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 | |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 27 | int can_do_mlock(void) |
| 28 | { |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 29 | if (rlimit(RLIMIT_MEMLOCK) != 0) |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 30 | return 1; |
Jeff Vander Stoep | a5a6579 | 2015-03-12 16:26:17 -0700 | [diff] [blame] | 31 | if (capable(CAP_IPC_LOCK)) |
| 32 | return 1; |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 33 | return 0; |
| 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 | |
Nick Piggin | 5344b7e | 2008-10-18 20:26:51 -0700 | [diff] [blame] | 85 | if (!TestSetPageMlocked(page)) { |
David Rientjes | 8449d21 | 2012-10-08 16:34:06 -0700 | [diff] [blame] | 86 | mod_zone_page_state(page_zone(page), NR_MLOCK, |
| 87 | hpage_nr_pages(page)); |
Nick Piggin | 5344b7e | 2008-10-18 20:26:51 -0700 | [diff] [blame] | 88 | count_vm_event(UNEVICTABLE_PGMLOCKED); |
| 89 | if (!isolate_lru_page(page)) |
| 90 | putback_lru_page(page); |
| 91 | } |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 92 | } |
| 93 | |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 94 | /* |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 95 | * Isolate a page from LRU with optional get_page() pin. |
| 96 | * Assumes lru_lock already held and page already pinned. |
| 97 | */ |
| 98 | static bool __munlock_isolate_lru_page(struct page *page, bool getpage) |
| 99 | { |
| 100 | if (PageLRU(page)) { |
| 101 | struct lruvec *lruvec; |
| 102 | |
| 103 | lruvec = mem_cgroup_page_lruvec(page, page_zone(page)); |
| 104 | if (getpage) |
| 105 | get_page(page); |
| 106 | ClearPageLRU(page); |
| 107 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 108 | return true; |
| 109 | } |
| 110 | |
| 111 | return false; |
| 112 | } |
| 113 | |
| 114 | /* |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 115 | * Finish munlock after successful page isolation |
| 116 | * |
| 117 | * Page must be locked. This is a wrapper for try_to_munlock() |
| 118 | * and putback_lru_page() with munlock accounting. |
| 119 | */ |
| 120 | static void __munlock_isolated_page(struct page *page) |
| 121 | { |
| 122 | int ret = SWAP_AGAIN; |
| 123 | |
| 124 | /* |
| 125 | * Optimization: if the page was mapped just once, that's our mapping |
| 126 | * and we don't need to check all the other vmas. |
| 127 | */ |
| 128 | if (page_mapcount(page) > 1) |
| 129 | ret = try_to_munlock(page); |
| 130 | |
| 131 | /* Did try_to_unlock() succeed or punt? */ |
| 132 | if (ret != SWAP_MLOCK) |
| 133 | count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
| 134 | |
| 135 | putback_lru_page(page); |
| 136 | } |
| 137 | |
| 138 | /* |
| 139 | * Accounting for page isolation fail during munlock |
| 140 | * |
| 141 | * Performs accounting when page isolation fails in munlock. There is nothing |
| 142 | * else to do because it means some other task has already removed the page |
| 143 | * from the LRU. putback_lru_page() will take care of removing the page from |
| 144 | * the unevictable list, if necessary. vmscan [page_referenced()] will move |
| 145 | * the page back to the unevictable list if some other vma has it mlocked. |
| 146 | */ |
| 147 | static void __munlock_isolation_failed(struct page *page) |
| 148 | { |
| 149 | if (PageUnevictable(page)) |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 150 | __count_vm_event(UNEVICTABLE_PGSTRANDED); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 151 | else |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 152 | __count_vm_event(UNEVICTABLE_PGMUNLOCKED); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 153 | } |
| 154 | |
Lee Schermerhorn | 6927c1d | 2009-12-14 17:59:55 -0800 | [diff] [blame] | 155 | /** |
| 156 | * munlock_vma_page - munlock a vma page |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 157 | * @page - page to be unlocked, either a normal page or THP page head |
| 158 | * |
| 159 | * returns the size of the page as a page mask (0 for normal page, |
| 160 | * HPAGE_PMD_NR - 1 for THP head page) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 161 | * |
Lee Schermerhorn | 6927c1d | 2009-12-14 17:59:55 -0800 | [diff] [blame] | 162 | * called from munlock()/munmap() path with page supposedly on the LRU. |
| 163 | * When we munlock a page, because the vma where we found the page is being |
| 164 | * munlock()ed or munmap()ed, we want to check whether other vmas hold the |
| 165 | * page locked so that we can leave it on the unevictable lru list and not |
| 166 | * bother vmscan with it. However, to walk the page's rmap list in |
| 167 | * try_to_munlock() we must isolate the page from the LRU. If some other |
| 168 | * task has removed the page from the LRU, we won't be able to do that. |
| 169 | * So we clear the PageMlocked as we might not get another chance. If we |
| 170 | * can't isolate the page, we leave it for putback_lru_page() and vmscan |
