Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Memory Migration functionality - linux/mm/migration.c |
| 3 | * |
| 4 | * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter |
| 5 | * |
| 6 | * Page migration was first developed in the context of the memory hotplug |
| 7 | * project. The main authors of the migration code are: |
| 8 | * |
| 9 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> |
| 10 | * Hirokazu Takahashi <taka@valinux.co.jp> |
| 11 | * Dave Hansen <haveblue@us.ibm.com> |
Christoph Lameter | cde5353 | 2008-07-04 09:59:22 -0700 | [diff] [blame] | 12 | * Christoph Lameter |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 13 | */ |
| 14 | |
| 15 | #include <linux/migrate.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 16 | #include <linux/export.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 17 | #include <linux/swap.h> |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 18 | #include <linux/swapops.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 19 | #include <linux/pagemap.h> |
Christoph Lameter | e23ca00 | 2006-04-10 22:52:57 -0700 | [diff] [blame] | 20 | #include <linux/buffer_head.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 21 | #include <linux/mm_inline.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 22 | #include <linux/nsproxy.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 23 | #include <linux/pagevec.h> |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 24 | #include <linux/ksm.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 25 | #include <linux/rmap.h> |
| 26 | #include <linux/topology.h> |
| 27 | #include <linux/cpu.h> |
| 28 | #include <linux/cpuset.h> |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 29 | #include <linux/writeback.h> |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 30 | #include <linux/mempolicy.h> |
| 31 | #include <linux/vmalloc.h> |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 32 | #include <linux/security.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 33 | #include <linux/memcontrol.h> |
Adrian Bunk | 4f5ca26 | 2008-07-23 21:27:02 -0700 | [diff] [blame] | 34 | #include <linux/syscalls.h> |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 35 | #include <linux/hugetlb.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 36 | #include <linux/gfp.h> |
Liam Mark | cc2d4bd | 2013-01-16 10:14:40 -0800 | [diff] [blame^] | 37 | #include <trace/events/kmem.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 38 | |
Michal Nazarewicz | 0d1836c | 2010-12-21 17:24:26 -0800 | [diff] [blame] | 39 | #include <asm/tlbflush.h> |
| 40 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 41 | #include "internal.h" |
| 42 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 43 | /* |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 44 | * migrate_prep() needs to be called before we start compiling a list of pages |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 45 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
| 46 | * undesirable, use migrate_prep_local() |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 47 | */ |
| 48 | int migrate_prep(void) |
| 49 | { |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 50 | /* |
| 51 | * Clear the LRU lists so pages can be isolated. |
| 52 | * Note that pages may be moved off the LRU after we have |
| 53 | * drained them. Those pages will fail to migrate like other |
| 54 | * pages that may be busy. |
| 55 | */ |
| 56 | lru_add_drain_all(); |
| 57 | |
| 58 | return 0; |
| 59 | } |
| 60 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 61 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
| 62 | int migrate_prep_local(void) |
| 63 | { |
| 64 | lru_add_drain(); |
| 65 | |
| 66 | return 0; |
| 67 | } |
| 68 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 69 | /* |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 70 | * Add isolated pages on the list back to the LRU under page lock |
| 71 | * to avoid leaking evictable pages back onto unevictable list. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 72 | */ |
Minchan Kim | e13861d | 2010-05-24 14:31:59 -0700 | [diff] [blame] | 73 | void putback_lru_pages(struct list_head *l) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 74 | { |
| 75 | struct page *page; |
| 76 | struct page *page2; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 77 | |
| 78 | list_for_each_entry_safe(page, page2, l, lru) { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 79 | list_del(&page->lru); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 80 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
Johannes Weiner | 6c0b135 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 81 | page_is_file_cache(page)); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 82 | putback_lru_page(page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 83 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 84 | } |
| 85 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 86 | /* |
| 87 | * Restore a potential migration pte to a working pte entry |
| 88 | */ |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 89 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, |
| 90 | unsigned long addr, void *old) |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 91 | { |
| 92 | struct mm_struct *mm = vma->vm_mm; |
| 93 | swp_entry_t entry; |
| 94 | pgd_t *pgd; |
| 95 | pud_t *pud; |
| 96 | pmd_t *pmd; |
| 97 | pte_t *ptep, pte; |
| 98 | spinlock_t *ptl; |
| 99 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 100 | if (unlikely(PageHuge(new))) { |
| 101 | ptep = huge_pte_offset(mm, addr); |
| 102 | if (!ptep) |
| 103 | goto out; |
| 104 | ptl = &mm->page_table_lock; |
| 105 | } else { |
| 106 | pgd = pgd_offset(mm, addr); |
| 107 | if (!pgd_present(*pgd)) |
| 108 | goto out; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 109 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 110 | pud = pud_offset(pgd, addr); |
| 111 | if (!pud_present(*pud)) |
| 112 | goto out; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 113 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 114 | pmd = pmd_offset(pud, addr); |
Andrea Arcangeli | 500d65d | 2011-01-13 15:46:55 -0800 | [diff] [blame] | 115 | if (pmd_trans_huge(*pmd)) |
| 116 | goto out; |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 117 | if (!pmd_present(*pmd)) |
| 118 | goto out; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 119 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 120 | ptep = pte_offset_map(pmd, addr); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 121 | |
Hugh Dickins | 486cf46 | 2011-10-19 12:50:35 -0700 | [diff] [blame] | 122 | /* |
| 123 | * Peek to check is_swap_pte() before taking ptlock? No, we |
| 124 | * can race mremap's move_ptes(), which skips anon_vma lock. |
| 125 | */ |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 126 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 127 | ptl = pte_lockptr(mm, pmd); |
| 128 | } |
| 129 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 130 | spin_lock(ptl); |
| 131 | pte = *ptep; |
| 132 | if (!is_swap_pte(pte)) |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 133 | goto unlock; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 134 | |
| 135 | entry = pte_to_swp_entry(pte); |
| 136 | |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 137 | if (!is_migration_entry(entry) || |
| 138 | migration_entry_to_page(entry) != old) |
| 139 | goto unlock; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 140 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 141 | get_page(new); |
| 142 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); |
| 143 | if (is_write_migration_entry(entry)) |
| 144 | pte = pte_mkwrite(pte); |
Andi Kleen | 3ef8fd7 | 2010-10-11 16:03:21 +0200 | [diff] [blame] | 145 | #ifdef CONFIG_HUGETLB_PAGE |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 146 | if (PageHuge(new)) |
| 147 | pte = pte_mkhuge(pte); |
Andi Kleen | 3ef8fd7 | 2010-10-11 16:03:21 +0200 | [diff] [blame] | 148 | #endif |
KAMEZAWA Hiroyuki | 97ee052 | 2007-10-16 01:25:43 -0700 | [diff] [blame] | 149 | flush_cache_page(vma, addr, pte_pfn(pte)); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 150 | set_pte_at(mm, addr, ptep, pte); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 151 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 152 | if (PageHuge(new)) { |
| 153 | if (PageAnon(new)) |
| 154 | hugepage_add_anon_rmap(new, vma, addr); |
| 155 | else |
| 156 | page_dup_rmap(new); |
| 157 | } else if (PageAnon(new)) |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 158 | page_add_anon_rmap(new, vma, addr); |
| 159 | else |
| 160 | page_add_file_rmap(new); |
| 161 | |
| 162 | /* No need to invalidate - it was non-present before */ |
Russell King | 4b3073e | 2009-12-18 16:40:18 +0000 | [diff] [blame] | 163 | update_mmu_cache(vma, addr, ptep); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 164 | unlock: |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 165 | pte_unmap_unlock(ptep, ptl); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 166 | out: |
