blob: 9cdd063b7d323718b5d43f4713fe3ede2e7d7f92
1 | /* |
2 | * linux/mm/mlock.c |
3 | * |
4 | * (C) Copyright 1995 Linus Torvalds |
5 | * (C) Copyright 2002 Christoph Hellwig |
6 | */ |
7 | |
8 | #include <linux/capability.h> |
9 | #include <linux/mman.h> |
10 | #include <linux/mm.h> |
11 | #include <linux/swap.h> |
12 | #include <linux/swapops.h> |
13 | #include <linux/pagemap.h> |
14 | #include <linux/pagevec.h> |
15 | #include <linux/mempolicy.h> |
16 | #include <linux/syscalls.h> |
17 | #include <linux/sched.h> |
18 | #include <linux/export.h> |
19 | #include <linux/rmap.h> |
20 | #include <linux/mmzone.h> |
21 | #include <linux/hugetlb.h> |
22 | #include <linux/memcontrol.h> |
23 | #include <linux/mm_inline.h> |
24 | |
25 | #include "internal.h" |
26 | |
27 | bool can_do_mlock(void) |
28 | { |
29 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
30 | return true; |
31 | if (capable(CAP_IPC_LOCK)) |
32 | return true; |
33 | return false; |
34 | } |
35 | EXPORT_SYMBOL(can_do_mlock); |
36 | |
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 | */ |
57 | void clear_page_mlock(struct page *page) |
58 | { |
59 | if (!TestClearPageMlocked(page)) |
60 | return; |
61 | |
62 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
63 | -hpage_nr_pages(page)); |
64 | count_vm_event(UNEVICTABLE_PGCLEARED); |
65 | if (!isolate_lru_page(page)) { |
66 | putback_lru_page(page); |
67 | } else { |
68 | /* |
69 | * We lost the race. the page already moved to evictable list. |
70 | */ |
71 | if (PageUnevictable(page)) |
72 | count_vm_event(UNEVICTABLE_PGSTRANDED); |
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 | /* Serialize with page migration */ |
83 | BUG_ON(!PageLocked(page)); |
84 | |
85 | VM_BUG_ON_PAGE(PageTail(page), page); |
86 | VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page); |
87 | |
88 | if (!TestSetPageMlocked(page)) { |
89 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
90 | hpage_nr_pages(page)); |
91 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
92 | if (!isolate_lru_page(page)) |
93 | putback_lru_page(page); |
94 | } |
95 | } |
96 | |
97 | /* |
98 | * Isolate a page from LRU with optional get_page() pin. |
99 | * Assumes lru_lock already held and page already pinned. |
100 | */ |
101 | static bool __munlock_isolate_lru_page(struct page *page, bool getpage) |
102 | { |
103 | if (PageLRU(page)) { |
104 | struct lruvec *lruvec; |
105 | |
106 | lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page)); |
107 | if (getpage) |
108 | get_page(page); |
109 | ClearPageLRU(page); |
110 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
111 | return true; |
112 | } |
113 | |
114 | return false; |
115 | } |
116 | |
117 | /* |
118 | * Finish munlock after successful page isolation |
119 | * |
120 | * Page must be locked. This is a wrapper for try_to_munlock() |
121 | * and putback_lru_page() with munlock accounting. |
122 | */ |
123 | static void __munlock_isolated_page(struct page *page) |
124 | { |
125 | int ret = SWAP_AGAIN; |
126 | |
127 | /* |
128 | * Optimization: if the page was mapped just once, that's our mapping |
129 | * and we don't need to check all the other vmas. |
130 | */ |
131 | if (page_mapcount(page) > 1) |
132 | ret = try_to_munlock(page); |
133 | |
134 | /* Did try_to_unlock() succeed or punt? */ |
135 | if (ret != SWAP_MLOCK) |
136 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
137 | |
138 | putback_lru_page(page); |
139 | } |
140 | |
141 | /* |
142 | * Accounting for page isolation fail during munlock |
143 | * |
144 | * Performs accounting when page isolation fails in munlock. There is nothing |
145 | * else to do because it means some other task has already removed the page |
146 | * from the LRU. putback_lru_page() will take care of removing the page from |
147 | * the unevictable list, if necessary. vmscan [page_referenced()] will move |
