blob: 008ea70b7afa9a2baed56cdc9b6b43ef0b05a6ff [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#ifdef CONFIG_UNEVICTABLE_LRU
35/*
36 * Mlocked pages are marked with PageMlocked() flag for efficient testing
37 * in vmscan and, possibly, the fault path; and to support semi-accurate
38 * statistics.
39 *
40 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
41 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
42 * The unevictable list is an LRU sibling list to the [in]active lists.
43 * PageUnevictable is set to indicate the unevictable state.
44 *
45 * When lazy mlocking via vmscan, it is important to ensure that the
46 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
47 * may have mlocked a page that is being munlocked. So lazy mlock must take
48 * the mmap_sem for read, and verify that the vma really is locked
49 * (see mm/rmap.c).
50 */
51
52/*
53 * LRU accounting for clear_page_mlock()
54 */
55void __clear_page_mlock(struct page *page)
56{
57 VM_BUG_ON(!PageLocked(page));
58
59 if (!page->mapping) { /* truncated ? */
60 return;
61 }
62
Nick Piggin5344b7e2008-10-18 20:26:51 -070063 dec_zone_page_state(page, NR_MLOCK);
64 count_vm_event(UNEVICTABLE_PGCLEARED);
Nick Pigginb291f002008-10-18 20:26:44 -070065 if (!isolate_lru_page(page)) {
66 putback_lru_page(page);
67 } else {
68 /*
69 * Page not on the LRU yet. Flush all pagevecs and retry.
70 */
71 lru_add_drain_all();
72 if (!isolate_lru_page(page))
73 putback_lru_page(page);
Nick Piggin5344b7e2008-10-18 20:26:51 -070074 else if (PageUnevictable(page))
75 count_vm_event(UNEVICTABLE_PGSTRANDED);
76
Nick Pigginb291f002008-10-18 20:26:44 -070077 }
78}
79
80/*
81 * Mark page as mlocked if not already.
82 * If page on LRU, isolate and putback to move to unevictable list.
83 */
84void mlock_vma_page(struct page *page)
85{
86 BUG_ON(!PageLocked(page));
87
Nick Piggin5344b7e2008-10-18 20:26:51 -070088 if (!TestSetPageMlocked(page)) {
89 inc_zone_page_state(page, NR_MLOCK);
90 count_vm_event(UNEVICTABLE_PGMLOCKED);
91 if (!isolate_lru_page(page))
92 putback_lru_page(page);
93 }
Nick Pigginb291f002008-10-18 20:26:44 -070094}
95
96/*
97 * called from munlock()/munmap() path with page supposedly on the LRU.
98 *
99 * Note: unlike mlock_vma_page(), we can't just clear the PageMlocked
100 * [in try_to_munlock()] and then attempt to isolate the page. We must
101 * isolate the page to keep others from messing with its unevictable
102 * and mlocked state while trying to munlock. However, we pre-clear the
103 * mlocked state anyway as we might lose the isolation race and we might
104 * not get another chance to clear PageMlocked. If we successfully
105 * isolate the page and try_to_munlock() detects other VM_LOCKED vmas
106 * mapping the page, it will restore the PageMlocked state, unless the page
107 * is mapped in a non-linear vma. So, we go ahead and SetPageMlocked(),
108 * perhaps redundantly.
109 * If we lose the isolation race, and the page is mapped by other VM_LOCKED
110 * vmas, we'll detect this in vmscan--via try_to_munlock() or try_to_unmap()
111 * either of which will restore the PageMlocked state by calling
112 * mlock_vma_page() above, if it can grab the vma's mmap sem.
113 */
114static void munlock_vma_page(struct page *page)
115{
116 BUG_ON(!PageLocked(page));
117
Nick Piggin5344b7e2008-10-18 20:26:51 -0700118 if (TestClearPageMlocked(page)) {
119 dec_zone_page_state(page, NR_MLOCK);
120 if (!isolate_lru_page(page)) {
121 int ret = try_to_munlock(page);
122 /*
123 * did try_to_unlock() succeed or punt?
