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> | |
7225522b | 14 | #include <linux/pagevec.h> |
1da177e4 LT |
15 | #include <linux/mempolicy.h> |
16 | #include <linux/syscalls.h> | |
e8edc6e0 | 17 | #include <linux/sched.h> |
b95f1b31 | 18 | #include <linux/export.h> |
b291f000 NP |
19 | #include <linux/rmap.h> |
20 | #include <linux/mmzone.h> | |
21 | #include <linux/hugetlb.h> | |
7225522b VB |
22 | #include <linux/memcontrol.h> |
23 | #include <linux/mm_inline.h> | |
b291f000 NP |
24 | |
25 | #include "internal.h" | |
1da177e4 | 26 | |
e8edc6e0 AD |
27 | int can_do_mlock(void) |
28 | { | |
29 | if (capable(CAP_IPC_LOCK)) | |
30 | return 1; | |
59e99e5b | 31 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
e8edc6e0 AD |
32 | return 1; |
33 | return 0; | |
34 | } | |
35 | EXPORT_SYMBOL(can_do_mlock); | |
1da177e4 | 36 | |
b291f000 NP |
37 | /* |
38 | * Mlocked pages are marked with PageMlocked() flag for efficient testing | |
39 | * in vmscan and, possibly, the fault path; and to support semi-accurate | |
40 | * statistics. | |
41 | * | |
42 | * An mlocked page [PageMlocked(page)] is unevictable. As such, it will | |
43 | * be placed on the LRU "unevictable" list, rather than the [in]active lists. | |
44 | * The unevictable list is an LRU sibling list to the [in]active lists. | |
45 | * PageUnevictable is set to indicate the unevictable state. | |
46 | * | |
47 | * When lazy mlocking via vmscan, it is important to ensure that the | |
48 | * vma's VM_LOCKED status is not concurrently being modified, otherwise we | |
49 | * may have mlocked a page that is being munlocked. So lazy mlock must take | |
50 | * the mmap_sem for read, and verify that the vma really is locked | |
51 | * (see mm/rmap.c). | |
52 | */ | |
53 | ||
54 | /* | |
55 | * LRU accounting for clear_page_mlock() | |
56 | */ | |
e6c509f8 | 57 | void clear_page_mlock(struct page *page) |
b291f000 | 58 | { |
e6c509f8 | 59 | if (!TestClearPageMlocked(page)) |
b291f000 | 60 | return; |
b291f000 | 61 | |
8449d21f DR |
62 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
63 | -hpage_nr_pages(page)); | |
5344b7e6 | 64 | count_vm_event(UNEVICTABLE_PGCLEARED); |
b291f000 NP |
65 | if (!isolate_lru_page(page)) { |
66 | putback_lru_page(page); | |
67 | } else { | |
68 | /* | |
8891d6da | 69 | * We lost the race. the page already moved to evictable list. |
b291f000 | 70 | */ |
8891d6da | 71 | if (PageUnevictable(page)) |
5344b7e6 | 72 | count_vm_event(UNEVICTABLE_PGSTRANDED); |
b291f000 NP |
73 | } |
74 | } | |
75 | ||
76 | /* | |
77 | * Mark page as mlocked if not already. | |
78 | * If page on LRU, isolate and putback to move to unevictable list. | |
79 | */ | |
80 | void mlock_vma_page(struct page *page) | |
81 | { | |
82 | BUG_ON(!PageLocked(page)); | |
83 | ||
5344b7e6 | 84 | if (!TestSetPageMlocked(page)) { |
8449d21f DR |
85 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
86 | hpage_nr_pages(page)); | |
5344b7e6 NP |
87 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
88 | if (!isolate_lru_page(page)) | |
89 | putback_lru_page(page); | |
90 | } | |
b291f000 NP |
91 | } |
92 | ||
7225522b VB |
93 | /* |
94 | * Finish munlock after successful page isolation | |
95 | * | |
96 | * Page must be locked. This is a wrapper for try_to_munlock() | |
97 | * and putback_lru_page() with munlock accounting. | |
98 | */ | |
99 | static void __munlock_isolated_page(struct page *page) | |
100 | { | |
101 | int ret = SWAP_AGAIN; | |
102 | ||
103 | /* | |
104 | * Optimization: if the page was mapped just once, that's our mapping | |
105 | * and we don't need to check all the other vmas. | |
106 | */ | |
107 | if (page_mapcount(page) > 1) | |
108 | ret = try_to_munlock(page); | |
109 | ||
110 | /* Did try_to_unlock() succeed or punt? */ | |
