Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/mlock.c |
| 3 | * |
| 4 | * (C) Copyright 1995 Linus Torvalds |
| 5 | * (C) Copyright 2002 Christoph Hellwig |
| 6 | */ |
| 7 | |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 8 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | #include <linux/mman.h> |
| 10 | #include <linux/mm.h> |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 11 | #include <linux/swap.h> |
| 12 | #include <linux/swapops.h> |
| 13 | #include <linux/pagemap.h> |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 14 | #include <linux/pagevec.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 15 | #include <linux/mempolicy.h> |
| 16 | #include <linux/syscalls.h> |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 17 | #include <linux/sched.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 18 | #include <linux/export.h> |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 19 | #include <linux/rmap.h> |
| 20 | #include <linux/mmzone.h> |
| 21 | #include <linux/hugetlb.h> |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 22 | #include <linux/memcontrol.h> |
| 23 | #include <linux/mm_inline.h> |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 24 | |
| 25 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | |
Wang Xiaoqiang | 7f43add | 2016-01-15 16:57:22 -0800 | [diff] [blame] | 27 | bool can_do_mlock(void) |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 28 | { |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 29 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
Wang Xiaoqiang | 7f43add | 2016-01-15 16:57:22 -0800 | [diff] [blame] | 30 | return true; |
Jeff Vander Stoep | a5a6579 | 2015-03-12 16:26:17 -0700 | [diff] [blame] | 31 | if (capable(CAP_IPC_LOCK)) |
Wang Xiaoqiang | 7f43add | 2016-01-15 16:57:22 -0800 | [diff] [blame] | 32 | return true; |
| 33 | return false; |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 34 | } |
| 35 | EXPORT_SYMBOL(can_do_mlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 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 | */ |
Hugh Dickins | e6c509f | 2012-10-08 16:33:19 -0700 | [diff] [blame] | 57 | void clear_page_mlock(struct page *page) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 58 | { |
Hugh Dickins | e6c509f | 2012-10-08 16:33:19 -0700 | [diff] [blame] | 59 | if (!TestClearPageMlocked(page)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 60 | return; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 61 | |
David Rientjes | 8449d21 | 2012-10-08 16:34:06 -0700 | [diff] [blame] | 62 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
| 63 | -hpage_nr_pages(page)); |
Nick Piggin | 5344b7e | 2008-10-18 20:26:51 -0700 | [diff] [blame] | 64 | count_vm_event(UNEVICTABLE_PGCLEARED); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 65 | if (!isolate_lru_page(page)) { |
| 66 | putback_lru_page(page); |
| 67 | } else { |
| 68 | /* |
KOSAKI Motohiro | 8891d6d | 2008-11-12 13:26:53 -0800 | [diff] [blame] | 69 | * We lost the race. the page already moved to evictable list. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 70 | */ |
KOSAKI Motohiro | 8891d6d | 2008-11-12 13:26:53 -0800 | [diff] [blame] | 71 | if (PageUnevictable(page)) |
Nick Piggin | 5344b7e | 2008-10-18 20:26:51 -0700 | [diff] [blame] | 72 | count_vm_event(UNEVICTABLE_PGSTRANDED); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 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 | { |
Vlastimil Babka | 57e68e9 | 2014-04-07 15:37:50 -0700 | [diff] [blame] | 82 | /* Serialize with page migration */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 83 | BUG_ON(!PageLocked(page)); |
| 84 | |
Kirill A. Shutemov | e90309c | 2016-01-15 16:54:33 -0800 | [diff] [blame] | 85 | VM_BUG_ON_PAGE(PageTail(page), page); |
| 86 | VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page); |
| 87 | |
Nick Piggin | 5344b7e | 2008-10-18 20:26:51 -0700 | [diff] [blame] | 88 | if (!TestSetPageMlocked(page)) { |
David Rientjes | 8449d21 | 2012-10-08 16:34:06 -0700 | [diff] [blame] | 89 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
| 90 | hpage_nr_pages(page)); |
Nick Piggin | 5344b7e | 2008-10-18 20:26:51 -0700 | [diff] [blame] | 91 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
| 92 | if (!isolate_lru_page(page)) |
| 93 | putback_lru_page(page); |
| 94 | } |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 95 | } |
