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 | ||
01cc2e58 VB |
93 | /* |
94 | * Isolate a page from LRU with optional get_page() pin. | |
95 | * Assumes lru_lock already held and page already pinned. | |
96 | */ | |
97 | static bool __munlock_isolate_lru_page(struct page *page, bool getpage) | |
98 | { | |
99 | if (PageLRU(page)) { | |
100 | struct lruvec *lruvec; | |
101 | ||
102 | lruvec = mem_cgroup_page_lruvec(page, page_zone(page)); | |
103 | if (getpage) | |
104 | get_page(page); | |
105 | ClearPageLRU(page); | |
106 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
107 | return true; | |
108 | } | |
109 | ||
110 | return false; | |
111 | } | |
112 | ||
7225522b VB |
113 | /* |
114 | * Finish munlock after successful page isolation | |
115 | * | |
116 | * Page must be locked. This is a wrapper for try_to_munlock() | |
117 | * and putback_lru_page() with munlock accounting. | |
118 | */ | |
119 | static void __munlock_isolated_page(struct page *page) | |
120 | { | |
121 | int ret = SWAP_AGAIN; | |
122 | ||
123 | /* | |
124 | * Optimization: if the page was mapped just once, that's our mapping | |
125 | * and we don't need to check all the other vmas. | |
126 | */ | |
127 | if (page_mapcount(page) > 1) | |
128 | ret = try_to_munlock(page); | |
129 | ||
130 | /* Did try_to_unlock() succeed or punt? */ | |
131 | if (ret != SWAP_MLOCK) | |
132 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); | |
133 | ||
134 | putback_lru_page(page); | |
135 | } | |
136 | ||
137 | /* | |
138 | * Accounting for page isolation fail during munlock | |
139 | * | |
140 | * Performs accounting when page isolation fails in munlock. There is nothing | |
141 | * else to do because it means some other task has already removed the page | |
142 | * from the LRU. putback_lru_page() will take care of removing the page from | |
143 | * the unevictable list, if necessary. vmscan [page_referenced()] will move | |
144 | * the page back to the unevictable list if some other vma has it mlocked. | |
145 | */ | |
146 | static void __munlock_isolation_failed(struct page *page) | |
147 | { | |
148 | if (PageUnevictable(page)) | |
01cc2e58 | 149 | __count_vm_event(UNEVICTABLE_PGSTRANDED); |
7225522b | 150 | else |
01cc2e58 | 151 | __count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
7225522b VB |
152 | } |
153 | ||
6927c1dd LS |
154 | /** |
155 | * munlock_vma_page - munlock a vma page | |
c424be1c VB |
156 | * @page - page to be unlocked, either a normal page or THP page head |
157 | * | |
158 | * returns the size of the page as a page mask (0 for normal page, | |
159 | * HPAGE_PMD_NR - 1 for THP head page) | |
b291f000 | 160 | * |
6927c1dd LS |
161 | * called from munlock()/munmap() path with page supposedly on the LRU. |
162 | * When we munlock a page, because the vma where we found the page is being | |
163 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the | |
164 | * page locked so that we can leave it on the unevictable lru list and not | |
165 | * bother vmscan with it. However, to walk the page's rmap list in | |
166 | * try_to_munlock() we must isolate the page from the LRU. If some other | |
167 | * task has removed the page from the LRU, we won't be able to do that. | |
168 | * So we clear the PageMlocked as we might not get another chance. If we | |
169 | * can't isolate the page, we leave it for putback_lru_page() and vmscan | |
170 | * [page_referenced()/try_to_unmap()] to deal with. | |
b291f000 | 171 | */ |
ff6a6da6 | 172 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 173 | { |
c424be1c | 174 | unsigned int nr_pages; |
