Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/mlock.c | |
3 | * | |
4 | * (C) Copyright 1995 Linus Torvalds | |
5 | * (C) Copyright 2002 Christoph Hellwig | |
6 | */ | |
7 | ||
c59ede7b | 8 | #include <linux/capability.h> |
1da177e4 LT |
9 | #include <linux/mman.h> |
10 | #include <linux/mm.h> | |
b291f000 NP |
11 | #include <linux/swap.h> |
12 | #include <linux/swapops.h> | |
13 | #include <linux/pagemap.h> | |
1da177e4 LT |
14 | #include <linux/mempolicy.h> |
15 | #include <linux/syscalls.h> | |
e8edc6e0 AD |
16 | #include <linux/sched.h> |
17 | #include <linux/module.h> | |
b291f000 NP |
18 | #include <linux/rmap.h> |
19 | #include <linux/mmzone.h> | |
20 | #include <linux/hugetlb.h> | |
21 | ||
22 | #include "internal.h" | |
1da177e4 | 23 | |
e8edc6e0 AD |
24 | int 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 | } | |
32 | EXPORT_SYMBOL(can_do_mlock); | |
1da177e4 | 33 | |
b291f000 NP |
34 | /* |
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 | */ | |
54 | void __clear_page_mlock(struct page *page) | |
55 | { | |
56 | VM_BUG_ON(!PageLocked(page)); | |
57 | ||
58 | if (!page->mapping) { /* truncated ? */ | |
59 | return; | |
60 | } | |
61 | ||
5344b7e6 NP |
62 | dec_zone_page_state(page, NR_MLOCK); |
63 | count_vm_event(UNEVICTABLE_PGCLEARED); | |
b291f000 NP |
64 | if (!isolate_lru_page(page)) { |
65 | putback_lru_page(page); | |
66 | } else { | |
67 | /* | |
8891d6da | 68 | * We lost the race. the page already moved to evictable list. |
b291f000 | 69 | */ |
8891d6da | 70 | if (PageUnevictable(page)) |
5344b7e6 | 71 | count_vm_event(UNEVICTABLE_PGSTRANDED); |
b291f000 NP |
72 | } |
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 | */ | |
79 | void mlock_vma_page(struct page *page) | |
80 | { | |
81 | BUG_ON(!PageLocked(page)); | |
82 | ||
5344b7e6 NP |
83 | 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 | } | |
b291f000 NP |
89 | } |
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 | */ | |
109 | static void munlock_vma_page(struct page *page) | |
110 | { | |
111 | BUG_ON(!PageLocked(page)); | |
112 | ||
5344b7e6 NP |
113 | 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 | } | |
b291f000 NP |
138 | } |
139 | } | |
140 | ||
ba470de4 | 141 | /** |
408e82b7 | 142 | * __mlock_vma_pages_range() - mlock a range of pages in the vma. |
ba470de4 RR |
143 | * @vma: target vma |
144 | * @start: start address | |
145 | * @end: end address | |
ba470de4 | 146 | * |
408e82b7 | 147 | * This takes care of making the pages present too. |
b291f000 | 148 | * |
ba470de4 | 149 | * return 0 on success, negative error code on error. |
b291f000 | 150 | * |
ba470de4 | 151 | * vma->vm_mm->mmap_sem must be held for at least read. |
b291f000 | 152 | */ |
ba470de4 | 153 | static long __mlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 154 | unsigned long start, unsigned long end) |
b291f000 NP |
155 | { |
156 | struct mm_struct *mm = vma->vm_mm; | |
157 | unsigned long addr = start; | |
158 | struct page *pages[16]; /* 16 gives a reasonable batch */ | |
b291f000 | 159 | int nr_pages = (end - start) / PAGE_SIZE; |
72eb8c67 | 160 | int ret = 0; |
408e82b7 | 161 | int gup_flags; |
ba470de4 RR |
162 | |
163 | 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); | |
408e82b7 | 167 | VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem)); |
b291f000 | 168 | |
58fa879e | 169 | gup_flags = FOLL_TOUCH | FOLL_GET; |
ba470de4 | 170 | if (vma->vm_flags & VM_WRITE) |
58fa879e | 171 | gup_flags |= FOLL_WRITE; |
b291f000 | 172 | |
b291f000 NP |
173 | 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 | */ | |
ba470de4 | 184 | ret = __get_user_pages(current, mm, addr, |
b291f000 | 185 | min_t(int, nr_pages, ARRAY_SIZE(pages)), |
ba470de4 | 186 | gup_flags, pages, NULL); |
b291f000 NP |
187 | /* |
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. | |