111 | if (ret != SWAP_MLOCK) | |
112 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); | |
113 | ||
114 | putback_lru_page(page); | |
115 | } | |
116 | ||
117 | /* | |
118 | * Accounting for page isolation fail during munlock | |
119 | * | |
120 | * Performs accounting when page isolation fails in munlock. There is nothing | |
121 | * else to do because it means some other task has already removed the page | |
122 | * from the LRU. putback_lru_page() will take care of removing the page from | |
123 | * the unevictable list, if necessary. vmscan [page_referenced()] will move | |
124 | * the page back to the unevictable list if some other vma has it mlocked. | |
125 | */ | |
126 | static void __munlock_isolation_failed(struct page *page) | |
127 | { | |
128 | if (PageUnevictable(page)) | |
129 | count_vm_event(UNEVICTABLE_PGSTRANDED); | |
130 | else | |
131 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); | |
132 | } | |
133 | ||
6927c1dd LS |
134 | /** |
135 | * munlock_vma_page - munlock a vma page | |
136 | * @page - page to be unlocked | |
b291f000 | 137 | * |
6927c1dd LS |
138 | * called from munlock()/munmap() path with page supposedly on the LRU. |
139 | * When we munlock a page, because the vma where we found the page is being | |
140 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the | |
141 | * page locked so that we can leave it on the unevictable lru list and not | |
142 | * bother vmscan with it. However, to walk the page's rmap list in | |
143 | * try_to_munlock() we must isolate the page from the LRU. If some other | |
144 | * task has removed the page from the LRU, we won't be able to do that. | |
145 | * So we clear the PageMlocked as we might not get another chance. If we | |
146 | * can't isolate the page, we leave it for putback_lru_page() and vmscan | |
147 | * [page_referenced()/try_to_unmap()] to deal with. | |
b291f000 | 148 | */ |
ff6a6da6 | 149 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 150 | { |
ff6a6da6 ML |
151 | unsigned int page_mask = 0; |
152 | ||
b291f000 NP |
153 | BUG_ON(!PageLocked(page)); |
154 | ||
5344b7e6 | 155 | if (TestClearPageMlocked(page)) { |
ff6a6da6 ML |
156 | unsigned int nr_pages = hpage_nr_pages(page); |
157 | mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); | |
158 | page_mask = nr_pages - 1; | |
7225522b VB |
159 | if (!isolate_lru_page(page)) |
160 | __munlock_isolated_page(page); | |
161 | else | |
162 | __munlock_isolation_failed(page); | |
b291f000 | 163 | } |
ff6a6da6 ML |
164 | |
165 | return page_mask; | |
b291f000 NP |
166 | } |
167 | ||
ba470de4 | 168 | /** |
408e82b7 | 169 | * __mlock_vma_pages_range() - mlock a range of pages in the vma. |
ba470de4 RR |
170 | * @vma: target vma |
171 | * @start: start address | |
172 | * @end: end address | |
ba470de4 | 173 | * |
408e82b7 | 174 | * This takes care of making the pages present too. |
b291f000 | 175 | * |
ba470de4 | 176 | * return 0 on success, negative error code on error. |
b291f000 | 177 | * |
ba470de4 | 178 | * vma->vm_mm->mmap_sem must be held for at least read. |
b291f000 | 179 | */ |
cea10a19 ML |
180 | long __mlock_vma_pages_range(struct vm_area_struct *vma, |
181 | unsigned long start, unsigned long end, int *nonblocking) | |
b291f000 NP |
182 | { |
183 | struct mm_struct *mm = vma->vm_mm; | |
28a35716 | 184 | unsigned long nr_pages = (end - start) / PAGE_SIZE; |
408e82b7 | 185 | int gup_flags; |
ba470de4 RR |
186 | |
187 | VM_BUG_ON(start & ~PAGE_MASK); | |
188 | VM_BUG_ON(end & ~PAGE_MASK); | |
189 | VM_BUG_ON(start < vma->vm_start); | |
190 | VM_BUG_ON(end > vma->vm_end); | |
408e82b7 | 191 | VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem)); |
b291f000 | 192 | |
a1fde08c | 193 | gup_flags = FOLL_TOUCH | FOLL_MLOCK; |