01cc2e58 | 175 | struct zone *zone = page_zone(page); |
ff6a6da6 | 176 | |
b291f000 NP |
177 | BUG_ON(!PageLocked(page)); |
178 | ||
c424be1c | 179 | /* |
01cc2e58 VB |
180 | * Serialize with any parallel __split_huge_page_refcount() which |
181 | * might otherwise copy PageMlocked to part of the tail pages before | |
182 | * we clear it in the head page. It also stabilizes hpage_nr_pages(). | |
c424be1c | 183 | */ |
01cc2e58 VB |
184 | spin_lock_irq(&zone->lru_lock); |
185 | ||
186 | nr_pages = hpage_nr_pages(page); | |
187 | if (!TestClearPageMlocked(page)) | |
188 | goto unlock_out; | |
189 | ||
190 | __mod_zone_page_state(zone, NR_MLOCK, -nr_pages); | |
191 | ||
192 | if (__munlock_isolate_lru_page(page, true)) { | |
193 | spin_unlock_irq(&zone->lru_lock); | |
194 | __munlock_isolated_page(page); | |
195 | goto out; | |
196 | } | |
197 | __munlock_isolation_failed(page); | |
198 | ||
199 | unlock_out: | |
200 | spin_unlock_irq(&zone->lru_lock); | |
201 | ||
202 | out: | |
c424be1c | 203 | return nr_pages - 1; |
b291f000 NP |
204 | } |
205 | ||
ba470de4 | 206 | /** |
408e82b7 | 207 | * __mlock_vma_pages_range() - mlock a range of pages in the vma. |
ba470de4 RR |
208 | * @vma: target vma |
209 | * @start: start address | |
210 | * @end: end address | |
ba470de4 | 211 | * |
408e82b7 | 212 | * This takes care of making the pages present too. |
b291f000 | 213 | * |
ba470de4 | 214 | * return 0 on success, negative error code on error. |
b291f000 | 215 | * |
ba470de4 | 216 | * vma->vm_mm->mmap_sem must be held for at least read. |
b291f000 | 217 | */ |
cea10a19 ML |
218 | long __mlock_vma_pages_range(struct vm_area_struct *vma, |
219 | unsigned long start, unsigned long end, int *nonblocking) | |
b291f000 NP |
220 | { |
221 | struct mm_struct *mm = vma->vm_mm; | |
28a35716 | 222 | unsigned long nr_pages = (end - start) / PAGE_SIZE; |
408e82b7 | 223 | int gup_flags; |
ba470de4 RR |
224 | |
225 | VM_BUG_ON(start & ~PAGE_MASK); | |
226 | VM_BUG_ON(end & ~PAGE_MASK); | |
227 | VM_BUG_ON(start < vma->vm_start); | |
228 | VM_BUG_ON(end > vma->vm_end); | |
408e82b7 | 229 | VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem)); |
b291f000 | 230 | |
a1fde08c | 231 | gup_flags = FOLL_TOUCH | FOLL_MLOCK; |
5ecfda04 ML |
232 | /* |
233 | * We want to touch writable mappings with a write fault in order | |
234 | * to break COW, except for shared mappings because these don't COW | |
235 | * and we would not want to dirty them for nothing. | |
236 | */ | |
237 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
58fa879e | 238 | gup_flags |= FOLL_WRITE; |
b291f000 | 239 | |
fdf4c587 ML |
240 | /* |
241 | * We want mlock to succeed for regions that have any permissions | |
242 | * other than PROT_NONE. | |
243 | */ | |
244 | if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)) | |
245 | gup_flags |= FOLL_FORCE; | |
246 | ||
4805b02e JW |
247 | /* |
248 | * We made sure addr is within a VMA, so the following will | |
249 | * not result in a stack expansion that recurses back here. | |
250 | */ | |
ff6a6da6 | 251 | return __get_user_pages(current, mm, start, nr_pages, gup_flags, |
53a7706d | 252 | NULL, NULL, nonblocking); |
9978ad58 LS |
253 | } |
254 | ||
255 | /* | |
256 | * convert get_user_pages() return value to posix mlock() error | |
257 | */ | |
258 | static int __mlock_posix_error_return(long retval) | |
259 | { | |
260 | if (retval == -EFAULT) | |
261 | retval = -ENOMEM; | |
262 | else if (retval == -ENOMEM) | |
263 | retval = -EAGAIN; | |
264 | return retval; | |
b291f000 NP |
265 | } |
266 | ||
56afe477 VB |
267 | /* |
