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 | 16 | #include <linux/sched.h> |
b95f1b31 | 17 | #include <linux/export.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; | |
59e99e5b | 28 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
e8edc6e0 AD |
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 | */ | |
e6c509f8 | 54 | void clear_page_mlock(struct page *page) |
b291f000 | 55 | { |
e6c509f8 | 56 | if (!TestClearPageMlocked(page)) |
b291f000 | 57 | return; |
b291f000 | 58 | |
8449d21f DR |
59 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
60 | -hpage_nr_pages(page)); | |
5344b7e6 | 61 | count_vm_event(UNEVICTABLE_PGCLEARED); |
b291f000 NP |
62 | if (!isolate_lru_page(page)) { |
63 | putback_lru_page(page); | |
64 | } else { | |
65 | /* | |
8891d6da | 66 | * We lost the race. the page already moved to evictable list. |
b291f000 | 67 | */ |
8891d6da | 68 | if (PageUnevictable(page)) |
5344b7e6 | 69 | count_vm_event(UNEVICTABLE_PGSTRANDED); |
b291f000 NP |
70 | } |
71 | } | |
72 | ||
73 | /* | |
74 | * Mark page as mlocked if not already. | |
75 | * If page on LRU, isolate and putback to move to unevictable list. | |
76 | */ | |
77 | void mlock_vma_page(struct page *page) | |
78 | { | |
77552735 | 79 | /* Serialize with page migration */ |
b291f000 NP |
80 | BUG_ON(!PageLocked(page)); |
81 | ||
5344b7e6 | 82 | if (!TestSetPageMlocked(page)) { |
8449d21f DR |
83 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
84 | hpage_nr_pages(page)); | |
5344b7e6 NP |
85 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
86 | if (!isolate_lru_page(page)) | |
87 | putback_lru_page(page); | |
88 | } | |
b291f000 NP |
89 | } |
90 | ||
6927c1dd LS |
91 | /** |
92 | * munlock_vma_page - munlock a vma page | |
93 | * @page - page to be unlocked | |
b291f000 | 94 | * |
6927c1dd LS |
95 | * 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. | |
b291f000 | 105 | */ |
ff6a6da6 | 106 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 107 | { |
ff6a6da6 ML |
108 | unsigned int page_mask = 0; |
109 | ||
77552735 | 110 | /* For try_to_munlock() and to serialize with page migration */ |
b291f000 NP |
111 | BUG_ON(!PageLocked(page)); |
112 | ||
5344b7e6 | 113 | if (TestClearPageMlocked(page)) { |
ff6a6da6 ML |
114 | unsigned int nr_pages = hpage_nr_pages(page); |
115 | mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); | |
116 | page_mask = nr_pages - 1; | |
5344b7e6 | 117 | if (!isolate_lru_page(page)) { |
3d470fc3 HD |
118 | int ret = SWAP_AGAIN; |
119 | ||
120 | /* | |
121 | * Optimization: if the page was mapped just once, | |
122 | * that's our mapping and we don't need to check all the | |
123 | * other vmas. | |
124 | */ | |
125 | if (page_mapcount(page) > 1) | |
126 | ret = try_to_munlock(page); | |
5344b7e6 NP |
127 | /* |
128 | * did try_to_unlock() succeed or punt? | |
129 | */ | |
53f79acb | 130 | if (ret != SWAP_MLOCK) |
5344b7e6 NP |
131 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
132 | ||
133 | putback_lru_page(page); | |
134 | } else { | |
135 | /* | |
6927c1dd LS |
136 | * Some other task has removed the page from the LRU. |
137 | * putback_lru_page() will take care of removing the | |
138 | * page from the unevictable list, if necessary. | |
139 | * vmscan [page_referenced()] will move the page back | |
140 | * to the unevictable list if some other vma has it | |
141 | * mlocked. | |
5344b7e6 NP |
142 | */ |
143 | if (PageUnevictable(page)) | |
144 | count_vm_event(UNEVICTABLE_PGSTRANDED); | |
145 | else | |
146 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); | |
147 | } | |
b291f000 | 148 | } |
ff6a6da6 ML |
149 | |
150 | return page_mask; | |
b291f000 NP |
151 | } |
152 | ||
ba470de4 | 153 | /** |
408e82b7 | 154 | * __mlock_vma_pages_range() - mlock a range of pages in the vma. |
