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> |
| 12 | * Christoph Lameter <clameter@sgi.com> |
| 13 | */ |
| 14 | |
| 15 | #include <linux/migrate.h> |
| 16 | #include <linux/module.h> |
| 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> |
| 22 | #include <linux/pagevec.h> |
| 23 | #include <linux/rmap.h> |
| 24 | #include <linux/topology.h> |
| 25 | #include <linux/cpu.h> |
| 26 | #include <linux/cpuset.h> |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 27 | #include <linux/writeback.h> |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 28 | #include <linux/mempolicy.h> |
| 29 | #include <linux/vmalloc.h> |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 30 | #include <linux/security.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 31 | |
| 32 | #include "internal.h" |
| 33 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 34 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
| 35 | |
| 36 | /* |
| 37 | * Isolate one page from the LRU lists. If successful put it onto |
| 38 | * the indicated list with elevated page count. |
| 39 | * |
| 40 | * Result: |
| 41 | * -EBUSY: page not on LRU list |
| 42 | * 0: page removed from LRU list and added to the specified list. |
| 43 | */ |
| 44 | int isolate_lru_page(struct page *page, struct list_head *pagelist) |
| 45 | { |
| 46 | int ret = -EBUSY; |
| 47 | |
| 48 | if (PageLRU(page)) { |
| 49 | struct zone *zone = page_zone(page); |
| 50 | |
| 51 | spin_lock_irq(&zone->lru_lock); |
KAMEZAWA Hiroyuki | 3dd9fe8 | 2007-07-26 10:41:08 -0700 | [diff] [blame] | 52 | if (PageLRU(page) && get_page_unless_zero(page)) { |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 53 | ret = 0; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 54 | ClearPageLRU(page); |
| 55 | if (PageActive(page)) |
| 56 | del_page_from_active_list(zone, page); |
| 57 | else |
| 58 | del_page_from_inactive_list(zone, page); |
| 59 | list_add_tail(&page->lru, pagelist); |
| 60 | } |
| 61 | spin_unlock_irq(&zone->lru_lock); |
| 62 | } |
| 63 | return ret; |
| 64 | } |
| 65 | |
| 66 | /* |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 67 | * migrate_prep() needs to be called before we start compiling a list of pages |
| 68 | * to be migrated using isolate_lru_page(). |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 69 | */ |
| 70 | int migrate_prep(void) |
| 71 | { |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 72 | /* |
| 73 | * Clear the LRU lists so pages can be isolated. |
| 74 | * Note that pages may be moved off the LRU after we have |
| 75 | * drained them. Those pages will fail to migrate like other |
| 76 | * pages that may be busy. |
| 77 | */ |
| 78 | lru_add_drain_all(); |
| 79 | |
| 80 | return 0; |
| 81 | } |
| 82 | |
| 83 | static inline void move_to_lru(struct page *page) |
| 84 | { |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 85 | if (PageActive(page)) { |
| 86 | /* |
| 87 | * lru_cache_add_active checks that |
| 88 | * the PG_active bit is off. |
| 89 | */ |
| 90 | ClearPageActive(page); |
| 91 | lru_cache_add_active(page); |
| 92 | } else { |
| 93 | lru_cache_add(page); |
| 94 | } |
| 95 | put_page(page); |
| 96 | } |
| 97 | |
| 98 | /* |
| 99 | * Add isolated pages on the list back to the LRU. |
| 100 | * |
| 101 | * returns the number of pages put back. |
| 102 | */ |
| 103 | int putback_lru_pages(struct list_head *l) |
| 104 | { |
| 105 | struct page *page; |
| 106 | struct page *page2; |
| 107 | int count = 0; |
| 108 | |
| 109 | list_for_each_entry_safe(page, page2, l, lru) { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 110 | list_del(&page->lru); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 111 | move_to_lru(page); |
| 112 | count++; |
| 113 | } |
| 114 | return count; |
| 115 | } |
| 116 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 117 | static inline int is_swap_pte(pte_t pte) |
| 118 | { |
| 119 | return !pte_none(pte) && !pte_present(pte) && !pte_file(pte); |
| 120 | } |
| 121 | |
| 122 | /* |
| 123 | * Restore a potential migration pte to a working pte entry |
| 124 | */ |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 125 | static void remove_migration_pte(struct vm_area_struct *vma, |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 126 | struct page *old, struct page *new) |
| 127 | { |
| 128 | struct mm_struct *mm = vma->vm_mm; |
| 129 | swp_entry_t entry; |
| 130 | pgd_t *pgd; |
| 131 | pud_t *pud; |
| 132 | pmd_t *pmd; |
| 133 | pte_t *ptep, pte; |
| 134 | spinlock_t *ptl; |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 135 | unsigned long addr = page_address_in_vma(new, vma); |
| 136 | |
| 137 | if (addr == -EFAULT) |
| 138 | return; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 139 | |
| 140 | pgd = pgd_offset(mm, addr); |
| 141 | if (!pgd_present(*pgd)) |
| 142 | return; |
| 143 | |
| 144 | pud = pud_offset(pgd, addr); |
| 145 | if (!pud_present(*pud)) |
| 146 | return; |
| 147 | |
| 148 | pmd = pmd_offset(pud, addr); |
| 149 | if (!pmd_present(*pmd)) |
| 150 | return; |
| 151 | |
| 152 | ptep = pte_offset_map(pmd, addr); |
| 153 | |