| 171 | * [page_referenced()/try_to_unmap()] to deal with. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 172 | */ |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 173 | unsigned int munlock_vma_page(struct page *page) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 174 | { |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 175 | unsigned int nr_pages; |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 176 | struct zone *zone = page_zone(page); |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 177 | |
Vlastimil Babka | 57e68e9 | 2014-04-07 15:37:50 -0700 | [diff] [blame] | 178 | /* For try_to_munlock() and to serialize with page migration */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 179 | BUG_ON(!PageLocked(page)); |
| 180 | |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 181 | /* |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 182 | * Serialize with any parallel __split_huge_page_refcount() which |
| 183 | * might otherwise copy PageMlocked to part of the tail pages before |
| 184 | * 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] | 185 | */ |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 186 | spin_lock_irq(&zone->lru_lock); |
| 187 | |
| 188 | nr_pages = hpage_nr_pages(page); |
| 189 | if (!TestClearPageMlocked(page)) |
| 190 | goto unlock_out; |
| 191 | |
| 192 | __mod_zone_page_state(zone, NR_MLOCK, -nr_pages); |
| 193 | |
| 194 | if (__munlock_isolate_lru_page(page, true)) { |
| 195 | spin_unlock_irq(&zone->lru_lock); |
| 196 | __munlock_isolated_page(page); |
| 197 | goto out; |
| 198 | } |
| 199 | __munlock_isolation_failed(page); |
| 200 | |
| 201 | unlock_out: |
| 202 | spin_unlock_irq(&zone->lru_lock); |
| 203 | |
| 204 | out: |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 205 | return nr_pages - 1; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 206 | } |
| 207 | |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 208 | /** |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 209 | * __mlock_vma_pages_range() - mlock a range of pages in the vma. |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 210 | * @vma: target vma |
| 211 | * @start: start address |
| 212 | * @end: end address |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 213 | * @nonblocking: |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 214 | * |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 215 | * This takes care of making the pages present too. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 216 | * |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 217 | * return 0 on success, negative error code on error. |
| 218 | * |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 219 | * vma->vm_mm->mmap_sem must be held. |
| 220 | * |
| 221 | * If @nonblocking is NULL, it may be held for read or write and will |
| 222 | * be unperturbed. |
| 223 | * |
| 224 | * If @nonblocking is non-NULL, it must held for read only and may be |
| 225 | * released. If it's released, *@nonblocking will be set to 0. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 226 | */ |
Michel Lespinasse | cea10a1 | 2013-02-22 16:32:44 -0800 | [diff] [blame] | 227 | long __mlock_vma_pages_range(struct vm_area_struct *vma, |
| 228 | unsigned long start, unsigned long end, int *nonblocking) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 229 | { |
| 230 | struct mm_struct *mm = vma->vm_mm; |
Michel Lespinasse | 28a3571 | 2013-02-22 16:35:55 -0800 | [diff] [blame] | 231 | unsigned long nr_pages = (end - start) / PAGE_SIZE; |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 232 | int gup_flags; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 233 | |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 234 | VM_BUG_ON(start & ~PAGE_MASK); |
| 235 | VM_BUG_ON(end & ~PAGE_MASK); |
Sasha Levin | 81d1b09 | 2014-10-09 15:28:10 -0700 | [diff] [blame] | 236 | VM_BUG_ON_VMA(start < vma->vm_start, vma); |
| 237 | VM_BUG_ON_VMA(end > vma->vm_end, vma); |
Sasha Levin | 96dad67 | 2014-10-09 15:28:39 -0700 | [diff] [blame] | 238 | VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm); |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 239 | |
Linus Torvalds | a1fde08 | 2011-05-04 21:30:28 -0700 | [diff] [blame] | 240 | gup_flags = FOLL_TOUCH | FOLL_MLOCK; |
Michel Lespinasse | 5ecfda0 | 2011-01-13 15:46:09 -0800 | [diff] [blame] | 241 | /* |
| 242 | * We want to touch writable mappings with a write fault in order |
| 243 | * to break COW, except for shared mappings because these don't COW |
| 244 | * and we would not want to dirty them for nothing. |
| 245 | */ |
| 246 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) |
Hugh Dickins | 58fa879 | 2009-09-21 17:03:31 -0700 | [diff] [blame] | 247 | gup_flags |= FOLL_WRITE; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 248 | |
Michel Lespinasse | fdf4c58 | 2011-01-31 17:03:41 -0800 | [diff] [blame] | 249 | /* |
| 250 | * We want mlock to succeed for regions that have any permissions |
| 251 | * other than PROT_NONE. |
| 252 | */ |