ba470de4 RR |
155 | * @vma: target vma |
156 | * @start: start address | |
157 | * @end: end address | |
ba470de4 | 158 | * |
408e82b7 | 159 | * This takes care of making the pages present too. |
b291f000 | 160 | * |
ba470de4 | 161 | * return 0 on success, negative error code on error. |
b291f000 | 162 | * |
ba470de4 | 163 | * vma->vm_mm->mmap_sem must be held for at least read. |
b291f000 | 164 | */ |
cea10a19 ML |
165 | long __mlock_vma_pages_range(struct vm_area_struct *vma, |
166 | unsigned long start, unsigned long end, int *nonblocking) | |
b291f000 NP |
167 | { |
168 | struct mm_struct *mm = vma->vm_mm; | |
28a35716 | 169 | unsigned long nr_pages = (end - start) / PAGE_SIZE; |
408e82b7 | 170 | int gup_flags; |
ba470de4 RR |
171 | |
172 | 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); | |
408e82b7 | 176 | VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem)); |
b291f000 | 177 | |
a1fde08c | 178 | gup_flags = FOLL_TOUCH | FOLL_MLOCK; |
5ecfda04 ML |
179 | /* |
180 | * We want to touch writable mappings with a write fault in order | |
181 | * to break COW, except for shared mappings because these don't COW | |
182 | * and we would not want to dirty them for nothing. | |
183 | */ | |
184 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
58fa879e | 185 | gup_flags |= FOLL_WRITE; |
b291f000 | 186 | |
fdf4c587 ML |
187 | /* |
188 | * We want mlock to succeed for regions that have any permissions | |
189 | * other than PROT_NONE. | |
190 | */ | |
191 | if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)) | |
192 | gup_flags |= FOLL_FORCE; | |
193 | ||
4805b02e JW |
194 | /* |
195 | * We made sure addr is within a VMA, so the following will | |
196 | * not result in a stack expansion that recurses back here. | |
197 | */ | |
ff6a6da6 | 198 | return __get_user_pages(current, mm, start, nr_pages, gup_flags, |
53a7706d | 199 | NULL, NULL, nonblocking); |
9978ad58 LS |
200 | } |
201 | ||
202 | /* | |
203 | * convert get_user_pages() return value to posix mlock() error | |
204 | */ | |
205 | static int __mlock_posix_error_return(long retval) | |
206 | { | |
207 | if (retval == -EFAULT) | |
208 | retval = -ENOMEM; | |
209 | else if (retval == -ENOMEM) | |
210 | retval = -EAGAIN; | |
211 | return retval; | |
b291f000 NP |
212 | } |
213 | ||
b291f000 | 214 | /* |
ba470de4 RR |
215 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
216 | * @vma - vma containing range to be munlock()ed. | |
217 | * @start - start address in @vma of the range | |
218 | * @end - end of range in @vma. | |
219 | * | |
220 | * For mremap(), munmap() and exit(). | |
221 | * | |
222 | * Called with @vma VM_LOCKED. | |
223 | * | |
224 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
225 | * deal with this. | |
226 | * | |
227 | * We don't save and restore VM_LOCKED here because pages are | |
228 | * still on lru. In unmap path, pages might be scanned by reclaim | |
229 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
230 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 231 | */ |
ba470de4 | 232 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 233 | unsigned long start, unsigned long end) |
b291f000 NP |
234 | { |
235 | vma->vm_flags &= ~VM_LOCKED; | |
408e82b7 | 236 | |
ff6a6da6 | 237 | while (start < end) { |
6e919717 | 238 | struct page *page; |
ff6a6da6 ML |
239 | unsigned int page_mask, page_increm; |
240 | ||
6e919717 HD |
241 | /* |
242 | * Although FOLL_DUMP is intended for get_dump_page(), | |
243 | * it just so happens that its special treatment of the | |
244 | * ZERO_PAGE (returning an error instead of doing get_page) | |
245 | * suits munlock very well (and if somehow an abnormal page | |
246 | * has sneaked into the range, we won't oops here: great). | |
247 | */ | |
ff6a6da6 ML |