268 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() | |
269 | * | |
270 | * The fast path is available only for evictable pages with single mapping. | |
271 | * Then we can bypass the per-cpu pvec and get better performance. | |
272 | * when mapcount > 1 we need try_to_munlock() which can fail. | |
273 | * when !page_evictable(), we need the full redo logic of putback_lru_page to | |
274 | * avoid leaving evictable page in unevictable list. | |
275 | * | |
276 | * In case of success, @page is added to @pvec and @pgrescued is incremented | |
277 | * in case that the page was previously unevictable. @page is also unlocked. | |
278 | */ | |
279 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, | |
280 | int *pgrescued) | |
281 | { | |
309381fe SL |
282 | VM_BUG_ON_PAGE(PageLRU(page), page); |
283 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
56afe477 VB |
284 | |
285 | if (page_mapcount(page) <= 1 && page_evictable(page)) { | |
286 | pagevec_add(pvec, page); | |
287 | if (TestClearPageUnevictable(page)) | |
288 | (*pgrescued)++; | |
289 | unlock_page(page); | |
290 | return true; | |
291 | } | |
292 | ||
293 | return false; | |
294 | } | |
295 | ||
296 | /* | |
297 | * Putback multiple evictable pages to the LRU | |
298 | * | |
299 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of | |
300 | * the pages might have meanwhile become unevictable but that is OK. | |
301 | */ | |
302 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) | |
303 | { | |
304 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); | |
305 | /* | |
306 | *__pagevec_lru_add() calls release_pages() so we don't call | |
307 | * put_page() explicitly | |
308 | */ | |
309 | __pagevec_lru_add(pvec); | |
310 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
311 | } | |
312 | ||
7225522b VB |
313 | /* |
314 | * Munlock a batch of pages from the same zone | |
315 | * | |
316 | * The work is split to two main phases. First phase clears the Mlocked flag | |
317 | * and attempts to isolate the pages, all under a single zone lru lock. | |
318 | * The second phase finishes the munlock only for pages where isolation | |
319 | * succeeded. | |
320 | * | |
7a8010cd | 321 | * Note that the pagevec may be modified during the process. |
7225522b VB |
322 | */ |
323 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) | |
324 | { | |
325 | int i; | |
326 | int nr = pagevec_count(pvec); | |
3b25df93 | 327 | int delta_munlocked; |
56afe477 VB |
328 | struct pagevec pvec_putback; |
329 | int pgrescued = 0; | |
7225522b | 330 | |
3b25df93 VB |
331 | pagevec_init(&pvec_putback, 0); |
332 | ||
7225522b VB |
333 | /* Phase 1: page isolation */ |
334 | spin_lock_irq(&zone->lru_lock); | |
335 | for (i = 0; i < nr; i++) { | |
336 | struct page *page = pvec->pages[i]; | |
337 | ||
338 | if (TestClearPageMlocked(page)) { | |
7225522b | 339 | /* |
01cc2e58 VB |
340 | * We already have pin from follow_page_mask() |
341 | * so we can spare the get_page() here. | |
7225522b | 342 | */ |
01cc2e58 VB |
343 | if (__munlock_isolate_lru_page(page, false)) |
344 | continue; | |
345 | else | |
346 | __munlock_isolation_failed(page); | |
7225522b | 347 | } |
01cc2e58 VB |
348 | |
349 | /* | |
350 | * We won't be munlocking this page in the next phase | |
351 | * but we still need to release the follow_page_mask() | |
352 | * pin. We cannot do it under lru_lock however. If it's | |
353 | * the last pin, __page_cache_release() would deadlock. | |
354 | */ | |
355 | pagevec_add(&pvec_putback, pvec->pages[i]); | |
356 | pvec->pages[i] = NULL; | |
7225522b | 357 | } |
3b25df93 | 358 | delta_munlocked = -nr + pagevec_count(&pvec_putback); |
