]>
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 | |
c424be1c VB |
136 | * @page - page to be unlocked, either a normal page or THP page head |
137 | * | |
138 | * returns the size of the page as a page mask (0 for normal page, | |
139 | * HPAGE_PMD_NR - 1 for THP head page) | |
b291f000 | 140 | * |
6927c1dd LS |
141 | * called from munlock()/munmap() path with page supposedly on the LRU. |
142 | * When we munlock a page, because the vma where we found the page is being | |
143 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the | |
144 | * page locked so that we can leave it on the unevictable lru list and not | |
145 | * bother vmscan with it. However, to walk the page's rmap list in | |
146 | * try_to_munlock() we must isolate the page from the LRU. If some other | |
147 | * task has removed the page from the LRU, we won't be able to do that. | |
148 | * So we clear the PageMlocked as we might not get another chance. If we | |
149 | * can't isolate the page, we leave it for putback_lru_page() and vmscan | |
150 | * [page_referenced()/try_to_unmap()] to deal with. | |
b291f000 | 151 | */ |
ff6a6da6 | 152 | unsigned int munlock_vma_page(struct page *page) |
b291f000 | 153 | { |
c424be1c | 154 | unsigned int nr_pages; |
ff6a6da6 | 155 | |
b291f000 NP |
156 | BUG_ON(!PageLocked(page)); |
157 | ||
5344b7e6 | 158 | if (TestClearPageMlocked(page)) { |
c424be1c | 159 | nr_pages = hpage_nr_pages(page); |
ff6a6da6 | 160 | mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); |
7225522b VB |
161 | if (!isolate_lru_page(page)) |
162 | __munlock_isolated_page(page); | |
163 | else | |
164 | __munlock_isolation_failed(page); | |
c424be1c VB |
165 | } else { |
166 | nr_pages = hpage_nr_pages(page); | |
b291f000 | 167 | } |
ff6a6da6 | 168 | |
c424be1c VB |
169 | /* |
170 | * Regardless of the original PageMlocked flag, we determine nr_pages | |
171 | * after touching the flag. This leaves a possible race with a THP page | |
172 | * split, such that a whole THP page was munlocked, but nr_pages == 1. | |
173 | * Returning a smaller mask due to that is OK, the worst that can | |
174 | * happen is subsequent useless scanning of the former tail pages. | |
175 | * The NR_MLOCK accounting can however become broken. | |
176 | */ | |
177 | return nr_pages - 1; | |
b291f000 NP |
178 | } |
179 | ||
ba470de4 | 180 | /** |
408e82b7 | 181 | * __mlock_vma_pages_range() - mlock a range of pages in the vma. |
ba470de4 RR |
182 | * @vma: target vma |
183 | * @start: start address | |
184 | * @end: end address | |
ba470de4 | 185 | * |
408e82b7 | 186 | * This takes care of making the pages present too. |
b291f000 | 187 | * |
ba470de4 | 188 | * return 0 on success, negative error code on error. |
b291f000 | 189 | * |
ba470de4 | 190 | * vma->vm_mm->mmap_sem must be held for at least read. |
b291f000 | 191 | */ |
cea10a19 ML |
192 | long __mlock_vma_pages_range(struct vm_area_struct *vma, |
193 | unsigned long start, unsigned long end, int *nonblocking) | |
b291f000 NP |
194 | { |
195 | struct mm_struct *mm = vma->vm_mm; | |
28a35716 | 196 | unsigned long nr_pages = (end - start) / PAGE_SIZE; |
408e82b7 | 197 | int gup_flags; |
ba470de4 RR |
198 | |
199 | VM_BUG_ON(start & ~PAGE_MASK); | |
200 | VM_BUG_ON(end & ~PAGE_MASK); | |
201 | VM_BUG_ON(start < vma->vm_start); | |
202 | VM_BUG_ON(end > vma->vm_end); | |
408e82b7 | 203 | VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem)); |
b291f000 | 204 | |
a1fde08c | 205 | gup_flags = FOLL_TOUCH | FOLL_MLOCK; |
5ecfda04 ML |
206 | /* |
207 | * We want to touch writable mappings with a write fault in order | |
208 | * to break COW, except for shared mappings because these don't COW | |
209 | * and we would not want to dirty them for nothing. | |
210 | */ | |