| 154 | if (!is_swap_pte(*ptep)) { |
| 155 | pte_unmap(ptep); |
| 156 | return; |
| 157 | } |
| 158 | |
| 159 | ptl = pte_lockptr(mm, pmd); |
| 160 | spin_lock(ptl); |
| 161 | pte = *ptep; |
| 162 | if (!is_swap_pte(pte)) |
| 163 | goto out; |
| 164 | |
| 165 | entry = pte_to_swp_entry(pte); |
| 166 | |
| 167 | if (!is_migration_entry(entry) || migration_entry_to_page(entry) != old) |
| 168 | goto out; |
| 169 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 170 | get_page(new); |
| 171 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); |
| 172 | if (is_write_migration_entry(entry)) |
| 173 | pte = pte_mkwrite(pte); |
KAMEZAWA Hiroyuki | 97ee052 | 2007-10-16 01:25:43 -0700 | [diff] [blame^] | 174 | flush_cache_page(vma, addr, pte_pfn(pte)); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 175 | set_pte_at(mm, addr, ptep, pte); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 176 | |
| 177 | if (PageAnon(new)) |
| 178 | page_add_anon_rmap(new, vma, addr); |
| 179 | else |
| 180 | page_add_file_rmap(new); |
| 181 | |
| 182 | /* No need to invalidate - it was non-present before */ |
| 183 | update_mmu_cache(vma, addr, pte); |
| 184 | lazy_mmu_prot_update(pte); |
| 185 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 186 | out: |
| 187 | pte_unmap_unlock(ptep, ptl); |
| 188 | } |
| 189 | |
| 190 | /* |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 191 | * Note that remove_file_migration_ptes will only work on regular mappings, |
| 192 | * Nonlinear mappings do not use migration entries. |
| 193 | */ |
| 194 | static void remove_file_migration_ptes(struct page *old, struct page *new) |
| 195 | { |
| 196 | struct vm_area_struct *vma; |
| 197 | struct address_space *mapping = page_mapping(new); |
| 198 | struct prio_tree_iter iter; |
| 199 | pgoff_t pgoff = new->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 200 | |
| 201 | if (!mapping) |
| 202 | return; |
| 203 | |
| 204 | spin_lock(&mapping->i_mmap_lock); |
| 205 | |
| 206 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) |
| 207 | remove_migration_pte(vma, old, new); |
| 208 | |
| 209 | spin_unlock(&mapping->i_mmap_lock); |
| 210 | } |
| 211 | |
| 212 | /* |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 213 | * Must hold mmap_sem lock on at least one of the vmas containing |
| 214 | * the page so that the anon_vma cannot vanish. |
| 215 | */ |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 216 | static void remove_anon_migration_ptes(struct page *old, struct page *new) |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 217 | { |
| 218 | struct anon_vma *anon_vma; |
| 219 | struct vm_area_struct *vma; |
| 220 | unsigned long mapping; |
| 221 | |
| 222 | mapping = (unsigned long)new->mapping; |
| 223 | |
| 224 | if (!mapping || (mapping & PAGE_MAPPING_ANON) == 0) |
| 225 | return; |
| 226 | |
| 227 | /* |
| 228 | * We hold the mmap_sem lock. So no need to call page_lock_anon_vma. |
| 229 | */ |
| 230 | anon_vma = (struct anon_vma *) (mapping - PAGE_MAPPING_ANON); |
| 231 | spin_lock(&anon_vma->lock); |
| 232 | |
| 233 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 234 | remove_migration_pte(vma, old, new); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 235 | |
| 236 | spin_unlock(&anon_vma->lock); |
| 237 | } |
| 238 | |
| 239 | /* |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 240 | * Get rid of all migration entries and replace them by |
| 241 | * references to the indicated page. |
| 242 | */ |
| 243 | static void remove_migration_ptes(struct page *old, struct page *new) |
| 244 | { |
| 245 | if (PageAnon(new)) |
| 246 | remove_anon_migration_ptes(old, new); |
| 247 | else |
| 248 | remove_file_migration_ptes(old, new); |
| 249 | } |
| 250 | |
| 251 | /* |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 252 | * Something used the pte of a page under migration. We need to |
| 253 | * get to the page and wait until migration is finished. |
| 254 | * When we return from this function the fault will be retried. |
| 255 | * |
| 256 | * This function is called from do_swap_page(). |
| 257 | */ |
| 258 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, |
| 259 | unsigned long address) |
| 260 | { |
| 261 | pte_t *ptep, pte; |
| 262 | spinlock_t *ptl; |
| 263 | swp_entry_t entry; |
| 264 | struct page *page; |
| 265 | |
| 266 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
| 267 | pte = *ptep; |
| 268 | if (!is_swap_pte(pte)) |
| 269 | goto out; |
| 270 | |
| 271 | entry = pte_to_swp_entry(pte); |
| 272 | if (!is_migration_entry(entry)) |
| 273 | goto out; |
| 274 | |
| 275 | page = migration_entry_to_page(entry); |
| 276 | |
| 277 | get_page(page); |
| 278 | pte_unmap_unlock(ptep, ptl); |
| 279 | wait_on_page_locked(page); |
| 280 | put_page(page); |
| 281 | return; |
| 282 | out: |
| 283 | pte_unmap_unlock(ptep, ptl); |
| 284 | } |
| 285 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 286 | /* |
Christoph Lameter | c3fcf8a | 2006-06-23 02:03:32 -0700 | [diff] [blame] | 287 | * Replace the page in the mapping. |
Christoph Lameter | 5b5c712 | 2006-06-23 02:03:29 -0700 | [diff] [blame] | 288 | * |