408e82b7 | 191 | * We'll mlock the pages if/when they get faulted in. |
b291f000 NP |
192 | */ |
193 | if (ret < 0) | |
194 | break; | |
b291f000 NP |
195 | |
196 | lru_add_drain(); /* push cached pages to LRU */ | |
197 | ||
198 | for (i = 0; i < ret; i++) { | |
199 | struct page *page = pages[i]; | |
200 | ||
6e919717 HD |
201 | if (page->mapping) { |
202 | /* | |
203 | * That preliminary check is mainly to avoid | |
204 | * the pointless overhead of lock_page on the | |
205 | * ZERO_PAGE: which might bounce very badly if | |
206 | * there is contention. However, we're still | |
207 | * dirtying its cacheline with get/put_page: | |
208 | * we'll add another __get_user_pages flag to | |
209 | * avoid it if that case turns out to matter. | |
210 | */ | |
211 | lock_page(page); | |
212 | /* | |
213 | * Because we lock page here and migration is | |
214 | * blocked by the elevated reference, we need | |
215 | * only check for file-cache page truncation. | |
216 | */ | |
217 | if (page->mapping) | |
218 | mlock_vma_page(page); | |
219 | unlock_page(page); | |
220 | } | |
408e82b7 | 221 | put_page(page); /* ref from get_user_pages() */ |
b291f000 | 222 | } |
408e82b7 HD |
223 | |
224 | addr += ret * PAGE_SIZE; | |
225 | nr_pages -= ret; | |
9978ad58 | 226 | ret = 0; |
b291f000 NP |
227 | } |
228 | ||
408e82b7 | 229 | return ret; /* 0 or negative error code */ |
9978ad58 LS |
230 | } |
231 | ||
232 | /* | |
233 | * convert get_user_pages() return value to posix mlock() error | |
234 | */ | |
235 | static int __mlock_posix_error_return(long retval) | |
236 | { | |
237 | if (retval == -EFAULT) | |
238 | retval = -ENOMEM; | |
239 | else if (retval == -ENOMEM) | |
240 | retval = -EAGAIN; | |
241 | return retval; | |
b291f000 NP |
242 | } |
243 | ||
ba470de4 RR |
244 | /** |
245 | * mlock_vma_pages_range() - mlock pages in specified vma range. | |
246 | * @vma - the vma containing the specfied address range | |
247 | * @start - starting address in @vma to mlock | |
248 | * @end - end address [+1] in @vma to mlock | |
249 | * | |
250 | * For mmap()/mremap()/expansion of mlocked vma. | |
251 | * | |
252 | * return 0 on success for "normal" vmas. | |
253 | * | |
254 | * return number of pages [> 0] to be removed from locked_vm on success | |
255 | * of "special" vmas. | |
b291f000 | 256 | */ |
ba470de4 | 257 | long mlock_vma_pages_range(struct vm_area_struct *vma, |
b291f000 NP |
258 | unsigned long start, unsigned long end) |
259 | { | |
260 | int nr_pages = (end - start) / PAGE_SIZE; | |
261 | BUG_ON(!(vma->vm_flags & VM_LOCKED)); | |
262 | ||
263 | /* | |
264 | * filter unlockable vmas | |
265 | */ | |
266 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
267 | goto no_mlock; | |
268 | ||
269 | if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) || | |
270 | is_vm_hugetlb_page(vma) || | |
8edb08ca | 271 | vma == get_gate_vma(current))) { |
8edb08ca | 272 | |
408e82b7 | 273 | __mlock_vma_pages_range(vma, start, end); |
d5b56233 HD |
274 | |
275 | /* Hide errors from mmap() and other callers */ | |
276 | return 0; | |
8edb08ca | 277 | } |
b291f000 NP |
278 | |
279 | /* | |
280 | * User mapped kernel pages or huge pages: | |
281 | * make these pages present to populate the ptes, but | |
282 | * fall thru' to reset VM_LOCKED--no need to unlock, and | |
283 | * return nr_pages so these don't get counted against task's | |
284 | * locked limit. huge pages are already counted against | |
285 | * locked vm limit. | |
286 | */ | |
287 | make_pages_present(start, end); | |
288 | ||
289 | no_mlock: | |
290 | vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */ | |
ba470de4 | 291 | return nr_pages; /* error or pages NOT mlocked */ |
b291f000 NP |
292 | } |
293 | ||
b291f000 | 294 | /* |
ba470de4 RR |
295 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
296 | * @vma - vma containing range to be munlock()ed. | |