1ebb7cc6 | 359 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
7225522b VB |
360 | spin_unlock_irq(&zone->lru_lock); |
361 | ||
3b25df93 VB |
362 | /* Now we can release pins of pages that we are not munlocking */ |
363 | pagevec_release(&pvec_putback); | |
364 | ||
56afe477 | 365 | /* Phase 2: page munlock */ |
7225522b VB |
366 | for (i = 0; i < nr; i++) { |
367 | struct page *page = pvec->pages[i]; | |
368 | ||
369 | if (page) { | |
370 | lock_page(page); | |
56afe477 VB |
371 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
372 | &pgrescued)) { | |
5b40998a VB |
373 | /* |
374 | * Slow path. We don't want to lose the last | |
375 | * pin before unlock_page() | |
376 | */ | |
377 | get_page(page); /* for putback_lru_page() */ | |
56afe477 VB |
378 | __munlock_isolated_page(page); |
379 | unlock_page(page); | |
5b40998a | 380 | put_page(page); /* from follow_page_mask() */ |
56afe477 | 381 | } |
7225522b VB |
382 | } |
383 | } | |
56afe477 | 384 | |
5b40998a VB |
385 | /* |
386 | * Phase 3: page putback for pages that qualified for the fast path | |
387 | * This will also call put_page() to return pin from follow_page_mask() | |
388 | */ | |
56afe477 VB |
389 | if (pagevec_count(&pvec_putback)) |
390 | __putback_lru_fast(&pvec_putback, pgrescued); | |
7a8010cd VB |
391 | } |
392 | ||
393 | /* | |
394 | * Fill up pagevec for __munlock_pagevec using pte walk | |
395 | * | |
396 | * The function expects that the struct page corresponding to @start address is | |
397 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. | |
398 | * | |
399 | * The rest of @pvec is filled by subsequent pages within the same pmd and same | |
400 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These | |
401 | * pages also get pinned. | |
402 | * | |
403 | * Returns the address of the next page that should be scanned. This equals | |
404 | * @start + PAGE_SIZE when no page could be added by the pte walk. | |
405 | */ | |
406 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, | |
407 | struct vm_area_struct *vma, int zoneid, unsigned long start, | |
408 | unsigned long end) | |
409 | { | |
410 | pte_t *pte; | |
411 | spinlock_t *ptl; | |
412 | ||
413 | /* | |
414 | * Initialize pte walk starting at the already pinned page where we | |
eadb41ae VB |
415 | * are sure that there is a pte, as it was pinned under the same |
416 | * mmap_sem write op. | |
7a8010cd VB |
417 | */ |
418 | pte = get_locked_pte(vma->vm_mm, start, &ptl); | |
eadb41ae VB |
419 | /* Make sure we do not cross the page table boundary */ |
420 | end = pgd_addr_end(start, end); | |
421 | end = pud_addr_end(start, end); | |
422 | end = pmd_addr_end(start, end); | |
7a8010cd VB |
423 | |
424 | /* The page next to the pinned page is the first we will try to get */ | |
425 | start += PAGE_SIZE; | |
426 | while (start < end) { | |
427 | struct page *page = NULL; | |
428 | pte++; | |
429 | if (pte_present(*pte)) | |
430 | page = vm_normal_page(vma, start, *pte); | |
431 | /* | |
432 | * Break if page could not be obtained or the page's node+zone does not | |
433 | * match | |
434 | */ | |
435 | if (!page || page_zone_id(page) != zoneid) | |
436 | break; | |
56afe477 | 437 | |
7a8010cd VB |
438 | get_page(page); |
439 | /* | |
440 | * Increase the address that will be returned *before* the | |
441 | * eventual break due to pvec becoming full by adding the page | |
442 | */ | |
443 | start += PAGE_SIZE; | |
444 | if (pagevec_add(pvec, page) == 0) | |
445 | break; | |
446 | } | |
447 | pte_unmap_unlock(pte, ptl); | |
448 | return start; | |
7225522b VB |
449 | } |
450 | ||