| 289 | * The number of remaining references must be: |
| 290 | * 1 for anonymous pages without a mapping |
| 291 | * 2 for pages with a mapping |
| 292 | * 3 for pages with a mapping and PagePrivate set. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 293 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 294 | static int migrate_page_move_mapping(struct address_space *mapping, |
| 295 | struct page *newpage, struct page *page) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 296 | { |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 297 | void **pslot; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 298 | |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 299 | if (!mapping) { |
Christoph Lameter | 0e8c7d0 | 2007-04-23 14:41:09 -0700 | [diff] [blame] | 300 | /* Anonymous page without mapping */ |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 301 | if (page_count(page) != 1) |
| 302 | return -EAGAIN; |
| 303 | return 0; |
| 304 | } |
| 305 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 306 | write_lock_irq(&mapping->tree_lock); |
| 307 | |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 308 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
| 309 | page_index(page)); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 310 | |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 311 | if (page_count(page) != 2 + !!PagePrivate(page) || |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 312 | (struct page *)radix_tree_deref_slot(pslot) != page) { |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 313 | write_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | e23ca00 | 2006-04-10 22:52:57 -0700 | [diff] [blame] | 314 | return -EAGAIN; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 315 | } |
| 316 | |
| 317 | /* |
| 318 | * Now we know that no one else is looking at the page. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 319 | */ |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 320 | get_page(newpage); /* add cache reference */ |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 321 | #ifdef CONFIG_SWAP |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 322 | if (PageSwapCache(page)) { |
| 323 | SetPageSwapCache(newpage); |
| 324 | set_page_private(newpage, page_private(page)); |
| 325 | } |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 326 | #endif |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 327 | |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 328 | radix_tree_replace_slot(pslot, newpage); |
| 329 | |
| 330 | /* |
| 331 | * Drop cache reference from old page. |
| 332 | * We know this isn't the last reference. |
| 333 | */ |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 334 | __put_page(page); |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 335 | |
Christoph Lameter | 0e8c7d0 | 2007-04-23 14:41:09 -0700 | [diff] [blame] | 336 | /* |
| 337 | * If moved to a different zone then also account |
| 338 | * the page for that zone. Other VM counters will be |
| 339 | * taken care of when we establish references to the |
| 340 | * new page and drop references to the old page. |
| 341 | * |
| 342 | * Note that anonymous pages are accounted for |
| 343 | * via NR_FILE_PAGES and NR_ANON_PAGES if they |
| 344 | * are mapped to swap space. |
| 345 | */ |
| 346 | __dec_zone_page_state(page, NR_FILE_PAGES); |
| 347 | __inc_zone_page_state(newpage, NR_FILE_PAGES); |
| 348 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 349 | write_unlock_irq(&mapping->tree_lock); |
| 350 | |
| 351 | return 0; |
| 352 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 353 | |
| 354 | /* |
| 355 | * Copy the page to its new location |
| 356 | */ |
Christoph Lameter | e7340f7 | 2006-06-23 02:03:29 -0700 | [diff] [blame] | 357 | static void migrate_page_copy(struct page *newpage, struct page *page) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 358 | { |
| 359 | copy_highpage(newpage, page); |
| 360 | |
| 361 | if (PageError(page)) |
| 362 | SetPageError(newpage); |
| 363 | if (PageReferenced(page)) |
| 364 | SetPageReferenced(newpage); |
| 365 | if (PageUptodate(page)) |
| 366 | SetPageUptodate(newpage); |
| 367 | if (PageActive(page)) |
| 368 | SetPageActive(newpage); |
| 369 | if (PageChecked(page)) |
| 370 | SetPageChecked(newpage); |
| 371 | if (PageMappedToDisk(page)) |
| 372 | SetPageMappedToDisk(newpage); |
| 373 | |
| 374 | if (PageDirty(page)) { |
| 375 | clear_page_dirty_for_io(page); |
| 376 | set_page_dirty(newpage); |
| 377 | } |
| 378 | |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 379 | #ifdef CONFIG_SWAP |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 380 | ClearPageSwapCache(page); |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 381 | #endif |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 382 | ClearPageActive(page); |
| 383 | ClearPagePrivate(page); |
| 384 | set_page_private(page, 0); |
| 385 | page->mapping = NULL; |
| 386 | |
| 387 | /* |
| 388 | * If any waiters have accumulated on the new page then |
| 389 | * wake them up. |
| 390 | */ |
| 391 | if (PageWriteback(newpage)) |