| 167 | return SWAP_AGAIN; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 168 | } |
| 169 | |
| 170 | /* |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 171 | * Get rid of all migration entries and replace them by |
| 172 | * references to the indicated page. |
| 173 | */ |
| 174 | static void remove_migration_ptes(struct page *old, struct page *new) |
| 175 | { |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 176 | rmap_walk(new, remove_migration_pte, old); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 177 | } |
| 178 | |
| 179 | /* |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 180 | * Something used the pte of a page under migration. We need to |
| 181 | * get to the page and wait until migration is finished. |
| 182 | * When we return from this function the fault will be retried. |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 183 | */ |
| 184 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, |
| 185 | unsigned long address) |
| 186 | { |
| 187 | pte_t *ptep, pte; |
| 188 | spinlock_t *ptl; |
| 189 | swp_entry_t entry; |
| 190 | struct page *page; |
| 191 | |
| 192 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
| 193 | pte = *ptep; |
| 194 | if (!is_swap_pte(pte)) |
| 195 | goto out; |
| 196 | |
| 197 | entry = pte_to_swp_entry(pte); |
| 198 | if (!is_migration_entry(entry)) |
| 199 | goto out; |
| 200 | |
| 201 | page = migration_entry_to_page(entry); |
| 202 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 203 | /* |
| 204 | * Once radix-tree replacement of page migration started, page_count |
| 205 | * *must* be zero. And, we don't want to call wait_on_page_locked() |
| 206 | * against a page without get_page(). |
| 207 | * So, we use get_page_unless_zero(), here. Even failed, page fault |
| 208 | * will occur again. |
| 209 | */ |
| 210 | if (!get_page_unless_zero(page)) |
| 211 | goto out; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 212 | pte_unmap_unlock(ptep, ptl); |
| 213 | wait_on_page_locked(page); |
| 214 | put_page(page); |
| 215 | return; |
| 216 | out: |
| 217 | pte_unmap_unlock(ptep, ptl); |
| 218 | } |
| 219 | |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 220 | #ifdef CONFIG_BLOCK |
| 221 | /* Returns true if all buffers are successfully locked */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 222 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, |
| 223 | enum migrate_mode mode) |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 224 | { |
| 225 | struct buffer_head *bh = head; |
| 226 | |
| 227 | /* Simple case, sync compaction */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 228 | if (mode != MIGRATE_ASYNC) { |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 229 | do { |
| 230 | get_bh(bh); |
| 231 | lock_buffer(bh); |
| 232 | bh = bh->b_this_page; |
| 233 | |
| 234 | } while (bh != head); |
| 235 | |
| 236 | return true; |
| 237 | } |
| 238 | |
| 239 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ |
| 240 | do { |
| 241 | get_bh(bh); |
| 242 | if (!trylock_buffer(bh)) { |
| 243 | /* |
| 244 | * We failed to lock the buffer and cannot stall in |
| 245 | * async migration. Release the taken locks |
| 246 | */ |
| 247 | struct buffer_head *failed_bh = bh; |
| 248 | put_bh(failed_bh); |
| 249 | bh = head; |
| 250 | while (bh != failed_bh) { |
| 251 | unlock_buffer(bh); |
| 252 | put_bh(bh); |
| 253 | bh = bh->b_this_page; |
| 254 | } |
| 255 | return false; |
| 256 | } |
| 257 | |
| 258 | bh = bh->b_this_page; |
| 259 | } while (bh != head); |
| 260 | return true; |
| 261 | } |
| 262 | #else |
| 263 | static inline bool buffer_migrate_lock_buffers(struct buffer_head *head, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 264 | enum migrate_mode mode) |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 265 | { |
| 266 | return true; |
| 267 | } |
| 268 | #endif /* CONFIG_BLOCK */ |
| 269 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 270 | /* |
Christoph Lameter | c3fcf8a | 2006-06-23 02:03:32 -0700 | [diff] [blame] | 271 | * Replace the page in the mapping. |
Christoph Lameter | 5b5c712 | 2006-06-23 02:03:29 -0700 | [diff] [blame] | 272 | * |
| 273 | * The number of remaining references must be: |
| 274 | * 1 for anonymous pages without a mapping |
| 275 | * 2 for pages with a mapping |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 276 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 277 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 278 | static int migrate_page_move_mapping(struct address_space *mapping, |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 279 | struct page *newpage, struct page *page, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 280 | struct buffer_head *head, enum migrate_mode mode) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 281 | { |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 282 | int expected_count; |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 283 | void **pslot; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 284 | |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 285 | if (!mapping) { |
Christoph Lameter | 0e8c7d0 | 2007-04-23 14:41:09 -0700 | [diff] [blame] | 286 | /* Anonymous page without mapping */ |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 287 | if (page_count(page) != 1) |
| 288 | return -EAGAIN; |
| 289 | return 0; |
| 290 | } |
| 291 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 292 | spin_lock_irq(&mapping->tree_lock); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 293 | |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 294 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
| 295 | page_index(page)); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 296 | |
Johannes Weiner | edcf474 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 297 | expected_count = 2 + page_has_private(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 298 | if (page_count(page) != expected_count || |
Mel Gorman | 29c1f67 | 2011-01-13 15:47:21 -0800 | [diff] [blame] | 299 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 300 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | e23ca00 | 2006-04-10 22:52:57 -0700 | [diff] [blame] | 301 | return -EAGAIN; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 302 | } |
| 303 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 304 | if (!page_freeze_refs(page, expected_count)) { |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 305 | spin_unlock_irq(&mapping->tree_lock); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 306 | return -EAGAIN; |
| 307 | } |
| 308 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 309 | /* |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 310 | * In the async migration case of moving a page with buffers, lock the |
| 311 | * buffers using trylock before the mapping is moved. If the mapping |
| 312 | * was moved, we later failed to lock the buffers and could not move |
| 313 | * the mapping back due to an elevated page count, we would have to |
| 314 | * block waiting on other references to be dropped. |
| 315 | */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 316 | if (mode == MIGRATE_ASYNC && head && |
| 317 | !buffer_migrate_lock_buffers(head, mode)) { |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 318 | page_unfreeze_refs(page, expected_count); |
| 319 | spin_unlock_irq(&mapping->tree_lock); |
| 320 | return -EAGAIN; |
| 321 | } |
| 322 | |
| 323 | /* |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 324 | * Now we know that no one else is looking at the page. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 325 | */ |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 326 | get_page(newpage); /* add cache reference */ |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 327 | if (PageSwapCache(page)) { |
| 328 | SetPageSwapCache(newpage); |
| 329 | set_page_private(newpage, page_private(page)); |
| 330 | } |
| 331 | |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 332 | radix_tree_replace_slot(pslot, newpage); |
| 333 | |
| 334 | /* |
Jacobo Giralt | 937a94c | 2012-01-10 15:07:11 -0800 | [diff] [blame] | 335 | * Drop cache reference from old page by unfreezing |
| 336 | * to one less reference. |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 337 | * We know this isn't the last reference. |
| 338 | */ |
Jacobo Giralt | 937a94c | 2012-01-10 15:07:11 -0800 | [diff] [blame] | 339 | page_unfreeze_refs(page, expected_count - 1); |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 340 | |
Christoph Lameter | 0e8c7d0 | 2007-04-23 14:41:09 -0700 | [diff] [blame] | 341 | /* |
| 342 | * If moved to a different zone then also account |
| 343 | * the page for that zone. Other VM counters will be |
| 344 | * taken care of when we establish references to the |
| 345 | * new page and drop references to the old page. |
| 346 | * |
| 347 | * Note that anonymous pages are accounted for |
| 348 | * via NR_FILE_PAGES and NR_ANON_PAGES if they |
| 349 | * are mapped to swap space. |
| 350 | */ |