248 | page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP, |
249 | &page_mask); | |
6e919717 | 250 | if (page && !IS_ERR(page)) { |
408e82b7 | 251 | lock_page(page); |
ff6a6da6 ML |
252 | lru_add_drain(); |
253 | /* | |
254 | * Any THP page found by follow_page_mask() may have | |
255 | * gotten split before reaching munlock_vma_page(), | |
256 | * so we need to recompute the page_mask here. | |
257 | */ | |
258 | page_mask = munlock_vma_page(page); | |
408e82b7 HD |
259 | unlock_page(page); |
260 | put_page(page); | |
261 | } | |
ff6a6da6 ML |
262 | page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask); |
263 | start += page_increm * PAGE_SIZE; | |
408e82b7 HD |
264 | cond_resched(); |
265 | } | |
b291f000 NP |
266 | } |
267 | ||
268 | /* | |
269 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
270 | * | |
271 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
272 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 273 | * populate the ptes. |
b291f000 NP |
274 | * |
275 | * For vmas that pass the filters, merge/split as appropriate. | |
276 | */ | |
1da177e4 | 277 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 278 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 279 | { |
b291f000 | 280 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 281 | pgoff_t pgoff; |
b291f000 | 282 | int nr_pages; |
1da177e4 | 283 | int ret = 0; |
ca16d140 | 284 | int lock = !!(newflags & VM_LOCKED); |
1da177e4 | 285 | |
fed067da | 286 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
31db58b3 | 287 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
b291f000 NP |
288 | goto out; /* don't set VM_LOCKED, don't count */ |
289 | ||
1da177e4 LT |
290 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
291 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
6fa3eb70 S |
292 | vma->vm_file, pgoff, vma_policy(vma), |
293 | vma_get_anon_name(vma)); | |
1da177e4 LT |
294 | if (*prev) { |
295 | vma = *prev; | |
296 | goto success; | |
297 | } | |
298 | ||
1da177e4 LT |
299 | if (start != vma->vm_start) { |
300 | ret = split_vma(mm, vma, start, 1); | |
301 | if (ret) | |
302 | goto out; | |
303 | } | |
304 | ||
305 | if (end != vma->vm_end) { | |
306 | ret = split_vma(mm, vma, end, 0); | |
307 | if (ret) | |
308 | goto out; | |
309 | } | |
310 | ||
311 | success: | |
b291f000 NP |
312 | /* |
313 | * Keep track of amount of locked VM. | |
314 | */ | |
315 | nr_pages = (end - start) >> PAGE_SHIFT; | |
316 | if (!lock) | |
317 | nr_pages = -nr_pages; | |
318 | mm->locked_vm += nr_pages; | |
319 | ||
1da177e4 LT |
320 | /* |
321 | * vm_flags is protected by the mmap_sem held in write mode. | |
322 | * It's okay if try_to_unmap_one unmaps a page just after we | |
b291f000 | 323 | * set VM_LOCKED, __mlock_vma_pages_range will bring it back. |
1da177e4 | 324 | */ |
1da177e4 | 325 | |
fed067da | 326 | if (lock) |
408e82b7 | 327 | vma->vm_flags = newflags; |
fed067da | 328 | else |
408e82b7 | 329 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 330 | |
1da177e4 | 331 | out: |
b291f000 | 332 | *prev = vma; |
1da177e4 LT |
333 | return ret; |
334 | } | |
335 | ||
336 | static int do_mlock(unsigned long start, size_t len, int on) | |
337 | { | |
338 | unsigned long nstart, end, tmp; | |
339 | struct vm_area_struct * vma, * prev; | |
340 | int error; | |
341 | ||
fed067da ML |
342 | VM_BUG_ON(start & ~PAGE_MASK); |
343 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
1da177e4 LT |
344 | end = start + len; |
345 | if (end < start) | |
346 | return -EINVAL; | |
347 | if (end == start) | |
348 | return 0; | |
097d5910 | 349 | vma = find_vma(current->mm, start); |
1da177e4 LT |
350 | if (!vma || vma->vm_start > start) |
351 | return -ENOMEM; | |
352 | ||
097d5910 | 353 | prev = vma->vm_prev; |
1da177e4 LT |
354 | if (start > vma->vm_start) |