148 | * the page back to the unevictable list if some other vma has it mlocked. |
149 | */ |
150 | static void __munlock_isolation_failed(struct page *page) |
151 | { |
152 | if (PageUnevictable(page)) |
153 | __count_vm_event(UNEVICTABLE_PGSTRANDED); |
154 | else |
155 | __count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
156 | } |
157 | |
158 | /** |
159 | * munlock_vma_page - munlock a vma page |
160 | * @page - page to be unlocked, either a normal page or THP page head |
161 | * |
162 | * returns the size of the page as a page mask (0 for normal page, |
163 | * HPAGE_PMD_NR - 1 for THP head page) |
164 | * |
165 | * called from munlock()/munmap() path with page supposedly on the LRU. |
166 | * When we munlock a page, because the vma where we found the page is being |
167 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the |
168 | * page locked so that we can leave it on the unevictable lru list and not |
169 | * bother vmscan with it. However, to walk the page's rmap list in |
170 | * try_to_munlock() we must isolate the page from the LRU. If some other |
171 | * task has removed the page from the LRU, we won't be able to do that. |
172 | * So we clear the PageMlocked as we might not get another chance. If we |
173 | * can't isolate the page, we leave it for putback_lru_page() and vmscan |
174 | * [page_referenced()/try_to_unmap()] to deal with. |
175 | */ |
176 | unsigned int munlock_vma_page(struct page *page) |
177 | { |
178 | int nr_pages; |
179 | struct zone *zone = page_zone(page); |
180 | |
181 | /* For try_to_munlock() and to serialize with page migration */ |
182 | BUG_ON(!PageLocked(page)); |
183 | |
184 | VM_BUG_ON_PAGE(PageTail(page), page); |
185 | |
186 | /* |
187 | * Serialize with any parallel __split_huge_page_refcount() which |
188 | * might otherwise copy PageMlocked to part of the tail pages before |
189 | * we clear it in the head page. It also stabilizes hpage_nr_pages(). |
190 | */ |
191 | spin_lock_irq(zone_lru_lock(zone)); |
192 | |
193 | if (!TestClearPageMlocked(page)) { |
194 | /* Potentially, PTE-mapped THP: do not skip the rest PTEs */ |
195 | nr_pages = 1; |
196 | goto unlock_out; |
197 | } |
198 | |
199 | nr_pages = hpage_nr_pages(page); |
200 | __mod_zone_page_state(zone, NR_MLOCK, -nr_pages); |
201 | |
202 | if (__munlock_isolate_lru_page(page, true)) { |
203 | spin_unlock_irq(zone_lru_lock(zone)); |
204 | __munlock_isolated_page(page); |
205 | goto out; |
206 | } |
207 | __munlock_isolation_failed(page); |
208 | |
209 | unlock_out: |
210 | spin_unlock_irq(zone_lru_lock(zone)); |
211 | |
212 | out: |
213 | return nr_pages - 1; |
214 | } |
215 | |
216 | /* |
217 | * convert get_user_pages() return value to posix mlock() error |
218 | */ |
219 | static int __mlock_posix_error_return(long retval) |
220 | { |
221 | if (retval == -EFAULT) |
222 | retval = -ENOMEM; |
223 | else if (retval == -ENOMEM) |
224 | retval = -EAGAIN; |
225 | return retval; |
226 | } |
227 | |
228 | /* |
229 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() |
230 | * |
231 | * The fast path is available only for evictable pages with single mapping. |
232 | * Then we can bypass the per-cpu pvec and get better performance. |
233 | * when mapcount > 1 we need try_to_munlock() which can fail. |
234 | * when !page_evictable(), we need the full redo logic of putback_lru_page to |
235 | * avoid leaving evictable page in unevictable list. |
236 | * |
237 | * In case of success, @page is added to @pvec and @pgrescued is incremented |
238 | * in case that the page was previously unevictable. @page is also unlocked. |
239 | */ |
240 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, |
241 | int *pgrescued) |
242 | { |
243 | VM_BUG_ON_PAGE(PageLRU(page), page); |
244 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
245 | |
246 | if (page_mapcount(page) <= 1 && page_evictable(page)) { |
247 | pagevec_add(pvec, page); |