124 */
125 if (ret == SWAP_SUCCESS || ret == SWAP_AGAIN)
126 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
127
128 putback_lru_page(page);
129 } else {
130 /*
131 * We lost the race. let try_to_unmap() deal
132 * with it. At least we get the page state and
133 * mlock stats right. However, page is still on
134 * the noreclaim list. We'll fix that up when
135 * the page is eventually freed or we scan the
136 * noreclaim list.
137 */
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/**
147 * __mlock_vma_pages_range() - mlock/munlock a range of pages in the vma.
148 * @vma: target vma
149 * @start: start address
150 * @end: end address
151 * @mlock: 0 indicate munlock, otherwise mlock.
Nick Pigginb291f002008-10-18 20:26:44 -0700152 *
Rik van Rielba470de2008-10-18 20:26:50 -0700153 * If @mlock == 0, unlock an mlocked range;
154 * else mlock the range of pages. This takes care of making the pages present ,
155 * too.
Nick Pigginb291f002008-10-18 20:26:44 -0700156 *
Rik van Rielba470de2008-10-18 20:26:50 -0700157 * return 0 on success, negative error code on error.
158 *
159 * vma->vm_mm->mmap_sem must be held for at least read.
Nick Pigginb291f002008-10-18 20:26:44 -0700160 */
Rik van Rielba470de2008-10-18 20:26:50 -0700161static long __mlock_vma_pages_range(struct vm_area_struct *vma,
162 unsigned long start, unsigned long end,
163 int mlock)
Nick Pigginb291f002008-10-18 20:26:44 -0700164{
165 struct mm_struct *mm = vma->vm_mm;
166 unsigned long addr = start;
167 struct page *pages[16]; /* 16 gives a reasonable batch */
Nick Pigginb291f002008-10-18 20:26:44 -0700168 int nr_pages = (end - start) / PAGE_SIZE;
169 int ret;
Rik van Rielba470de2008-10-18 20:26:50 -0700170 int gup_flags = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700171
Rik van Rielba470de2008-10-18 20:26:50 -0700172 VM_BUG_ON(start & ~PAGE_MASK);
173 VM_BUG_ON(end & ~PAGE_MASK);
174 VM_BUG_ON(start < vma->vm_start);
175 VM_BUG_ON(end > vma->vm_end);
176 VM_BUG_ON((!rwsem_is_locked(&mm->mmap_sem)) &&
177 (atomic_read(&mm->mm_users) != 0));
178
179 /*
180 * mlock: don't page populate if page has PROT_NONE permission.
181 * munlock: the pages always do munlock althrough
182 * its has PROT_NONE permission.
183 */
184 if (!mlock)
185 gup_flags |= GUP_FLAGS_IGNORE_VMA_PERMISSIONS;
186
187 if (vma->vm_flags & VM_WRITE)
188 gup_flags |= GUP_FLAGS_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700189
190 lru_add_drain_all(); /* push cached pages to LRU */
191
192 while (nr_pages > 0) {
193 int i;
194
195 cond_resched();
196
197 /*
198 * get_user_pages makes pages present if we are
199 * setting mlock. and this extra reference count will
200 * disable migration of this page. However, page may
201 * still be truncated out from under us.
202 */
Rik van Rielba470de2008-10-18 20:26:50 -0700203 ret = __get_user_pages(current, mm, addr,
Nick Pigginb291f002008-10-18 20:26:44 -0700204 min_t(int, nr_pages, ARRAY_SIZE(pages)),
Rik van Rielba470de2008-10-18 20:26:50 -0700205 gup_flags, pages, NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700206 /*
207 * This can happen for, e.g., VM_NONLINEAR regions before
208 * a page has been allocated and mapped at a given offset,
209 * or for addresses that map beyond end of a file.