| 96 | |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 97 | /* |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 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_zone(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 | /* |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 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)) |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 153 | __count_vm_event(UNEVICTABLE_PGSTRANDED); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 154 | else |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 155 | __count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 156 | } |
| 157 | |
Lee Schermerhorn | 6927c1d | 2009-12-14 17:59:55 -0800 | [diff] [blame] | 158 | /** |
| 159 | * munlock_vma_page - munlock a vma page |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 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) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 164 | * |
Lee Schermerhorn | 6927c1d | 2009-12-14 17:59:55 -0800 | [diff] [blame] | 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. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 175 | */ |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 176 | unsigned int munlock_vma_page(struct page *page) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 177 | { |
Kirill A. Shutemov | 7162a1e | 2016-01-21 16:40:27 -0800 | [diff] [blame] | 178 | int nr_pages; |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 179 | struct zone *zone = page_zone(page); |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 180 | |
Vlastimil Babka | 57e68e9 | 2014-04-07 15:37:50 -0700 | [diff] [blame] | 181 | /* For try_to_munlock() and to serialize with page migration */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 182 | BUG_ON(!PageLocked(page)); |
| 183 | |
Kirill A. Shutemov | e90309c | 2016-01-15 16:54:33 -0800 | [diff] [blame] | 184 | VM_BUG_ON_PAGE(PageTail(page), page); |
| 185 | |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 186 | /* |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 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(). |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 190 | */ |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 191 | spin_lock_irq(&zone->lru_lock); |
| 192 | |
| 193 | nr_pages = hpage_nr_pages(page); |
| 194 | if (!TestClearPageMlocked(page)) |
| 195 | goto unlock_out; |
| 196 | |
| 197 | __mod_zone_page_state(zone, NR_MLOCK, -nr_pages); |
| 198 | |
| 199 | if (__munlock_isolate_lru_page(page, true)) { |
| 200 | spin_unlock_irq(&zone->lru_lock); |
| 201 | __munlock_isolated_page(page); |
| 202 | goto out; |
| 203 | } |
| 204 | __munlock_isolation_failed(page); |
| 205 | |
| 206 | unlock_out: |
| 207 | spin_unlock_irq(&zone->lru_lock); |
| 208 | |
| 209 | out: |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 210 | return nr_pages - 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 211 | } |
| 212 | |
Lee Schermerhorn | 9978ad5 | 2008-10-18 20:26:56 -0700 | [diff] [blame] | 213 | /* |
| 214 | * convert get_user_pages() return value to posix mlock() error |
| 215 | */ |
| 216 | static int __mlock_posix_error_return(long retval) |
| 217 | { |
| 218 | if (retval == -EFAULT) |
| 219 | retval = -ENOMEM; |
| 220 | else if (retval == -ENOMEM) |
| 221 | retval = -EAGAIN; |
| 222 | return retval; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 223 | } |
| 224 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 225 | /* |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 226 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() |
| 227 | * |
| 228 | * The fast path is available only for evictable pages with single mapping. |
| 229 | * Then we can bypass the per-cpu pvec and get better performance. |
| 230 | * when mapcount > 1 we need try_to_munlock() which can fail. |
| 231 | * when !page_evictable(), we need the full redo logic of putback_lru_page to |
| 232 | * avoid leaving evictable page in unevictable list. |
| 233 | * |
| 234 | * In case of success, @page is added to @pvec and @pgrescued is incremented |
| 235 | * in case that the page was previously unevictable. @page is also unlocked. |
| 236 | */ |
| 237 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, |
| 238 | int *pgrescued) |
| 239 | { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 240 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 241 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 242 | |
| 243 | if (page_mapcount(page) <= 1 && page_evictable(page)) { |
| 244 | pagevec_add(pvec, page); |
| 245 | if (TestClearPageUnevictable(page)) |
| 246 | (*pgrescued)++; |
| 247 | unlock_page(page); |
| 248 | return true; |