5ecfda04 ML |
194 | /* |
195 | * We want to touch writable mappings with a write fault in order | |
196 | * to break COW, except for shared mappings because these don't COW | |
197 | * and we would not want to dirty them for nothing. | |
198 | */ | |
199 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
58fa879e | 200 | gup_flags |= FOLL_WRITE; |
b291f000 | 201 | |
fdf4c587 ML |
202 | /* |
203 | * We want mlock to succeed for regions that have any permissions | |
204 | * other than PROT_NONE. | |
205 | */ | |
206 | if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)) | |
207 | gup_flags |= FOLL_FORCE; | |
208 | ||
4805b02e JW |
209 | /* |
210 | * We made sure addr is within a VMA, so the following will | |
211 | * not result in a stack expansion that recurses back here. | |
212 | */ | |
ff6a6da6 | 213 | return __get_user_pages(current, mm, start, nr_pages, gup_flags, |
53a7706d | 214 | NULL, NULL, nonblocking); |
9978ad58 LS |
215 | } |
216 | ||
217 | /* | |
218 | * convert get_user_pages() return value to posix mlock() error | |
219 | */ | |
220 | static int __mlock_posix_error_return(long retval) | |
221 | { | |
222 | if (retval == -EFAULT) | |
223 | retval = -ENOMEM; | |
224 | else if (retval == -ENOMEM) | |
225 | retval = -EAGAIN; | |
226 | return retval; | |
b291f000 NP |
227 | } |
228 | ||
56afe477 VB |
229 | /* |
230 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() | |
231 | * | |
232 | * The fast path is available only for evictable pages with single mapping. | |
233 | * Then we can bypass the per-cpu pvec and get better performance. | |
234 | * when mapcount > 1 we need try_to_munlock() which can fail. | |
235 | * when !page_evictable(), we need the full redo logic of putback_lru_page to | |
236 | * avoid leaving evictable page in unevictable list. | |
237 | * | |
238 | * In case of success, @page is added to @pvec and @pgrescued is incremented | |
239 | * in case that the page was previously unevictable. @page is also unlocked. | |
240 | */ | |
241 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, | |
242 | int *pgrescued) | |
243 | { | |
244 | VM_BUG_ON(PageLRU(page)); | |
245 | VM_BUG_ON(!PageLocked(page)); | |
246 | ||
247 | if (page_mapcount(page) <= 1 && page_evictable(page)) { | |
248 | pagevec_add(pvec, page); | |
249 | if (TestClearPageUnevictable(page)) | |
250 | (*pgrescued)++; | |
251 | unlock_page(page); | |
252 | return true; | |
253 | } | |
254 | ||
255 | return false; | |
256 | } | |
257 | ||
258 | /* | |
259 | * Putback multiple evictable pages to the LRU | |
260 | * | |
261 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of | |
262 | * the pages might have meanwhile become unevictable but that is OK. | |
263 | */ | |
264 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) | |
265 | { | |
266 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); | |
267 | /* | |
268 | *__pagevec_lru_add() calls release_pages() so we don't call | |
269 | * put_page() explicitly | |
270 | */ | |
271 | __pagevec_lru_add(pvec); | |
272 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
273 | } | |
274 | ||
7225522b VB |
275 | /* |
276 | * Munlock a batch of pages from the same zone | |
277 | * | |
278 | * The work is split to two main phases. First phase clears the Mlocked flag | |
279 | * and attempts to isolate the pages, all under a single zone lru lock. | |
280 | * The second phase finishes the munlock only for pages where isolation | |
281 | * succeeded. | |
282 | * | |
7a8010cd | 283 | * Note that the pagevec may be modified during the process. |
7225522b VB |
284 | */ |
285 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) | |
286 | { | |
287 | int i; | |
288 | int nr = pagevec_count(pvec); | |
1ebb7cc6 | 289 | int delta_munlocked = -nr; |
56afe477 VB |