355 | prev = vma; | |
356 | ||
357 | for (nstart = start ; ; ) { | |
ca16d140 | 358 | vm_flags_t newflags; |
1da177e4 LT |
359 | |
360 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ | |
361 | ||
18693050 ML |
362 | newflags = vma->vm_flags & ~VM_LOCKED; |
363 | if (on) | |
09a9f1d2 | 364 | newflags |= VM_LOCKED; |
1da177e4 LT |
365 | |
366 | tmp = vma->vm_end; | |
367 | if (tmp > end) | |
368 | tmp = end; | |
369 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
370 | if (error) | |
371 | break; | |
372 | nstart = tmp; | |
373 | if (nstart < prev->vm_end) | |
374 | nstart = prev->vm_end; | |
375 | if (nstart >= end) | |
376 | break; | |
377 | ||
378 | vma = prev->vm_next; | |
379 | if (!vma || vma->vm_start != nstart) { | |
380 | error = -ENOMEM; | |
381 | break; | |
382 | } | |
383 | } | |
384 | return error; | |
385 | } | |
386 | ||
bebeb3d6 ML |
387 | /* |
388 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
389 | * | |
390 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
391 | * flags. VMAs must be already marked with the desired vm_flags, and | |
392 | * mmap_sem must not be held. | |
393 | */ | |
394 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
fed067da ML |
395 | { |
396 | struct mm_struct *mm = current->mm; | |
397 | unsigned long end, nstart, nend; | |
398 | struct vm_area_struct *vma = NULL; | |
53a7706d | 399 | int locked = 0; |
28a35716 | 400 | long ret = 0; |
fed067da ML |
401 | |
402 | VM_BUG_ON(start & ~PAGE_MASK); | |
403 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
404 | end = start + len; | |
405 | ||
fed067da ML |
406 | for (nstart = start; nstart < end; nstart = nend) { |
407 | /* | |
408 | * We want to fault in pages for [nstart; end) address range. | |
409 | * Find first corresponding VMA. | |
410 | */ | |
53a7706d ML |
411 | if (!locked) { |
412 | locked = 1; | |
413 | down_read(&mm->mmap_sem); | |
fed067da | 414 | vma = find_vma(mm, nstart); |
53a7706d | 415 | } else if (nstart >= vma->vm_end) |
fed067da ML |
416 | vma = vma->vm_next; |
417 | if (!vma || vma->vm_start >= end) | |
418 | break; | |
419 | /* | |
420 | * Set [nstart; nend) to intersection of desired address | |
421 | * range with the first VMA. Also, skip undesirable VMA types. | |
422 | */ | |
423 | nend = min(end, vma->vm_end); | |
09a9f1d2 | 424 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
fed067da ML |
425 | continue; |
426 | if (nstart < vma->vm_start) | |
427 | nstart = vma->vm_start; | |
428 | /* | |
53a7706d ML |
429 | * Now fault in a range of pages. __mlock_vma_pages_range() |
430 | * double checks the vma flags, so that it won't mlock pages | |
431 | * if the vma was already munlocked. | |
fed067da | 432 | */ |
53a7706d ML |
433 | ret = __mlock_vma_pages_range(vma, nstart, nend, &locked); |
434 | if (ret < 0) { | |
435 | if (ignore_errors) { | |
436 | ret = 0; | |
437 | continue; /* continue at next VMA */ | |
438 | } | |
5fdb2002 ML |
439 | ret = __mlock_posix_error_return(ret); |
440 | break; | |
441 | } | |
53a7706d ML |
442 | nend = nstart + ret * PAGE_SIZE; |
443 | ret = 0; | |
fed067da | 444 | } |
53a7706d ML |
445 | if (locked) |
446 | up_read(&mm->mmap_sem); | |
fed067da ML |
447 | return ret; /* 0 or negative error code */ |
448 | } | |
449 | ||
6a6160a7 | 450 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
1da177e4 LT |
451 | { |
452 | unsigned long locked; | |
453 | unsigned long lock_limit; | |
454 | int error = -ENOMEM; | |
455 | ||
456 | if (!can_do_mlock()) | |
457 | return -EPERM; | |
458 | ||
8891d6da KM |
459 | lru_add_drain_all(); /* flush pagevec */ |
460 | ||
1da177e4 LT |
461 | down_write(¤t->mm->mmap_sem); |
462 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); | |
463 | start &= PAGE_MASK; | |
464 | ||
465 | locked = len >> PAGE_SHIFT; | |