| 249 | } |
| 250 | |
| 251 | return false; |
| 252 | } |
| 253 | |
| 254 | /* |
| 255 | * Putback multiple evictable pages to the LRU |
| 256 | * |
| 257 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of |
| 258 | * the pages might have meanwhile become unevictable but that is OK. |
| 259 | */ |
| 260 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) |
| 261 | { |
| 262 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); |
| 263 | /* |
| 264 | *__pagevec_lru_add() calls release_pages() so we don't call |
| 265 | * put_page() explicitly |
| 266 | */ |
| 267 | __pagevec_lru_add(pvec); |
| 268 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); |
| 269 | } |
| 270 | |
| 271 | /* |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 272 | * Munlock a batch of pages from the same zone |
| 273 | * |
| 274 | * The work is split to two main phases. First phase clears the Mlocked flag |
| 275 | * and attempts to isolate the pages, all under a single zone lru lock. |
| 276 | * The second phase finishes the munlock only for pages where isolation |
| 277 | * succeeded. |
| 278 | * |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 279 | * Note that the pagevec may be modified during the process. |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 280 | */ |
| 281 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) |
| 282 | { |
| 283 | int i; |
| 284 | int nr = pagevec_count(pvec); |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 285 | int delta_munlocked; |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 286 | struct pagevec pvec_putback; |
| 287 | int pgrescued = 0; |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 288 | |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 289 | pagevec_init(&pvec_putback, 0); |
| 290 | |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 291 | /* Phase 1: page isolation */ |
| 292 | spin_lock_irq(&zone->lru_lock); |
| 293 | for (i = 0; i < nr; i++) { |
| 294 | struct page *page = pvec->pages[i]; |
| 295 | |
| 296 | if (TestClearPageMlocked(page)) { |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 297 | /* |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 298 | * We already have pin from follow_page_mask() |
| 299 | * so we can spare the get_page() here. |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 300 | */ |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 301 | if (__munlock_isolate_lru_page(page, false)) |
| 302 | continue; |
| 303 | else |
| 304 | __munlock_isolation_failed(page); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 305 | } |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 306 | |
| 307 | /* |
| 308 | * We won't be munlocking this page in the next phase |
| 309 | * but we still need to release the follow_page_mask() |
| 310 | * pin. We cannot do it under lru_lock however. If it's |
| 311 | * the last pin, __page_cache_release() would deadlock. |
| 312 | */ |
| 313 | pagevec_add(&pvec_putback, pvec->pages[i]); |
| 314 | pvec->pages[i] = NULL; |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 315 | } |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 316 | delta_munlocked = -nr + pagevec_count(&pvec_putback); |
Vlastimil Babka | 1ebb7cc | 2013-09-11 14:22:30 -0700 | [diff] [blame] | 317 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 318 | spin_unlock_irq(&zone->lru_lock); |
| 319 | |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 320 | /* Now we can release pins of pages that we are not munlocking */ |
| 321 | pagevec_release(&pvec_putback); |
| 322 | |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 323 | /* Phase 2: page munlock */ |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 324 | for (i = 0; i < nr; i++) { |
| 325 | struct page *page = pvec->pages[i]; |
| 326 | |
| 327 | if (page) { |
| 328 | lock_page(page); |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 329 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
| 330 | &pgrescued)) { |
Vlastimil Babka | 5b40998 | 2013-09-11 14:22:33 -0700 | [diff] [blame] | 331 | /* |
| 332 | * Slow path. We don't want to lose the last |
| 333 | * pin before unlock_page() |
| 334 | */ |
| 335 | get_page(page); /* for putback_lru_page() */ |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 336 | __munlock_isolated_page(page); |
| 337 | unlock_page(page); |
Vlastimil Babka | 5b40998 | 2013-09-11 14:22:33 -0700 | [diff] [blame] | 338 | put_page(page); /* from follow_page_mask() */ |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 339 | } |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 340 | } |