211 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
58fa879e | 212 | gup_flags |= FOLL_WRITE; |
b291f000 | 213 | |
fdf4c587 ML |
214 | /* |
215 | * We want mlock to succeed for regions that have any permissions | |
216 | * other than PROT_NONE. | |
217 | */ | |
218 | if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)) | |
219 | gup_flags |= FOLL_FORCE; | |
220 | ||
4805b02e JW |
221 | /* |
222 | * We made sure addr is within a VMA, so the following will | |
223 | * not result in a stack expansion that recurses back here. | |
224 | */ | |
ff6a6da6 | 225 | return __get_user_pages(current, mm, start, nr_pages, gup_flags, |
53a7706d | 226 | NULL, NULL, nonblocking); |
9978ad58 LS |
227 | } |
228 | ||
229 | /* | |
230 | * convert get_user_pages() return value to posix mlock() error | |
231 | */ | |
232 | static int __mlock_posix_error_return(long retval) | |
233 | { | |
234 | if (retval == -EFAULT) | |
235 | retval = -ENOMEM; | |
236 | else if (retval == -ENOMEM) | |
237 | retval = -EAGAIN; | |
238 | return retval; | |
b291f000 NP |
239 | } |
240 | ||
56afe477 VB |
241 | /* |
242 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() | |
243 | * | |
244 | * The fast path is available only for evictable pages with single mapping. | |
245 | * Then we can bypass the per-cpu pvec and get better performance. | |
246 | * when mapcount > 1 we need try_to_munlock() which can fail. | |
247 | * when !page_evictable(), we need the full redo logic of putback_lru_page to | |
248 | * avoid leaving evictable page in unevictable list. | |
249 | * | |
250 | * In case of success, @page is added to @pvec and @pgrescued is incremented | |
251 | * in case that the page was previously unevictable. @page is also unlocked. | |
252 | */ | |
253 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, | |
254 | int *pgrescued) | |
255 | { | |
256 | VM_BUG_ON(PageLRU(page)); | |
257 | VM_BUG_ON(!PageLocked(page)); | |
258 | ||
259 | if (page_mapcount(page) <= 1 && page_evictable(page)) { | |
260 | pagevec_add(pvec, page); | |
261 | if (TestClearPageUnevictable(page)) | |
262 | (*pgrescued)++; | |
263 | unlock_page(page); | |
264 | return true; | |
265 | } | |
266 | ||
267 | return false; | |
268 | } | |
269 | ||
270 | /* | |
271 | * Putback multiple evictable pages to the LRU | |
272 | * | |
273 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of | |
274 | * the pages might have meanwhile become unevictable but that is OK. | |
275 | */ | |
276 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) | |
277 | { | |
278 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); | |
279 | /* | |
280 | *__pagevec_lru_add() calls release_pages() so we don't call | |
281 | * put_page() explicitly | |
282 | */ | |
283 | __pagevec_lru_add(pvec); | |
284 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
285 | } | |
286 | ||
7225522b VB |
287 | /* |
288 | * Munlock a batch of pages from the same zone | |
289 | * | |
290 | * The work is split to two main phases. First phase clears the Mlocked flag | |
291 | * and attempts to isolate the pages, all under a single zone lru lock. | |
292 | * The second phase finishes the munlock only for pages where isolation | |
293 | * succeeded. | |
294 | * | |
7a8010cd | 295 | * Note that the pagevec may be modified during the process. |
7225522b VB |
296 | */ |
297 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) | |
298 | { | |
299 | int i; | |
300 | int nr = pagevec_count(pvec); | |
1ebb7cc6 | 301 | int delta_munlocked = -nr; |
56afe477 VB |
302 | struct pagevec pvec_putback; |
303 | int pgrescued = 0; | |
7225522b VB |
304 | |
305 | /* Phase 1: page isolation */ | |
306 | spin_lock_irq(&zone->lru_lock); | |
307 | for (i = 0; i < nr; i++) { | |
308 | struct page *page = pvec->pages[i]; | |
309 | ||