| 351 | __dec_zone_page_state(page, NR_FILE_PAGES); |
| 352 | __inc_zone_page_state(newpage, NR_FILE_PAGES); |
Andrea Arcangeli | 99a15e2 | 2011-06-16 12:56:19 -0700 | [diff] [blame] | 353 | if (!PageSwapCache(page) && PageSwapBacked(page)) { |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 354 | __dec_zone_page_state(page, NR_SHMEM); |
| 355 | __inc_zone_page_state(newpage, NR_SHMEM); |
| 356 | } |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 357 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 358 | |
| 359 | return 0; |
| 360 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 361 | |
| 362 | /* |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 363 | * The expected number of remaining references is the same as that |
| 364 | * of migrate_page_move_mapping(). |
| 365 | */ |
| 366 | int migrate_huge_page_move_mapping(struct address_space *mapping, |
| 367 | struct page *newpage, struct page *page) |
| 368 | { |
| 369 | int expected_count; |
| 370 | void **pslot; |
| 371 | |
| 372 | if (!mapping) { |
| 373 | if (page_count(page) != 1) |
| 374 | return -EAGAIN; |
| 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | spin_lock_irq(&mapping->tree_lock); |
| 379 | |
| 380 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
| 381 | page_index(page)); |
| 382 | |
| 383 | expected_count = 2 + page_has_private(page); |
| 384 | if (page_count(page) != expected_count || |
Mel Gorman | 29c1f67 | 2011-01-13 15:47:21 -0800 | [diff] [blame] | 385 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 386 | spin_unlock_irq(&mapping->tree_lock); |
| 387 | return -EAGAIN; |
| 388 | } |
| 389 | |
| 390 | if (!page_freeze_refs(page, expected_count)) { |
| 391 | spin_unlock_irq(&mapping->tree_lock); |
| 392 | return -EAGAIN; |
| 393 | } |
| 394 | |
| 395 | get_page(newpage); |
| 396 | |
| 397 | radix_tree_replace_slot(pslot, newpage); |
| 398 | |
Jacobo Giralt | 937a94c | 2012-01-10 15:07:11 -0800 | [diff] [blame] | 399 | page_unfreeze_refs(page, expected_count - 1); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 400 | |
| 401 | spin_unlock_irq(&mapping->tree_lock); |
| 402 | return 0; |
| 403 | } |
| 404 | |
| 405 | /* |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 406 | * Copy the page to its new location |
| 407 | */ |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 408 | void migrate_page_copy(struct page *newpage, struct page *page) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 409 | { |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 410 | if (PageHuge(page)) |
| 411 | copy_huge_page(newpage, page); |
| 412 | else |
| 413 | copy_highpage(newpage, page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 414 | |
| 415 | if (PageError(page)) |
| 416 | SetPageError(newpage); |
| 417 | if (PageReferenced(page)) |
| 418 | SetPageReferenced(newpage); |
| 419 | if (PageUptodate(page)) |
| 420 | SetPageUptodate(newpage); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 421 | if (TestClearPageActive(page)) { |
| 422 | VM_BUG_ON(PageUnevictable(page)); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 423 | SetPageActive(newpage); |
Lee Schermerhorn | 418b27e | 2009-12-14 17:59:54 -0800 | [diff] [blame] | 424 | } else if (TestClearPageUnevictable(page)) |
| 425 | SetPageUnevictable(newpage); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 426 | if (PageChecked(page)) |
| 427 | SetPageChecked(newpage); |
| 428 | if (PageMappedToDisk(page)) |
| 429 | SetPageMappedToDisk(newpage); |
| 430 | |
| 431 | if (PageDirty(page)) { |
| 432 | clear_page_dirty_for_io(page); |
Nick Piggin | 3a902c5 | 2008-04-30 00:55:16 -0700 | [diff] [blame] | 433 | /* |
| 434 | * Want to mark the page and the radix tree as dirty, and |
| 435 | * redo the accounting that clear_page_dirty_for_io undid, |
| 436 | * but we can't use set_page_dirty because that function |
| 437 | * is actually a signal that all of the page has become dirty. |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 438 | * Whereas only part of our page may be dirty. |
Nick Piggin | 3a902c5 | 2008-04-30 00:55:16 -0700 | [diff] [blame] | 439 | */ |
| 440 | __set_page_dirty_nobuffers(newpage); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 441 | } |
| 442 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 443 | mlock_migrate_page(newpage, page); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 444 | ksm_migrate_page(newpage, page); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 445 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 446 | ClearPageSwapCache(page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 447 | ClearPagePrivate(page); |
| 448 | set_page_private(page, 0); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 449 | |
| 450 | /* |
| 451 | * If any waiters have accumulated on the new page then |
| 452 | * wake them up. |
| 453 | */ |
| 454 | if (PageWriteback(newpage)) |
| 455 | end_page_writeback(newpage); |
| 456 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 457 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 458 | /************************************************************ |
| 459 | * Migration functions |
| 460 | ***********************************************************/ |
| 461 | |
| 462 | /* Always fail migration. Used for mappings that are not movable */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 463 | int fail_migrate_page(struct address_space *mapping, |
| 464 | struct page *newpage, struct page *page) |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 465 | { |
| 466 | return -EIO; |
| 467 | } |
| 468 | EXPORT_SYMBOL(fail_migrate_page); |
| 469 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 470 | /* |
| 471 | * Common logic to directly migrate a single page suitable for |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 472 | * pages that do not use PagePrivate/PagePrivate2. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 473 | * |
| 474 | * Pages are locked upon entry and exit. |
| 475 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 476 | int migrate_page(struct address_space *mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 477 | struct page *newpage, struct page *page, |
| 478 | enum migrate_mode mode) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 479 | { |
| 480 | int rc; |
| 481 | |
| 482 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ |
| 483 | |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 484 | rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 485 | |
| 486 | if (rc) |
| 487 | return rc; |
| 488 | |
| 489 | migrate_page_copy(newpage, page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 490 | return 0; |
| 491 | } |
| 492 | EXPORT_SYMBOL(migrate_page); |
| 493 | |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 494 | #ifdef CONFIG_BLOCK |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 495 | /* |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 496 | * Migration function for pages with buffers. This function can only be used |
| 497 | * if the underlying filesystem guarantees that no other references to "page" |
| 498 | * exist. |
| 499 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 500 | int buffer_migrate_page(struct address_space *mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 501 | struct page *newpage, struct page *page, enum migrate_mode mode) |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 502 | { |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 503 | struct buffer_head *bh, *head; |
| 504 | int rc; |
| 505 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 506 | if (!page_has_buffers(page)) |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 507 | return migrate_page(mapping, newpage, page, mode); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 508 | |
| 509 | head = page_buffers(page); |
| 510 | |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 511 | rc = migrate_page_move_mapping(mapping, newpage, page, head, mode); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 512 | |
| 513 | if (rc) |
| 514 | return rc; |
| 515 | |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 516 | /* |
| 517 | * In the async case, migrate_page_move_mapping locked the buffers |
| 518 | * with an IRQ-safe spinlock held. In the sync case, the buffers |
| 519 | * need to be locked now |
| 520 | */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 521 | if (mode != MIGRATE_ASYNC) |
| 522 | BUG_ON(!buffer_migrate_lock_buffers(head, mode)); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 523 | |
| 524 | ClearPagePrivate(page); |
| 525 | set_page_private(newpage, page_private(page)); |
| 526 | set_page_private(page, 0); |
| 527 | put_page(page); |
| 528 | get_page(newpage); |
| 529 | |
| 530 | bh = head; |
| 531 | do { |
| 532 | set_bh_page(bh, newpage, bh_offset(bh)); |
| 533 | bh = bh->b_this_page; |
| 534 | |
| 535 | } while (bh != head); |
| 536 | |
| 537 | SetPagePrivate(newpage); |
| 538 | |
| 539 | migrate_page_copy(newpage, page); |
| 540 | |
| 541 | bh = head; |
| 542 | do { |
| 543 | unlock_buffer(bh); |
| 544 | put_bh(bh); |
| 545 | bh = bh->b_this_page; |
| 546 | |