| 392 | end_page_writeback(newpage); |
| 393 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 394 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 395 | /************************************************************ |
| 396 | * Migration functions |
| 397 | ***********************************************************/ |
| 398 | |
| 399 | /* Always fail migration. Used for mappings that are not movable */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 400 | int fail_migrate_page(struct address_space *mapping, |
| 401 | struct page *newpage, struct page *page) |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 402 | { |
| 403 | return -EIO; |
| 404 | } |
| 405 | EXPORT_SYMBOL(fail_migrate_page); |
| 406 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 407 | /* |
| 408 | * Common logic to directly migrate a single page suitable for |
| 409 | * pages that do not use PagePrivate. |
| 410 | * |
| 411 | * Pages are locked upon entry and exit. |
| 412 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 413 | int migrate_page(struct address_space *mapping, |
| 414 | struct page *newpage, struct page *page) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 415 | { |
| 416 | int rc; |
| 417 | |
| 418 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ |
| 419 | |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 420 | rc = migrate_page_move_mapping(mapping, newpage, page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 421 | |
| 422 | if (rc) |
| 423 | return rc; |
| 424 | |
| 425 | migrate_page_copy(newpage, page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 426 | return 0; |
| 427 | } |
| 428 | EXPORT_SYMBOL(migrate_page); |
| 429 | |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 430 | #ifdef CONFIG_BLOCK |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 431 | /* |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 432 | * Migration function for pages with buffers. This function can only be used |
| 433 | * if the underlying filesystem guarantees that no other references to "page" |
| 434 | * exist. |
| 435 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 436 | int buffer_migrate_page(struct address_space *mapping, |
| 437 | struct page *newpage, struct page *page) |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 438 | { |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 439 | struct buffer_head *bh, *head; |
| 440 | int rc; |
| 441 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 442 | if (!page_has_buffers(page)) |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 443 | return migrate_page(mapping, newpage, page); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 444 | |
| 445 | head = page_buffers(page); |
| 446 | |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 447 | rc = migrate_page_move_mapping(mapping, newpage, page); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 448 | |
| 449 | if (rc) |
| 450 | return rc; |
| 451 | |
| 452 | bh = head; |
| 453 | do { |
| 454 | get_bh(bh); |
| 455 | lock_buffer(bh); |
| 456 | bh = bh->b_this_page; |
| 457 | |
| 458 | } while (bh != head); |
| 459 | |
| 460 | ClearPagePrivate(page); |
| 461 | set_page_private(newpage, page_private(page)); |
| 462 | set_page_private(page, 0); |
| 463 | put_page(page); |
| 464 | get_page(newpage); |
| 465 | |
| 466 | bh = head; |
| 467 | do { |
| 468 | set_bh_page(bh, newpage, bh_offset(bh)); |
| 469 | bh = bh->b_this_page; |
| 470 | |
| 471 | } while (bh != head); |
| 472 | |
| 473 | SetPagePrivate(newpage); |
| 474 | |
| 475 | migrate_page_copy(newpage, page); |
| 476 | |
| 477 | bh = head; |
| 478 | do { |
| 479 | unlock_buffer(bh); |
| 480 | put_bh(bh); |
| 481 | bh = bh->b_this_page; |
| 482 | |
| 483 | } while (bh != head); |
| 484 | |
| 485 | return 0; |
| 486 | } |
| 487 | EXPORT_SYMBOL(buffer_migrate_page); |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 488 | #endif |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 489 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 490 | /* |
| 491 | * Writeback a page to clean the dirty state |
| 492 | */ |
| 493 | static int writeout(struct address_space *mapping, struct page *page) |
| 494 | { |
| 495 | struct writeback_control wbc = { |
| 496 | .sync_mode = WB_SYNC_NONE, |
| 497 | .nr_to_write = 1, |
| 498 | .range_start = 0, |
| 499 | .range_end = LLONG_MAX, |
| 500 | .nonblocking = 1, |
| 501 | .for_reclaim = 1 |
| 502 | }; |
| 503 | int rc; |
| 504 | |
| 505 | if (!mapping->a_ops->writepage) |
| 506 | /* No write method for the address space */ |
| 507 | return -EINVAL; |
| 508 | |
| 509 | if (!clear_page_dirty_for_io(page)) |
| 510 | /* Someone else already triggered a write */ |
| 511 | return -EAGAIN; |
| 512 | |
| 513 | /* |
| 514 | * A dirty page may imply that the underlying filesystem has |
| 515 | * the page on some queue. So the page must be clean for |
| 516 | * migration. Writeout may mean we loose the lock and the |
| 517 | * page state is no longer what we checked for earlier. |
| 518 | * At this point we know that the migration attempt cannot |
| 519 | * be successful. |