310 | if (TestClearPageMlocked(page)) { | |
311 | struct lruvec *lruvec; | |
312 | int lru; | |
313 | ||
7225522b VB |
314 | if (PageLRU(page)) { |
315 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
316 | lru = page_lru(page); | |
5b40998a VB |
317 | /* |
318 | * We already have pin from follow_page_mask() | |
319 | * so we can spare the get_page() here. | |
320 | */ | |
7225522b VB |
321 | ClearPageLRU(page); |
322 | del_page_from_lru_list(page, lruvec, lru); | |
323 | } else { | |
324 | __munlock_isolation_failed(page); | |
325 | goto skip_munlock; | |
326 | } | |
327 | ||
328 | } else { | |
329 | skip_munlock: | |
330 | /* | |
331 | * We won't be munlocking this page in the next phase | |
332 | * but we still need to release the follow_page_mask() | |
333 | * pin. | |
334 | */ | |
335 | pvec->pages[i] = NULL; | |
336 | put_page(page); | |
1ebb7cc6 | 337 | delta_munlocked++; |
7225522b VB |
338 | } |
339 | } | |
1ebb7cc6 | 340 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
7225522b VB |
341 | spin_unlock_irq(&zone->lru_lock); |
342 | ||
56afe477 VB |
343 | /* Phase 2: page munlock */ |
344 | pagevec_init(&pvec_putback, 0); | |
7225522b VB |
345 | for (i = 0; i < nr; i++) { |
346 | struct page *page = pvec->pages[i]; | |
347 | ||
348 | if (page) { | |
349 | lock_page(page); | |
56afe477 VB |
350 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
351 | &pgrescued)) { | |
5b40998a VB |
352 | /* |
353 | * Slow path. We don't want to lose the last | |
354 | * pin before unlock_page() | |
355 | */ | |
356 | get_page(page); /* for putback_lru_page() */ | |
56afe477 VB |
357 | __munlock_isolated_page(page); |
358 | unlock_page(page); | |
5b40998a | 359 | put_page(page); /* from follow_page_mask() */ |
56afe477 | 360 | } |
7225522b VB |
361 | } |
362 | } | |
56afe477 | 363 | |
5b40998a VB |
364 | /* |
365 | * Phase 3: page putback for pages that qualified for the fast path | |
366 | * This will also call put_page() to return pin from follow_page_mask() | |
367 | */ | |
56afe477 VB |
368 | if (pagevec_count(&pvec_putback)) |
369 | __putback_lru_fast(&pvec_putback, pgrescued); | |
7a8010cd VB |
370 | } |
371 | ||
372 | /* | |
373 | * Fill up pagevec for __munlock_pagevec using pte walk | |
374 | * | |
375 | * The function expects that the struct page corresponding to @start address is | |
376 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. | |
377 | * | |
378 | * The rest of @pvec is filled by subsequent pages within the same pmd and same | |
379 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These | |
380 | * pages also get pinned. | |
381 | * | |
382 | * Returns the address of the next page that should be scanned. This equals | |
383 | * @start + PAGE_SIZE when no page could be added by the pte walk. | |
384 | */ | |
385 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, | |
386 | struct vm_area_struct *vma, int zoneid, unsigned long start, | |
387 | unsigned long end) | |
388 | { | |
389 | pte_t *pte; | |
390 | spinlock_t *ptl; | |
391 | ||
392 | /* | |
393 | * Initialize pte walk starting at the already pinned page where we | |
eadb41ae VB |
394 | * are sure that there is a pte, as it was pinned under the same |
395 | * mmap_sem write op. | |
7a8010cd VB |
396 | */ |
397 | pte = get_locked_pte(vma->vm_mm, start, &ptl); | |
eadb41ae VB |
398 | /* Make sure we do not cross the page table boundary */ |
399 | end = pgd_addr_end(start, end); | |
400 | end = pud_addr_end(start, end); | |
401 | end = pmd_addr_end(start, end); | |
7a8010cd VB |
402 | |
403 | /* The page next to the pinned page is the first we will try to get */ | |
404 | start += PAGE_SIZE; | |
405 | while (start < end) { | |
406 | struct page *page = NULL; | |
407 | pte++; | |
408 | if (pte_present(*pte)) | |
409 | page = vm_normal_page(vma, start, *pte); | |