297 | * @start - start address in @vma of the range | |
298 | * @end - end of range in @vma. | |
299 | * | |
300 | * For mremap(), munmap() and exit(). | |
301 | * | |
302 | * Called with @vma VM_LOCKED. | |
303 | * | |
304 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
305 | * deal with this. | |
306 | * | |
307 | * We don't save and restore VM_LOCKED here because pages are | |
308 | * still on lru. In unmap path, pages might be scanned by reclaim | |
309 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
310 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 311 | */ |
ba470de4 | 312 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 313 | unsigned long start, unsigned long end) |
b291f000 | 314 | { |
408e82b7 HD |
315 | unsigned long addr; |
316 | ||
317 | lru_add_drain(); | |
b291f000 | 318 | vma->vm_flags &= ~VM_LOCKED; |
408e82b7 HD |
319 | |
320 | for (addr = start; addr < end; addr += PAGE_SIZE) { | |
6e919717 HD |
321 | struct page *page; |
322 | /* | |
323 | * Although FOLL_DUMP is intended for get_dump_page(), | |
324 | * it just so happens that its special treatment of the | |
325 | * ZERO_PAGE (returning an error instead of doing get_page) | |
326 | * suits munlock very well (and if somehow an abnormal page | |
327 | * has sneaked into the range, we won't oops here: great). | |
328 | */ | |
329 | page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP); | |
330 | if (page && !IS_ERR(page)) { | |
408e82b7 | 331 | lock_page(page); |
6e919717 HD |
332 | /* |
333 | * Like in __mlock_vma_pages_range(), | |
334 | * because we lock page here and migration is | |
335 | * blocked by the elevated reference, we need | |
336 | * only check for file-cache page truncation. | |
337 | */ | |
408e82b7 HD |
338 | if (page->mapping) |
339 | munlock_vma_page(page); | |
340 | unlock_page(page); | |
341 | put_page(page); | |
342 | } | |
343 | cond_resched(); | |
344 | } | |
b291f000 NP |
345 | } |
346 | ||
347 | /* | |
348 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
349 | * | |
350 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
351 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
352 | * populate the ptes via make_pages_present(). | |
353 | * | |
354 | * For vmas that pass the filters, merge/split as appropriate. | |
355 | */ | |
1da177e4 LT |
356 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
357 | unsigned long start, unsigned long end, unsigned int newflags) | |
358 | { | |
b291f000 | 359 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 360 | pgoff_t pgoff; |
b291f000 | 361 | int nr_pages; |
1da177e4 | 362 | int ret = 0; |
b291f000 | 363 | int lock = newflags & VM_LOCKED; |
1da177e4 | 364 | |
b291f000 NP |
365 | if (newflags == vma->vm_flags || |
366 | (vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
367 | goto out; /* don't set VM_LOCKED, don't count */ | |
368 | ||
369 | if ((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) || | |
370 | is_vm_hugetlb_page(vma) || | |
371 | vma == get_gate_vma(current)) { | |
372 | if (lock) | |
373 | make_pages_present(start, end); | |
374 | goto out; /* don't set VM_LOCKED, don't count */ | |
1da177e4 LT |
375 | } |
376 | ||
377 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
378 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
379 | vma->vm_file, pgoff, vma_policy(vma)); | |
380 | if (*prev) { | |
381 | vma = *prev; | |
382 | goto success; | |
383 | } | |
384 | ||
1da177e4 LT |
385 | if (start != vma->vm_start) { |
386 | ret = split_vma(mm, vma, start, 1); | |
387 | if (ret) | |
388 | goto out; | |
389 | } | |
390 | ||
391 | if (end != vma->vm_end) { | |
392 | ret = split_vma(mm, vma, end, 0); | |
393 | if (ret) | |
394 | goto out; | |
395 | } | |
396 | ||
397 | success: | |
b291f000 NP |
398 | /* |
399 | * Keep track of amount of locked VM. | |
400 | */ | |