b291f000 | 451 | /* |
ba470de4 RR |
452 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
453 | * @vma - vma containing range to be munlock()ed. | |
454 | * @start - start address in @vma of the range | |
455 | * @end - end of range in @vma. | |
456 | * | |
457 | * For mremap(), munmap() and exit(). | |
458 | * | |
459 | * Called with @vma VM_LOCKED. | |
460 | * | |
461 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
462 | * deal with this. | |
463 | * | |
464 | * We don't save and restore VM_LOCKED here because pages are | |
465 | * still on lru. In unmap path, pages might be scanned by reclaim | |
466 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
467 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 468 | */ |
ba470de4 | 469 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 470 | unsigned long start, unsigned long end) |
b291f000 NP |
471 | { |
472 | vma->vm_flags &= ~VM_LOCKED; | |
408e82b7 | 473 | |
ff6a6da6 | 474 | while (start < end) { |
7a8010cd | 475 | struct page *page = NULL; |
c424be1c VB |
476 | unsigned int page_mask; |
477 | unsigned long page_increm; | |
7a8010cd VB |
478 | struct pagevec pvec; |
479 | struct zone *zone; | |
480 | int zoneid; | |
ff6a6da6 | 481 | |
7a8010cd | 482 | pagevec_init(&pvec, 0); |
6e919717 HD |
483 | /* |
484 | * Although FOLL_DUMP is intended for get_dump_page(), | |
485 | * it just so happens that its special treatment of the | |
486 | * ZERO_PAGE (returning an error instead of doing get_page) | |
487 | * suits munlock very well (and if somehow an abnormal page | |
488 | * has sneaked into the range, we won't oops here: great). | |
489 | */ | |
ff6a6da6 | 490 | page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP, |
7a8010cd VB |
491 | &page_mask); |
492 | ||
6e919717 | 493 | if (page && !IS_ERR(page)) { |
7225522b | 494 | if (PageTransHuge(page)) { |
7225522b VB |
495 | lock_page(page); |
496 | /* | |
497 | * Any THP page found by follow_page_mask() may | |
498 | * have gotten split before reaching | |
499 | * munlock_vma_page(), so we need to recompute | |
500 | * the page_mask here. | |
501 | */ | |
502 | page_mask = munlock_vma_page(page); | |
503 | unlock_page(page); | |
504 | put_page(page); /* follow_page_mask() */ | |
505 | } else { | |
506 | /* | |
7a8010cd VB |
507 | * Non-huge pages are handled in batches via |
508 | * pagevec. The pin from follow_page_mask() | |
509 | * prevents them from collapsing by THP. | |
510 | */ | |
511 | pagevec_add(&pvec, page); | |
512 | zone = page_zone(page); | |
513 | zoneid = page_zone_id(page); | |
514 | ||
515 | /* | |
516 | * Try to fill the rest of pagevec using fast | |
517 | * pte walk. This will also update start to | |
518 | * the next page to process. Then munlock the | |
519 | * pagevec. | |
7225522b | 520 | */ |
7a8010cd VB |
521 | start = __munlock_pagevec_fill(&pvec, vma, |
522 | zoneid, start, end); | |
523 | __munlock_pagevec(&pvec, zone); | |
524 | goto next; | |
7225522b | 525 | } |
408e82b7 | 526 | } |
c424be1c VB |
527 | /* It's a bug to munlock in the middle of a THP page */ |
528 | VM_BUG_ON((start >> PAGE_SHIFT) & page_mask); | |
529 | page_increm = 1 + page_mask; | |
ff6a6da6 | 530 | start += page_increm * PAGE_SIZE; |
7a8010cd | 531 | next: |
408e82b7 HD |
532 | cond_resched(); |
533 | } | |
b291f000 NP |
534 | } |
535 | ||
536 | /* | |
537 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
538 | * | |
539 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
540 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 541 | * populate the ptes. |
b291f000 NP |
542 | * |