248 | if (TestClearPageUnevictable(page)) |
249 | (*pgrescued)++; |
250 | unlock_page(page); |
251 | return true; |
252 | } |
253 | |
254 | return false; |
255 | } |
256 | |
257 | /* |
258 | * Putback multiple evictable pages to the LRU |
259 | * |
260 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of |
261 | * the pages might have meanwhile become unevictable but that is OK. |
262 | */ |
263 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) |
264 | { |
265 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); |
266 | /* |
267 | *__pagevec_lru_add() calls release_pages() so we don't call |
268 | * put_page() explicitly |
269 | */ |
270 | __pagevec_lru_add(pvec); |
271 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); |
272 | } |
273 | |
274 | /* |
275 | * Munlock a batch of pages from the same zone |
276 | * |
277 | * The work is split to two main phases. First phase clears the Mlocked flag |
278 | * and attempts to isolate the pages, all under a single zone lru lock. |
279 | * The second phase finishes the munlock only for pages where isolation |
280 | * succeeded. |
281 | * |
282 | * Note that the pagevec may be modified during the process. |
283 | */ |
284 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) |
285 | { |
286 | int i; |
287 | int nr = pagevec_count(pvec); |
288 | int delta_munlocked = -nr; |
289 | struct pagevec pvec_putback; |
290 | int pgrescued = 0; |
291 | |
292 | pagevec_init(&pvec_putback, 0); |
293 | |
294 | /* Phase 1: page isolation */ |
295 | spin_lock_irq(zone_lru_lock(zone)); |
296 | for (i = 0; i < nr; i++) { |
297 | struct page *page = pvec->pages[i]; |
298 | |
299 | if (TestClearPageMlocked(page)) { |
300 | /* |
301 | * We already have pin from follow_page_mask() |
302 | * so we can spare the get_page() here. |
303 | */ |
304 | if (__munlock_isolate_lru_page(page, false)) |
305 | continue; |
306 | else |
307 | __munlock_isolation_failed(page); |
308 | } else { |
309 | delta_munlocked++; |
310 | } |
311 | |
312 | /* |
313 | * We won't be munlocking this page in the next phase |
314 | * but we still need to release the follow_page_mask() |
315 | * pin. We cannot do it under lru_lock however. If it's |
316 | * the last pin, __page_cache_release() would deadlock. |
317 | */ |
318 | pagevec_add(&pvec_putback, pvec->pages[i]); |
319 | pvec->pages[i] = NULL; |
320 | } |
321 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
322 | spin_unlock_irq(zone_lru_lock(zone)); |
323 | |
324 | /* Now we can release pins of pages that we are not munlocking */ |
325 | pagevec_release(&pvec_putback); |
326 | |
327 | /* Phase 2: page munlock */ |
328 | for (i = 0; i < nr; i++) { |
329 | struct page *page = pvec->pages[i]; |
330 | |
331 | if (page) { |
332 | lock_page(page); |
333 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
334 | &pgrescued)) { |
335 | /* |
336 | * Slow path. We don't want to lose the last |
337 | * pin before unlock_page() |
338 | */ |
339 | get_page(page); /* for putback_lru_page() */ |
340 | __munlock_isolated_page(page); |
341 | unlock_page(page); |
342 | put_page(page); /* from follow_page_mask() */ |
343 | } |
344 | } |
345 | } |
346 | |
347 | /* |
348 | * Phase 3: page putback for pages that qualified for the fast path |
349 | * This will also call put_page() to return pin from follow_page_mask() |
350 | */ |
351 | if (pagevec_count(&pvec_putback)) |
352 | __putback_lru_fast(&pvec_putback, pgrescued); |
353 | } |
354 | |
355 | /* |
356 | * Fill up pagevec for __munlock_pagevec using pte walk |
357 | * |
358 | * The function expects that the struct page corresponding to @start address is |
359 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. |
360 | * |
361 | * The rest of @pvec is filled by subsequent pages within the same pmd and same |
362 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These |
363 | * pages also get pinned. |