401 | nr_pages = (end - start) >> PAGE_SHIFT; | |
402 | if (!lock) | |
403 | nr_pages = -nr_pages; | |
404 | mm->locked_vm += nr_pages; | |
405 | ||
1da177e4 LT |
406 | /* |
407 | * vm_flags is protected by the mmap_sem held in write mode. | |
408 | * It's okay if try_to_unmap_one unmaps a page just after we | |
b291f000 | 409 | * set VM_LOCKED, __mlock_vma_pages_range will bring it back. |
1da177e4 | 410 | */ |
1da177e4 | 411 | |
b291f000 | 412 | if (lock) { |
408e82b7 HD |
413 | vma->vm_flags = newflags; |
414 | ret = __mlock_vma_pages_range(vma, start, end); | |
415 | if (ret < 0) | |
416 | ret = __mlock_posix_error_return(ret); | |
8edb08ca | 417 | } else { |
408e82b7 | 418 | munlock_vma_pages_range(vma, start, end); |
8edb08ca | 419 | } |
1da177e4 | 420 | |
1da177e4 | 421 | out: |
b291f000 | 422 | *prev = vma; |
1da177e4 LT |
423 | return ret; |
424 | } | |
425 | ||
426 | static int do_mlock(unsigned long start, size_t len, int on) | |
427 | { | |
428 | unsigned long nstart, end, tmp; | |
429 | struct vm_area_struct * vma, * prev; | |
430 | int error; | |
431 | ||
432 | len = PAGE_ALIGN(len); | |
433 | end = start + len; | |
434 | if (end < start) | |
435 | return -EINVAL; | |
436 | if (end == start) | |
437 | return 0; | |
438 | vma = find_vma_prev(current->mm, start, &prev); | |
439 | if (!vma || vma->vm_start > start) | |
440 | return -ENOMEM; | |
441 | ||
442 | if (start > vma->vm_start) | |
443 | prev = vma; | |
444 | ||
445 | for (nstart = start ; ; ) { | |
446 | unsigned int newflags; | |
447 | ||
448 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ | |
449 | ||
450 | newflags = vma->vm_flags | VM_LOCKED; | |
451 | if (!on) | |
452 | newflags &= ~VM_LOCKED; | |
453 | ||
454 | tmp = vma->vm_end; | |
455 | if (tmp > end) | |
456 | tmp = end; | |
457 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
458 | if (error) | |
459 | break; | |
460 | nstart = tmp; | |
461 | if (nstart < prev->vm_end) | |
462 | nstart = prev->vm_end; | |
463 | if (nstart >= end) | |
464 | break; | |
465 | ||
466 | vma = prev->vm_next; | |
467 | if (!vma || vma->vm_start != nstart) { | |
468 | error = -ENOMEM; | |
469 | break; | |
470 | } | |
471 | } | |
472 | return error; | |
473 | } | |
474 | ||
6a6160a7 | 475 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
1da177e4 LT |
476 | { |
477 | unsigned long locked; | |
478 | unsigned long lock_limit; | |
479 | int error = -ENOMEM; | |
480 | ||
481 | if (!can_do_mlock()) | |
482 | return -EPERM; | |
483 | ||
8891d6da KM |
484 | lru_add_drain_all(); /* flush pagevec */ |
485 | ||
1da177e4 LT |
486 | down_write(¤t->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 | ||
493 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; | |
494 | 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(¤t->mm->mmap_sem); | |
500 | return error; | |
501 | } | |
502 | ||
6a6160a7 | 503 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
504 | { |
505 | int ret; | |
506 | ||
507 | down_write(¤t->mm->mmap_sem); | |
508 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); | |
509 | start &= PAGE_MASK; | |
510 | ret = do_mlock(start, len, 0); | |
511 | up_write(¤t->mm->mmap_sem); | |
512 | return ret; | |
513 | } | |
514 | ||
515 | static int do_mlockall(int flags) | |
516 | { | |
517 | struct vm_area_struct * vma, * prev = NULL; | |
518 | unsigned int def_flags = 0; | |
519 | ||
520 | if (flags & MCL_FUTURE) | |
521 | def_flags = VM_LOCKED; | |
522 | current->mm->def_flags = def_flags; | |
523 | if (flags == MCL_FUTURE) | |
524 | goto out; | |
525 | ||
526 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
527 | unsigned int newflags; | |
528 | ||
529 | newflags = vma->vm_flags | VM_LOCKED; | |
530 | if (!(flags & MCL_CURRENT)) | |
531 | newflags &= ~VM_LOCKED; | |
532 | ||
533 | /* Ignore errors */ | |
534 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