| 547 | } while (bh != head); |
| 548 | |
| 549 | return 0; |
| 550 | } |
| 551 | EXPORT_SYMBOL(buffer_migrate_page); |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 552 | #endif |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 553 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 554 | /* |
| 555 | * Writeback a page to clean the dirty state |
| 556 | */ |
| 557 | static int writeout(struct address_space *mapping, struct page *page) |
| 558 | { |
| 559 | struct writeback_control wbc = { |
| 560 | .sync_mode = WB_SYNC_NONE, |
| 561 | .nr_to_write = 1, |
| 562 | .range_start = 0, |
| 563 | .range_end = LLONG_MAX, |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 564 | .for_reclaim = 1 |
| 565 | }; |
| 566 | int rc; |
| 567 | |
| 568 | if (!mapping->a_ops->writepage) |
| 569 | /* No write method for the address space */ |
| 570 | return -EINVAL; |
| 571 | |
| 572 | if (!clear_page_dirty_for_io(page)) |
| 573 | /* Someone else already triggered a write */ |
| 574 | return -EAGAIN; |
| 575 | |
| 576 | /* |
| 577 | * A dirty page may imply that the underlying filesystem has |
| 578 | * the page on some queue. So the page must be clean for |
| 579 | * migration. Writeout may mean we loose the lock and the |
| 580 | * page state is no longer what we checked for earlier. |
| 581 | * At this point we know that the migration attempt cannot |
| 582 | * be successful. |
| 583 | */ |
| 584 | remove_migration_ptes(page, page); |
| 585 | |
| 586 | rc = mapping->a_ops->writepage(page, &wbc); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 587 | |
| 588 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
| 589 | /* unlocked. Relock */ |
| 590 | lock_page(page); |
| 591 | |
Hugh Dickins | bda8550 | 2008-11-19 15:36:36 -0800 | [diff] [blame] | 592 | return (rc < 0) ? -EIO : -EAGAIN; |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 593 | } |
| 594 | |
| 595 | /* |
| 596 | * Default handling if a filesystem does not provide a migration function. |
| 597 | */ |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 598 | static int fallback_migrate_page(struct address_space *mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 599 | struct page *newpage, struct page *page, enum migrate_mode mode) |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 600 | { |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 601 | if (PageDirty(page)) { |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 602 | /* Only writeback pages in full synchronous migration */ |
| 603 | if (mode != MIGRATE_SYNC) |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 604 | return -EBUSY; |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 605 | return writeout(mapping, page); |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 606 | } |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 607 | |
| 608 | /* |
| 609 | * Buffers may be managed in a filesystem specific way. |
| 610 | * We must have no buffers or drop them. |
| 611 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 612 | if (page_has_private(page) && |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 613 | !try_to_release_page(page, GFP_KERNEL)) |
| 614 | return -EAGAIN; |
| 615 | |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 616 | return migrate_page(mapping, newpage, page, mode); |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 617 | } |
| 618 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 619 | /* |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 620 | * Move a page to a newly allocated page |
| 621 | * The page is locked and all ptes have been successfully removed. |
| 622 | * |
| 623 | * The new page will have replaced the old page if this function |
| 624 | * is successful. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 625 | * |
| 626 | * Return value: |
| 627 | * < 0 - error code |
| 628 | * == 0 - success |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 629 | */ |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 630 | static int move_to_new_page(struct page *newpage, struct page *page, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 631 | int remap_swapcache, enum migrate_mode mode) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 632 | { |
| 633 | struct address_space *mapping; |
| 634 | int rc; |
| 635 | |
| 636 | /* |
| 637 | * Block others from accessing the page when we get around to |
| 638 | * establishing additional references. We are the only one |
| 639 | * holding a reference to the new page at this point. |
| 640 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 641 | if (!trylock_page(newpage)) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 642 | BUG(); |
| 643 | |
| 644 | /* Prepare mapping for the new page.*/ |
| 645 | newpage->index = page->index; |
| 646 | newpage->mapping = page->mapping; |
Rik van Riel | b2e1853 | 2008-10-18 20:26:30 -0700 | [diff] [blame] | 647 | if (PageSwapBacked(page)) |
| 648 | SetPageSwapBacked(newpage); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 649 | |
| 650 | mapping = page_mapping(page); |
| 651 | if (!mapping) |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 652 | rc = migrate_page(mapping, newpage, page, mode); |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 653 | else if (mapping->a_ops->migratepage) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 654 | /* |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 655 | * Most pages have a mapping and most filesystems provide a |
| 656 | * migratepage callback. Anonymous pages are part of swap |
| 657 | * space which also has its own migratepage callback. This |
| 658 | * is the most common path for page migration. |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 659 | */ |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 660 | rc = mapping->a_ops->migratepage(mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 661 | newpage, page, mode); |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 662 | else |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 663 | rc = fallback_migrate_page(mapping, newpage, page, mode); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 664 | |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 665 | if (rc) { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 666 | newpage->mapping = NULL; |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 667 | } else { |
| 668 | if (remap_swapcache) |
| 669 | remove_migration_ptes(page, newpage); |
Konstantin Khlebnikov | 35512ec | 2012-02-03 15:37:13 -0800 | [diff] [blame] | 670 | page->mapping = NULL; |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 671 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 672 | |
| 673 | unlock_page(newpage); |
| 674 | |
| 675 | return rc; |
| 676 | } |
| 677 | |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 678 | static int __unmap_and_move(struct page *page, struct page *newpage, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 679 | int force, bool offlining, enum migrate_mode mode) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 680 | { |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 681 | int rc = -EAGAIN; |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 682 | int remap_swapcache = 1; |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 683 | int charge = 0; |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 684 | struct mem_cgroup *mem; |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 685 | struct anon_vma *anon_vma = NULL; |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 686 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 687 | if (!trylock_page(page)) { |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 688 | if (!force || mode == MIGRATE_ASYNC) |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 689 | goto out; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 690 | |
| 691 | /* |
| 692 | * It's not safe for direct compaction to call lock_page. |
| 693 | * For example, during page readahead pages are added locked |
| 694 | * to the LRU. Later, when the IO completes the pages are |
| 695 | * marked uptodate and unlocked. However, the queueing |
| 696 | * could be merging multiple pages for one bio (e.g. |
| 697 | * mpage_readpages). If an allocation happens for the |
| 698 | * second or third page, the process can end up locking |
| 699 | * the same page twice and deadlocking. Rather than |
| 700 | * trying to be clever about what pages can be locked, |
| 701 | * avoid the use of lock_page for direct compaction |
| 702 | * altogether. |
| 703 | */ |
| 704 | if (current->flags & PF_MEMALLOC) |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 705 | goto out; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 706 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 707 | lock_page(page); |
| 708 | } |
| 709 | |
Hugh Dickins | 62b61f6 | 2009-12-14 17:59:33 -0800 | [diff] [blame] | 710 | /* |
| 711 | * Only memory hotplug's offline_pages() caller has locked out KSM, |
| 712 | * and can safely migrate a KSM page. The other cases have skipped |