| 341 | } |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 342 | |
Vlastimil Babka | 5b40998 | 2013-09-11 14:22:33 -0700 | [diff] [blame] | 343 | /* |
| 344 | * Phase 3: page putback for pages that qualified for the fast path |
| 345 | * This will also call put_page() to return pin from follow_page_mask() |
| 346 | */ |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 347 | if (pagevec_count(&pvec_putback)) |
| 348 | __putback_lru_fast(&pvec_putback, pgrescued); |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 349 | } |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 350 | |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 351 | /* |
| 352 | * Fill up pagevec for __munlock_pagevec using pte walk |
| 353 | * |
| 354 | * The function expects that the struct page corresponding to @start address is |
| 355 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. |
| 356 | * |
| 357 | * The rest of @pvec is filled by subsequent pages within the same pmd and same |
| 358 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These |
| 359 | * pages also get pinned. |
| 360 | * |
| 361 | * Returns the address of the next page that should be scanned. This equals |
| 362 | * @start + PAGE_SIZE when no page could be added by the pte walk. |
| 363 | */ |
| 364 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, |
| 365 | struct vm_area_struct *vma, int zoneid, unsigned long start, |
| 366 | unsigned long end) |
| 367 | { |
| 368 | pte_t *pte; |
| 369 | spinlock_t *ptl; |
| 370 | |
| 371 | /* |
| 372 | * Initialize pte walk starting at the already pinned page where we |
Vlastimil Babka | eadb41a | 2013-09-30 13:45:18 -0700 | [diff] [blame] | 373 | * are sure that there is a pte, as it was pinned under the same |
| 374 | * mmap_sem write op. |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 375 | */ |
| 376 | pte = get_locked_pte(vma->vm_mm, start, &ptl); |
Vlastimil Babka | eadb41a | 2013-09-30 13:45:18 -0700 | [diff] [blame] | 377 | /* Make sure we do not cross the page table boundary */ |
| 378 | end = pgd_addr_end(start, end); |
| 379 | end = pud_addr_end(start, end); |
| 380 | end = pmd_addr_end(start, end); |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 381 | |
| 382 | /* The page next to the pinned page is the first we will try to get */ |
| 383 | start += PAGE_SIZE; |
| 384 | while (start < end) { |
| 385 | struct page *page = NULL; |
| 386 | pte++; |
| 387 | if (pte_present(*pte)) |
| 388 | page = vm_normal_page(vma, start, *pte); |
| 389 | /* |
| 390 | * Break if page could not be obtained or the page's node+zone does not |
| 391 | * match |
| 392 | */ |
| 393 | if (!page || page_zone_id(page) != zoneid) |
| 394 | break; |
| 395 | |
Kirill A. Shutemov | e90309c | 2016-01-15 16:54:33 -0800 | [diff] [blame] | 396 | /* |
| 397 | * Do not use pagevec for PTE-mapped THP, |
| 398 | * munlock_vma_pages_range() will handle them. |
| 399 | */ |
| 400 | if (PageTransCompound(page)) |
| 401 | break; |
| 402 | |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 403 | get_page(page); |
| 404 | /* |
| 405 | * Increase the address that will be returned *before* the |
| 406 | * eventual break due to pvec becoming full by adding the page |
| 407 | */ |
| 408 | start += PAGE_SIZE; |
| 409 | if (pagevec_add(pvec, page) == 0) |
| 410 | break; |
| 411 | } |
| 412 | pte_unmap_unlock(pte, ptl); |
| 413 | return start; |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 414 | } |
| 415 | |
| 416 | /* |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 417 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
| 418 | * @vma - vma containing range to be munlock()ed. |
| 419 | * @start - start address in @vma of the range |
| 420 | * @end - end of range in @vma. |
| 421 | * |
| 422 | * For mremap(), munmap() and exit(). |
| 423 | * |
| 424 | * Called with @vma VM_LOCKED. |
| 425 | * |
| 426 | * Returns with VM_LOCKED cleared. Callers must be prepared to |
| 427 | * deal with this. |
| 428 | * |
| 429 | * We don't save and restore VM_LOCKED here because pages are |
| 430 | * still on lru. In unmap path, pages might be scanned by reclaim |
| 431 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and |
| 432 | * free them. This will result in freeing mlocked pages. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 433 | */ |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 434 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 435 | unsigned long start, unsigned long end) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 436 | { |