535 | } | |
536 | out: | |
537 | return 0; | |
538 | } | |
539 | ||
3480b257 | 540 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
541 | { |
542 | unsigned long lock_limit; | |
543 | int ret = -EINVAL; | |
544 | ||
545 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE))) | |
546 | goto out; | |
547 | ||
548 | ret = -EPERM; | |
549 | if (!can_do_mlock()) | |
550 | goto out; | |
551 | ||
8891d6da KM |
552 | lru_add_drain_all(); /* flush pagevec */ |
553 | ||
1da177e4 LT |
554 | down_write(¤t->mm->mmap_sem); |
555 | ||
556 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; | |
557 | lock_limit >>= PAGE_SHIFT; | |
558 | ||
559 | ret = -ENOMEM; | |
560 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || | |
561 | capable(CAP_IPC_LOCK)) | |
562 | ret = do_mlockall(flags); | |
563 | up_write(¤t->mm->mmap_sem); | |
564 | out: | |
565 | return ret; | |
566 | } | |
567 | ||
3480b257 | 568 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
569 | { |
570 | int ret; | |
571 | ||
572 | down_write(¤t->mm->mmap_sem); | |
573 | ret = do_mlockall(0); | |
574 | up_write(¤t->mm->mmap_sem); | |
575 | return ret; | |
576 | } | |
577 | ||
578 | /* | |
579 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
580 | * shm segments) get accounted against the user_struct instead. | |
581 | */ | |
582 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
583 | ||
584 | int user_shm_lock(size_t size, struct user_struct *user) | |
585 | { | |
586 | unsigned long lock_limit, locked; | |
587 | int allowed = 0; | |
588 | ||
589 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
590 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; | |
5ed44a40 HB |
591 | if (lock_limit == RLIM_INFINITY) |
592 | allowed = 1; | |
1da177e4 LT |
593 | lock_limit >>= PAGE_SHIFT; |
594 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
595 | if (!allowed && |
596 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
597 | goto out; |
598 | get_uid(user); | |
599 | user->locked_shm += locked; | |
600 | allowed = 1; | |
601 | out: | |
602 | spin_unlock(&shmlock_user_lock); | |
603 | return allowed; | |
604 | } | |
605 | ||
606 | void user_shm_unlock(size_t size, struct user_struct *user) | |
607 | { | |
608 | spin_lock(&shmlock_user_lock); | |
609 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
610 | spin_unlock(&shmlock_user_lock); | |
611 | free_uid(user); | |
612 | } | |
c5dee617 | 613 | |
1cb81b14 MM |
614 | int account_locked_memory(struct mm_struct *mm, struct rlimit *rlim, |
615 | size_t size) | |
c5dee617 | 616 | { |
1cb81b14 MM |
617 | unsigned long lim, vm, pgsz; |
618 | int error = -ENOMEM; | |
c5dee617 MM |
619 | |
620 | pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT; | |
621 | ||
1cb81b14 | 622 | down_write(&mm->mmap_sem); |
c5dee617 | 623 | |
1cb81b14 MM |
624 | lim = rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT; |
625 | vm = mm->total_vm + pgsz; | |
626 | if (lim < vm) | |
c5dee617 MM |
627 | goto out; |
628 | ||
1cb81b14 MM |
629 | lim = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; |
630 | vm = mm->locked_vm + pgsz; | |
631 | if (lim < vm) | |
c5dee617 MM |
632 | goto out; |
633 | ||
1cb81b14 MM |
634 | mm->total_vm += pgsz; |
635 | mm->locked_vm += pgsz; | |
c5dee617 | 636 | |
1cb81b14 | 637 | error = 0; |
c5dee617 | 638 | out: |
1cb81b14 MM |
639 | up_write(&mm->mmap_sem); |
640 | return error; | |
c5dee617 MM |
641 | } |
642 | ||
1cb81b14 | 643 | void refund_locked_memory(struct mm_struct *mm, size_t size) |
c5dee617 MM |
644 | { |
645 | unsigned long pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT; | |
646 | ||
e2b371f0 | 647 | down_write(&mm->mmap_sem); |
c5dee617 | 648 | |
e2b371f0 MM |
649 | mm->total_vm -= pgsz; |
650 | mm->locked_vm -= pgsz; | |
c5dee617 | 651 | |
e2b371f0 | 652 | up_write(&mm->mmap_sem); |
9f339e70 | 653 | } |