| 520 | */ |
| 521 | remove_migration_ptes(page, page); |
| 522 | |
| 523 | rc = mapping->a_ops->writepage(page, &wbc); |
| 524 | if (rc < 0) |
| 525 | /* I/O Error writing */ |
| 526 | return -EIO; |
| 527 | |
| 528 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
| 529 | /* unlocked. Relock */ |
| 530 | lock_page(page); |
| 531 | |
| 532 | return -EAGAIN; |
| 533 | } |
| 534 | |
| 535 | /* |
| 536 | * Default handling if a filesystem does not provide a migration function. |
| 537 | */ |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 538 | static int fallback_migrate_page(struct address_space *mapping, |
| 539 | struct page *newpage, struct page *page) |
| 540 | { |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 541 | if (PageDirty(page)) |
| 542 | return writeout(mapping, page); |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 543 | |
| 544 | /* |
| 545 | * Buffers may be managed in a filesystem specific way. |
| 546 | * We must have no buffers or drop them. |
| 547 | */ |
David Howells | b398f6b | 2006-08-29 19:05:58 +0100 | [diff] [blame] | 548 | if (PagePrivate(page) && |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 549 | !try_to_release_page(page, GFP_KERNEL)) |
| 550 | return -EAGAIN; |
| 551 | |
| 552 | return migrate_page(mapping, newpage, page); |
| 553 | } |
| 554 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 555 | /* |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 556 | * Move a page to a newly allocated page |
| 557 | * The page is locked and all ptes have been successfully removed. |
| 558 | * |
| 559 | * The new page will have replaced the old page if this function |
| 560 | * is successful. |
| 561 | */ |
| 562 | static int move_to_new_page(struct page *newpage, struct page *page) |
| 563 | { |
| 564 | struct address_space *mapping; |
| 565 | int rc; |
| 566 | |
| 567 | /* |
| 568 | * Block others from accessing the page when we get around to |
| 569 | * establishing additional references. We are the only one |
| 570 | * holding a reference to the new page at this point. |
| 571 | */ |
| 572 | if (TestSetPageLocked(newpage)) |
| 573 | BUG(); |
| 574 | |
| 575 | /* Prepare mapping for the new page.*/ |
| 576 | newpage->index = page->index; |
| 577 | newpage->mapping = page->mapping; |
| 578 | |
| 579 | mapping = page_mapping(page); |
| 580 | if (!mapping) |
| 581 | rc = migrate_page(mapping, newpage, page); |
| 582 | else if (mapping->a_ops->migratepage) |
| 583 | /* |
| 584 | * Most pages have a mapping and most filesystems |
| 585 | * should provide a migration function. Anonymous |
| 586 | * pages are part of swap space which also has its |
| 587 | * own migration function. This is the most common |
| 588 | * path for page migration. |
| 589 | */ |
| 590 | rc = mapping->a_ops->migratepage(mapping, |
| 591 | newpage, page); |
| 592 | else |
| 593 | rc = fallback_migrate_page(mapping, newpage, page); |
| 594 | |
| 595 | if (!rc) |
| 596 | remove_migration_ptes(page, newpage); |
| 597 | else |
| 598 | newpage->mapping = NULL; |
| 599 | |
| 600 | unlock_page(newpage); |
| 601 | |
| 602 | return rc; |
| 603 | } |
| 604 | |
| 605 | /* |
| 606 | * Obtain the lock on page, remove all ptes and migrate the page |
| 607 | * to the newly allocated page in newpage. |
| 608 | */ |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 609 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, |
| 610 | struct page *page, int force) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 611 | { |
| 612 | int rc = 0; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 613 | int *result = NULL; |
| 614 | struct page *newpage = get_new_page(page, private, &result); |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 615 | int rcu_locked = 0; |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 616 | |
| 617 | if (!newpage) |
| 618 | return -ENOMEM; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 619 | |
| 620 | if (page_count(page) == 1) |
| 621 | /* page was freed from under us. So we are done. */ |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 622 | goto move_newpage; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 623 | |
| 624 | rc = -EAGAIN; |
| 625 | if (TestSetPageLocked(page)) { |
| 626 | if (!force) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 627 | goto move_newpage; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 628 | lock_page(page); |
| 629 | } |
| 630 | |
| 631 | if (PageWriteback(page)) { |
| 632 | if (!force) |
| 633 | goto unlock; |
| 634 | wait_on_page_writeback(page); |
| 635 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 636 | /* |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 637 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
| 638 | * we cannot notice that anon_vma is freed while we migrates a page. |
| 639 | * This rcu_read_lock() delays freeing anon_vma pointer until the end |
| 640 | * of migration. File cache pages are no problem because of page_lock() |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 641 | * File Caches may use write_page() or lock_page() in migration, then, |