| 253 | if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)) |
| 254 | gup_flags |= FOLL_FORCE; |
| 255 | |
Johannes Weiner | 4805b02 | 2013-02-22 16:35:20 -0800 | [diff] [blame] | 256 | /* |
| 257 | * We made sure addr is within a VMA, so the following will |
| 258 | * not result in a stack expansion that recurses back here. |
| 259 | */ |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 260 | return __get_user_pages(current, mm, start, nr_pages, gup_flags, |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 261 | NULL, NULL, nonblocking); |
Lee Schermerhorn | 9978ad5 | 2008-10-18 20:26:56 -0700 | [diff] [blame] | 262 | } |
| 263 | |
| 264 | /* |
| 265 | * convert get_user_pages() return value to posix mlock() error |
| 266 | */ |
| 267 | static int __mlock_posix_error_return(long retval) |
| 268 | { |
| 269 | if (retval == -EFAULT) |
| 270 | retval = -ENOMEM; |
| 271 | else if (retval == -ENOMEM) |
| 272 | retval = -EAGAIN; |
| 273 | return retval; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 274 | } |
| 275 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 276 | /* |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 277 | * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec() |
| 278 | * |
| 279 | * The fast path is available only for evictable pages with single mapping. |
| 280 | * Then we can bypass the per-cpu pvec and get better performance. |
| 281 | * when mapcount > 1 we need try_to_munlock() which can fail. |
| 282 | * when !page_evictable(), we need the full redo logic of putback_lru_page to |
| 283 | * avoid leaving evictable page in unevictable list. |
| 284 | * |
| 285 | * In case of success, @page is added to @pvec and @pgrescued is incremented |
| 286 | * in case that the page was previously unevictable. @page is also unlocked. |
| 287 | */ |
| 288 | static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec, |
| 289 | int *pgrescued) |
| 290 | { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 291 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 292 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 293 | |
| 294 | if (page_mapcount(page) <= 1 && page_evictable(page)) { |
| 295 | pagevec_add(pvec, page); |
| 296 | if (TestClearPageUnevictable(page)) |
| 297 | (*pgrescued)++; |
| 298 | unlock_page(page); |
| 299 | return true; |
| 300 | } |
| 301 | |
| 302 | return false; |
| 303 | } |
| 304 | |
| 305 | /* |
| 306 | * Putback multiple evictable pages to the LRU |
| 307 | * |
| 308 | * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of |
| 309 | * the pages might have meanwhile become unevictable but that is OK. |
| 310 | */ |
| 311 | static void __putback_lru_fast(struct pagevec *pvec, int pgrescued) |
| 312 | { |
| 313 | count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec)); |
| 314 | /* |
| 315 | *__pagevec_lru_add() calls release_pages() so we don't call |
| 316 | * put_page() explicitly |
| 317 | */ |
| 318 | __pagevec_lru_add(pvec); |
| 319 | count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); |
| 320 | } |
| 321 | |
| 322 | /* |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 323 | * Munlock a batch of pages from the same zone |
| 324 | * |
| 325 | * The work is split to two main phases. First phase clears the Mlocked flag |
| 326 | * and attempts to isolate the pages, all under a single zone lru lock. |
| 327 | * The second phase finishes the munlock only for pages where isolation |
| 328 | * succeeded. |
| 329 | * |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 330 | * Note that the pagevec may be modified during the process. |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 331 | */ |
| 332 | static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone) |
| 333 | { |
| 334 | int i; |
| 335 | int nr = pagevec_count(pvec); |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 336 | int delta_munlocked; |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 337 | struct pagevec pvec_putback; |
| 338 | int pgrescued = 0; |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 339 | |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 340 | pagevec_init(&pvec_putback, 0); |
| 341 | |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 342 | /* Phase 1: page isolation */ |
| 343 | spin_lock_irq(&zone->lru_lock); |
| 344 | for (i = 0; i < nr; i++) { |
| 345 | struct page *page = pvec->pages[i]; |
| 346 | |
| 347 | if (TestClearPageMlocked(page)) { |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 348 | /* |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 349 | * We already have pin from follow_page_mask() |
| 350 | * so we can spare the get_page() here. |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 351 | */ |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 352 | if (__munlock_isolate_lru_page(page, false)) |
| 353 | continue; |
| 354 | else |
| 355 | __munlock_isolation_failed(page); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 356 | } |
Vlastimil Babka | 01cc2e5 | 2014-01-23 15:52:50 -0800 | [diff] [blame] | 357 | |