543 | * For vmas that pass the filters, merge/split as appropriate. | |
544 | */ | |
1da177e4 | 545 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 546 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 547 | { |
b291f000 | 548 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 549 | pgoff_t pgoff; |
b291f000 | 550 | int nr_pages; |
1da177e4 | 551 | int ret = 0; |
ca16d140 | 552 | int lock = !!(newflags & VM_LOCKED); |
1da177e4 | 553 | |
fed067da | 554 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
31db58b3 | 555 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
b291f000 NP |
556 | goto out; /* don't set VM_LOCKED, don't count */ |
557 | ||
1da177e4 LT |
558 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
559 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
560 | vma->vm_file, pgoff, vma_policy(vma)); | |
561 | if (*prev) { | |
562 | vma = *prev; | |
563 | goto success; | |
564 | } | |
565 | ||
1da177e4 LT |
566 | if (start != vma->vm_start) { |
567 | ret = split_vma(mm, vma, start, 1); | |
568 | if (ret) | |
569 | goto out; | |
570 | } | |
571 | ||
572 | if (end != vma->vm_end) { | |
573 | ret = split_vma(mm, vma, end, 0); | |
574 | if (ret) | |
575 | goto out; | |
576 | } | |
577 | ||
578 | success: | |
b291f000 NP |
579 | /* |
580 | * Keep track of amount of locked VM. | |
581 | */ | |
582 | nr_pages = (end - start) >> PAGE_SHIFT; | |
583 | if (!lock) | |
584 | nr_pages = -nr_pages; | |
585 | mm->locked_vm += nr_pages; | |
586 | ||
1da177e4 LT |
587 | /* |
588 | * vm_flags is protected by the mmap_sem held in write mode. | |
589 | * It's okay if try_to_unmap_one unmaps a page just after we | |
b291f000 | 590 | * set VM_LOCKED, __mlock_vma_pages_range will bring it back. |
1da177e4 | 591 | */ |
1da177e4 | 592 | |
fed067da | 593 | if (lock) |
408e82b7 | 594 | vma->vm_flags = newflags; |
fed067da | 595 | else |
408e82b7 | 596 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 597 | |
1da177e4 | 598 | out: |
b291f000 | 599 | *prev = vma; |
1da177e4 LT |
600 | return ret; |
601 | } | |
602 | ||
603 | static int do_mlock(unsigned long start, size_t len, int on) | |
604 | { | |
605 | unsigned long nstart, end, tmp; | |
606 | struct vm_area_struct * vma, * prev; | |
607 | int error; | |
608 | ||
fed067da ML |
609 | VM_BUG_ON(start & ~PAGE_MASK); |
610 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
1da177e4 LT |
611 | end = start + len; |
612 | if (end < start) | |
613 | return -EINVAL; | |
614 | if (end == start) | |
615 | return 0; | |
097d5910 | 616 | vma = find_vma(current->mm, start); |
1da177e4 LT |
617 | if (!vma || vma->vm_start > start) |
618 | return -ENOMEM; | |
619 | ||
097d5910 | 620 | prev = vma->vm_prev; |
1da177e4 LT |
621 | if (start > vma->vm_start) |
622 | prev = vma; | |
623 | ||
624 | for (nstart = start ; ; ) { | |
ca16d140 | 625 | vm_flags_t newflags; |
1da177e4 LT |
626 | |
627 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ | |
628 | ||
18693050 ML |
629 | newflags = vma->vm_flags & ~VM_LOCKED; |
630 | if (on) | |
09a9f1d2 | 631 | newflags |= VM_LOCKED; |
1da177e4 LT |
632 | |
633 | tmp = vma->vm_end; | |
634 | if (tmp > end) | |
635 | tmp = end; | |
636 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
637 | if (error) | |
638 | break; | |
639 | nstart = tmp; | |
640 | if (nstart < prev->vm_end) | |
641 | nstart = prev->vm_end; | |
642 | if (nstart >= end) | |
643 | break; | |
644 | ||
645 | vma = prev->vm_next; | |
646 | if (!vma || vma->vm_start != nstart) { | |
647 | error = -ENOMEM; | |
648 | break; | |
649 | } | |
650 | } | |
651 | return error; | |
652 | } | |
653 | ||
bebeb3d6 ML |
654 | /* |