210 * We'll mlock the the pages if/when they get faulted in.
211 */
212 if (ret < 0)
213 break;
214 if (ret == 0) {
215 /*
216 * We know the vma is there, so the only time
217 * we cannot get a single page should be an
218 * error (ret < 0) case.
219 */
220 WARN_ON(1);
221 break;
222 }
223
224 lru_add_drain(); /* push cached pages to LRU */
225
226 for (i = 0; i < ret; i++) {
227 struct page *page = pages[i];
228
229 lock_page(page);
230 /*
231 * Because we lock page here and migration is blocked
232 * by the elevated reference, we need only check for
233 * page truncation (file-cache only).
234 */
Rik van Rielba470de2008-10-18 20:26:50 -0700235 if (page->mapping) {
236 if (mlock)
237 mlock_vma_page(page);
238 else
239 munlock_vma_page(page);
240 }
Nick Pigginb291f002008-10-18 20:26:44 -0700241 unlock_page(page);
242 put_page(page); /* ref from get_user_pages() */
243
244 /*
245 * here we assume that get_user_pages() has given us
246 * a list of virtually contiguous pages.
247 */
248 addr += PAGE_SIZE; /* for next get_user_pages() */
249 nr_pages--;
250 }
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700251 ret = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700252 }
253
254 lru_add_drain_all(); /* to update stats */
255
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700256 return ret; /* count entire vma as locked_vm */
257}
258
259/*
260 * convert get_user_pages() return value to posix mlock() error
261 */
262static int __mlock_posix_error_return(long retval)
263{
264 if (retval == -EFAULT)
265 retval = -ENOMEM;
266 else if (retval == -ENOMEM)
267 retval = -EAGAIN;
268 return retval;
Nick Pigginb291f002008-10-18 20:26:44 -0700269}
270
Nick Pigginb291f002008-10-18 20:26:44 -0700271#else /* CONFIG_UNEVICTABLE_LRU */
272
273/*
274 * Just make pages present if VM_LOCKED. No-op if unlocking.
275 */
Rik van Rielba470de2008-10-18 20:26:50 -0700276static long __mlock_vma_pages_range(struct vm_area_struct *vma,
277 unsigned long start, unsigned long end,
278 int mlock)
Nick Pigginb291f002008-10-18 20:26:44 -0700279{
Rik van Rielba470de2008-10-18 20:26:50 -0700280 if (mlock && (vma->vm_flags & VM_LOCKED))
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700281 return make_pages_present(start, end);
Nick Pigginb291f002008-10-18 20:26:44 -0700282 return 0;
283}
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700284
285static inline int __mlock_posix_error_return(long retval)
286{
287 return 0;
288}
289
Nick Pigginb291f002008-10-18 20:26:44 -0700290#endif /* CONFIG_UNEVICTABLE_LRU */
291
Rik van Rielba470de2008-10-18 20:26:50 -0700292/**
293 * mlock_vma_pages_range() - mlock pages in specified vma range.
294 * @vma - the vma containing the specfied address range
295 * @start - starting address in @vma to mlock
296 * @end - end address [+1] in @vma to mlock
297 *
298 * For mmap()/mremap()/expansion of mlocked vma.
299 *
300 * return 0 on success for "normal" vmas.
301 *
302 * return number of pages [> 0] to be removed from locked_vm on success
303 * of "special" vmas.
304 *
305 * return negative error if vma spanning @start-@range disappears while
306 * mmap semaphore is dropped. Unlikely?