290 | struct pagevec pvec_putback; |
291 | int pgrescued = 0; | |
7225522b VB |
292 | |
293 | /* Phase 1: page isolation */ | |
294 | spin_lock_irq(&zone->lru_lock); | |
295 | for (i = 0; i < nr; i++) { | |
296 | struct page *page = pvec->pages[i]; | |
297 | ||
298 | if (TestClearPageMlocked(page)) { | |
299 | struct lruvec *lruvec; | |
300 | int lru; | |
301 | ||
7225522b VB |
302 | if (PageLRU(page)) { |
303 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
304 | lru = page_lru(page); | |
5b40998a VB |
305 | /* |
306 | * We already have pin from follow_page_mask() | |
307 | * so we can spare the get_page() here. | |
308 | */ | |
7225522b VB |
309 | ClearPageLRU(page); |
310 | del_page_from_lru_list(page, lruvec, lru); | |
311 | } else { | |
312 | __munlock_isolation_failed(page); | |
313 | goto skip_munlock; | |
314 | } | |
315 | ||
316 | } else { | |
317 | skip_munlock: | |
318 | /* | |
319 | * We won't be munlocking this page in the next phase | |
320 | * but we still need to release the follow_page_mask() | |
321 | * pin. | |
322 | */ | |
323 | pvec->pages[i] = NULL; | |
324 | put_page(page); | |
1ebb7cc6 | 325 | delta_munlocked++; |
7225522b VB |
326 | } |
327 | } | |
1ebb7cc6 | 328 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
7225522b VB |
329 | spin_unlock_irq(&zone->lru_lock); |
330 | ||
56afe477 VB |
331 | /* Phase 2: page munlock */ |
332 | pagevec_init(&pvec_putback, 0); | |
7225522b VB |
333 | for (i = 0; i < nr; i++) { |
334 | struct page *page = pvec->pages[i]; | |
335 | ||
336 | if (page) { | |
337 | lock_page(page); | |
56afe477 VB |
338 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
339 | &pgrescued)) { | |
5b40998a VB |
340 | /* |
341 | * Slow path. We don't want to lose the last | |
342 | * pin before unlock_page() | |
343 | */ | |
344 | get_page(page); /* for putback_lru_page() */ | |
56afe477 VB |
345 | __munlock_isolated_page(page); |
346 | unlock_page(page); | |
5b40998a | 347 | put_page(page); /* from follow_page_mask() */ |
56afe477 | 348 | } |
7225522b VB |
349 | } |
350 | } | |
56afe477 | 351 | |
5b40998a VB |
352 | /* |
353 | * Phase 3: page putback for pages that qualified for the fast path | |
354 | * This will also call put_page() to return pin from follow_page_mask() | |
355 | */ | |
56afe477 VB |
356 | if (pagevec_count(&pvec_putback)) |
357 | __putback_lru_fast(&pvec_putback, pgrescued); | |
7a8010cd VB |
358 | } |
359 | ||
360 | /* | |
361 | * Fill up pagevec for __munlock_pagevec using pte walk | |
362 | * | |
363 | * The function expects that the struct page corresponding to @start address is | |
364 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. | |
365 | * | |
366 | * The rest of @pvec is filled by subsequent pages within the same pmd and same | |
367 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These | |
368 | * pages also get pinned. | |
369 | * | |
370 | * Returns the address of the next page that should be scanned. This equals | |
371 | * @start + PAGE_SIZE when no page could be added by the pte walk. | |
372 | */ | |
373 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, | |
374 | struct vm_area_struct *vma, int zoneid, unsigned long start, | |
375 | unsigned long end) | |
376 | { | |
377 | pte_t *pte; | |
378 | spinlock_t *ptl; | |
379 | ||
380 | /* | |
381 | * Initialize pte walk starting at the already pinned page where we | |
eadb41ae VB |
382 | * are sure that there is a pte, as it was pinned under the same |
383 | * mmap_sem write op. | |
7a8010cd VB |
384 | */ |
385 | pte = get_locked_pte(vma->vm_mm, start, &ptl); | |
eadb41ae VB |
386 | /* Make sure we do not cross the page table boundary */ |
387 | end = pgd_addr_end(start, end); | |
388 | end = pud_addr_end(start, end); | |
389 | end = pmd_addr_end(start, end); | |