655 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
656 | * | |
657 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
658 | * flags. VMAs must be already marked with the desired vm_flags, and | |
659 | * mmap_sem must not be held. | |
660 | */ | |
661 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
fed067da ML |
662 | { |
663 | struct mm_struct *mm = current->mm; | |
664 | unsigned long end, nstart, nend; | |
665 | struct vm_area_struct *vma = NULL; | |
53a7706d | 666 | int locked = 0; |
28a35716 | 667 | long ret = 0; |
fed067da ML |
668 | |
669 | VM_BUG_ON(start & ~PAGE_MASK); | |
670 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
671 | end = start + len; | |
672 | ||
fed067da ML |
673 | for (nstart = start; nstart < end; nstart = nend) { |
674 | /* | |
675 | * We want to fault in pages for [nstart; end) address range. | |
676 | * Find first corresponding VMA. | |
677 | */ | |
53a7706d ML |
678 | if (!locked) { |
679 | locked = 1; | |
680 | down_read(&mm->mmap_sem); | |
fed067da | 681 | vma = find_vma(mm, nstart); |
53a7706d | 682 | } else if (nstart >= vma->vm_end) |
fed067da ML |
683 | vma = vma->vm_next; |
684 | if (!vma || vma->vm_start >= end) | |
685 | break; | |
686 | /* | |
687 | * Set [nstart; nend) to intersection of desired address | |
688 | * range with the first VMA. Also, skip undesirable VMA types. | |
689 | */ | |
690 | nend = min(end, vma->vm_end); | |
09a9f1d2 | 691 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
fed067da ML |
692 | continue; |
693 | if (nstart < vma->vm_start) | |
694 | nstart = vma->vm_start; | |
695 | /* | |
53a7706d ML |
696 | * Now fault in a range of pages. __mlock_vma_pages_range() |
697 | * double checks the vma flags, so that it won't mlock pages | |
698 | * if the vma was already munlocked. | |
fed067da | 699 | */ |
53a7706d ML |
700 | ret = __mlock_vma_pages_range(vma, nstart, nend, &locked); |
701 | if (ret < 0) { | |
702 | if (ignore_errors) { | |
703 | ret = 0; | |
704 | continue; /* continue at next VMA */ | |
705 | } | |
5fdb2002 ML |
706 | ret = __mlock_posix_error_return(ret); |
707 | break; | |
708 | } | |
53a7706d ML |
709 | nend = nstart + ret * PAGE_SIZE; |
710 | ret = 0; | |
fed067da | 711 | } |
53a7706d ML |
712 | if (locked) |
713 | up_read(&mm->mmap_sem); | |
fed067da ML |
714 | return ret; /* 0 or negative error code */ |
715 | } | |
716 | ||
6a6160a7 | 717 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
1da177e4 LT |
718 | { |
719 | unsigned long locked; | |
720 | unsigned long lock_limit; | |
721 | int error = -ENOMEM; | |
722 | ||
723 | if (!can_do_mlock()) | |
724 | return -EPERM; | |
725 | ||
8891d6da KM |
726 | lru_add_drain_all(); /* flush pagevec */ |
727 | ||
1da177e4 LT |
728 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); |
729 | start &= PAGE_MASK; | |
730 | ||
59e99e5b | 731 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 | 732 | lock_limit >>= PAGE_SHIFT; |
1f1cd705 DB |
733 | locked = len >> PAGE_SHIFT; |
734 | ||
735 | down_write(¤t->mm->mmap_sem); | |
736 | ||
737 | locked += current->mm->locked_vm; | |
1da177e4 LT |
738 | |
739 | /* check against resource limits */ | |
740 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
741 | error = do_mlock(start, len, 1); | |
1f1cd705 | 742 | |
1da177e4 | 743 | up_write(¤t->mm->mmap_sem); |
fed067da | 744 | if (!error) |
bebeb3d6 | 745 | error = __mm_populate(start, len, 0); |
1da177e4 LT |
746 | return error; |
747 | } | |
748 | ||
6a6160a7 | 749 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
750 | { |
751 | int ret; | |
752 | ||
1da177e4 LT |