| 358 | /* |
| 359 | * We won't be munlocking this page in the next phase |
| 360 | * but we still need to release the follow_page_mask() |
| 361 | * pin. We cannot do it under lru_lock however. If it's |
| 362 | * the last pin, __page_cache_release() would deadlock. |
| 363 | */ |
| 364 | pagevec_add(&pvec_putback, pvec->pages[i]); |
| 365 | pvec->pages[i] = NULL; |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 366 | } |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 367 | delta_munlocked = -nr + pagevec_count(&pvec_putback); |
Vlastimil Babka | 1ebb7cc | 2013-09-11 14:22:30 -0700 | [diff] [blame] | 368 | __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked); |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 369 | spin_unlock_irq(&zone->lru_lock); |
| 370 | |
Vlastimil Babka | 3b25df9 | 2014-01-02 12:58:44 -0800 | [diff] [blame] | 371 | /* Now we can release pins of pages that we are not munlocking */ |
| 372 | pagevec_release(&pvec_putback); |
| 373 | |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 374 | /* Phase 2: page munlock */ |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 375 | for (i = 0; i < nr; i++) { |
| 376 | struct page *page = pvec->pages[i]; |
| 377 | |
| 378 | if (page) { |
| 379 | lock_page(page); |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 380 | if (!__putback_lru_fast_prepare(page, &pvec_putback, |
| 381 | &pgrescued)) { |
Vlastimil Babka | 5b40998 | 2013-09-11 14:22:33 -0700 | [diff] [blame] | 382 | /* |
| 383 | * Slow path. We don't want to lose the last |
| 384 | * pin before unlock_page() |
| 385 | */ |
| 386 | get_page(page); /* for putback_lru_page() */ |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 387 | __munlock_isolated_page(page); |
| 388 | unlock_page(page); |
Vlastimil Babka | 5b40998 | 2013-09-11 14:22:33 -0700 | [diff] [blame] | 389 | put_page(page); /* from follow_page_mask() */ |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 390 | } |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 391 | } |
| 392 | } |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 393 | |
Vlastimil Babka | 5b40998 | 2013-09-11 14:22:33 -0700 | [diff] [blame] | 394 | /* |
| 395 | * Phase 3: page putback for pages that qualified for the fast path |
| 396 | * This will also call put_page() to return pin from follow_page_mask() |
| 397 | */ |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 398 | if (pagevec_count(&pvec_putback)) |
| 399 | __putback_lru_fast(&pvec_putback, pgrescued); |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 400 | } |
Vlastimil Babka | 56afe47 | 2013-09-11 14:22:32 -0700 | [diff] [blame] | 401 | |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 402 | /* |
| 403 | * Fill up pagevec for __munlock_pagevec using pte walk |
| 404 | * |
| 405 | * The function expects that the struct page corresponding to @start address is |
| 406 | * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone. |
| 407 | * |
| 408 | * The rest of @pvec is filled by subsequent pages within the same pmd and same |
| 409 | * zone, as long as the pte's are present and vm_normal_page() succeeds. These |
| 410 | * pages also get pinned. |
| 411 | * |
| 412 | * Returns the address of the next page that should be scanned. This equals |
| 413 | * @start + PAGE_SIZE when no page could be added by the pte walk. |
| 414 | */ |
| 415 | static unsigned long __munlock_pagevec_fill(struct pagevec *pvec, |
| 416 | struct vm_area_struct *vma, int zoneid, unsigned long start, |
| 417 | unsigned long end) |
| 418 | { |
| 419 | pte_t *pte; |
| 420 | spinlock_t *ptl; |
| 421 | |
| 422 | /* |
| 423 | * Initialize pte walk starting at the already pinned page where we |
Vlastimil Babka | eadb41a | 2013-09-30 13:45:18 -0700 | [diff] [blame] | 424 | * are sure that there is a pte, as it was pinned under the same |
| 425 | * mmap_sem write op. |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 426 | */ |
| 427 | pte = get_locked_pte(vma->vm_mm, start, &ptl); |
Vlastimil Babka | eadb41a | 2013-09-30 13:45:18 -0700 | [diff] [blame] | 428 | /* Make sure we do not cross the page table boundary */ |
| 429 | end = pgd_addr_end(start, end); |
| 430 | end = pud_addr_end(start, end); |
| 431 | end = pmd_addr_end(start, end); |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 432 | |
| 433 | /* The page next to the pinned page is the first we will try to get */ |
| 434 | start += PAGE_SIZE; |
| 435 | while (start < end) { |
| 436 | struct page *page = NULL; |
| 437 | pte++; |
| 438 | if (pte_present(*pte)) |
| 439 | page = vm_normal_page(vma, start, *pte); |
| 440 | /* |
| 441 | * Break if page could not be obtained or the page's node+zone does not |
| 442 | * match |
| 443 | */ |
| 444 | if (!page || page_zone_id(page) != zoneid) |
| 445 | break; |
| 446 | |
| 447 | get_page(page); |
| 448 | /* |
| 449 | * Increase the address that will be returned *before* the |
| 450 | * eventual break due to pvec becoming full by adding the page |
| 451 | */ |