| 642 | * just care Anon page here. |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 643 | */ |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 644 | if (PageAnon(page)) { |
| 645 | rcu_read_lock(); |
| 646 | rcu_locked = 1; |
| 647 | } |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 648 | /* |
| 649 | * This is a corner case handling. |
| 650 | * When a new swap-cache is read into, it is linked to LRU |
| 651 | * and treated as swapcache but has no rmap yet. |
| 652 | * Calling try_to_unmap() against a page->mapping==NULL page is |
| 653 | * BUG. So handle it here. |
| 654 | */ |
| 655 | if (!page->mapping) |
| 656 | goto rcu_unlock; |
| 657 | /* Establish migration ptes or remove ptes */ |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 658 | try_to_unmap(page, 1); |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 659 | |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 660 | if (!page_mapped(page)) |
| 661 | rc = move_to_new_page(newpage, page); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 662 | |
| 663 | if (rc) |
| 664 | remove_migration_ptes(page, page); |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 665 | rcu_unlock: |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 666 | if (rcu_locked) |
| 667 | rcu_read_unlock(); |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 668 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 669 | unlock: |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 670 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 671 | unlock_page(page); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 672 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 673 | if (rc != -EAGAIN) { |
Christoph Lameter | aaa994b | 2006-06-23 02:03:52 -0700 | [diff] [blame] | 674 | /* |
| 675 | * A page that has been migrated has all references |
| 676 | * removed and will be freed. A page that has not been |
| 677 | * migrated will have kepts its references and be |
| 678 | * restored. |
| 679 | */ |
| 680 | list_del(&page->lru); |
| 681 | move_to_lru(page); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 682 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 683 | |
| 684 | move_newpage: |
| 685 | /* |
| 686 | * Move the new page to the LRU. If migration was not successful |
| 687 | * then this will free the page. |
| 688 | */ |
| 689 | move_to_lru(newpage); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 690 | if (result) { |
| 691 | if (rc) |
| 692 | *result = rc; |
| 693 | else |
| 694 | *result = page_to_nid(newpage); |
| 695 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 696 | return rc; |
| 697 | } |
| 698 | |
| 699 | /* |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 700 | * migrate_pages |
| 701 | * |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 702 | * The function takes one list of pages to migrate and a function |
| 703 | * that determines from the page to be migrated and the private data |
| 704 | * the target of the move and allocates the page. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 705 | * |
| 706 | * The function returns after 10 attempts or if no pages |
| 707 | * are movable anymore because to has become empty |
Christoph Lameter | aaa994b | 2006-06-23 02:03:52 -0700 | [diff] [blame] | 708 | * or no retryable pages exist anymore. All pages will be |
| 709 | * retruned to the LRU or freed. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 710 | * |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 711 | * Return: Number of pages not migrated or error code. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 712 | */ |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 713 | int migrate_pages(struct list_head *from, |
| 714 | new_page_t get_new_page, unsigned long private) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 715 | { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 716 | int retry = 1; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 717 | int nr_failed = 0; |
| 718 | int pass = 0; |
| 719 | struct page *page; |
| 720 | struct page *page2; |
| 721 | int swapwrite = current->flags & PF_SWAPWRITE; |
| 722 | int rc; |
| 723 | |
| 724 | if (!swapwrite) |
| 725 | current->flags |= PF_SWAPWRITE; |
| 726 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 727 | for(pass = 0; pass < 10 && retry; pass++) { |
| 728 | retry = 0; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 729 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 730 | list_for_each_entry_safe(page, page2, from, lru) { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 731 | cond_resched(); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 732 | |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 733 | rc = unmap_and_move(get_new_page, private, |
| 734 | page, pass > 2); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 735 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 736 | switch(rc) { |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 737 | case -ENOMEM: |
| 738 | goto out; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 739 | case -EAGAIN: |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 740 | retry++; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 741 | break; |