753 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); |
754 | start &= PAGE_MASK; | |
1f1cd705 DB |
755 | |
756 | down_write(¤t->mm->mmap_sem); | |
1da177e4 LT |
757 | ret = do_mlock(start, len, 0); |
758 | up_write(¤t->mm->mmap_sem); | |
1f1cd705 | 759 | |
1da177e4 LT |
760 | return ret; |
761 | } | |
762 | ||
763 | static int do_mlockall(int flags) | |
764 | { | |
765 | struct vm_area_struct * vma, * prev = NULL; | |
1da177e4 LT |
766 | |
767 | if (flags & MCL_FUTURE) | |
09a9f1d2 | 768 | current->mm->def_flags |= VM_LOCKED; |
9977f0f1 | 769 | else |
09a9f1d2 | 770 | current->mm->def_flags &= ~VM_LOCKED; |
1da177e4 LT |
771 | if (flags == MCL_FUTURE) |
772 | goto out; | |
773 | ||
774 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 775 | vm_flags_t newflags; |
1da177e4 | 776 | |
18693050 ML |
777 | newflags = vma->vm_flags & ~VM_LOCKED; |
778 | if (flags & MCL_CURRENT) | |
09a9f1d2 | 779 | newflags |= VM_LOCKED; |
1da177e4 LT |
780 | |
781 | /* Ignore errors */ | |
782 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
22356f44 | 783 | cond_resched(); |
1da177e4 LT |
784 | } |
785 | out: | |
786 | return 0; | |
787 | } | |
788 | ||
3480b257 | 789 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
790 | { |
791 | unsigned long lock_limit; | |
792 | int ret = -EINVAL; | |
793 | ||
794 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE))) | |
795 | goto out; | |
796 | ||
797 | ret = -EPERM; | |
798 | if (!can_do_mlock()) | |
799 | goto out; | |
800 | ||
df9d6985 CL |
801 | if (flags & MCL_CURRENT) |
802 | lru_add_drain_all(); /* flush pagevec */ | |
8891d6da | 803 | |
59e99e5b | 804 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
805 | lock_limit >>= PAGE_SHIFT; |
806 | ||
807 | ret = -ENOMEM; | |
1f1cd705 DB |
808 | down_write(¤t->mm->mmap_sem); |
809 | ||
1da177e4 LT |
810 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
811 | capable(CAP_IPC_LOCK)) | |
812 | ret = do_mlockall(flags); | |
813 | up_write(¤t->mm->mmap_sem); | |
bebeb3d6 ML |
814 | if (!ret && (flags & MCL_CURRENT)) |
815 | mm_populate(0, TASK_SIZE); | |
1da177e4 LT |
816 | out: |
817 | return ret; | |
818 | } | |
819 | ||
3480b257 | 820 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
821 | { |
822 | int ret; | |
823 | ||
824 | down_write(¤t->mm->mmap_sem); | |
825 | ret = do_mlockall(0); | |
826 | up_write(¤t->mm->mmap_sem); | |
827 | return ret; | |
828 | } | |
829 | ||
830 | /* | |
831 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
832 | * shm segments) get accounted against the user_struct instead. | |
833 | */ | |
834 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
835 | ||
836 | int user_shm_lock(size_t size, struct user_struct *user) | |
837 | { | |
838 | unsigned long lock_limit, locked; | |
839 | int allowed = 0; | |
840 | ||
841 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 842 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
843 | if (lock_limit == RLIM_INFINITY) |
844 | allowed = 1; | |
1da177e4 LT |
845 | lock_limit >>= PAGE_SHIFT; |
846 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
847 | if (!allowed && |
848 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
849 | goto out; |
850 | get_uid(user); | |
851 | user->locked_shm += locked; | |
852 | allowed = 1; | |
853 | out: | |
854 | spin_unlock(&shmlock_user_lock); | |
855 | return allowed; | |
856 | } | |
857 | ||
858 | void user_shm_unlock(size_t size, struct user_struct *user) | |
859 | { | |
860 | spin_lock(&shmlock_user_lock); | |
861 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
862 | spin_unlock(&shmlock_user_lock); | |
863 | free_uid(user); | |
864 | } |