| 452 | start += PAGE_SIZE; |
| 453 | if (pagevec_add(pvec, page) == 0) |
| 454 | break; |
| 455 | } |
| 456 | pte_unmap_unlock(pte, ptl); |
| 457 | return start; |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 458 | } |
| 459 | |
| 460 | /* |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 461 | * munlock_vma_pages_range() - munlock all pages in the vma range.' |
| 462 | * @vma - vma containing range to be munlock()ed. |
| 463 | * @start - start address in @vma of the range |
| 464 | * @end - end of range in @vma. |
| 465 | * |
| 466 | * For mremap(), munmap() and exit(). |
| 467 | * |
| 468 | * Called with @vma VM_LOCKED. |
| 469 | * |
| 470 | * Returns with VM_LOCKED cleared. Callers must be prepared to |
| 471 | * deal with this. |
| 472 | * |
| 473 | * We don't save and restore VM_LOCKED here because pages are |
| 474 | * still on lru. In unmap path, pages might be scanned by reclaim |
| 475 | * and re-mlocked by try_to_{munlock|unmap} before we unmap and |
| 476 | * free them. This will result in freeing mlocked pages. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 477 | */ |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 478 | void munlock_vma_pages_range(struct vm_area_struct *vma, |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 479 | unsigned long start, unsigned long end) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 480 | { |
| 481 | vma->vm_flags &= ~VM_LOCKED; |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 482 | |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 483 | while (start < end) { |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 484 | struct page *page = NULL; |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 485 | unsigned int page_mask; |
| 486 | unsigned long page_increm; |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 487 | struct pagevec pvec; |
| 488 | struct zone *zone; |
| 489 | int zoneid; |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 490 | |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 491 | pagevec_init(&pvec, 0); |
Hugh Dickins | 6e91971 | 2009-09-21 17:03:32 -0700 | [diff] [blame] | 492 | /* |
| 493 | * Although FOLL_DUMP is intended for get_dump_page(), |
| 494 | * it just so happens that its special treatment of the |
| 495 | * ZERO_PAGE (returning an error instead of doing get_page) |
| 496 | * suits munlock very well (and if somehow an abnormal page |
| 497 | * has sneaked into the range, we won't oops here: great). |
| 498 | */ |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 499 | page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP, |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 500 | &page_mask); |
| 501 | |
Hugh Dickins | 6e91971 | 2009-09-21 17:03:32 -0700 | [diff] [blame] | 502 | if (page && !IS_ERR(page)) { |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 503 | if (PageTransHuge(page)) { |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 504 | lock_page(page); |
| 505 | /* |
| 506 | * Any THP page found by follow_page_mask() may |
| 507 | * have gotten split before reaching |
| 508 | * munlock_vma_page(), so we need to recompute |
| 509 | * the page_mask here. |
| 510 | */ |
| 511 | page_mask = munlock_vma_page(page); |
| 512 | unlock_page(page); |
| 513 | put_page(page); /* follow_page_mask() */ |
| 514 | } else { |
| 515 | /* |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 516 | * Non-huge pages are handled in batches via |
| 517 | * pagevec. The pin from follow_page_mask() |
| 518 | * prevents them from collapsing by THP. |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 519 | */ |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 520 | pagevec_add(&pvec, page); |
| 521 | zone = page_zone(page); |
| 522 | zoneid = page_zone_id(page); |
| 523 | |
| 524 | /* |
| 525 | * Try to fill the rest of pagevec using fast |
| 526 | * pte walk. This will also update start to |
| 527 | * the next page to process. Then munlock the |
| 528 | * pagevec. |
| 529 | */ |
| 530 | start = __munlock_pagevec_fill(&pvec, vma, |
| 531 | zoneid, start, end); |
| 532 | __munlock_pagevec(&pvec, zone); |
| 533 | goto next; |
Vlastimil Babka | 7225522 | 2013-09-11 14:22:29 -0700 | [diff] [blame] | 534 | } |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 535 | } |
Vlastimil Babka | c424be1 | 2014-01-02 12:58:43 -0800 | [diff] [blame] | 536 | /* It's a bug to munlock in the middle of a THP page */ |
| 537 | VM_BUG_ON((start >> PAGE_SHIFT) & page_mask); |
| 538 | page_increm = 1 + page_mask; |
Michel Lespinasse | ff6a6da | 2013-02-27 17:02:44 -0800 | [diff] [blame] | 539 | start += page_increm * PAGE_SIZE; |
Vlastimil Babka | 7a8010c | 2013-09-11 14:22:35 -0700 | [diff] [blame] | 540 | next: |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 541 | cond_resched(); |
| 542 | } |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 543 | } |
| 544 | |
| 545 | /* |
| 546 | * mlock_fixup - handle mlock[all]/munlock[all] requests. |
| 547 | * |
| 548 | * Filters out "special" vmas -- VM_LOCKED never gets set for these, and |