| 742 | case 0: |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 743 | break; |
| 744 | default: |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 745 | /* Permanent failure */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 746 | nr_failed++; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 747 | break; |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 748 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 749 | } |
| 750 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 751 | rc = 0; |
| 752 | out: |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 753 | if (!swapwrite) |
| 754 | current->flags &= ~PF_SWAPWRITE; |
| 755 | |
Christoph Lameter | aaa994b | 2006-06-23 02:03:52 -0700 | [diff] [blame] | 756 | putback_lru_pages(from); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 757 | |
| 758 | if (rc) |
| 759 | return rc; |
| 760 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 761 | return nr_failed + retry; |
| 762 | } |
| 763 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 764 | #ifdef CONFIG_NUMA |
| 765 | /* |
| 766 | * Move a list of individual pages |
| 767 | */ |
| 768 | struct page_to_node { |
| 769 | unsigned long addr; |
| 770 | struct page *page; |
| 771 | int node; |
| 772 | int status; |
| 773 | }; |
| 774 | |
| 775 | static struct page *new_page_node(struct page *p, unsigned long private, |
| 776 | int **result) |
| 777 | { |
| 778 | struct page_to_node *pm = (struct page_to_node *)private; |
| 779 | |
| 780 | while (pm->node != MAX_NUMNODES && pm->page != p) |
| 781 | pm++; |
| 782 | |
| 783 | if (pm->node == MAX_NUMNODES) |
| 784 | return NULL; |
| 785 | |
| 786 | *result = &pm->status; |
| 787 | |
Mel Gorman | 769848c | 2007-07-17 04:03:05 -0700 | [diff] [blame] | 788 | return alloc_pages_node(pm->node, |
| 789 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 790 | } |
| 791 | |
| 792 | /* |
| 793 | * Move a set of pages as indicated in the pm array. The addr |
| 794 | * field must be set to the virtual address of the page to be moved |
| 795 | * and the node number must contain a valid target node. |
| 796 | */ |
| 797 | static int do_move_pages(struct mm_struct *mm, struct page_to_node *pm, |
| 798 | int migrate_all) |
| 799 | { |
| 800 | int err; |
| 801 | struct page_to_node *pp; |
| 802 | LIST_HEAD(pagelist); |
| 803 | |
| 804 | down_read(&mm->mmap_sem); |
| 805 | |
| 806 | /* |
| 807 | * Build a list of pages to migrate |
| 808 | */ |
| 809 | migrate_prep(); |
| 810 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
| 811 | struct vm_area_struct *vma; |
| 812 | struct page *page; |
| 813 | |
| 814 | /* |
| 815 | * A valid page pointer that will not match any of the |
| 816 | * pages that will be moved. |
| 817 | */ |
| 818 | pp->page = ZERO_PAGE(0); |
| 819 | |
| 820 | err = -EFAULT; |
| 821 | vma = find_vma(mm, pp->addr); |
Christoph Lameter | 0dc952d | 2007-03-05 00:30:33 -0800 | [diff] [blame] | 822 | if (!vma || !vma_migratable(vma)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 823 | goto set_status; |
| 824 | |
| 825 | page = follow_page(vma, pp->addr, FOLL_GET); |
| 826 | err = -ENOENT; |
| 827 | if (!page) |
| 828 | goto set_status; |
| 829 | |
| 830 | if (PageReserved(page)) /* Check for zero page */ |
| 831 | goto put_and_set; |
| 832 | |
| 833 | pp->page = page; |
| 834 | err = page_to_nid(page); |
| 835 | |
| 836 | if (err == pp->node) |
| 837 | /* |
| 838 | * Node already in the right place |
| 839 | */ |
| 840 | goto put_and_set; |
| 841 | |
| 842 | err = -EACCES; |
| 843 | if (page_mapcount(page) > 1 && |
| 844 | !migrate_all) |
| 845 | goto put_and_set; |
| 846 | |
| 847 | err = isolate_lru_page(page, &pagelist); |
| 848 | put_and_set: |
| 849 | /* |
| 850 | * Either remove the duplicate refcount from |
| 851 | * isolate_lru_page() or drop the page ref if it was |
| 852 | * not isolated. |
| 853 | */ |
| 854 | put_page(page); |
| 855 | set_status: |
| 856 | pp->status = err; |
| 857 | } |
| 858 | |
| 859 | if (!list_empty(&pagelist)) |
| 860 | err = migrate_pages(&pagelist, new_page_node, |
| 861 | (unsigned long)pm); |
| 862 | else |
| 863 | err = -ENOENT; |
| 864 | |
| 865 | up_read(&mm->mmap_sem); |
| 866 | return err; |
| 867 | } |
| 868 | |
| 869 | /* |
| 870 | * Determine the nodes of a list of pages. The addr in the pm array |
| 871 | * must have been set to the virtual address of which we want to determine |
| 872 | * the node number. |
| 873 | */ |
| 874 | static int do_pages_stat(struct mm_struct *mm, struct page_to_node *pm) |
| 875 | { |
| 876 | down_read(&mm->mmap_sem); |
| 877 | |
| 878 | for ( ; pm->node != MAX_NUMNODES; pm++) { |
| 879 | struct vm_area_struct *vma; |
| 880 | struct page *page; |
| 881 | int err; |
| 882 | |
| 883 | err = -EFAULT; |
| 884 | vma = find_vma(mm, pm->addr); |
| 885 | if (!vma) |
| 886 | goto set_status; |
| 887 | |
| 888 | page = follow_page(vma, pm->addr, 0); |
| 889 | err = -ENOENT; |
| 890 | /* Use PageReserved to check for zero page */ |
| 891 | if (!page || PageReserved(page)) |