410 | /* | |
411 | * Break if page could not be obtained or the page's node+zone does not | |
412 | * match | |
413 | */ | |
414 | if (!page || page_zone_id(page) != zoneid) | |
415 | break; | |
56afe477 | 416 | |
7a8010cd VB |
417 | get_page(page); |
418 | /* | |
419 | * Increase the address that will be returned *before* the | |
420 | * eventual break due to pvec becoming full by adding the page | |
421 | */ | |
422 | start += PAGE_SIZE; | |
423 | if (pagevec_add(pvec, page) == 0) | |
424 | break; | |
425 | } | |
426 | pte_unmap_unlock(pte, ptl); | |
427 | return start; | |
7225522b VB |
428 | } |
429 | ||
b291f000 | 430 | /* |
ba470de4 RR |
431 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
432 | * @vma - vma containing range to be munlock()ed. | |
433 | * @start - start address in @vma of the range | |
434 | * @end - end of range in @vma. | |
435 | * | |
436 | * For mremap(), munmap() and exit(). | |
437 | * | |
438 | * Called with @vma VM_LOCKED. | |
439 | * | |
440 | * Returns with VM_LOCKED cleared. Callers must be prepared to | |
441 | * deal with this. | |
442 | * | |
443 | * We don't save and restore VM_LOCKED here because pages are | |
444 | * still on lru. In unmap path, pages might be scanned by reclaim | |
445 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and | |
446 | * free them. This will result in freeing mlocked pages. | |
b291f000 | 447 | */ |
ba470de4 | 448 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
408e82b7 | 449 | unsigned long start, unsigned long end) |
b291f000 NP |
450 | { |
451 | vma->vm_flags &= ~VM_LOCKED; | |
408e82b7 | 452 | |
ff6a6da6 | 453 | while (start < end) { |
7a8010cd | 454 | struct page *page = NULL; |
c424be1c VB |
455 | unsigned int page_mask; |
456 | unsigned long page_increm; | |
7a8010cd VB |
457 | struct pagevec pvec; |
458 | struct zone *zone; | |
459 | int zoneid; | |
ff6a6da6 | 460 | |
7a8010cd | 461 | pagevec_init(&pvec, 0); |
6e919717 HD |
462 | /* |
463 | * Although FOLL_DUMP is intended for get_dump_page(), | |
464 | * it just so happens that its special treatment of the | |
465 | * ZERO_PAGE (returning an error instead of doing get_page) | |
466 | * suits munlock very well (and if somehow an abnormal page | |
467 | * has sneaked into the range, we won't oops here: great). | |
468 | */ | |
ff6a6da6 | 469 | page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP, |
7a8010cd VB |
470 | &page_mask); |
471 | ||
6e919717 | 472 | if (page && !IS_ERR(page)) { |
7225522b | 473 | if (PageTransHuge(page)) { |
7225522b VB |
474 | lock_page(page); |
475 | /* | |
476 | * Any THP page found by follow_page_mask() may | |
477 | * have gotten split before reaching | |
478 | * munlock_vma_page(), so we need to recompute | |
479 | * the page_mask here. | |
480 | */ | |
481 | page_mask = munlock_vma_page(page); | |
482 | unlock_page(page); | |
483 | put_page(page); /* follow_page_mask() */ | |
484 | } else { | |
485 | /* | |
7a8010cd VB |
486 | * Non-huge pages are handled in batches via |
487 | * pagevec. The pin from follow_page_mask() | |
488 | * prevents them from collapsing by THP. | |
489 | */ | |
490 | pagevec_add(&pvec, page); | |
491 | zone = page_zone(page); | |
492 | zoneid = page_zone_id(page); | |
493 | ||
494 | /* | |
495 | * Try to fill the rest of pagevec using fast | |
496 | * pte walk. This will also update start to | |
497 | * the next page to process. Then munlock the | |
498 | * pagevec. | |
7225522b | 499 | */ |
7a8010cd VB |
500 | start = __munlock_pagevec_fill(&pvec, vma, |
501 | zoneid, start, end); | |
502 | __munlock_pagevec(&pvec, zone); | |
503 | goto next; | |
7225522b | 504 | } |
408e82b7 | 505 | } |
c424be1c VB |
506 | /* It's a bug to munlock in the middle of a THP page */ |