Eric B Munson | de60f5f | 2015-11-05 18:51:36 -0800 | [diff] [blame] | 437 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 438 | |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 439 | while (start < end) { |
Alexey Klimov | ab7a5af | 2016-01-14 15:18:24 -0800 | [diff] [blame] | 440 | struct page *page; |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 441 | unsigned int page_mask; |
| 442 | unsigned long page_increm; |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 443 | struct pagevec pvec; |
| 444 | struct zone *zone; |
| 445 | int zoneid; |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 446 | |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 447 | pagevec_init(&pvec, 0); |
Hugh Dickins | 6e91971 | 2009-09-21 17:03:32 -0700 | [diff] [blame] | 448 | /* |
| 449 | * Although FOLL_DUMP is intended for get_dump_page(), |
| 450 | * it just so happens that its special treatment of the |
| 451 | * ZERO_PAGE (returning an error instead of doing get_page) |
| 452 | * suits munlock very well (and if somehow an abnormal page |
| 453 | * has sneaked into the range, we won't oops here: great). |
| 454 | */ |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 455 | page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP, |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 456 | &page_mask); |
| 457 | |
Kirill A. Shutemov | e90309c | 2016-01-15 16:54:33 -0800 | [diff] [blame] | 458 | if (page && !IS_ERR(page)) { |
| 459 | if (PageTransTail(page)) { |
| 460 | VM_BUG_ON_PAGE(PageMlocked(page), page); |
| 461 | put_page(page); /* follow_page_mask() */ |
| 462 | } else if (PageTransHuge(page)) { |
| 463 | lock_page(page); |
| 464 | /* |
| 465 | * Any THP page found by follow_page_mask() may |
| 466 | * have gotten split before reaching |
| 467 | * munlock_vma_page(), so we need to recompute |
| 468 | * the page_mask here. |
| 469 | */ |
| 470 | page_mask = munlock_vma_page(page); |
| 471 | unlock_page(page); |
| 472 | put_page(page); /* follow_page_mask() */ |
| 473 | } else { |
| 474 | /* |
| 475 | * Non-huge pages are handled in batches via |
| 476 | * pagevec. The pin from follow_page_mask() |
| 477 | * prevents them from collapsing by THP. |
| 478 | */ |
| 479 | pagevec_add(&pvec, page); |
| 480 | zone = page_zone(page); |
| 481 | zoneid = page_zone_id(page); |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 482 | |
Kirill A. Shutemov | e90309c | 2016-01-15 16:54:33 -0800 | [diff] [blame] | 483 | /* |
| 484 | * Try to fill the rest of pagevec using fast |
| 485 | * pte walk. This will also update start to |
| 486 | * the next page to process. Then munlock the |
| 487 | * pagevec. |
| 488 | */ |
| 489 | start = __munlock_pagevec_fill(&pvec, vma, |
| 490 | zoneid, start, end); |
| 491 | __munlock_pagevec(&pvec, zone); |
| 492 | goto next; |
| 493 | } |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 494 | } |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 495 | page_increm = 1 + page_mask; |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 496 | start += page_increm * PAGE_SIZE; |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 497 | next: |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 498 | cond_resched(); |
| 499 | } |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 500 | } |
| 501 | |
| 502 | /* |
| 503 | * mlock_fixup - handle mlock[all]/munlock[all] requests. |
| 504 | * |
| 505 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and |
| 506 | * munlock is a no-op. However, for some special vmas, we go ahead and |
Michel Lespinasse | cea10a1 | 2013-02-22 16:32:44 -0800 | [diff] [blame] | 507 | * populate the ptes. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 508 | * |
| 509 | * For vmas that pass the filters, merge/split as appropriate. |
| 510 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 512 | unsigned long start, unsigned long end, vm_flags_t newflags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 514 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 515 | pgoff_t pgoff; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 516 | int nr_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | int ret = 0; |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 518 | int lock = !!(newflags & VM_LOCKED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 520 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
Stephen Wilson | 31db58b | 2011-03-13 15:49:15 -0400 | [diff] [blame] | 521 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 522 | /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */ |
| 523 | goto out; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 524 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 525 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
| 526 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 527 | vma->vm_file, pgoff, vma_policy(vma), |
| 528 | vma->vm_userfaultfd_ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | if (*prev) { |
| 530 | vma = *prev; |
| 531 | goto success; |
| 532 | } |
| 533 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | if (start != vma->vm_start) { |
| 535 | ret = split_vma(mm, vma, start, 1); |
| 536 | if (ret) |
| 537 | goto out; |
| 538 | } |
| 539 | |
| 540 | if (end != vma->vm_end) { |
| 541 | ret = split_vma(mm, vma, end, 0); |
| 542 | if (ret) |
| 543 | goto out; |
| 544 | } |
| 545 | |
| 546 | success: |
| 547 | /* |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 548 | * Keep track of amount of locked VM. |
| 549 | */ |
| 550 | nr_pages = (end - start) >> PAGE_SHIFT; |
| 551 | if (!lock) |
| 552 | nr_pages = -nr_pages; |
| 553 | mm->locked_vm += nr_pages; |
| 554 | |
| 555 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | * vm_flags is protected by the mmap_sem held in write mode. |
| 557 | * It's okay if try_to_unmap_one unmaps a page just after we |
Kirill A. Shutemov | fc05f56 | 2015-04-14 15:44:39 -0700 | [diff] [blame] | 558 | * set VM_LOCKED, populate_vma_page_range will bring it back. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 561 | if (lock) |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 562 | vma->vm_flags = newflags; |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 563 | else |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 564 | munlock_vma_pages_range(vma, start, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | out: |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 567 | *prev = vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | return ret; |
| 569 | } |
| 570 | |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 571 | static int apply_vma_lock_flags(unsigned long start, size_t len, |
| 572 | vm_flags_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | { |
| 574 | unsigned long nstart, end, tmp; |
| 575 | struct vm_area_struct * vma, * prev; |
| 576 | int error; |
| 577 | |
Alexander Kuleshov | 8fd9e48 | 2015-11-05 18:46:49 -0800 | [diff] [blame] | 578 | VM_BUG_ON(offset_in_page(start)); |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 579 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 580 | end = start + len; |
| 581 | if (end < start) |
| 582 | return -EINVAL; |
| 583 | if (end == start) |
| 584 | return 0; |
Linus Torvalds | 097d591 | 2012-03-06 18:23:36 -0800 | [diff] [blame] | 585 | vma = find_vma(current->mm, start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 586 | if (!vma || vma->vm_start > start) |
| 587 | return -ENOMEM; |
| 588 | |
Linus Torvalds | 097d591 | 2012-03-06 18:23:36 -0800 | [diff] [blame] | 589 | prev = vma->vm_prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 590 | if (start > vma->vm_start) |
| 591 | prev = vma; |
| 592 | |
| 593 | for (nstart = start ; ; ) { |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 594 | vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 595 | |
| 596 | newflags |= flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 597 | |
| 598 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 599 | tmp = vma->vm_end; |
| 600 | if (tmp > end) |
| 601 | tmp = end; |
| 602 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); |
| 603 | if (error) |
| 604 | break; |
| 605 | nstart = tmp; |
| 606 | if (nstart < prev->vm_end) |
| 607 | nstart = prev->vm_end; |
| 608 | if (nstart >= end) |
| 609 | break; |
| 610 | |
| 611 | vma = prev->vm_next; |
| 612 | if (!vma || vma->vm_start != nstart) { |
| 613 | error = -ENOMEM; |
| 614 | break; |
| 615 | } |
| 616 | } |
| 617 | return error; |
| 618 | } |
| 619 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 620 | static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 621 | { |
| 622 | unsigned long locked; |
| 623 | unsigned long lock_limit; |
| 624 | int error = -ENOMEM; |
| 625 | |
| 626 | if (!can_do_mlock()) |
| 627 | return -EPERM; |
| 628 | |
KOSAKI Motohiro | 8891d6d | 2008-11-12 13:26:53 -0800 | [diff] [blame] | 629 | lru_add_drain_all(); /* flush pagevec */ |
| 630 | |
Alexander Kuleshov | 8fd9e48 | 2015-11-05 18:46:49 -0800 | [diff] [blame] | 631 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | start &= PAGE_MASK; |