Nick Pigginb291f002008-10-18 20:26:44 -0700307 */
Rik van Rielba470de2008-10-18 20:26:50 -0700308long mlock_vma_pages_range(struct vm_area_struct *vma,
Nick Pigginb291f002008-10-18 20:26:44 -0700309 unsigned long start, unsigned long end)
310{
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700311 struct mm_struct *mm = vma->vm_mm;
Nick Pigginb291f002008-10-18 20:26:44 -0700312 int nr_pages = (end - start) / PAGE_SIZE;
313 BUG_ON(!(vma->vm_flags & VM_LOCKED));
314
315 /*
316 * filter unlockable vmas
317 */
318 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
319 goto no_mlock;
320
321 if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
322 is_vm_hugetlb_page(vma) ||
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700323 vma == get_gate_vma(current))) {
Rik van Rielba470de2008-10-18 20:26:50 -0700324 long error;
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700325 downgrade_write(&mm->mmap_sem);
Rik van Rielba470de2008-10-18 20:26:50 -0700326
327 error = __mlock_vma_pages_range(vma, start, end, 1);
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700328
329 up_read(&mm->mmap_sem);
330 /* vma can change or disappear */
331 down_write(&mm->mmap_sem);
332 vma = find_vma(mm, start);
333 /* non-NULL vma must contain @start, but need to check @end */
334 if (!vma || end > vma->vm_end)
Rik van Rielba470de2008-10-18 20:26:50 -0700335 return -ENOMEM;
336
337 return 0; /* hide other errors from mmap(), et al */
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700338 }
Nick Pigginb291f002008-10-18 20:26:44 -0700339
340 /*
341 * User mapped kernel pages or huge pages:
342 * make these pages present to populate the ptes, but
343 * fall thru' to reset VM_LOCKED--no need to unlock, and
344 * return nr_pages so these don't get counted against task's
345 * locked limit. huge pages are already counted against
346 * locked vm limit.
347 */
348 make_pages_present(start, end);
349
350no_mlock:
351 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
Rik van Rielba470de2008-10-18 20:26:50 -0700352 return nr_pages; /* error or pages NOT mlocked */
Nick Pigginb291f002008-10-18 20:26:44 -0700353}
354
355
356/*
Rik van Rielba470de2008-10-18 20:26:50 -0700357 * munlock_vma_pages_range() - munlock all pages in the vma range.'
358 * @vma - vma containing range to be munlock()ed.
359 * @start - start address in @vma of the range
360 * @end - end of range in @vma.
361 *
362 * For mremap(), munmap() and exit().
363 *
364 * Called with @vma VM_LOCKED.
365 *
366 * Returns with VM_LOCKED cleared. Callers must be prepared to
367 * deal with this.
368 *
369 * We don't save and restore VM_LOCKED here because pages are
370 * still on lru. In unmap path, pages might be scanned by reclaim
371 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
372 * free them. This will result in freeing mlocked pages.
Nick Pigginb291f002008-10-18 20:26:44 -0700373 */
Rik van Rielba470de2008-10-18 20:26:50 -0700374void munlock_vma_pages_range(struct vm_area_struct *vma,
375 unsigned long start, unsigned long end)
Nick Pigginb291f002008-10-18 20:26:44 -0700376{
377 vma->vm_flags &= ~VM_LOCKED;
Rik van Rielba470de2008-10-18 20:26:50 -0700378 __mlock_vma_pages_range(vma, start, end, 0);
Nick Pigginb291f002008-10-18 20:26:44 -0700379}
380
381/*
382 * mlock_fixup - handle mlock[all]/munlock[all] requests.
383 *
384 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
385 * munlock is a no-op. However, for some special vmas, we go ahead and
386 * populate the ptes via make_pages_present().
387 *
388 * For vmas that pass the filters, merge/split as appropriate.
