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