| 549 | * 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] | 550 | * populate the ptes. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 551 | * |
| 552 | * For vmas that pass the filters, merge/split as appropriate. |
| 553 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | 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] | 555 | unsigned long start, unsigned long end, vm_flags_t newflags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 557 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | pgoff_t pgoff; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 559 | int nr_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | int ret = 0; |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 561 | int lock = !!(newflags & VM_LOCKED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 563 | if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) || |
Stephen Wilson | 31db58b | 2011-03-13 15:49:15 -0400 | [diff] [blame] | 564 | is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 565 | goto out; /* don't set VM_LOCKED, don't count */ |
| 566 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 567 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
| 568 | *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma, |
| 569 | vma->vm_file, pgoff, vma_policy(vma)); |
| 570 | if (*prev) { |
| 571 | vma = *prev; |
| 572 | goto success; |
| 573 | } |
| 574 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | if (start != vma->vm_start) { |
| 576 | ret = split_vma(mm, vma, start, 1); |
| 577 | if (ret) |
| 578 | goto out; |
| 579 | } |
| 580 | |
| 581 | if (end != vma->vm_end) { |
| 582 | ret = split_vma(mm, vma, end, 0); |
| 583 | if (ret) |
| 584 | goto out; |
| 585 | } |
| 586 | |
| 587 | success: |
| 588 | /* |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 589 | * Keep track of amount of locked VM. |
| 590 | */ |
| 591 | nr_pages = (end - start) >> PAGE_SHIFT; |
| 592 | if (!lock) |
| 593 | nr_pages = -nr_pages; |
| 594 | mm->locked_vm += nr_pages; |
| 595 | |
| 596 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 597 | * vm_flags is protected by the mmap_sem held in write mode. |
| 598 | * It's okay if try_to_unmap_one unmaps a page just after we |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 599 | * set VM_LOCKED, __mlock_vma_pages_range will bring it back. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 601 | |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 602 | if (lock) |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 603 | vma->vm_flags = newflags; |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 604 | else |
Hugh Dickins | 408e82b | 2009-09-21 17:03:23 -0700 | [diff] [blame] | 605 | munlock_vma_pages_range(vma, start, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 606 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | out: |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 608 | *prev = vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 609 | return ret; |
| 610 | } |
| 611 | |
| 612 | static int do_mlock(unsigned long start, size_t len, int on) |
| 613 | { |
| 614 | unsigned long nstart, end, tmp; |
| 615 | struct vm_area_struct * vma, * prev; |
| 616 | int error; |
| 617 | |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 618 | VM_BUG_ON(start & ~PAGE_MASK); |
| 619 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | end = start + len; |
| 621 | if (end < start) |
| 622 | return -EINVAL; |
| 623 | if (end == start) |
| 624 | return 0; |
Linus Torvalds | 097d591 | 2012-03-06 18:23:36 -0800 | [diff] [blame] | 625 | vma = find_vma(current->mm, start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | if (!vma || vma->vm_start > start) |
| 627 | return -ENOMEM; |
| 628 | |
Linus Torvalds | 097d591 | 2012-03-06 18:23:36 -0800 | [diff] [blame] | 629 | prev = vma->vm_prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | if (start > vma->vm_start) |
| 631 | prev = vma; |
| 632 | |
| 633 | for (nstart = start ; ; ) { |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 634 | vm_flags_t newflags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | |
| 636 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
| 637 | |
Michel Lespinasse | 1869305 | 2013-02-22 16:32:46 -0800 | [diff] [blame] | 638 | newflags = vma->vm_flags & ~VM_LOCKED; |
| 639 | if (on) |
Michel Lespinasse | 09a9f1d | 2013-03-28 16:26:23 -0700 | [diff] [blame] | 640 | newflags |= VM_LOCKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | |
| 642 | tmp = vma->vm_end; |
| 643 | if (tmp > end) |
| 644 | tmp = end; |
| 645 | error = mlock_fixup(vma, &prev, nstart, tmp, newflags); |
| 646 | if (error) |
| 647 | break; |
| 648 | nstart = tmp; |
| 649 | if (nstart < prev->vm_end) |
| 650 | nstart = prev->vm_end; |
| 651 | if (nstart >= end) |
| 652 | break; |
| 653 | |
| 654 | vma = prev->vm_next; |
| 655 | if (!vma || vma->vm_start != nstart) { |
| 656 | error = -ENOMEM; |
| 657 | break; |
| 658 | } |
| 659 | } |
