82b2009-09-21 17:03:23 -0700395 munlock_vma_pages_range(vma, start, end);
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398out:
Nick Pigginb291f002008-10-18 20:26:44 -0700399 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 return ret;
401}
402
403static int do_mlock(unsigned long start, size_t len, int on)
404{
405 unsigned long nstart, end, tmp;
406 struct vm_area_struct * vma, * prev;
407 int error;
408
409 len = PAGE_ALIGN(len);
410 end = start + len;
411 if (end < start)
412 return -EINVAL;
413 if (end == start)
414 return 0;
415 vma = find_vma_prev(current->mm, start, &prev);
416 if (!vma || vma->vm_start > start)
417 return -ENOMEM;
418
419 if (start > vma->vm_start)
420 prev = vma;
421
422 for (nstart = start ; ; ) {
423 unsigned int newflags;
424
425 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
426
427 newflags = vma->vm_flags | VM_LOCKED;
428 if (!on)
429 newflags &= ~VM_LOCKED;
430
431 tmp = vma->vm_end;
432 if (tmp > end)
433 tmp = end;
434 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
435 if (error)
436 break;
437 nstart = tmp;
438 if (nstart < prev->vm_end)
439 nstart = prev->vm_end;
440 if (nstart >= end)
441 break;
442
443 vma = prev->vm_next;
444 if (!vma || vma->vm_start != nstart) {
445 error = -ENOMEM;
446 break;
447 }
448 }
449 return error;
450}
451
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100452SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 unsigned long locked;
455 unsigned long lock_limit;
456 int error = -ENOMEM;
457
458 if (!can_do_mlock())
459 return -EPERM;
460
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800461 lru_add_drain_all(); /* flush pagevec */
462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 down_write(&current->mm->mmap_sem);
464 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
465 start &= PAGE_MASK;
466
467 locked = len >> PAGE_SHIFT;
468 locked += current->mm->locked_vm;
469
470 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
471 lock_limit >>= PAGE_SHIFT;
472
473 /* check against resource limits */
474 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
475 error = do_mlock(start, len, 1);
476 up_write(&current->mm->mmap_sem);
477 return error;
478}
479
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100480SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
482 int ret;
483
484 down_write(&current->mm->mmap_sem);
485 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
486 start &= PAGE_MASK;
487 ret = do_mlock(start, len, 0);
488 up_write(&current->mm->mmap_sem);
489 return ret;
490}
491
492static int do_mlockall(int flags)
493{
494 struct vm_area_struct * vma, * prev = NULL;
495 unsigned int def_flags = 0;
496
497 if (flags & MCL_FUTURE)
498 def_flags = VM_LOCKED;
499 current->mm->def_flags = def_flags;
500 if (flags == MCL_FUTURE)
501 goto out;
502
503 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
504 unsigned int newflags;
505
506 newflags = vma->vm_flags | VM_LOCKED;
507 if (!(flags & MCL_CURRENT))
508 newflags &= ~VM_LOCKED;
509
510 /* Ignore errors */
511 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
512 }
513out:
514 return 0;
515}
516
Heiko Carstens3480b252009-01-14 14:14:16 +0100517SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518{
519 unsigned long lock_limit;
520 int ret = -EINVAL;
521
522 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
523 goto out;
524
525 ret = -EPERM;
526 if (!can_do_mlock())
527 goto out;
528
KOSAKI Motohiro8891d6d2008-11-12 13:26:53 -0800529 lru_add_drain_all(); /* flush pagevec */
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 down_write(&current->mm->mmap_sem);
532
533 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
534 lock_limit >>= PAGE_SHIFT;
535
536 ret = -ENOMEM;
537 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
538 capable(CAP_IPC_LOCK))
539 ret = do_mlockall(flags);
540 up_write(&current->mm->mmap_sem);
541out:
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;
567 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
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}
Markus Metzgerc5dee612008-12-19 15:17:02 +0100590
Markus Metzger1cb81b12009-04-24 09:51:43 +0200591int account_locked_memory(struct mm_struct *mm, struct rlimit *rlim,
592 size_t size)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100593{
Markus Metzger1cb81b12009-04-24 09:51:43 +0200594 unsigned long lim, vm, pgsz;
595 int error = -ENOMEM;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100596
597 pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
598
Markus Metzger1cb81b12009-04-24 09:51:43 +0200599 down_write(&mm->mmap_sem);
Markus Metzgerc5dee612008-12-19 15:17:02 +0100600
Markus Metzger1cb81b12009-04-24 09:51:43 +0200601 lim = rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
602 vm = mm->total_vm + pgsz;
603 if (lim < vm)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100604 goto out;
605
Markus Metzger1cb81b12009-04-24 09:51:43 +0200606 lim = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
607 vm = mm->locked_vm + pgsz;
608 if (lim < vm)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100609 goto out;
610
Markus Metzger1cb81b12009-04-24 09:51:43 +0200611 mm->total_vm += pgsz;
612 mm->locked_vm += pgsz;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100613
Markus Metzger1cb81b12009-04-24 09:51:43 +0200614 error = 0;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100615 out:
Markus Metzger1cb81b12009-04-24 09:51:43 +0200616 up_write(&mm->mmap_sem);
617 return error;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100618}
619
Markus Metzger1cb81b12009-04-24 09:51:43 +0200620void refund_locked_memory(struct mm_struct *mm, size_t size)
Markus Metzgerc5dee612008-12-19 15:17:02 +0100621{
622 unsigned long pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
623
Markus Metzgere2b371f2009-04-03 16:43:35 +0200624 down_write(&mm->mmap_sem);
Markus Metzgerc5dee612008-12-19 15:17:02 +0100625
Markus Metzgere2b371f2009-04-03 16:43:35 +0200626 mm->total_vm -= pgsz;
627 mm->locked_vm -= pgsz;
Markus Metzgerc5dee612008-12-19 15:17:02 +0100628
Markus Metzgere2b371f2009-04-03 16:43:35 +0200629 up_write(&mm->mmap_sem);
Markus Metzger9f339e72009-02-11 15:10:27 +0100630}