389 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
391 unsigned long start, unsigned long end, unsigned int newflags)
392{
Nick Pigginb291f002008-10-18 20:26:44 -0700393 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700395 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 int ret = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700397 int lock = newflags & VM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Nick Pigginb291f002008-10-18 20:26:44 -0700399 if (newflags == vma->vm_flags ||
400 (vma->vm_flags & (VM_IO | VM_PFNMAP)))
401 goto out; /* don't set VM_LOCKED, don't count */
402
403 if ((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
404 is_vm_hugetlb_page(vma) ||
405 vma == get_gate_vma(current)) {
406 if (lock)
407 make_pages_present(start, end);
408 goto out; /* don't set VM_LOCKED, don't count */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
410
411 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
412 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
413 vma->vm_file, pgoff, vma_policy(vma));
414 if (*prev) {
415 vma = *prev;
416 goto success;
417 }
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 if (start != vma->vm_start) {
420 ret = split_vma(mm, vma, start, 1);
421 if (ret)
422 goto out;
423 }
424
425 if (end != vma->vm_end) {
426 ret = split_vma(mm, vma, end, 0);
427 if (ret)
428 goto out;
429 }
430
431success:
432 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700433 * Keep track of amount of locked VM.
434 */
435 nr_pages = (end - start) >> PAGE_SHIFT;
436 if (!lock)
437 nr_pages = -nr_pages;
438 mm->locked_vm += nr_pages;
439
440 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * vm_flags is protected by the mmap_sem held in write mode.
442 * It's okay if try_to_unmap_one unmaps a page just after we
Nick Pigginb291f002008-10-18 20:26:44 -0700443 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 */
445 vma->vm_flags = newflags;
446
Nick Pigginb291f002008-10-18 20:26:44 -0700447 if (lock) {
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700448 /*
449 * mmap_sem is currently held for write. Downgrade the write
450 * lock to a read lock so that other faults, mmap scans, ...
451 * while we fault in all pages.
452 */
453 downgrade_write(&mm->mmap_sem);
454
Rik van Rielba470de2008-10-18 20:26:50 -0700455 ret = __mlock_vma_pages_range(vma, start, end, 1);
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700456
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700457 /*
458 * Need to reacquire mmap sem in write mode, as our callers
459 * expect this. We have no support for atomically upgrading
460 * a sem to write, so we need to check for ranges while sem
461 * is unlocked.
462 */
463 up_read(&mm->mmap_sem);
464 /* vma can change or disappear */
465 down_write(&mm->mmap_sem);
466 *prev = find_vma(mm, start);
467 /* non-NULL *prev must contain @start, but need to check @end */
468 if (!(*prev) || end > (*prev)->vm_end)
Rik van Rielba470de2008-10-18 20:26:50 -0700469 ret = -ENOMEM;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -0700470 else if (ret > 0) {
471 mm->locked_vm -= ret;
472 ret = 0;
473 } else
474 ret = __mlock_posix_error_return(ret); /* translate if needed */
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700475 } else {
476 /*
477 * TODO: for unlocking, pages will already be resident, so
478 * we don't need to wait for allocations/reclaim/pagein, ...
479 * However, unlocking a very large region can still take a
480 * while. Should we downgrade the semaphore for both lock
481 * AND unlock ?