| 713 | * PageKsm along with PageReserved - but it is only now when we have |
| 714 | * the page lock that we can be certain it will not go KSM beneath us |
| 715 | * (KSM will not upgrade a page from PageAnon to PageKsm when it sees |
| 716 | * its pagecount raised, but only here do we take the page lock which |
| 717 | * serializes that). |
| 718 | */ |
| 719 | if (PageKsm(page) && !offlining) { |
| 720 | rc = -EBUSY; |
| 721 | goto unlock; |
| 722 | } |
| 723 | |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 724 | /* charge against new page */ |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 725 | charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL); |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 726 | if (charge == -ENOMEM) { |
| 727 | rc = -ENOMEM; |
| 728 | goto unlock; |
| 729 | } |
| 730 | BUG_ON(charge); |
| 731 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 732 | if (PageWriteback(page)) { |
Andrea Arcangeli | 11bc82d | 2011-03-22 16:33:11 -0700 | [diff] [blame] | 733 | /* |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 734 | * Only in the case of a full syncronous migration is it |
| 735 | * necessary to wait for PageWriteback. In the async case, |
| 736 | * the retry loop is too short and in the sync-light case, |
| 737 | * the overhead of stalling is too much |
Andrea Arcangeli | 11bc82d | 2011-03-22 16:33:11 -0700 | [diff] [blame] | 738 | */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 739 | if (mode != MIGRATE_SYNC) { |
Andrea Arcangeli | 11bc82d | 2011-03-22 16:33:11 -0700 | [diff] [blame] | 740 | rc = -EBUSY; |
| 741 | goto uncharge; |
| 742 | } |
| 743 | if (!force) |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 744 | goto uncharge; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 745 | wait_on_page_writeback(page); |
| 746 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 747 | /* |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 748 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
| 749 | * we cannot notice that anon_vma is freed while we migrates a page. |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 750 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 751 | * of migration. File cache pages are no problem because of page_lock() |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 752 | * File Caches may use write_page() or lock_page() in migration, then, |
| 753 | * just care Anon page here. |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 754 | */ |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 755 | if (PageAnon(page)) { |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 756 | /* |
| 757 | * Only page_lock_anon_vma() understands the subtleties of |
| 758 | * getting a hold on an anon_vma from outside one of its mms. |
| 759 | */ |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 760 | anon_vma = page_get_anon_vma(page); |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 761 | if (anon_vma) { |
| 762 | /* |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 763 | * Anon page |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 764 | */ |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 765 | } else if (PageSwapCache(page)) { |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 766 | /* |
| 767 | * We cannot be sure that the anon_vma of an unmapped |
| 768 | * swapcache page is safe to use because we don't |
| 769 | * know in advance if the VMA that this page belonged |
| 770 | * to still exists. If the VMA and others sharing the |
| 771 | * data have been freed, then the anon_vma could |
| 772 | * already be invalid. |
| 773 | * |
| 774 | * To avoid this possibility, swapcache pages get |
| 775 | * migrated but are not remapped when migration |
| 776 | * completes |
| 777 | */ |
| 778 | remap_swapcache = 0; |
| 779 | } else { |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 780 | goto uncharge; |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 781 | } |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 782 | } |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 783 | |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 784 | /* |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 785 | * Corner case handling: |
| 786 | * 1. When a new swap-cache page is read into, it is added to the LRU |
| 787 | * and treated as swapcache but it has no rmap yet. |
| 788 | * Calling try_to_unmap() against a page->mapping==NULL page will |
| 789 | * trigger a BUG. So handle it here. |
| 790 | * 2. An orphaned page (see truncate_complete_page) might have |
| 791 | * fs-private metadata. The page can be picked up due to memory |
| 792 | * offlining. Everywhere else except page reclaim, the page is |
| 793 | * invisible to the vm, so the page can not be migrated. So try to |
| 794 | * free the metadata, so the page can be freed. |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 795 | */ |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 796 | if (!page->mapping) { |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 797 | VM_BUG_ON(PageAnon(page)); |
| 798 | if (page_has_private(page)) { |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 799 | try_to_free_buffers(page); |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 800 | goto uncharge; |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 801 | } |
Shaohua Li | abfc348 | 2009-09-21 17:01:19 -0700 | [diff] [blame] | 802 | goto skip_unmap; |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 803 | } |
| 804 | |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 805 | /* Establish migration ptes or remove ptes */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 806 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 807 | |
Shaohua Li | abfc348 | 2009-09-21 17:01:19 -0700 | [diff] [blame] | 808 | skip_unmap: |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 809 | if (!page_mapped(page)) |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 810 | rc = move_to_new_page(newpage, page, remap_swapcache, mode); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 811 | |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 812 | if (rc && remap_swapcache) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 813 | remove_migration_ptes(page, page); |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 814 | |
| 815 | /* Drop an anon_vma reference if we took one */ |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 816 | if (anon_vma) |
Peter Zijlstra | 9e60109 | 2011-03-22 16:32:46 -0700 | [diff] [blame] | 817 | put_anon_vma(anon_vma); |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 818 | |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 819 | uncharge: |
| 820 | if (!charge) |
Daisuke Nishimura | 50de1dd | 2011-01-13 15:47:43 -0800 | [diff] [blame] | 821 | mem_cgroup_end_migration(mem, page, newpage, rc == 0); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 822 | unlock: |
| 823 | unlock_page(page); |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 824 | out: |
| 825 | return rc; |
| 826 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 827 | |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 828 | /* |
| 829 | * Obtain the lock on page, remove all ptes and migrate the page |
| 830 | * to the newly allocated page in newpage. |
| 831 | */ |
| 832 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 833 | struct page *page, int force, bool offlining, |
| 834 | enum migrate_mode mode) |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 835 | { |
| 836 | int rc = 0; |
| 837 | int *result = NULL; |
| 838 | struct page *newpage = get_new_page(page, private, &result); |
| 839 | |
| 840 | if (!newpage) |
| 841 | return -ENOMEM; |
| 842 | |
| 843 | if (page_count(page) == 1) { |
| 844 | /* page was freed from under us. So we are done. */ |
| 845 | goto out; |
| 846 | } |
| 847 | |
| 848 | if (unlikely(PageTransHuge(page))) |
| 849 | if (unlikely(split_huge_page(page))) |
| 850 | goto out; |
| 851 | |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 852 | rc = __unmap_and_move(page, newpage, force, offlining, mode); |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 853 | out: |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 854 | if (rc != -EAGAIN) { |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 855 | /* |
| 856 | * A page that has been migrated has all references |
| 857 | * removed and will be freed. A page that has not been |
| 858 | * migrated will have kepts its references and be |
| 859 | * restored. |
| 860 | */ |
| 861 | list_del(&page->lru); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 862 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
Johannes Weiner | 6c0b135 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 863 | page_is_file_cache(page)); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 864 | putback_lru_page(page); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 865 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 866 | /* |
| 867 | * Move the new page to the LRU. If migration was not successful |
| 868 | * then this will free the page. |