364 | * |
365 | * Returns the address of the next page that should be scanned. This equals |
366 | * @start + PAGE_SIZE when no page could be added by the pte walk. |
367 | */ |
368 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, |
369 | struct vm_area_struct *vma, int zoneid, unsigned long start, |
370 | unsigned long end) |
371 | { |
372 | pte_t *pte; |
373 | spinlock_t *ptl; |
374 | |
375 | /* |
376 | * Initialize pte walk starting at the already pinned page where we |
377 | * are sure that there is a pte, as it was pinned under the same |
378 | * mmap_sem write op. |
379 | */ |
380 | pte = get_locked_pte(vma->vm_mm, start, &ptl); |
381 | /* Make sure we do not cross the page table boundary */ |
382 | end = pgd_addr_end(start, end); |
383 | end = pud_addr_end(start, end); |
384 | end = pmd_addr_end(start, end); |
385 | |
386 | /* The page next to the pinned page is the first we will try to get */ |
387 | start += PAGE_SIZE; |
388 | while (start < end) { |
389 | struct page *page = NULL; |
390 | pte++; |
391 | if (pte_present(*pte)) |
392 | page = vm_normal_page(vma, start, *pte); |
393 | /* |
394 | * Break if page could not be obtained or the page's node+zone does not |
395 | * match |
396 | */ |
397 | if (!page || page_zone_id(page) != zoneid) |
398 | break; |
399 | |
400 | /* |
401 | * Do not use pagevec for PTE-mapped THP, |
402 | * munlock_vma_pages_range() will handle them. |
403 | */ |
404 | if (PageTransCompound(page)) |
405 | break; |
406 | |
407 | get_page(page); |
408 | /* |
409 | * Increase the address that will be returned *before* the |
410 | * eventual break due to pvec becoming full by adding the page |
411 | */ |
412 | start += PAGE_SIZE; |
413 | if (pagevec_add(pvec, page) == 0) |
414 | break; |
415 | } |
416 | pte_unmap_unlock(pte, ptl); |
417 | return start; |
418 | } |
419 | |
420 | /* |
421 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
422 | * @vma - vma containing range to be munlock()ed. |
423 | * @start - start address in @vma of the range |
424 | * @end - end of range in @vma. |
425 | * |
426 | * For mremap(), munmap() and exit(). |
427 | * |
428 | * Called with @vma VM_LOCKED. |
429 | * |
430 | * Returns with VM_LOCKED cleared. Callers must be prepared to |
431 | * deal with this. |
432 | * |
433 | * We don't save and restore VM_LOCKED here because pages are |
434 | * still on lru. In unmap path, pages might be scanned by reclaim |
435 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and |
436 | * free them. This will result in freeing mlocked pages. |
437 | */ |
438 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
439 | unsigned long start, unsigned long end) |
440 | { |
441 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
442 | |
443 | while (start < end) { |
444 | struct page *page; |
445 | unsigned int page_mask = 0; |
446 | unsigned long page_increm; |
447 | struct pagevec pvec; |
448 | struct zone *zone; |
449 | int zoneid; |
450 | |
451 | pagevec_init(&pvec, 0); |
452 | /* |
453 | * Although FOLL_DUMP is intended for get_dump_page(), |
454 | * it just so happens that its special treatment of the |
455 | * ZERO_PAGE (returning an error instead of doing get_page) |
456 | * suits munlock very well (and if somehow an abnormal page |
457 | * has sneaked into the range, we won't oops here: great). |
458 | */ |
459 | page = follow_page(vma, start, FOLL_GET | FOLL_DUMP); |
460 | |
461 | if (page && !IS_ERR(page)) { |
462 | if (PageTransTail(page)) { |
463 | VM_BUG_ON_PAGE(PageMlocked(page), page); |
464 | put_page(page); /* follow_page_mask() */ |
465 | } else if (PageTransHuge(page)) { |
466 | lock_page(page); |
467 | /* |
468 | * Any THP page found by follow_page_mask() may |
469 | * have gotten split before reaching |
470 | * munlock_vma_page(), so we need to compute |
471 | * the page_mask here instead. |
472 | */ |
473 | page_mask = munlock_vma_page(page); |
