Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * mm/rmap.c - physical to virtual reverse mappings |
| 3 | * |
| 4 | * Copyright 2001, Rik van Riel <riel@conectiva.com.br> |
| 5 | * Released under the General Public License (GPL). |
| 6 | * |
| 7 | * Simple, low overhead reverse mapping scheme. |
| 8 | * Please try to keep this thing as modular as possible. |
| 9 | * |
| 10 | * Provides methods for unmapping each kind of mapped page: |
| 11 | * the anon methods track anonymous pages, and |
| 12 | * the file methods track pages belonging to an inode. |
| 13 | * |
| 14 | * Original design by Rik van Riel <riel@conectiva.com.br> 2001 |
| 15 | * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 |
| 16 | * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 |
Hugh Dickins | 98f3260 | 2009-05-21 20:33:58 +0100 | [diff] [blame] | 17 | * Contributions by Hugh Dickins 2003, 2004 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | */ |
| 19 | |
| 20 | /* |
| 21 | * Lock ordering in mm: |
| 22 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 23 | * inode->i_mutex (while writing or truncating, not reading or faulting) |
Nick Piggin | 82591e6 | 2006-10-19 23:29:10 -0700 | [diff] [blame] | 24 | * inode->i_alloc_sem (vmtruncate_range) |
| 25 | * mm->mmap_sem |
| 26 | * page->flags PG_locked (lock_page) |
| 27 | * mapping->i_mmap_lock |
| 28 | * anon_vma->lock |
| 29 | * mm->page_table_lock or pte_lock |
| 30 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) |
| 31 | * swap_lock (in swap_duplicate, swap_info_get) |
| 32 | * mmlist_lock (in mmput, drain_mmlist and others) |
| 33 | * mapping->private_lock (in __set_page_dirty_buffers) |
| 34 | * inode_lock (in set_page_dirty's __mark_inode_dirty) |
| 35 | * sb_lock (within inode_lock in fs/fs-writeback.c) |
| 36 | * mapping->tree_lock (widely used, in set_page_dirty, |
| 37 | * in arch-dependent flush_dcache_mmap_lock, |
| 38 | * within inode_lock in __sync_single_inode) |
Andi Kleen | 6a46079 | 2009-09-16 11:50:15 +0200 | [diff] [blame] | 39 | * |
| 40 | * (code doesn't rely on that order so it could be switched around) |
| 41 | * ->tasklist_lock |
| 42 | * anon_vma->lock (memory_failure, collect_procs_anon) |
| 43 | * pte map lock |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | */ |
| 45 | |
| 46 | #include <linux/mm.h> |
| 47 | #include <linux/pagemap.h> |
| 48 | #include <linux/swap.h> |
| 49 | #include <linux/swapops.h> |
| 50 | #include <linux/slab.h> |
| 51 | #include <linux/init.h> |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 52 | #include <linux/ksm.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 53 | #include <linux/rmap.h> |
| 54 | #include <linux/rcupdate.h> |
Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame] | 55 | #include <linux/module.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 56 | #include <linux/memcontrol.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 57 | #include <linux/mmu_notifier.h> |
KOSAKI Motohiro | 64cdd54 | 2009-01-06 14:39:16 -0800 | [diff] [blame] | 58 | #include <linux/migrate.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | |
| 60 | #include <asm/tlbflush.h> |
| 61 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 62 | #include "internal.h" |
| 63 | |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 64 | static struct kmem_cache *anon_vma_cachep; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 65 | static struct kmem_cache *anon_vma_chain_cachep; |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 66 | |
| 67 | static inline struct anon_vma *anon_vma_alloc(void) |
| 68 | { |
| 69 | return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); |
| 70 | } |
| 71 | |
Hugh Dickins | db114b8 | 2009-12-14 17:59:25 -0800 | [diff] [blame] | 72 | void anon_vma_free(struct anon_vma *anon_vma) |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 73 | { |
| 74 | kmem_cache_free(anon_vma_cachep, anon_vma); |
| 75 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 77 | static inline struct anon_vma_chain *anon_vma_chain_alloc(void) |
| 78 | { |
| 79 | return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL); |
| 80 | } |
| 81 | |
| 82 | void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
| 83 | { |
| 84 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); |
| 85 | } |
| 86 | |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 87 | /** |
| 88 | * anon_vma_prepare - attach an anon_vma to a memory region |
| 89 | * @vma: the memory region in question |
| 90 | * |
| 91 | * This makes sure the memory mapping described by 'vma' has |
| 92 | * an 'anon_vma' attached to it, so that we can associate the |
| 93 | * anonymous pages mapped into it with that anon_vma. |
| 94 | * |
| 95 | * The common case will be that we already have one, but if |
| 96 | * if not we either need to find an adjacent mapping that we |
| 97 | * can re-use the anon_vma from (very common when the only |
| 98 | * reason for splitting a vma has been mprotect()), or we |
| 99 | * allocate a new one. |
| 100 | * |
| 101 | * Anon-vma allocations are very subtle, because we may have |
| 102 | * optimistically looked up an anon_vma in page_lock_anon_vma() |
| 103 | * and that may actually touch the spinlock even in the newly |
| 104 | * allocated vma (it depends on RCU to make sure that the |
| 105 | * anon_vma isn't actually destroyed). |
| 106 | * |
| 107 | * As a result, we need to do proper anon_vma locking even |