7a8010cd VB |
390 | |
391 | /* The page next to the pinned page is the first we will try to get */ | |
392 | start += PAGE_SIZE; | |
393 | while (start < end) { | |
394 | struct page *page = NULL; | |
395 | pte++; | |
396 | if (pte_present(*pte)) | |
397 | page = vm_normal_page(vma, start, *pte); | |
398 | /* | |
399 | * Break if page could not be obtained or the page's node+zone does not | |
400 | * match | |
401 | */ | |
402 | if (!page || page_zone_id(page) != zoneid) | |
403 | break; | |
56afe477 | 404 | |
7a8010cd VB |
405 | get_page(page); |
406 | /* | |
407 | * Increase the address that will be returned *before* the | |
408 | * eventual break due to pvec becoming full by adding the page | |
409 | */ | |
410 | start += PAGE_SIZE; | |
411 | if (pagevec_add(pvec, page) == 0) | |
412 | break; | |
413 | } | |
414 | pte_unmap_unlock(pte, ptl); | |
415 | return start; | |
7225522b VB |
416 | } |
417 | ||
b291f000 | 418 | /* |
ba470de4 RR |
419 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
420 | * @vma - vma containing range to be munlock()ed. | |
421 | * @start - start address in @vma of the range | |
422 | * @end - end of range in @vma. | |
423 | * | |
424 | * For mremap(), munmap() and exit(). | |
425 | * | |
426 | * Called with @vma VM_LOCKED. | |
427 | * | |
428 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
429 | * deal with this. | |
430 | * | |
431 | * We don't save and restore VM_LOCKED here because pages are | |
432 | * still on lru. In unmap path, pages might be scanned by reclaim | |
433 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
434 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 435 | */ |
ba470de4 | 436 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 437 | unsigned long start, unsigned long end) |
b291f000 NP |
438 | { |
439 | vma->vm_flags &= ~VM_LOCKED; | |
408e82b7 | 440 | |
ff6a6da6 | 441 | while (start < end) { |
7a8010cd | 442 | struct page *page = NULL; |
ff6a6da6 | 443 | unsigned int page_mask, page_increm; |
7a8010cd VB |
444 | struct pagevec pvec; |
445 | struct zone *zone; | |
446 | int zoneid; | |
ff6a6da6 | 447 | |
7a8010cd | 448 | pagevec_init(&pvec, 0); |
6e919717 HD |
449 | /* |
450 | * Although FOLL_DUMP is intended for get_dump_page(), | |
451 | * it just so happens that its special treatment of the | |
452 | * ZERO_PAGE (returning an error instead of doing get_page) | |
453 | * suits munlock very well (and if somehow an abnormal page | |
454 | * has sneaked into the range, we won't oops here: great). | |
455 | */ | |
ff6a6da6 | 456 | page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP, |
7a8010cd VB |
457 | &page_mask); |
458 | ||
6e919717 | 459 | if (page && !IS_ERR(page)) { |
7225522b | 460 | if (PageTransHuge(page)) { |
7225522b VB |
461 | lock_page(page); |
462 | /* | |
463 | * Any THP page found by follow_page_mask() may | |
464 | * have gotten split before reaching | |
465 | * munlock_vma_page(), so we need to recompute | |
466 | * the page_mask here. | |
467 | */ | |
468 | page_mask = munlock_vma_page(page); | |
469 | unlock_page(page); | |
470 | put_page(page); /* follow_page_mask() */ | |
471 | } else { | |
472 | /* | |
7a8010cd VB |
473 | * Non-huge pages are handled in batches via |
474 | * pagevec. The pin from follow_page_mask() | |
475 | * prevents them from collapsing by THP. | |
476 | */ | |
477 | pagevec_add(&pvec, page); | |
478 | zone = page_zone(page); | |
479 | zoneid = page_zone_id(page); | |
480 | ||
481 | /* | |
482 | * Try to fill the rest of pagevec using fast | |
483 | * pte walk. This will also update start to | |
484 | * the next page to process. Then munlock the | |
485 | * pagevec. | |
7225522b | 486 | */ |
7a8010cd VB |
487 | start = __munlock_pagevec_fill(&pvec, vma, |