| 869 | */ |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 870 | putback_lru_page(newpage); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 871 | if (result) { |
| 872 | if (rc) |
| 873 | *result = rc; |
| 874 | else |
| 875 | *result = page_to_nid(newpage); |
| 876 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 877 | return rc; |
| 878 | } |
| 879 | |
| 880 | /* |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 881 | * Counterpart of unmap_and_move_page() for hugepage migration. |
| 882 | * |
| 883 | * This function doesn't wait the completion of hugepage I/O |
| 884 | * because there is no race between I/O and migration for hugepage. |
| 885 | * Note that currently hugepage I/O occurs only in direct I/O |
| 886 | * where no lock is held and PG_writeback is irrelevant, |
| 887 | * and writeback status of all subpages are counted in the reference |
| 888 | * count of the head page (i.e. if all subpages of a 2MB hugepage are |
| 889 | * under direct I/O, the reference of the head page is 512 and a bit more.) |
| 890 | * This means that when we try to migrate hugepage whose subpages are |
| 891 | * doing direct I/O, some references remain after try_to_unmap() and |
| 892 | * hugepage migration fails without data corruption. |
| 893 | * |
| 894 | * There is also no race when direct I/O is issued on the page under migration, |
| 895 | * because then pte is replaced with migration swap entry and direct I/O code |
| 896 | * will wait in the page fault for migration to complete. |
| 897 | */ |
| 898 | static int unmap_and_move_huge_page(new_page_t get_new_page, |
| 899 | unsigned long private, struct page *hpage, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 900 | int force, bool offlining, |
| 901 | enum migrate_mode mode) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 902 | { |
| 903 | int rc = 0; |
| 904 | int *result = NULL; |
| 905 | struct page *new_hpage = get_new_page(hpage, private, &result); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 906 | struct anon_vma *anon_vma = NULL; |
| 907 | |
| 908 | if (!new_hpage) |
| 909 | return -ENOMEM; |
| 910 | |
| 911 | rc = -EAGAIN; |
| 912 | |
| 913 | if (!trylock_page(hpage)) { |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 914 | if (!force || mode != MIGRATE_SYNC) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 915 | goto out; |
| 916 | lock_page(hpage); |
| 917 | } |
| 918 | |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 919 | if (PageAnon(hpage)) |
| 920 | anon_vma = page_get_anon_vma(hpage); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 921 | |
| 922 | try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
| 923 | |
| 924 | if (!page_mapped(hpage)) |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 925 | rc = move_to_new_page(new_hpage, hpage, 1, mode); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 926 | |
| 927 | if (rc) |
| 928 | remove_migration_ptes(hpage, hpage); |
| 929 | |
Hugh Dickins | fd4a466 | 2011-01-13 15:47:31 -0800 | [diff] [blame] | 930 | if (anon_vma) |
Peter Zijlstra | 9e60109 | 2011-03-22 16:32:46 -0700 | [diff] [blame] | 931 | put_anon_vma(anon_vma); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 932 | unlock_page(hpage); |
| 933 | |
Hillf Danton | 0976133 | 2011-12-08 14:34:20 -0800 | [diff] [blame] | 934 | out: |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 935 | if (rc != -EAGAIN) { |
| 936 | list_del(&hpage->lru); |
| 937 | put_page(hpage); |
| 938 | } |
| 939 | |
| 940 | put_page(new_hpage); |
| 941 | |
| 942 | if (result) { |
| 943 | if (rc) |
| 944 | *result = rc; |
| 945 | else |
| 946 | *result = page_to_nid(new_hpage); |
| 947 | } |
| 948 | return rc; |
| 949 | } |
| 950 | |
| 951 | /* |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 952 | * migrate_pages |
| 953 | * |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 954 | * The function takes one list of pages to migrate and a function |
| 955 | * that determines from the page to be migrated and the private data |
| 956 | * the target of the move and allocates the page. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 957 | * |
| 958 | * The function returns after 10 attempts or if no pages |
| 959 | * are movable anymore because to has become empty |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 960 | * or no retryable pages exist anymore. |
| 961 | * Caller should call putback_lru_pages to return pages to the LRU |
Minchan Kim | 28bd657 | 2011-01-25 15:07:26 -0800 | [diff] [blame] | 962 | * or free list only if ret != 0. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 963 | * |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 964 | * Return: Number of pages not migrated or error code. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 965 | */ |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 966 | int migrate_pages(struct list_head *from, |
Mel Gorman | 7f0f249 | 2011-01-13 15:45:58 -0800 | [diff] [blame] | 967 | new_page_t get_new_page, unsigned long private, bool offlining, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 968 | enum migrate_mode mode) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 969 | { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 970 | int retry = 1; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 971 | int nr_failed = 0; |
| 972 | int pass = 0; |
| 973 | struct page *page; |
| 974 | struct page *page2; |
| 975 | int swapwrite = current->flags & PF_SWAPWRITE; |
| 976 | int rc; |
| 977 | |
Liam Mark | cc2d4bd | 2013-01-16 10:14:40 -0800 | [diff] [blame^] | 978 | trace_migrate_pages_start(mode); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 979 | if (!swapwrite) |
| 980 | current->flags |= PF_SWAPWRITE; |
| 981 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 982 | for(pass = 0; pass < 10 && retry; pass++) { |
| 983 | retry = 0; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 984 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 985 | list_for_each_entry_safe(page, page2, from, lru) { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 986 | cond_resched(); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 987 | |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 988 | rc = unmap_and_move(get_new_page, private, |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 989 | page, pass > 2, offlining, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 990 | mode); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 991 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 992 | switch(rc) { |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 993 | case -ENOMEM: |
| 994 | goto out; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 995 | case -EAGAIN: |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 996 | retry++; |
Liam Mark | cc2d4bd | 2013-01-16 10:14:40 -0800 | [diff] [blame^] | 997 | trace_migrate_retry(retry); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 998 | break; |
| 999 | case 0: |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1000 | break; |
| 1001 | default: |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 1002 | /* Permanent failure */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 1003 | nr_failed++; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1004 | break; |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 1005 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1006 | } |
| 1007 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1008 | rc = 0; |
| 1009 | out: |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1010 | if (!swapwrite) |
| 1011 | current->flags &= ~PF_SWAPWRITE; |
| 1012 | |
Liam Mark | cc2d4bd | 2013-01-16 10:14:40 -0800 | [diff] [blame^] | 1013 | trace_migrate_pages_end(mode); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1014 | if (rc) |
| 1015 | return rc; |
| 1016 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1017 | return nr_failed + retry; |
| 1018 | } |
| 1019 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1020 | int migrate_huge_pages(struct list_head *from, |
Mel Gorman | 7f0f249 | 2011-01-13 15:45:58 -0800 | [diff] [blame] | 1021 | new_page_t get_new_page, unsigned long private, bool offlining, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 1022 | enum migrate_mode mode) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1023 | { |
| 1024 | int retry = 1; |
| 1025 | int nr_failed = 0; |
| 1026 | int pass = 0; |
| 1027 | struct page *page; |
| 1028 | struct page *page2; |
| 1029 | int rc; |
| 1030 | |
| 1031 | for (pass = 0; pass < 10 && retry; pass++) { |
| 1032 | retry = 0; |
| 1033 | |
| 1034 | list_for_each_entry_safe(page, page2, from, lru) { |
| 1035 | cond_resched(); |
| 1036 | |
| 1037 | rc = unmap_and_move_huge_page(get_new_page, |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 1038 | private, page, pass > 2, offlining, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 1039 | mode); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1040 | |
| 1041 | switch(rc) { |
| 1042 | case -ENOMEM: |
| 1043 | goto out; |
| 1044 | case -EAGAIN: |
| 1045 | retry++; |
| 1046 | break; |
| 1047 | case 0: |