| 633 | |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 634 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | lock_limit >>= PAGE_SHIFT; |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 636 | locked = len >> PAGE_SHIFT; |
| 637 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 638 | if (down_write_killable(¤t->mm->mmap_sem)) |
| 639 | return -EINTR; |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 640 | |
| 641 | locked += current->mm->locked_vm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | |
| 643 | /* check against resource limits */ |
| 644 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 645 | error = apply_vma_lock_flags(start, len, flags); |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 646 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | up_write(¤t->mm->mmap_sem); |
Kirill A. Shutemov | c561259 | 2015-04-14 15:44:42 -0700 | [diff] [blame] | 648 | if (error) |
| 649 | return error; |
| 650 | |
| 651 | error = __mm_populate(start, len, 0); |
| 652 | if (error) |
| 653 | return __mlock_posix_error_return(error); |
| 654 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | } |
| 656 | |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 657 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
| 658 | { |
| 659 | return do_mlock(start, len, VM_LOCKED); |
| 660 | } |
| 661 | |
Eric B Munson | a8ca5d0 | 2015-11-05 18:51:33 -0800 | [diff] [blame] | 662 | SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags) |
| 663 | { |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 664 | vm_flags_t vm_flags = VM_LOCKED; |
| 665 | |
| 666 | if (flags & ~MLOCK_ONFAULT) |
Eric B Munson | a8ca5d0 | 2015-11-05 18:51:33 -0800 | [diff] [blame] | 667 | return -EINVAL; |
| 668 | |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 669 | if (flags & MLOCK_ONFAULT) |
| 670 | vm_flags |= VM_LOCKONFAULT; |
| 671 | |
| 672 | return do_mlock(start, len, vm_flags); |
Eric B Munson | a8ca5d0 | 2015-11-05 18:51:33 -0800 | [diff] [blame] | 673 | } |
| 674 | |
Heiko Carstens | 6a6160a | 2009-01-14 14:14:15 +0100 | [diff] [blame] | 675 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | { |
| 677 | int ret; |
| 678 | |
Alexander Kuleshov | 8fd9e48 | 2015-11-05 18:46:49 -0800 | [diff] [blame] | 679 | len = PAGE_ALIGN(len + (offset_in_page(start))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | start &= PAGE_MASK; |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 681 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 682 | if (down_write_killable(¤t->mm->mmap_sem)) |
| 683 | return -EINTR; |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 684 | ret = apply_vma_lock_flags(start, len, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | up_write(¤t->mm->mmap_sem); |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 686 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | return ret; |
| 688 | } |
| 689 | |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 690 | /* |
| 691 | * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall) |
| 692 | * and translate into the appropriate modifications to mm->def_flags and/or the |
| 693 | * flags for all current VMAs. |
| 694 | * |
| 695 | * There are a couple of subtleties with this. If mlockall() is called multiple |
| 696 | * times with different flags, the values do not necessarily stack. If mlockall |
| 697 | * is called once including the MCL_FUTURE flag and then a second time without |
| 698 | * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags. |
| 699 | */ |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 700 | static int apply_mlockall_flags(int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 701 | { |
| 702 | struct vm_area_struct * vma, * prev = NULL; |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 703 | vm_flags_t to_add = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 705 | current->mm->def_flags &= VM_LOCKED_CLEAR_MASK; |
| 706 | if (flags & MCL_FUTURE) { |
Michel Lespinasse | 09a9f1d | 2013-03-28 16:26:23 -0700 | [diff] [blame] | 707 | current->mm->def_flags |= VM_LOCKED; |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 708 | |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 709 | if (flags & MCL_ONFAULT) |
| 710 | current->mm->def_flags |= VM_LOCKONFAULT; |
| 711 | |
| 712 | if (!(flags & MCL_CURRENT)) |
| 713 | goto out; |
| 714 | } |
| 715 | |
| 716 | if (flags & MCL_CURRENT) { |
| 717 | to_add |= VM_LOCKED; |
| 718 | if (flags & MCL_ONFAULT) |
| 719 | to_add |= VM_LOCKONFAULT; |