488 | zoneid, start, end); | |
489 | __munlock_pagevec(&pvec, zone); | |
490 | goto next; | |
7225522b | 491 | } |
408e82b7 | 492 | } |
ff6a6da6 ML |
493 | page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask); |
494 | start += page_increm * PAGE_SIZE; | |
7a8010cd | 495 | next: |
408e82b7 HD |
496 | cond_resched(); |
497 | } | |
b291f000 NP |
498 | } |
499 | ||
500 | /* | |
501 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
502 | * | |
503 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
504 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 505 | * populate the ptes. |
b291f000 NP |
506 | * |
507 | * For vmas that pass the filters, merge/split as appropriate. | |
508 | */ | |
1da177e4 | 509 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 510 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 511 | { |
b291f000 | 512 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 513 | pgoff_t pgoff; |
b291f000 | 514 | int nr_pages; |
1da177e4 | 515 | int ret = 0; |
ca16d140 | 516 | int lock = !!(newflags & VM_LOCKED); |
1da177e4 | 517 | |
fed067da | 518 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
31db58b3 | 519 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
b291f000 NP |
520 | goto out; /* don't set VM_LOCKED, don't count */ |
521 | ||
1da177e4 LT |
522 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
523 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
524 | vma->vm_file, pgoff, vma_policy(vma)); | |
525 | if (*prev) { | |
526 | vma = *prev; | |
527 | goto success; | |
528 | } | |
529 | ||
1da177e4 LT |
530 | if (start != vma->vm_start) { |
531 | ret = split_vma(mm, vma, start, 1); | |
532 | if (ret) | |
533 | goto out; | |
534 | } | |
535 | ||
536 | if (end != vma->vm_end) { | |
537 | ret = split_vma(mm, vma, end, 0); | |
538 | if (ret) | |
539 | goto out; | |
540 | } | |
541 | ||
542 | success: | |
b291f000 NP |
543 | /* |
544 | * Keep track of amount of locked VM. | |
545 | */ | |
546 | nr_pages = (end - start) >> PAGE_SHIFT; | |
547 | if (!lock) | |
548 | nr_pages = -nr_pages; | |
549 | mm->locked_vm += nr_pages; | |
550 | ||
1da177e4 LT |
551 | /* |
552 | * vm_flags is protected by the mmap_sem held in write mode. | |
553 | * It's okay if try_to_unmap_one unmaps a page just after we | |
b291f000 | 554 | * set VM_LOCKED, __mlock_vma_pages_range will bring it back. |
1da177e4 | 555 | */ |
1da177e4 | 556 | |
fed067da | 557 | if (lock) |
408e82b7 | 558 | vma->vm_flags = newflags; |
fed067da | 559 | else |
408e82b7 | 560 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 561 | |
1da177e4 | 562 | out: |
b291f000 | 563 | *prev = vma; |
1da177e4 LT |
564 | return ret; |
565 | } | |
566 | ||
567 | static int do_mlock(unsigned long start, size_t len, int on) | |
568 | { | |
569 | unsigned long nstart, end, tmp; | |
570 | struct vm_area_struct * vma, * prev; | |
571 | int error; | |
572 | ||
fed067da ML |
573 | VM_BUG_ON(start & ~PAGE_MASK); |
574 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
1da177e4 LT |
575 | end = start + len; |
576 | if (end < start) | |
577 | return -EINVAL; | |
578 | if (end == start) | |
579 | return 0; | |
097d5910 | 580 | vma = find_vma(current->mm, start); |
1da177e4 LT |
581 | if (!vma || vma->vm_start > start) |
582 | return -ENOMEM; | |
583 | ||
097d5910 | 584 | prev = vma->vm_prev; |
1da177e4 LT |
585 | if (start > vma->vm_start) |
586 | prev = vma; | |
587 | ||
588 | for (nstart = start ; ; ) { | |
ca16d140 | 589 | vm_flags_t newflags; |
1da177e4 LT |
590 | |
591 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ | |
592 | ||
18693050 ML |
593 | newflags = vma->vm_flags & ~VM_LOCKED; |
594 | if (on) | |
09a9f1d2 | 595 | newflags |= VM_LOCKED; |
1da177e4 LT |
596 | |
597 | tmp = vma->vm_end; | |