474 | unlock_page(page); |
475 | put_page(page); /* follow_page_mask() */ |
476 | } else { |
477 | /* |
478 | * Non-huge pages are handled in batches via |
479 | * pagevec. The pin from follow_page_mask() |
480 | * prevents them from collapsing by THP. |
481 | */ |
482 | pagevec_add(&pvec, page); |
483 | zone = page_zone(page); |
484 | zoneid = page_zone_id(page); |
485 | |
486 | /* |
487 | * Try to fill the rest of pagevec using fast |
488 | * pte walk. This will also update start to |
489 | * the next page to process. Then munlock the |
490 | * pagevec. |
491 | */ |
492 | start = __munlock_pagevec_fill(&pvec, vma, |
493 | zoneid, start, end); |
494 | __munlock_pagevec(&pvec, zone); |
495 | goto next; |
496 | } |
497 | } |
498 | page_increm = 1 + page_mask; |
499 | start += page_increm * PAGE_SIZE; |
500 | next: |
501 | cond_resched(); |
502 | } |
503 | } |
504 | |
505 | /* |
506 | * mlock_fixup - handle mlock[all]/munlock[all] requests. |
507 | * |
508 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and |
509 | * munlock is a no-op. However, for some special vmas, we go ahead and |
510 | * populate the ptes. |
511 | * |
512 | * For vmas that pass the filters, merge/split as appropriate. |
513 | */ |
514 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
515 | unsigned long start, unsigned long end, vm_flags_t newflags) |
516 | { |
517 | struct mm_struct *mm = vma->vm_mm; |
518 | pgoff_t pgoff; |
519 | int nr_pages; |
520 | int ret = 0; |
521 | int lock = !!(newflags & VM_LOCKED); |
522 | vm_flags_t old_flags = vma->vm_flags; |
523 | |
524 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
525 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
526 | /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */ |
527 | goto out; |
528 | |
529 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
530 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, |
531 | vma->vm_file, pgoff, vma_policy(vma), |
532 | vma->vm_userfaultfd_ctx, vma_get_anon_name(vma)); |
533 | if (*prev) { |
534 | vma = *prev; |
535 | goto success; |
536 | } |
537 | |
538 | if (start != vma->vm_start) { |
539 | ret = split_vma(mm, vma, start, 1); |
540 | if (ret) |
541 | goto out; |
542 | } |
543 | |
544 | if (end != vma->vm_end) { |
545 | ret = split_vma(mm, vma, end, 0); |
546 | if (ret) |
547 | goto out; |
548 | } |
549 | |
550 | success: |
551 | /* |
552 | * Keep track of amount of locked VM. |
553 | */ |
554 | nr_pages = (end - start) >> PAGE_SHIFT; |
555 | if (!lock) |
556 | nr_pages = -nr_pages; |
557 | else if (old_flags & VM_LOCKED) |
558 | nr_pages = 0; |
559 | mm->locked_vm += nr_pages; |
560 | |
561 | /* |
562 | * vm_flags is protected by the mmap_sem held in write mode. |
563 | * It's okay if try_to_unmap_one unmaps a page just after we |
564 | * set VM_LOCKED, populate_vma_page_range will bring it back. |
565 | */ |
566 | |
567 | if (lock) |
568 | vma->vm_flags = newflags; |
569 | else |
570 | munlock_vma_pages_range(vma, start, end); |
571 | |
572 | out: |
573 | *prev = vma; |
574 | return ret; |
575 | } |
576 | |
577 | static int apply_vma_lock_flags(unsigned long start, size_t len, |
578 | vm_flags_t flags) |
579 | { |
580 | unsigned long nstart, end, tmp; |
581 | struct vm_area_struct * vma, * prev; |
582 | int error; |
583 | |
584 | VM_BUG_ON(offset_in_page(start)); |
585 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
586 | end = start + len; |
587 | if (end < start) |
588 | return -EINVAL; |
589 | if (end == start) |
590 | return 0; |
591 | vma = find_vma(current->mm, start); |
592 | if (!vma || vma->vm_start > start) |
593 | return -ENOMEM; |
594 | |
595 | prev = vma->vm_prev; |
596 | if (start > vma->vm_start) |
597 | prev = vma; |
598 | |
599 | for (nstart = start ; ; ) { |
600 | vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
601 | |
602 | newflags |= flags; |
603 | |
604 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