507 | VM_BUG_ON((start >> PAGE_SHIFT) & page_mask); | |
508 | page_increm = 1 + page_mask; | |
ff6a6da6 | 509 | start += page_increm * PAGE_SIZE; |
7a8010cd | 510 | next: |
408e82b7 HD |
511 | cond_resched(); |
512 | } | |
b291f000 NP |
513 | } |
514 | ||
515 | /* | |
516 | * mlock_fixup - handle mlock[all]/munlock[all] requests. | |
517 | * | |
518 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and | |
519 | * munlock is a no-op. However, for some special vmas, we go ahead and | |
cea10a19 | 520 | * populate the ptes. |
b291f000 NP |
521 | * |
522 | * For vmas that pass the filters, merge/split as appropriate. | |
523 | */ | |
1da177e4 | 524 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
ca16d140 | 525 | unsigned long start, unsigned long end, vm_flags_t newflags) |
1da177e4 | 526 | { |
b291f000 | 527 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 528 | pgoff_t pgoff; |
b291f000 | 529 | int nr_pages; |
1da177e4 | 530 | int ret = 0; |
ca16d140 | 531 | int lock = !!(newflags & VM_LOCKED); |
1da177e4 | 532 | |
fed067da | 533 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
31db58b3 | 534 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
b291f000 NP |
535 | goto out; /* don't set VM_LOCKED, don't count */ |
536 | ||
1da177e4 LT |
537 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
538 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, | |
539 | vma->vm_file, pgoff, vma_policy(vma)); | |
540 | if (*prev) { | |
541 | vma = *prev; | |
542 | goto success; | |
543 | } | |
544 | ||
1da177e4 LT |
545 | if (start != vma->vm_start) { |
546 | ret = split_vma(mm, vma, start, 1); | |
547 | if (ret) | |
548 | goto out; | |
549 | } | |
550 | ||
551 | if (end != vma->vm_end) { | |
552 | ret = split_vma(mm, vma, end, 0); | |
553 | if (ret) | |
554 | goto out; | |
555 | } | |
556 | ||
557 | success: | |
b291f000 NP |
558 | /* |
559 | * Keep track of amount of locked VM. | |
560 | */ | |
561 | nr_pages = (end - start) >> PAGE_SHIFT; | |
562 | if (!lock) | |
563 | nr_pages = -nr_pages; | |
564 | mm->locked_vm += nr_pages; | |
565 | ||
1da177e4 LT |
566 | /* |
567 | * vm_flags is protected by the mmap_sem held in write mode. | |
568 | * It's okay if try_to_unmap_one unmaps a page just after we | |
b291f000 | 569 | * set VM_LOCKED, __mlock_vma_pages_range will bring it back. |
1da177e4 | 570 | */ |
1da177e4 | 571 | |
fed067da | 572 | if (lock) |
408e82b7 | 573 | vma->vm_flags = newflags; |
fed067da | 574 | else |
408e82b7 | 575 | munlock_vma_pages_range(vma, start, end); |
1da177e4 | 576 | |
1da177e4 | 577 | out: |
b291f000 | 578 | *prev = vma; |
1da177e4 LT |
579 | return ret; |
580 | } | |
581 | ||
582 | static int do_mlock(unsigned long start, size_t len, int on) | |
583 | { | |
584 | unsigned long nstart, end, tmp; | |
585 | struct vm_area_struct * vma, * prev; | |
586 | int error; | |
587 | ||
fed067da ML |
588 | VM_BUG_ON(start & ~PAGE_MASK); |
589 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
1da177e4 LT |
590 | end = start + len; |
591 | if (end < start) | |
592 | return -EINVAL; | |
593 | if (end == start) | |
594 | return 0; | |
097d5910 | 595 | vma = find_vma(current->mm, start); |
1da177e4 LT |
596 | if (!vma || vma->vm_start > start) |
597 | return -ENOMEM; | |
598 | ||
097d5910 | 599 | prev = vma->vm_prev; |
1da177e4 LT |
600 | if (start > vma->vm_start) |
601 | prev = vma; | |
602 | ||
603 | for (nstart = start ; ; ) { | |
ca16d140 | 604 | vm_flags_t newflags; |
1da177e4 LT |
605 | |
606 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ | |
607 | ||
18693050 ML |
608 | newflags = vma->vm_flags & ~VM_LOCKED; |
609 | if (on) | |
09a9f1d2 | 610 | newflags |= VM_LOCKED; |
1da177e4 LT |
611 | |
612 | tmp = vma->vm_end; | |
613 | if (tmp > end) | |
614 | tmp = end; | |