598 | if (tmp > end) | |
599 | tmp = end; | |
600 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
601 | if (error) | |
602 | break; | |
603 | nstart = tmp; | |
604 | if (nstart < prev->vm_end) | |
605 | nstart = prev->vm_end; | |
606 | if (nstart >= end) | |
607 | break; | |
608 | ||
609 | vma = prev->vm_next; | |
610 | if (!vma || vma->vm_start != nstart) { | |
611 | error = -ENOMEM; | |
612 | break; | |
613 | } | |
614 | } | |
615 | return error; | |
616 | } | |
617 | ||
bebeb3d6 ML |
618 | /* |
619 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
620 | * | |
621 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
622 | * flags. VMAs must be already marked with the desired vm_flags, and | |
623 | * mmap_sem must not be held. | |
624 | */ | |
625 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
fed067da ML |
626 | { |
627 | struct mm_struct *mm = current->mm; | |
628 | unsigned long end, nstart, nend; | |
629 | struct vm_area_struct *vma = NULL; | |
53a7706d | 630 | int locked = 0; |
28a35716 | 631 | long ret = 0; |
fed067da ML |
632 | |
633 | VM_BUG_ON(start & ~PAGE_MASK); | |
634 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
635 | end = start + len; | |
636 | ||
fed067da ML |
637 | for (nstart = start; nstart < end; nstart = nend) { |
638 | /* | |
639 | * We want to fault in pages for [nstart; end) address range. | |
640 | * Find first corresponding VMA. | |
641 | */ | |
53a7706d ML |
642 | if (!locked) { |
643 | locked = 1; | |
644 | down_read(&mm->mmap_sem); | |
fed067da | 645 | vma = find_vma(mm, nstart); |
53a7706d | 646 | } else if (nstart >= vma->vm_end) |
fed067da ML |
647 | vma = vma->vm_next; |
648 | if (!vma || vma->vm_start >= end) | |
649 | break; | |
650 | /* | |
651 | * Set [nstart; nend) to intersection of desired address | |
652 | * range with the first VMA. Also, skip undesirable VMA types. | |
653 | */ | |
654 | nend = min(end, vma->vm_end); | |
09a9f1d2 | 655 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
fed067da ML |
656 | continue; |
657 | if (nstart < vma->vm_start) | |
658 | nstart = vma->vm_start; | |
659 | /* | |
53a7706d ML |
660 | * Now fault in a range of pages. __mlock_vma_pages_range() |
661 | * double checks the vma flags, so that it won't mlock pages | |
662 | * if the vma was already munlocked. | |
fed067da | 663 | */ |
53a7706d ML |
664 | ret = __mlock_vma_pages_range(vma, nstart, nend, &locked); |
665 | if (ret < 0) { | |
666 | if (ignore_errors) { | |
667 | ret = 0; | |
668 | continue; /* continue at next VMA */ | |
669 | } | |
5fdb2002 ML |
670 | ret = __mlock_posix_error_return(ret); |
671 | break; | |
672 | } | |
53a7706d ML |
673 | nend = nstart + ret * PAGE_SIZE; |
674 | ret = 0; | |
fed067da | 675 | } |
53a7706d ML |
676 | if (locked) |
677 | up_read(&mm->mmap_sem); | |
fed067da ML |
678 | return ret; /* 0 or negative error code */ |
679 | } | |
680 | ||
6a6160a7 | 681 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
1da177e4 LT |
682 | { |
683 | unsigned long locked; | |
684 | unsigned long lock_limit; | |
685 | int error = -ENOMEM; | |
686 | ||
687 | if (!can_do_mlock()) | |
688 | return -EPERM; | |
689 | ||
8891d6da KM |
690 | lru_add_drain_all(); /* flush pagevec */ |
691 | ||
1da177e4 LT |
692 | down_write(¤t->mm->mmap_sem); |
693 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); | |
694 | start &= PAGE_MASK; | |
695 | ||
696 | locked = len >> PAGE_SHIFT; | |
697 | locked += current->mm->locked_vm; | |
698 | ||
59e99e5b | 699 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
700 | lock_limit >>= PAGE_SHIFT; |
701 | ||
702 | /* check against resource limits */ | |
703 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
704 | error = do_mlock(start, len, 1); | |
705 | up_write(¤t->mm->mmap_sem); | |