605 | tmp = vma->vm_end; |
606 | if (tmp > end) |
607 | tmp = end; |
608 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); |
609 | if (error) |
610 | break; |
611 | nstart = tmp; |
612 | if (nstart < prev->vm_end) |
613 | nstart = prev->vm_end; |
614 | if (nstart >= end) |
615 | break; |
616 | |
617 | vma = prev->vm_next; |
618 | if (!vma || vma->vm_start != nstart) { |
619 | error = -ENOMEM; |
620 | break; |
621 | } |
622 | } |
623 | return error; |
624 | } |
625 | |
626 | /* |
627 | * Go through vma areas and sum size of mlocked |
628 | * vma pages, as return value. |
629 | * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT) |
630 | * is also counted. |
631 | * Return value: previously mlocked page counts |
632 | */ |
633 | static int count_mm_mlocked_page_nr(struct mm_struct *mm, |
634 | unsigned long start, size_t len) |
635 | { |
636 | struct vm_area_struct *vma; |
637 | int count = 0; |
638 | |
639 | if (mm == NULL) |
640 | mm = current->mm; |
641 | |
642 | vma = find_vma(mm, start); |
643 | if (vma == NULL) |
644 | vma = mm->mmap; |
645 | |
646 | for (; vma ; vma = vma->vm_next) { |
647 | if (start >= vma->vm_end) |
648 | continue; |
649 | if (start + len <= vma->vm_start) |
650 | break; |
651 | if (vma->vm_flags & VM_LOCKED) { |
652 | if (start > vma->vm_start) |
653 | count -= (start - vma->vm_start); |
654 | if (start + len < vma->vm_end) { |
655 | count += start + len - vma->vm_start; |
656 | break; |
657 | } |
658 | count += vma->vm_end - vma->vm_start; |
659 | } |
660 | } |
661 | |
662 | return count >> PAGE_SHIFT; |
663 | } |
664 | |
665 | static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags) |
666 | { |
667 | unsigned long locked; |
668 | unsigned long lock_limit; |
669 | int error = -ENOMEM; |
670 | |
671 | if (!can_do_mlock()) |
672 | return -EPERM; |
673 | |
674 | lru_add_drain_all(); /* flush pagevec */ |
675 | |
676 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
677 | start &= PAGE_MASK; |
678 | |
679 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
680 | lock_limit >>= PAGE_SHIFT; |
681 | locked = len >> PAGE_SHIFT; |
682 | |
683 | if (down_write_killable(¤t->mm->mmap_sem)) |
684 | return -EINTR; |
685 | |
686 | locked += current->mm->locked_vm; |
687 | if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) { |
688 | /* |
689 | * It is possible that the regions requested intersect with |
690 | * previously mlocked areas, that part area in "mm->locked_vm" |
691 | * should not be counted to new mlock increment count. So check |
692 | * and adjust locked count if necessary. |
693 | */ |
694 | locked -= count_mm_mlocked_page_nr(current->mm, |
695 | start, len); |
696 | } |
697 | |
698 | /* check against resource limits */ |
699 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) |
700 | error = apply_vma_lock_flags(start, len, flags); |
701 | |
702 | up_write(¤t->mm->mmap_sem); |
703 | if (error) |
704 | return error; |
705 | |
706 | error = __mm_populate(start, len, 0); |
707 | if (error) |
708 | return __mlock_posix_error_return(error); |
709 | return 0; |
710 | } |
711 | |
712 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
713 | { |
714 | return do_mlock(start, len, VM_LOCKED); |
715 | } |
716 | |
717 | SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags) |
718 | { |
719 | vm_flags_t vm_flags = VM_LOCKED; |
720 | |
721 | if (flags & ~MLOCK_ONFAULT) |
722 | return -EINVAL; |
723 | |
724 | if (flags & MLOCK_ONFAULT) |
725 | vm_flags |= VM_LOCKONFAULT; |
726 | |
727 | return do_mlock(start, len, vm_flags); |
728 | } |
729 | |
730 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
731 | { |
732 | int ret; |
733 | |
734 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
735 | start &= PAGE_MASK; |
736 | |
737 | if (down_write_killable(¤t->mm->mmap_sem)) |
738 | return -EINTR; |
739 | ret = apply_vma_lock_flags(start, len, 0); |