466 | locked += current->mm->locked_vm; | |
467 | ||
59e99e5b | 468 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
469 | lock_limit >>= PAGE_SHIFT; |
470 | ||
471 | /* check against resource limits */ | |
472 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
473 | error = do_mlock(start, len, 1); | |
474 | up_write(¤t->mm->mmap_sem); | |
fed067da | 475 | if (!error) |
bebeb3d6 | 476 | error = __mm_populate(start, len, 0); |
1da177e4 LT |
477 | return error; |
478 | } | |
479 | ||
6a6160a7 | 480 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
481 | { |
482 | int ret; | |
483 | ||
484 | down_write(¤t->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(¤t->mm->mmap_sem); | |
489 | return ret; | |
490 | } | |
491 | ||
492 | static int do_mlockall(int flags) | |
493 | { | |
494 | struct vm_area_struct * vma, * prev = NULL; | |
1da177e4 LT |
495 | |
496 | if (flags & MCL_FUTURE) | |
09a9f1d2 | 497 | current->mm->def_flags |= VM_LOCKED; |
9977f0f1 | 498 | else |
09a9f1d2 | 499 | current->mm->def_flags &= ~VM_LOCKED; |
1da177e4 LT |
500 | if (flags == MCL_FUTURE) |
501 | goto out; | |
502 | ||
503 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 504 | vm_flags_t newflags; |
1da177e4 | 505 | |
18693050 ML |
506 | newflags = vma->vm_flags & ~VM_LOCKED; |
507 | if (flags & MCL_CURRENT) | |
09a9f1d2 | 508 | newflags |= VM_LOCKED; |
1da177e4 LT |
509 | |
510 | /* Ignore errors */ | |
511 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
512 | } | |
513 | out: | |
514 | return 0; | |
515 | } | |
516 | ||
3480b257 | 517 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
518 | { |
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 | ||
df9d6985 CL |
529 | if (flags & MCL_CURRENT) |
530 | lru_add_drain_all(); /* flush pagevec */ | |
8891d6da | 531 | |
1da177e4 LT |
532 | down_write(¤t->mm->mmap_sem); |
533 | ||
59e99e5b | 534 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
535 | lock_limit >>= PAGE_SHIFT; |
536 | ||
537 | ret = -ENOMEM; | |
538 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || | |
539 | capable(CAP_IPC_LOCK)) | |
540 | ret = do_mlockall(flags); | |
541 | up_write(¤t->mm->mmap_sem); | |
bebeb3d6 ML |
542 | if (!ret && (flags & MCL_CURRENT)) |
543 | mm_populate(0, TASK_SIZE); | |
1da177e4 LT |
544 | out: |
545 | return ret; | |
546 | } | |
547 | ||
3480b257 | 548 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
549 | { |
550 | int ret; | |
551 | ||
552 | down_write(¤t->mm->mmap_sem); | |
553 | ret = do_mlockall(0); | |
554 | up_write(¤t->mm->mmap_sem); | |
555 | return ret; | |
556 | } | |
557 | ||
558 | /* | |
559 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
560 | * shm segments) get accounted against the user_struct instead. | |
561 | */ | |
562 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
563 | ||
564 | int user_shm_lock(size_t size, struct user_struct *user) | |
565 | { | |
566 | unsigned long lock_limit, locked; | |
567 | int allowed = 0; | |
568 | ||
569 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 570 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
571 | if (lock_limit == RLIM_INFINITY) |
572 | allowed = 1; | |
1da177e4 LT |
573 | lock_limit >>= PAGE_SHIFT; |
574 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
575 | if (!allowed && |
576 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
577 | goto out; |
578 | get_uid(user); | |
579 | user->locked_shm += locked; | |
580 | allowed = 1; | |
581 | out: | |
582 | spin_unlock(&shmlock_user_lock); | |
583 | return allowed; | |
584 | } | |
585 | ||
586 | void user_shm_unlock(size_t size, struct user_struct *user) | |
587 | { | |
588 | spin_lock(&shmlock_user_lock); | |
589 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
590 | spin_unlock(&shmlock_user_lock); | |
591 | free_uid(user); | |
592 | } |