| 660 | return error; |
| 661 | } |
| 662 | |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 663 | /* |
| 664 | * __mm_populate - populate and/or mlock pages within a range of address space. |
| 665 | * |
| 666 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap |
| 667 | * flags. VMAs must be already marked with the desired vm_flags, and |
| 668 | * mmap_sem must not be held. |
| 669 | */ |
| 670 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 671 | { |
| 672 | struct mm_struct *mm = current->mm; |
| 673 | unsigned long end, nstart, nend; |
| 674 | struct vm_area_struct *vma = NULL; |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 675 | int locked = 0; |
Michel Lespinasse | 28a3571 | 2013-02-22 16:35:55 -0800 | [diff] [blame] | 676 | long ret = 0; |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 677 | |
| 678 | VM_BUG_ON(start & ~PAGE_MASK); |
| 679 | VM_BUG_ON(len != PAGE_ALIGN(len)); |
| 680 | end = start + len; |
| 681 | |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 682 | for (nstart = start; nstart < end; nstart = nend) { |
| 683 | /* |
| 684 | * We want to fault in pages for [nstart; end) address range. |
| 685 | * Find first corresponding VMA. |
| 686 | */ |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 687 | if (!locked) { |
| 688 | locked = 1; |
| 689 | down_read(&mm->mmap_sem); |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 690 | vma = find_vma(mm, nstart); |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 691 | } else if (nstart >= vma->vm_end) |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 692 | vma = vma->vm_next; |
| 693 | if (!vma || vma->vm_start >= end) |
| 694 | break; |
| 695 | /* |
| 696 | * Set [nstart; nend) to intersection of desired address |
| 697 | * range with the first VMA. Also, skip undesirable VMA types. |
| 698 | */ |
| 699 | nend = min(end, vma->vm_end); |
Michel Lespinasse | 09a9f1d | 2013-03-28 16:26:23 -0700 | [diff] [blame] | 700 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 701 | continue; |
| 702 | if (nstart < vma->vm_start) |
| 703 | nstart = vma->vm_start; |
| 704 | /* |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 705 | * Now fault in a range of pages. __mlock_vma_pages_range() |
| 706 | * double checks the vma flags, so that it won't mlock pages |
| 707 | * if the vma was already munlocked. |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 708 | */ |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 709 | ret = __mlock_vma_pages_range(vma, nstart, nend, &locked); |
| 710 | if (ret < 0) { |
| 711 | if (ignore_errors) { |
| 712 | ret = 0; |
| 713 | continue; /* continue at next VMA */ |
| 714 | } |
Michel Lespinasse | 5fdb200 | 2011-01-13 15:46:12 -0800 | [diff] [blame] | 715 | ret = __mlock_posix_error_return(ret); |
| 716 | break; |
| 717 | } |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 718 | nend = nstart + ret * PAGE_SIZE; |
| 719 | ret = 0; |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 720 | } |
Michel Lespinasse | 53a7706 | 2011-01-13 15:46:14 -0800 | [diff] [blame] | 721 | if (locked) |
| 722 | up_read(&mm->mmap_sem); |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 723 | return ret; /* 0 or negative error code */ |
| 724 | } |
| 725 | |
Heiko Carstens | 6a6160a | 2009-01-14 14:14:15 +0100 | [diff] [blame] | 726 | SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | { |
| 728 | unsigned long locked; |
| 729 | unsigned long lock_limit; |
| 730 | int error = -ENOMEM; |
| 731 | |
| 732 | if (!can_do_mlock()) |
| 733 | return -EPERM; |
| 734 | |
KOSAKI Motohiro | 8891d6d | 2008-11-12 13:26:53 -0800 | [diff] [blame] | 735 | lru_add_drain_all(); /* flush pagevec */ |
| 736 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); |
| 738 | start &= PAGE_MASK; |
| 739 | |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 740 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | lock_limit >>= PAGE_SHIFT; |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 742 | locked = len >> PAGE_SHIFT; |
| 743 | |
| 744 | down_write(¤t->mm->mmap_sem); |
| 745 | |
| 746 | locked += current->mm->locked_vm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 747 | |
| 748 | /* check against resource limits */ |
| 749 | if ((locked <= lock_limit) || capable(CAP_IPC_LOCK)) |
| 750 | error = do_mlock(start, len, 1); |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 751 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 752 | up_write(¤t->mm->mmap_sem); |
Michel Lespinasse | fed067d | 2011-01-13 15:46:10 -0800 | [diff] [blame] | 753 | if (!error) |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 754 | error = __mm_populate(start, len, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 755 | return error; |
| 756 | } |
| 757 | |
Heiko Carstens | 6a6160a | 2009-01-14 14:14:15 +0100 | [diff] [blame] | 758 | SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | { |
| 760 | int ret; |
| 761 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | len = PAGE_ALIGN(len + (start & ~PAGE_MASK)); |