615 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); | |
616 | if (error) | |
617 | break; | |
618 | nstart = tmp; | |
619 | if (nstart < prev->vm_end) | |
620 | nstart = prev->vm_end; | |
621 | if (nstart >= end) | |
622 | break; | |
623 | ||
624 | vma = prev->vm_next; | |
625 | if (!vma || vma->vm_start != nstart) { | |
626 | error = -ENOMEM; | |
627 | break; | |
628 | } | |
629 | } | |
630 | return error; | |
631 | } | |
632 | ||
bebeb3d6 ML |
633 | /* |
634 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
635 | * | |
636 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
637 | * flags. VMAs must be already marked with the desired vm_flags, and | |
638 | * mmap_sem must not be held. | |
639 | */ | |
640 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
fed067da ML |
641 | { |
642 | struct mm_struct *mm = current->mm; | |
643 | unsigned long end, nstart, nend; | |
644 | struct vm_area_struct *vma = NULL; | |
53a7706d | 645 | int locked = 0; |
28a35716 | 646 | long ret = 0; |
fed067da ML |
647 | |
648 | VM_BUG_ON(start & ~PAGE_MASK); | |
649 | VM_BUG_ON(len != PAGE_ALIGN(len)); | |
650 | end = start + len; | |
651 | ||
fed067da ML |
652 | for (nstart = start; nstart < end; nstart = nend) { |
653 | /* | |
654 | * We want to fault in pages for [nstart; end) address range. | |
655 | * Find first corresponding VMA. | |
656 | */ | |
53a7706d ML |
657 | if (!locked) { |
658 | locked = 1; | |
659 | down_read(&mm->mmap_sem); | |
fed067da | 660 | vma = find_vma(mm, nstart); |
53a7706d | 661 | } else if (nstart >= vma->vm_end) |
fed067da ML |
662 | vma = vma->vm_next; |
663 | if (!vma || vma->vm_start >= end) | |
664 | break; | |
665 | /* | |
666 | * Set [nstart; nend) to intersection of desired address | |
667 | * range with the first VMA. Also, skip undesirable VMA types. | |
668 | */ | |
669 | nend = min(end, vma->vm_end); | |
09a9f1d2 | 670 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
fed067da ML |
671 | continue; |
672 | if (nstart < vma->vm_start) | |
673 | nstart = vma->vm_start; | |
674 | /* | |
53a7706d ML |
675 | * Now fault in a range of pages. __mlock_vma_pages_range() |
676 | * double checks the vma flags, so that it won't mlock pages | |
677 | * if the vma was already munlocked. | |
fed067da | 678 | */ |
53a7706d ML |
679 | ret = __mlock_vma_pages_range(vma, nstart, nend, &locked); |
680 | if (ret < 0) { | |
681 | if (ignore_errors) { | |
682 | ret = 0; | |
683 | continue; /* continue at next VMA */ | |
684 | } | |
5fdb2002 ML |
685 | ret = __mlock_posix_error_return(ret); |
686 | break; | |
687 | } | |
53a7706d ML |
688 | nend = nstart + ret * PAGE_SIZE; |
689 | ret = 0; | |
fed067da | 690 | } |
53a7706d ML |
691 | if (locked) |
692 | up_read(&mm->mmap_sem); | |
fed067da ML |
693 | return ret; /* 0 or negative error code */ |
694 | } | |
695 | ||
6a6160a7 | 696 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
1da177e4 LT |
697 | { |
698 | unsigned long locked; | |
699 | unsigned long lock_limit; | |
700 | int error = -ENOMEM; | |
701 | ||
702 | if (!can_do_mlock()) | |
703 | return -EPERM; | |
704 | ||
8891d6da KM |
705 | lru_add_drain_all(); /* flush pagevec */ |
706 | ||
1da177e4 LT |
707 | down_write(¤t->mm->mmap_sem); |
708 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); | |
709 | start &= PAGE_MASK; | |
710 | ||
711 | locked = len >> PAGE_SHIFT; | |
712 | locked += current->mm->locked_vm; | |
713 | ||
59e99e5b | 714 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
715 | lock_limit >>= PAGE_SHIFT; |
716 | ||
717 | /* check against resource limits */ | |
718 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) | |
719 | error = do_mlock(start, len, 1); | |
720 | up_write(¤t->mm->mmap_sem); | |
fed067da | 721 | if (!error) |
bebeb3d6 | 722 | error = __mm_populate(start, len, 0); |