482 */
Rik van Rielba470de2008-10-18 20:26:50 -0700483 __mlock_vma_pages_range(vma, start, end, 0);
Lee Schermerhorn8edb08c2008-10-18 20:26:49 -0700484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486out:
Nick Pigginb291f002008-10-18 20:26:44 -0700487 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 return ret;
489}
490
491static int do_mlock(unsigned long start, size_t len, int on)
492{
493 unsigned long nstart, end, tmp;
494 struct vm_area_struct * vma, * prev;
495 int error;
496
497 len = PAGE_ALIGN(len);
498 end = start + len;
499 if (end < start)
500 return -EINVAL;
501 if (end == start)
502 return 0;
503 vma = find_vma_prev(current->mm, start, &prev);
504 if (!vma || vma->vm_start > start)
505 return -ENOMEM;
506
507 if (start > vma->vm_start)
508 prev = vma;
509
510 for (nstart = start ; ; ) {
511 unsigned int newflags;
512
513 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
514
515 newflags = vma->vm_flags | VM_LOCKED;
516 if (!on)
517 newflags &= ~VM_LOCKED;
518
519 tmp = vma->vm_end;
520 if (tmp > end)
521 tmp = end;
522 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
523 if (error)
524 break;
525 nstart = tmp;
526 if (nstart < prev->vm_end)
527 nstart = prev->vm_end;
528 if (nstart >= end)
529 break;
530
531 vma = prev->vm_next;
532 if (!vma || vma->vm_start != nstart) {
533 error = -ENOMEM;
534 break;
535 }
536 }
537 return error;
538}
539
540asmlinkage long sys_mlock(unsigned long start, size_t len)
541{
542 unsigned long locked;
543 unsigned long lock_limit;
544 int error = -ENOMEM;
545
546 if (!can_do_mlock())
547 return -EPERM;
548
549 down_write(&current->mm->mmap_sem);
550 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
551 start &= PAGE_MASK;
552
553 locked = len >> PAGE_SHIFT;
554 locked += current->mm->locked_vm;
555
556 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
557 lock_limit >>= PAGE_SHIFT;
558
559 /* check against resource limits */
560 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
561 error = do_mlock(start, len, 1);
562 up_write(&current->mm->mmap_sem);
563 return error;
564}
565
566asmlinkage long sys_munlock(unsigned long start, size_t len)
567{
568 int ret;
569
570 down_write(&current->mm->mmap_sem);
571 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
572 start &= PAGE_MASK;
573 ret = do_mlock(start, len, 0);
574 up_write(&current->mm->mmap_sem);
575 return ret;
576}
577
578static int do_mlockall(int flags)
579{
580 struct vm_area_struct * vma, * prev = NULL;
581 unsigned int def_flags = 0;
582
583 if (flags & MCL_FUTURE)
584 def_flags = VM_LOCKED;
585 current->mm->def_flags = def_flags;
586 if (flags == MCL_FUTURE)
587 goto out;
588
589 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
590 unsigned int newflags;
591
592 newflags = vma->vm_flags | VM_LOCKED;
593 if (!(flags & MCL_CURRENT))
594 newflags &= ~VM_LOCKED;
595
596 /* Ignore errors */
597 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
598 }
599out:
600 return 0;
601}
602
603asmlinkage long sys_mlockall(int flags)
604{
605 unsigned long lock_limit;
606 int ret = -EINVAL;
607
608 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
609 goto out;
610
611 ret = -EPERM;
612 if (!can_do_mlock())
613 goto out;
614
615 down_write(&current->mm->mmap_sem);
616
617 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
618 lock_limit >>= PAGE_SHIFT;
619
620 ret = -ENOMEM;
621 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
622 capable(CAP_IPC_LOCK))
623 ret = do_mlockall(flags);
624 up_write(&current->mm->mmap_sem);
625out:
626 return ret;
627}
628
629asmlinkage long sys_munlockall(void)
630{
631 int ret;
632
633 down_write(&current->mm->mmap_sem);
634 ret = do_mlockall(0);
635 up_write(&current->mm->mmap_sem);
636 return ret;
637}
638
639/*
640 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
641 * shm segments) get accounted against the user_struct instead.
642 */
643static DEFINE_SPINLOCK(shmlock_user_lock);
644
645int user_shm_lock(size_t size, struct user_struct *user)
646{
647 unsigned long lock_limit, locked;
648 int allowed = 0;
649
650 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
651 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700652 if (lock_limit == RLIM_INFINITY)
653 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 lock_limit >>= PAGE_SHIFT;
655 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700656 if (!allowed &&
657 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 goto out;
659 get_uid(user);
660 user->locked_shm += locked;
661 allowed = 1;
662out:
663 spin_unlock(&shmlock_user_lock);
664 return allowed;
665}
666
667void user_shm_unlock(size_t size, struct user_struct *user)
668{
669 spin_lock(&shmlock_user_lock);
670 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
671 spin_unlock(&shmlock_user_lock);
672 free_uid(user);
673}