| 1048 | break; |
| 1049 | default: |
| 1050 | /* Permanent failure */ |
| 1051 | nr_failed++; |
| 1052 | break; |
| 1053 | } |
| 1054 | } |
| 1055 | } |
| 1056 | rc = 0; |
| 1057 | out: |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1058 | if (rc) |
| 1059 | return rc; |
| 1060 | |
| 1061 | return nr_failed + retry; |
| 1062 | } |
| 1063 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1064 | #ifdef CONFIG_NUMA |
| 1065 | /* |
| 1066 | * Move a list of individual pages |
| 1067 | */ |
| 1068 | struct page_to_node { |
| 1069 | unsigned long addr; |
| 1070 | struct page *page; |
| 1071 | int node; |
| 1072 | int status; |
| 1073 | }; |
| 1074 | |
| 1075 | static struct page *new_page_node(struct page *p, unsigned long private, |
| 1076 | int **result) |
| 1077 | { |
| 1078 | struct page_to_node *pm = (struct page_to_node *)private; |
| 1079 | |
| 1080 | while (pm->node != MAX_NUMNODES && pm->page != p) |
| 1081 | pm++; |
| 1082 | |
| 1083 | if (pm->node == MAX_NUMNODES) |
| 1084 | return NULL; |
| 1085 | |
| 1086 | *result = &pm->status; |
| 1087 | |
Mel Gorman | 6484eb3 | 2009-06-16 15:31:54 -0700 | [diff] [blame] | 1088 | return alloc_pages_exact_node(pm->node, |
Mel Gorman | 769848c | 2007-07-17 04:03:05 -0700 | [diff] [blame] | 1089 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * Move a set of pages as indicated in the pm array. The addr |
| 1094 | * field must be set to the virtual address of the page to be moved |
| 1095 | * and the node number must contain a valid target node. |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1096 | * The pm array ends with node = MAX_NUMNODES. |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1097 | */ |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1098 | static int do_move_page_to_node_array(struct mm_struct *mm, |
| 1099 | struct page_to_node *pm, |
| 1100 | int migrate_all) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1101 | { |
| 1102 | int err; |
| 1103 | struct page_to_node *pp; |
| 1104 | LIST_HEAD(pagelist); |
| 1105 | |
| 1106 | down_read(&mm->mmap_sem); |
| 1107 | |
| 1108 | /* |
| 1109 | * Build a list of pages to migrate |
| 1110 | */ |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1111 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
| 1112 | struct vm_area_struct *vma; |
| 1113 | struct page *page; |
| 1114 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1115 | err = -EFAULT; |
| 1116 | vma = find_vma(mm, pp->addr); |
Gleb Natapov | 70384dc | 2010-10-26 14:22:07 -0700 | [diff] [blame] | 1117 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1118 | goto set_status; |
| 1119 | |
Andrea Arcangeli | 500d65d | 2011-01-13 15:46:55 -0800 | [diff] [blame] | 1120 | page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT); |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1121 | |
| 1122 | err = PTR_ERR(page); |
| 1123 | if (IS_ERR(page)) |
| 1124 | goto set_status; |
| 1125 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1126 | err = -ENOENT; |
| 1127 | if (!page) |
| 1128 | goto set_status; |
| 1129 | |
Hugh Dickins | 62b61f6 | 2009-12-14 17:59:33 -0800 | [diff] [blame] | 1130 | /* Use PageReserved to check for zero page */ |
| 1131 | if (PageReserved(page) || PageKsm(page)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1132 | goto put_and_set; |
| 1133 | |
| 1134 | pp->page = page; |
| 1135 | err = page_to_nid(page); |
| 1136 | |
| 1137 | if (err == pp->node) |
| 1138 | /* |
| 1139 | * Node already in the right place |
| 1140 | */ |
| 1141 | goto put_and_set; |
| 1142 | |
| 1143 | err = -EACCES; |
| 1144 | if (page_mapcount(page) > 1 && |
| 1145 | !migrate_all) |
| 1146 | goto put_and_set; |
| 1147 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1148 | err = isolate_lru_page(page); |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 1149 | if (!err) { |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1150 | list_add_tail(&page->lru, &pagelist); |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 1151 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
| 1152 | page_is_file_cache(page)); |
| 1153 | } |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1154 | put_and_set: |
| 1155 | /* |
| 1156 | * Either remove the duplicate refcount from |
| 1157 | * isolate_lru_page() or drop the page ref if it was |
| 1158 | * not isolated. |
| 1159 | */ |
| 1160 | put_page(page); |
| 1161 | set_status: |
| 1162 | pp->status = err; |
| 1163 | } |
| 1164 | |
Brice Goglin | e78bbfa | 2008-10-18 20:27:15 -0700 | [diff] [blame] | 1165 | err = 0; |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1166 | if (!list_empty(&pagelist)) { |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1167 | err = migrate_pages(&pagelist, new_page_node, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 1168 | (unsigned long)pm, 0, MIGRATE_SYNC); |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1169 | if (err) |
| 1170 | putback_lru_pages(&pagelist); |
| 1171 | } |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1172 | |
| 1173 | up_read(&mm->mmap_sem); |
| 1174 | return err; |
| 1175 | } |
| 1176 | |
| 1177 | /* |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1178 | * Migrate an array of page address onto an array of nodes and fill |
| 1179 | * the corresponding array of status. |
| 1180 | */ |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1181 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1182 | unsigned long nr_pages, |
| 1183 | const void __user * __user *pages, |
| 1184 | const int __user *nodes, |
| 1185 | int __user *status, int flags) |
| 1186 | { |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1187 | struct page_to_node *pm; |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1188 | unsigned long chunk_nr_pages; |
| 1189 | unsigned long chunk_start; |
| 1190 | int err; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1191 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1192 | err = -ENOMEM; |
| 1193 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); |
| 1194 | if (!pm) |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1195 | goto out; |
Brice Goglin | 35282a2 | 2009-06-16 15:32:43 -0700 | [diff] [blame] | 1196 | |
| 1197 | migrate_prep(); |
| 1198 | |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1199 | /* |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1200 | * Store a chunk of page_to_node array in a page, |
| 1201 | * but keep the last one as a marker |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1202 | */ |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1203 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1204 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1205 | for (chunk_start = 0; |
| 1206 | chunk_start < nr_pages; |
| 1207 | chunk_start += chunk_nr_pages) { |
| 1208 | int j; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1209 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1210 | if (chunk_start + chunk_nr_pages > nr_pages) |
| 1211 | chunk_nr_pages = nr_pages - chunk_start; |
| 1212 | |
| 1213 | /* fill the chunk pm with addrs and nodes from user-space */ |
| 1214 | for (j = 0; j < chunk_nr_pages; j++) { |
| 1215 | const void __user *p; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1216 | int node; |
| 1217 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1218 | err = -EFAULT; |
| 1219 | if (get_user(p, pages + j + chunk_start)) |
| 1220 | goto out_pm; |
| 1221 | pm[j].addr = (unsigned long) p; |
| 1222 | |
| 1223 | if (get_user(node, nodes + j + chunk_start)) |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1224 | goto out_pm; |
| 1225 | |
| 1226 | err = -ENODEV; |
Linus Torvalds | 6f5a55f | 2010-02-05 16:16:50 -0800 | [diff] [blame] | 1227 | if (node < 0 || node >= MAX_NUMNODES) |
| 1228 | goto out_pm; |
| 1229 | |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1230 | if (!node_state(node, N_HIGH_MEMORY)) |
| 1231 | goto out_pm; |
| 1232 | |
| 1233 | err = -EACCES; |
| 1234 | if (!node_isset(node, task_nodes)) |
| 1235 | goto out_pm; |
| 1236 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1237 | pm[j].node = node; |
| 1238 | } |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1239 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1240 | /* End marker for this chunk */ |
| 1241 | pm[chunk_nr_pages].node = MAX_NUMNODES; |
| 1242 | |
| 1243 | /* Migrate this chunk */ |
| 1244 | err = do_move_page_to_node_array(mm, pm, |
| 1245 | flags & MPOL_MF_MOVE_ALL); |
| 1246 | if (err < 0) |
| 1247 | goto out_pm; |
| 1248 | |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1249 | /* Return status information */ |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1250 | for (j = 0; j < chunk_nr_pages; j++) |
| 1251 | if (put_user(pm[j].status, status + j + chunk_start)) { |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1252 | err = -EFAULT; |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1253 | goto out_pm; |
| 1254 | } |
| 1255 | } |
| 1256 | err = 0; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1257 | |