| 720 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | |
| 722 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 723 | vm_flags_t newflags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 725 | newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; |
| 726 | newflags |= to_add; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | |
| 728 | /* Ignore errors */ |
| 729 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); |
Paul E. McKenney | bde6c3a | 2014-07-01 11:26:57 -0700 | [diff] [blame] | 730 | cond_resched_rcu_qs(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | } |
| 732 | out: |
| 733 | return 0; |
| 734 | } |
| 735 | |
Heiko Carstens | 3480b25 | 2009-01-14 14:14:16 +0100 | [diff] [blame] | 736 | SYSCALL_DEFINE1(mlockall, int, flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | { |
| 738 | unsigned long lock_limit; |
Alexey Klimov | 86d2adc | 2015-11-05 18:46:00 -0800 | [diff] [blame] | 739 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | |
Eric B Munson | b0f205c | 2015-11-05 18:51:39 -0800 | [diff] [blame] | 741 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT))) |
Alexey Klimov | 86d2adc | 2015-11-05 18:46:00 -0800 | [diff] [blame] | 742 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | if (!can_do_mlock()) |
Alexey Klimov | 86d2adc | 2015-11-05 18:46:00 -0800 | [diff] [blame] | 745 | return -EPERM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 746 | |
Christoph Lameter | df9d698 | 2011-10-31 17:09:35 -0700 | [diff] [blame] | 747 | if (flags & MCL_CURRENT) |
| 748 | lru_add_drain_all(); /* flush pagevec */ |
KOSAKI Motohiro | 8891d6d | 2008-11-12 13:26:53 -0800 | [diff] [blame] | 749 | |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 750 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | lock_limit >>= PAGE_SHIFT; |
| 752 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 753 | if (down_write_killable(¤t->mm->mmap_sem)) |
| 754 | return -EINTR; |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 755 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 756 | ret = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
| 758 | capable(CAP_IPC_LOCK)) |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 759 | ret = apply_mlockall_flags(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | up_write(¤t->mm->mmap_sem); |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 761 | if (!ret && (flags & MCL_CURRENT)) |
| 762 | mm_populate(0, TASK_SIZE); |
Alexey Klimov | 86d2adc | 2015-11-05 18:46:00 -0800 | [diff] [blame] | 763 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 764 | return ret; |
| 765 | } |
| 766 | |
Heiko Carstens | 3480b25 | 2009-01-14 14:14:16 +0100 | [diff] [blame] | 767 | SYSCALL_DEFINE0(munlockall) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | { |
| 769 | int ret; |
| 770 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 771 | if (down_write_killable(¤t->mm->mmap_sem)) |
| 772 | return -EINTR; |
Eric B Munson | 1aab92e | 2015-11-05 18:51:29 -0800 | [diff] [blame] | 773 | ret = apply_mlockall_flags(0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | up_write(¤t->mm->mmap_sem); |
| 775 | return ret; |
| 776 | } |
| 777 | |
| 778 | /* |
| 779 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB |
| 780 | * shm segments) get accounted against the user_struct instead. |
| 781 | */ |
| 782 | static DEFINE_SPINLOCK(shmlock_user_lock); |
| 783 | |
| 784 | int user_shm_lock(size_t size, struct user_struct *user) |
| 785 | { |
| 786 | unsigned long lock_limit, locked; |
| 787 | int allowed = 0; |
| 788 | |
| 789 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 790 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
Herbert van den Bergh | 5ed44a4 | 2007-07-15 23:38:25 -0700 | [diff] [blame] | 791 | if (lock_limit == RLIM_INFINITY) |
| 792 | allowed = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | lock_limit >>= PAGE_SHIFT; |
| 794 | spin_lock(&shmlock_user_lock); |
Herbert van den Bergh | 5ed44a4 | 2007-07-15 23:38:25 -0700 | [diff] [blame] | 795 | if (!allowed && |
| 796 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | goto out; |
| 798 | get_uid(user); |
| 799 | user->locked_shm += locked; |
| 800 | allowed = 1; |
| 801 | out: |
| 802 | spin_unlock(&shmlock_user_lock); |
| 803 | return allowed; |
| 804 | } |
| 805 | |
| 806 | void user_shm_unlock(size_t size, struct user_struct *user) |
| 807 | { |
| 808 | spin_lock(&shmlock_user_lock); |
| 809 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 810 | spin_unlock(&shmlock_user_lock); |
| 811 | free_uid(user); |
| 812 | } |