740 | up_write(¤t->mm->mmap_sem); |
741 | |
742 | return ret; |
743 | } |
744 | |
745 | /* |
746 | * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall) |
747 | * and translate into the appropriate modifications to mm->def_flags and/or the |
748 | * flags for all current VMAs. |
749 | * |
750 | * There are a couple of subtleties with this. If mlockall() is called multiple |
751 | * times with different flags, the values do not necessarily stack. If mlockall |
752 | * is called once including the MCL_FUTURE flag and then a second time without |
753 | * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags. |
754 | */ |
755 | static int apply_mlockall_flags(int flags) |
756 | { |
757 | struct vm_area_struct * vma, * prev = NULL; |
758 | vm_flags_t to_add = 0; |
759 | |
760 | current->mm->def_flags &= VM_LOCKED_CLEAR_MASK; |
761 | if (flags & MCL_FUTURE) { |
762 | current->mm->def_flags |= VM_LOCKED; |
763 | |
764 | if (flags & MCL_ONFAULT) |
765 | current->mm->def_flags |= VM_LOCKONFAULT; |
766 | |
767 | if (!(flags & MCL_CURRENT)) |
768 | goto out; |
769 | } |
770 | |
771 | if (flags & MCL_CURRENT) { |
772 | to_add |= VM_LOCKED; |
773 | if (flags & MCL_ONFAULT) |
774 | to_add |= VM_LOCKONFAULT; |
775 | } |
776 | |
777 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { |
778 | vm_flags_t newflags; |
779 | |
780 | newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
781 | newflags |= to_add; |
782 | |
783 | /* Ignore errors */ |
784 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); |
785 | cond_resched_rcu_qs(); |
786 | } |
787 | out: |
788 | return 0; |
789 | } |
790 | |
791 | SYSCALL_DEFINE1(mlockall, int, flags) |
792 | { |
793 | unsigned long lock_limit; |
794 | int ret; |
795 | |
796 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT))) |
797 | return -EINVAL; |
798 | |
799 | if (!can_do_mlock()) |
800 | return -EPERM; |
801 | |
802 | if (flags & MCL_CURRENT) |
803 | lru_add_drain_all(); /* flush pagevec */ |
804 | |
805 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
806 | lock_limit >>= PAGE_SHIFT; |
807 | |
808 | if (down_write_killable(¤t->mm->mmap_sem)) |
809 | return -EINTR; |
810 | |
811 | ret = -ENOMEM; |
812 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
813 | capable(CAP_IPC_LOCK)) |
814 | ret = apply_mlockall_flags(flags); |
815 | up_write(¤t->mm->mmap_sem); |
816 | if (!ret && (flags & MCL_CURRENT)) |
817 | mm_populate(0, TASK_SIZE); |
818 | |
819 | return ret; |
820 | } |
821 | |
822 | SYSCALL_DEFINE0(munlockall) |
823 | { |
824 | int ret; |
825 | |
826 | if (down_write_killable(¤t->mm->mmap_sem)) |
827 | return -EINTR; |
828 | ret = apply_mlockall_flags(0); |
829 | up_write(¤t->mm->mmap_sem); |
830 | return ret; |
831 | } |
832 | |
833 | /* |
834 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB |
835 | * shm segments) get accounted against the user_struct instead. |
836 | */ |
837 | static DEFINE_SPINLOCK(shmlock_user_lock); |
838 | |
839 | int user_shm_lock(size_t size, struct user_struct *user) |
840 | { |
841 | unsigned long lock_limit, locked; |
842 | int allowed = 0; |
843 | |
844 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
845 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
846 | if (lock_limit == RLIM_INFINITY) |
847 | allowed = 1; |
848 | lock_limit >>= PAGE_SHIFT; |
849 | spin_lock(&shmlock_user_lock); |
850 | if (!allowed && |
851 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) |
852 | goto out; |
853 | get_uid(user); |
854 | user->locked_shm += locked; |
855 | allowed = 1; |
856 | out: |
857 | spin_unlock(&shmlock_user_lock); |
858 | return allowed; |
859 | } |
860 | |
861 | void user_shm_unlock(size_t size, struct user_struct *user) |
862 | { |
863 | spin_lock(&shmlock_user_lock); |
864 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
865 | spin_unlock(&shmlock_user_lock); |
866 | free_uid(user); |
867 | } |
868 |