| 763 | start &= PAGE_MASK; |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 764 | |
| 765 | down_write(¤t->mm->mmap_sem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 766 | ret = do_mlock(start, len, 0); |
| 767 | up_write(¤t->mm->mmap_sem); |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 768 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 769 | return ret; |
| 770 | } |
| 771 | |
| 772 | static int do_mlockall(int flags) |
| 773 | { |
| 774 | struct vm_area_struct * vma, * prev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | |
| 776 | if (flags & MCL_FUTURE) |
Michel Lespinasse | 09a9f1d | 2013-03-28 16:26:23 -0700 | [diff] [blame] | 777 | current->mm->def_flags |= VM_LOCKED; |
Gerald Schaefer | 9977f0f | 2013-02-12 13:46:20 -0800 | [diff] [blame] | 778 | else |
Michel Lespinasse | 09a9f1d | 2013-03-28 16:26:23 -0700 | [diff] [blame] | 779 | current->mm->def_flags &= ~VM_LOCKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | if (flags == MCL_FUTURE) |
| 781 | goto out; |
| 782 | |
| 783 | for (vma = current->mm->mmap; vma ; vma = prev->vm_next) { |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 784 | vm_flags_t newflags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | |
Michel Lespinasse | 1869305 | 2013-02-22 16:32:46 -0800 | [diff] [blame] | 786 | newflags = vma->vm_flags & ~VM_LOCKED; |
| 787 | if (flags & MCL_CURRENT) |
Michel Lespinasse | 09a9f1d | 2013-03-28 16:26:23 -0700 | [diff] [blame] | 788 | newflags |= VM_LOCKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | |
| 790 | /* Ignore errors */ |
| 791 | mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags); |
Paul E. McKenney | bde6c3a | 2014-07-01 11:26:57 -0700 | [diff] [blame] | 792 | cond_resched_rcu_qs(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | } |
| 794 | out: |
| 795 | return 0; |
| 796 | } |
| 797 | |
Heiko Carstens | 3480b25 | 2009-01-14 14:14:16 +0100 | [diff] [blame] | 798 | SYSCALL_DEFINE1(mlockall, int, flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | { |
| 800 | unsigned long lock_limit; |
| 801 | int ret = -EINVAL; |
| 802 | |
| 803 | if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE))) |
| 804 | goto out; |
| 805 | |
| 806 | ret = -EPERM; |
| 807 | if (!can_do_mlock()) |
| 808 | goto out; |
| 809 | |
Christoph Lameter | df9d698 | 2011-10-31 17:09:35 -0700 | [diff] [blame] | 810 | if (flags & MCL_CURRENT) |
| 811 | lru_add_drain_all(); /* flush pagevec */ |
KOSAKI Motohiro | 8891d6d | 2008-11-12 13:26:53 -0800 | [diff] [blame] | 812 | |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 813 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 814 | lock_limit >>= PAGE_SHIFT; |
| 815 | |
| 816 | ret = -ENOMEM; |
Davidlohr Bueso | 1f1cd70 | 2014-01-21 15:49:16 -0800 | [diff] [blame] | 817 | down_write(¤t->mm->mmap_sem); |
| 818 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) || |
| 820 | capable(CAP_IPC_LOCK)) |
| 821 | ret = do_mlockall(flags); |
| 822 | up_write(¤t->mm->mmap_sem); |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 823 | if (!ret && (flags & MCL_CURRENT)) |
| 824 | mm_populate(0, TASK_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | out: |
| 826 | return ret; |
| 827 | } |
| 828 | |
Heiko Carstens | 3480b25 | 2009-01-14 14:14:16 +0100 | [diff] [blame] | 829 | SYSCALL_DEFINE0(munlockall) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | { |
| 831 | int ret; |
| 832 | |
| 833 | down_write(¤t->mm->mmap_sem); |
| 834 | ret = do_mlockall(0); |
| 835 | up_write(¤t->mm->mmap_sem); |
| 836 | return ret; |
| 837 | } |
| 838 | |
| 839 | /* |
| 840 | * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB |
| 841 | * shm segments) get accounted against the user_struct instead. |
| 842 | */ |
| 843 | static DEFINE_SPINLOCK(shmlock_user_lock); |
| 844 | |
| 845 | int user_shm_lock(size_t size, struct user_struct *user) |
| 846 | { |
| 847 | unsigned long lock_limit, locked; |
| 848 | int allowed = 0; |
| 849 | |
| 850 | locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 851 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
Herbert van den Bergh | 5ed44a4 | 2007-07-15 23:38:25 -0700 | [diff] [blame] | 852 | if (lock_limit == RLIM_INFINITY) |
| 853 | allowed = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | lock_limit >>= PAGE_SHIFT; |
| 855 | spin_lock(&shmlock_user_lock); |
Herbert van den Bergh | 5ed44a4 | 2007-07-15 23:38:25 -0700 | [diff] [blame] | 856 | if (!allowed && |
| 857 | locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | goto out; |
| 859 | get_uid(user); |
| 860 | user->locked_shm += locked; |
| 861 | allowed = 1; |
| 862 | out: |
| 863 | spin_unlock(&shmlock_user_lock); |
| 864 | return allowed; |
| 865 | } |
| 866 | |
| 867 | void user_shm_unlock(size_t size, struct user_struct *user) |
| 868 | { |
| 869 | spin_lock(&shmlock_user_lock); |
| 870 | user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 871 | spin_unlock(&shmlock_user_lock); |
| 872 | free_uid(user); |
| 873 | } |