| 1258 | out_pm: |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1259 | free_page((unsigned long)pm); |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1260 | out: |
| 1261 | return err; |
| 1262 | } |
| 1263 | |
| 1264 | /* |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1265 | * Determine the nodes of an array of pages and store it in an array of status. |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1266 | */ |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1267 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
| 1268 | const void __user **pages, int *status) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1269 | { |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1270 | unsigned long i; |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1271 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1272 | down_read(&mm->mmap_sem); |
| 1273 | |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1274 | for (i = 0; i < nr_pages; i++) { |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1275 | unsigned long addr = (unsigned long)(*pages); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1276 | struct vm_area_struct *vma; |
| 1277 | struct page *page; |
KOSAKI Motohiro | c095adb | 2008-12-16 16:06:43 +0900 | [diff] [blame] | 1278 | int err = -EFAULT; |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1279 | |
| 1280 | vma = find_vma(mm, addr); |
Gleb Natapov | 70384dc | 2010-10-26 14:22:07 -0700 | [diff] [blame] | 1281 | if (!vma || addr < vma->vm_start) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1282 | goto set_status; |
| 1283 | |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1284 | page = follow_page(vma, addr, 0); |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1285 | |
| 1286 | err = PTR_ERR(page); |
| 1287 | if (IS_ERR(page)) |
| 1288 | goto set_status; |
| 1289 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1290 | err = -ENOENT; |
| 1291 | /* Use PageReserved to check for zero page */ |
Hugh Dickins | 62b61f6 | 2009-12-14 17:59:33 -0800 | [diff] [blame] | 1292 | if (!page || PageReserved(page) || PageKsm(page)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1293 | goto set_status; |
| 1294 | |
| 1295 | err = page_to_nid(page); |
| 1296 | set_status: |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1297 | *status = err; |
| 1298 | |
| 1299 | pages++; |
| 1300 | status++; |
| 1301 | } |
| 1302 | |
| 1303 | up_read(&mm->mmap_sem); |
| 1304 | } |
| 1305 | |
| 1306 | /* |
| 1307 | * Determine the nodes of a user array of pages and store it in |
| 1308 | * a user array of status. |
| 1309 | */ |
| 1310 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, |
| 1311 | const void __user * __user *pages, |
| 1312 | int __user *status) |
| 1313 | { |
| 1314 | #define DO_PAGES_STAT_CHUNK_NR 16 |
| 1315 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; |
| 1316 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1317 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1318 | while (nr_pages) { |
| 1319 | unsigned long chunk_nr; |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1320 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1321 | chunk_nr = nr_pages; |
| 1322 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) |
| 1323 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; |
| 1324 | |
| 1325 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) |
| 1326 | break; |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1327 | |
| 1328 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); |
| 1329 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1330 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
| 1331 | break; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1332 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1333 | pages += chunk_nr; |
| 1334 | status += chunk_nr; |
| 1335 | nr_pages -= chunk_nr; |
| 1336 | } |
| 1337 | return nr_pages ? -EFAULT : 0; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1338 | } |
| 1339 | |
| 1340 | /* |
| 1341 | * Move a list of pages in the address space of the currently executing |
| 1342 | * process. |
| 1343 | */ |
Heiko Carstens | 938bb9f | 2009-01-14 14:14:30 +0100 | [diff] [blame] | 1344 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
| 1345 | const void __user * __user *, pages, |
| 1346 | const int __user *, nodes, |
| 1347 | int __user *, status, int, flags) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1348 | { |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1349 | const struct cred *cred = current_cred(), *tcred; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1350 | struct task_struct *task; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1351 | struct mm_struct *mm; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1352 | int err; |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1353 | nodemask_t task_nodes; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1354 | |
| 1355 | /* Check flags */ |
| 1356 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) |
| 1357 | return -EINVAL; |
| 1358 | |
| 1359 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
| 1360 | return -EPERM; |
| 1361 | |
| 1362 | /* Find the mm_struct */ |
Greg Thelen | a879bf5 | 2011-02-25 14:44:13 -0800 | [diff] [blame] | 1363 | rcu_read_lock(); |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 1364 | task = pid ? find_task_by_vpid(pid) : current; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1365 | if (!task) { |
Greg Thelen | a879bf5 | 2011-02-25 14:44:13 -0800 | [diff] [blame] | 1366 | rcu_read_unlock(); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1367 | return -ESRCH; |
| 1368 | } |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1369 | get_task_struct(task); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1370 | |
| 1371 | /* |
| 1372 | * Check if this process has the right to modify the specified |
| 1373 | * process. The right exists if the process has administrative |
| 1374 | * capabilities, superuser privileges or the same |
| 1375 | * userid as the target process. |
| 1376 | */ |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1377 | tcred = __task_cred(task); |
David Howells | b6dff3e | 2008-11-14 10:39:16 +1100 | [diff] [blame] | 1378 | if (cred->euid != tcred->suid && cred->euid != tcred->uid && |
| 1379 | cred->uid != tcred->suid && cred->uid != tcred->uid && |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1380 | !capable(CAP_SYS_NICE)) { |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1381 | rcu_read_unlock(); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1382 | err = -EPERM; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1383 | goto out; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1384 | } |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1385 | rcu_read_unlock(); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1386 | |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1387 | err = security_task_movememory(task); |
| 1388 | if (err) |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1389 | goto out; |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1390 | |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1391 | task_nodes = cpuset_mems_allowed(task); |
| 1392 | mm = get_task_mm(task); |
| 1393 | put_task_struct(task); |
| 1394 | |
Sasha Levin | 6e8b09e | 2012-04-25 16:01:53 -0700 | [diff] [blame] | 1395 | if (!mm) |
| 1396 | return -EINVAL; |
| 1397 | |
| 1398 | if (nodes) |
| 1399 | err = do_pages_move(mm, task_nodes, nr_pages, pages, |
| 1400 | nodes, status, flags); |
| 1401 | else |
| 1402 | err = do_pages_stat(mm, nr_pages, pages, status); |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1403 | |
| 1404 | mmput(mm); |
| 1405 | return err; |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1406 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1407 | out: |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1408 | put_task_struct(task); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1409 | return err; |
| 1410 | } |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1411 | |
Christoph Lameter | 7b2259b | 2006-06-25 05:46:48 -0700 | [diff] [blame] | 1412 | /* |
| 1413 | * Call migration functions in the vma_ops that may prepare |
| 1414 | * memory in a vm for migration. migration functions may perform |
| 1415 | * the migration for vmas that do not have an underlying page struct. |
| 1416 | */ |
| 1417 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, |
| 1418 | const nodemask_t *from, unsigned long flags) |
| 1419 | { |
| 1420 | struct vm_area_struct *vma; |
| 1421 | int err = 0; |
| 1422 | |
Daisuke Nishimura | 1001c9f | 2009-02-11 13:04:18 -0800 | [diff] [blame] | 1423 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { |
Christoph Lameter | 7b2259b | 2006-06-25 05:46:48 -0700 | [diff] [blame] | 1424 | if (vma->vm_ops && vma->vm_ops->migrate) { |
| 1425 | err = vma->vm_ops->migrate(vma, to, from, flags); |
| 1426 | if (err) |
| 1427 | break; |
| 1428 | } |
| 1429 | } |
| 1430 | return err; |
| 1431 | } |
Gerald Schaefer | 83d1674 | 2008-07-23 21:28:22 -0700 | [diff] [blame] | 1432 | #endif |