1da177e4 LT |
723 | return error; |
724 | } | |
725 | ||
6a6160a7 | 726 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
1da177e4 LT |
727 | { |
728 | int ret; | |
729 | ||
730 | down_write(¤t->mm->mmap_sem); | |
731 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); | |
732 | start &= PAGE_MASK; | |
733 | ret = do_mlock(start, len, 0); | |
734 | up_write(¤t->mm->mmap_sem); | |
735 | return ret; | |
736 | } | |
737 | ||
738 | static int do_mlockall(int flags) | |
739 | { | |
740 | struct vm_area_struct * vma, * prev = NULL; | |
1da177e4 LT |
741 | |
742 | if (flags & MCL_FUTURE) | |
09a9f1d2 | 743 | current->mm->def_flags |= VM_LOCKED; |
9977f0f1 | 744 | else |
09a9f1d2 | 745 | current->mm->def_flags &= ~VM_LOCKED; |
1da177e4 LT |
746 | if (flags == MCL_FUTURE) |
747 | goto out; | |
748 | ||
749 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { | |
ca16d140 | 750 | vm_flags_t newflags; |
1da177e4 | 751 | |
18693050 ML |
752 | newflags = vma->vm_flags & ~VM_LOCKED; |
753 | if (flags & MCL_CURRENT) | |
09a9f1d2 | 754 | newflags |= VM_LOCKED; |
1da177e4 LT |
755 | |
756 | /* Ignore errors */ | |
757 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); | |
22356f44 | 758 | cond_resched(); |
1da177e4 LT |
759 | } |
760 | out: | |
761 | return 0; | |
762 | } | |
763 | ||
3480b257 | 764 | SYSCALL_DEFINE1(mlockall, int, flags) |
1da177e4 LT |
765 | { |
766 | unsigned long lock_limit; | |
767 | int ret = -EINVAL; | |
768 | ||
769 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE))) | |
770 | goto out; | |
771 | ||
772 | ret = -EPERM; | |
773 | if (!can_do_mlock()) | |
774 | goto out; | |
775 | ||
df9d6985 CL |
776 | if (flags & MCL_CURRENT) |
777 | lru_add_drain_all(); /* flush pagevec */ | |
8891d6da | 778 | |
1da177e4 LT |
779 | down_write(¤t->mm->mmap_sem); |
780 | ||
59e99e5b | 781 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
1da177e4 LT |
782 | lock_limit >>= PAGE_SHIFT; |
783 | ||
784 | ret = -ENOMEM; | |
785 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || | |
786 | capable(CAP_IPC_LOCK)) | |
787 | ret = do_mlockall(flags); | |
788 | up_write(¤t->mm->mmap_sem); | |
bebeb3d6 ML |
789 | if (!ret && (flags & MCL_CURRENT)) |
790 | mm_populate(0, TASK_SIZE); | |
1da177e4 LT |
791 | out: |
792 | return ret; | |
793 | } | |
794 | ||
3480b257 | 795 | SYSCALL_DEFINE0(munlockall) |
1da177e4 LT |
796 | { |
797 | int ret; | |
798 | ||
799 | down_write(¤t->mm->mmap_sem); | |
800 | ret = do_mlockall(0); | |
801 | up_write(¤t->mm->mmap_sem); | |
802 | return ret; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB | |
807 | * shm segments) get accounted against the user_struct instead. | |
808 | */ | |
809 | static DEFINE_SPINLOCK(shmlock_user_lock); | |
810 | ||
811 | int user_shm_lock(size_t size, struct user_struct *user) | |
812 | { | |
813 | unsigned long lock_limit, locked; | |
814 | int allowed = 0; | |
815 | ||
816 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
59e99e5b | 817 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
5ed44a40 HB |
818 | if (lock_limit == RLIM_INFINITY) |
819 | allowed = 1; | |
1da177e4 LT |
820 | lock_limit >>= PAGE_SHIFT; |
821 | spin_lock(&shmlock_user_lock); | |
5ed44a40 HB |
822 | if (!allowed && |
823 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) | |
1da177e4 LT |
824 | goto out; |
825 | get_uid(user); | |
826 | user->locked_shm += locked; | |
827 | allowed = 1; | |
828 | out: | |
829 | spin_unlock(&shmlock_user_lock); | |
830 | return allowed; | |
831 | } | |
832 | ||
833 | void user_shm_unlock(size_t size, struct user_struct *user) | |
834 | { | |
835 | spin_lock(&shmlock_user_lock); | |
836 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
837 | spin_unlock(&shmlock_user_lock); | |
838 | free_uid(user); | |
839 | } |