| 892 | goto set_status; |
| 893 | |
| 894 | err = page_to_nid(page); |
| 895 | set_status: |
| 896 | pm->status = err; |
| 897 | } |
| 898 | |
| 899 | up_read(&mm->mmap_sem); |
| 900 | return 0; |
| 901 | } |
| 902 | |
| 903 | /* |
| 904 | * Move a list of pages in the address space of the currently executing |
| 905 | * process. |
| 906 | */ |
| 907 | asmlinkage long sys_move_pages(pid_t pid, unsigned long nr_pages, |
| 908 | const void __user * __user *pages, |
| 909 | const int __user *nodes, |
| 910 | int __user *status, int flags) |
| 911 | { |
| 912 | int err = 0; |
| 913 | int i; |
| 914 | struct task_struct *task; |
| 915 | nodemask_t task_nodes; |
| 916 | struct mm_struct *mm; |
| 917 | struct page_to_node *pm = NULL; |
| 918 | |
| 919 | /* Check flags */ |
| 920 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) |
| 921 | return -EINVAL; |
| 922 | |
| 923 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
| 924 | return -EPERM; |
| 925 | |
| 926 | /* Find the mm_struct */ |
| 927 | read_lock(&tasklist_lock); |
| 928 | task = pid ? find_task_by_pid(pid) : current; |
| 929 | if (!task) { |
| 930 | read_unlock(&tasklist_lock); |
| 931 | return -ESRCH; |
| 932 | } |
| 933 | mm = get_task_mm(task); |
| 934 | read_unlock(&tasklist_lock); |
| 935 | |
| 936 | if (!mm) |
| 937 | return -EINVAL; |
| 938 | |
| 939 | /* |
| 940 | * Check if this process has the right to modify the specified |
| 941 | * process. The right exists if the process has administrative |
| 942 | * capabilities, superuser privileges or the same |
| 943 | * userid as the target process. |
| 944 | */ |
| 945 | if ((current->euid != task->suid) && (current->euid != task->uid) && |
| 946 | (current->uid != task->suid) && (current->uid != task->uid) && |
| 947 | !capable(CAP_SYS_NICE)) { |
| 948 | err = -EPERM; |
| 949 | goto out2; |
| 950 | } |
| 951 | |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 952 | err = security_task_movememory(task); |
| 953 | if (err) |
| 954 | goto out2; |
| 955 | |
| 956 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 957 | task_nodes = cpuset_mems_allowed(task); |
| 958 | |
| 959 | /* Limit nr_pages so that the multiplication may not overflow */ |
| 960 | if (nr_pages >= ULONG_MAX / sizeof(struct page_to_node) - 1) { |
| 961 | err = -E2BIG; |
| 962 | goto out2; |
| 963 | } |
| 964 | |
| 965 | pm = vmalloc((nr_pages + 1) * sizeof(struct page_to_node)); |
| 966 | if (!pm) { |
| 967 | err = -ENOMEM; |
| 968 | goto out2; |
| 969 | } |
| 970 | |
| 971 | /* |
| 972 | * Get parameters from user space and initialize the pm |
| 973 | * array. Return various errors if the user did something wrong. |
| 974 | */ |
| 975 | for (i = 0; i < nr_pages; i++) { |
Al Viro | 9d966d4 | 2007-10-14 19:34:10 +0100 | [diff] [blame] | 976 | const void __user *p; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 977 | |
| 978 | err = -EFAULT; |
| 979 | if (get_user(p, pages + i)) |
| 980 | goto out; |
| 981 | |
| 982 | pm[i].addr = (unsigned long)p; |
| 983 | if (nodes) { |
| 984 | int node; |
| 985 | |
| 986 | if (get_user(node, nodes + i)) |
| 987 | goto out; |
| 988 | |
| 989 | err = -ENODEV; |
Christoph Lameter | 56bbd65 | 2007-10-16 01:25:35 -0700 | [diff] [blame] | 990 | if (!node_state(node, N_HIGH_MEMORY)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 991 | goto out; |
| 992 | |
| 993 | err = -EACCES; |
| 994 | if (!node_isset(node, task_nodes)) |
| 995 | goto out; |
| 996 | |
| 997 | pm[i].node = node; |
Stephen Rothwell | 8ce0846 | 2006-11-02 22:07:28 -0800 | [diff] [blame] | 998 | } else |
| 999 | pm[i].node = 0; /* anything to not match MAX_NUMNODES */ |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1000 | } |
| 1001 | /* End marker */ |
| 1002 | pm[nr_pages].node = MAX_NUMNODES; |
| 1003 | |
| 1004 | if (nodes) |
| 1005 | err = do_move_pages(mm, pm, flags & MPOL_MF_MOVE_ALL); |
| 1006 | else |
| 1007 | err = do_pages_stat(mm, pm); |
| 1008 | |
| 1009 | if (err >= 0) |
| 1010 | /* Return status information */ |
| 1011 | for (i = 0; i < nr_pages; i++) |
| 1012 | if (put_user(pm[i].status, status + i)) |
| 1013 | err = -EFAULT; |
| 1014 | |
| 1015 | out: |
| 1016 | vfree(pm); |
| 1017 | out2: |
| 1018 | mmput(mm); |
| 1019 | return err; |
| 1020 | } |
| 1021 | #endif |
| 1022 | |
Christoph Lameter | 7b2259b | 2006-06-25 05:46:48 -0700 | [diff] [blame] | 1023 | /* |
| 1024 | * Call migration functions in the vma_ops that may prepare |
| 1025 | * memory in a vm for migration. migration functions may perform |
| 1026 | * the migration for vmas that do not have an underlying page struct. |
| 1027 | */ |
| 1028 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, |
| 1029 | const nodemask_t *from, unsigned long flags) |
| 1030 | { |
| 1031 | struct vm_area_struct *vma; |
| 1032 | int err = 0; |
| 1033 | |
| 1034 | for(vma = mm->mmap; vma->vm_next && !err; vma = vma->vm_next) { |
| 1035 | if (vma->vm_ops && vma->vm_ops->migrate) { |
| 1036 | err = vma->vm_ops->migrate(vma, to, from, flags); |
| 1037 | if (err) |
| 1038 | break; |
| 1039 | } |
| 1040 | } |
| 1041 | return err; |
| 1042 | } |