fed067da | 706 | if (!error) |
bebeb3d6 | 707 | error = __mm_populate(start, len, 0); |
1da177e4 LT |
708 | return error; |
709 | } | |
710 | ||
6a6160a7 | 711 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
712 | { |
713 | int ret; | |
714 | ||
715 | down_write(¤t->mm->mmap_sem); | |
716 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); | |
717 | start &= PAGE_MASK; | |
718 | ret = do_mlock(start, len, 0); | |
719 | up_write(¤t->mm->mmap_sem); | |
720 | return ret; | |
721 | } | |
722 | ||
723 | static int do_mlockall(int flags) | |
724 | { | |
725 | struct vm_area_struct * vma, * prev = NULL; | |
1da177e4 LT |
726 | |
727 | if (flags & MCL_FUTURE) | |
09a9f1d2 | 728 | current->mm->def_flags |= VM_LOCKED; |
9977f0f1 | 729 | else |
09a9f1d2 | 730 | current->mm->def_flags &= ~VM_LOCKED; |
1da177e4 LT |
731 | if (flags == MCL_FUTURE) |
732 | goto out; | |
733 | ||
734 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 735 | vm_flags_t newflags; |
1da177e4 | 736 | |
18693050 ML |
737 | newflags = vma->vm_flags & ~VM_LOCKED; |
738 | if (flags & MCL_CURRENT) | |
09a9f1d2 | 739 | newflags |= VM_LOCKED; |
1da177e4 LT |
740 | |
741 | /* Ignore errors */ | |
742 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
22356f44 | 743 | cond_resched(); |
1da177e4 LT |
744 | } |
745 | out: | |
746 | return 0; | |
747 | } | |
748 | ||
3480b257 | 749 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
750 | { |
751 | unsigned long lock_limit; | |
752 | int ret = -EINVAL; | |
753 | ||
754 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE))) | |
755 | goto out; | |
756 | ||
757 | ret = -EPERM; | |
758 | if (!can_do_mlock()) | |
759 | goto out; | |
760 | ||
df9d6985 CL |
761 | if (flags & MCL_CURRENT) |
762 | lru_add_drain_all(); /* flush pagevec */ | |
8891d6da | 763 | |
1da177e4 LT |
764 | down_write(¤t->mm->mmap_sem); |
765 | ||
59e99e5b | 766 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
767 | lock_limit >>= PAGE_SHIFT; |
768 | ||
769 | ret = -ENOMEM; | |
770 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || | |
771 | capable(CAP_IPC_LOCK)) | |
772 | ret = do_mlockall(flags); | |
773 | up_write(¤t->mm->mmap_sem); | |
bebeb3d6 ML |
774 | if (!ret && (flags & MCL_CURRENT)) |
775 | mm_populate(0, TASK_SIZE); | |
1da177e4 LT |
776 | out: |
777 | return ret; | |
778 | } | |
779 | ||
3480b257 | 780 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
781 | { |
782 | int ret; | |
783 | ||
784 | down_write(¤t->mm->mmap_sem); | |
785 | ret = do_mlockall(0); | |
786 | up_write(¤t->mm->mmap_sem); | |
787 | return ret; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
792 | * shm segments) get accounted against the user_struct instead. | |
793 | */ | |
794 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
795 | ||
796 | int user_shm_lock(size_t size, struct user_struct *user) | |
797 | { | |
798 | unsigned long lock_limit, locked; | |
799 | int allowed = 0; | |
800 | ||
801 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 802 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
803 | if (lock_limit == RLIM_INFINITY) |
804 | allowed = 1; | |
1da177e4 LT |
805 | lock_limit >>= PAGE_SHIFT; |
806 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
807 | if (!allowed && |
808 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
809 | goto out; |
810 | get_uid(user); | |
811 | user->locked_shm += locked; | |
812 | allowed = 1; | |
813 | out: | |
814 | spin_unlock(&shmlock_user_lock); | |
815 | return allowed; | |
816 | } | |
817 | ||
818 | void user_shm_unlock(size_t size, struct user_struct *user) | |
819 | { | |
820 | spin_lock(&shmlock_user_lock); | |
821 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
822 | spin_unlock(&shmlock_user_lock); | |
823 | free_uid(user); | |
824 | } |