| 108 | * for the new allocation. At the same time, we do not want |
| 109 | * to do any locking for the common case of already having |
| 110 | * an anon_vma. |
| 111 | * |
| 112 | * This must be called with the mmap_sem held for reading. |
| 113 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | int anon_vma_prepare(struct vm_area_struct *vma) |
| 115 | { |
| 116 | struct anon_vma *anon_vma = vma->anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 117 | struct anon_vma_chain *avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | |
| 119 | might_sleep(); |
| 120 | if (unlikely(!anon_vma)) { |
| 121 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 122 | struct anon_vma *allocated; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 124 | avc = anon_vma_chain_alloc(); |
| 125 | if (!avc) |
| 126 | goto out_enomem; |
| 127 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | anon_vma = find_mergeable_anon_vma(vma); |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 129 | allocated = NULL; |
| 130 | if (!anon_vma) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | anon_vma = anon_vma_alloc(); |
| 132 | if (unlikely(!anon_vma)) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 133 | goto out_enomem_free_avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | allocated = anon_vma; |
Rik van Riel | 5c341ee | 2010-08-09 17:18:39 -0700 | [diff] [blame] | 135 | /* |
| 136 | * This VMA had no anon_vma yet. This anon_vma is |
| 137 | * the root of any anon_vma tree that might form. |
| 138 | */ |
| 139 | anon_vma->root = anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | } |
| 141 | |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 142 | anon_vma_lock(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 143 | /* page_table_lock to protect against threads */ |
| 144 | spin_lock(&mm->page_table_lock); |
| 145 | if (likely(!vma->anon_vma)) { |
| 146 | vma->anon_vma = anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 147 | avc->anon_vma = anon_vma; |
| 148 | avc->vma = vma; |
| 149 | list_add(&avc->same_vma, &vma->anon_vma_chain); |
Andrea Arcangeli | 26ba0cb | 2010-08-09 17:19:08 -0700 | [diff] [blame] | 150 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | allocated = NULL; |
Oleg Nesterov | 31f2b0e | 2010-04-23 13:18:01 -0400 | [diff] [blame] | 152 | avc = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | } |
| 154 | spin_unlock(&mm->page_table_lock); |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 155 | anon_vma_unlock(anon_vma); |
Oleg Nesterov | 31f2b0e | 2010-04-23 13:18:01 -0400 | [diff] [blame] | 156 | |
| 157 | if (unlikely(allocated)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | anon_vma_free(allocated); |
Oleg Nesterov | 31f2b0e | 2010-04-23 13:18:01 -0400 | [diff] [blame] | 159 | if (unlikely(avc)) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 160 | anon_vma_chain_free(avc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | } |
| 162 | return 0; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 163 | |
| 164 | out_enomem_free_avc: |
| 165 | anon_vma_chain_free(avc); |
| 166 | out_enomem: |
| 167 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | } |
| 169 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 170 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
| 171 | struct anon_vma_chain *avc, |
| 172 | struct anon_vma *anon_vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 174 | avc->vma = vma; |
| 175 | avc->anon_vma = anon_vma; |
| 176 | list_add(&avc->same_vma, &vma->anon_vma_chain); |
| 177 | |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 178 | anon_vma_lock(anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 179 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 180 | anon_vma_unlock(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | } |
| 182 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 183 | /* |
| 184 | * Attach the anon_vmas from src to dst. |
| 185 | * Returns 0 on success, -ENOMEM on failure. |
| 186 | */ |
| 187 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 189 | struct anon_vma_chain *avc, *pavc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
Linus Torvalds | 646d87b | 2010-04-11 17:15:03 -0700 | [diff] [blame] | 191 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 192 | avc = anon_vma_chain_alloc(); |
| 193 | if (!avc) |
| 194 | goto enomem_failure; |
| 195 | anon_vma_chain_link(dst, avc, pavc->anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | } |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 197 | return 0; |
| 198 | |
| 199 | enomem_failure: |
| 200 | unlink_anon_vmas(dst); |
| 201 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | } |
| 203 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 204 | /* |
| 205 | * Attach vma to its own anon_vma, as well as to the anon_vmas that |
| 206 | * the corresponding VMA in the parent process is attached to. |
| 207 | * Returns 0 on success, non-zero on failure. |
| 208 | */ |
| 209 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 211 | struct anon_vma_chain *avc; |
| 212 | struct anon_vma *anon_vma; |
| 213 | |
| 214 | /* Don't bother if the parent process has no anon_vma here. */ |
| 215 | if (!pvma->anon_vma) |
| 216 | return 0; |
| 217 | |
| 218 | /* |
| 219 | * First, attach the new VMA to the parent VMA's anon_vmas, |
| 220 | * so rmap can find non-COWed pages in child processes. |
| 221 | */ |
| 222 | if (anon_vma_clone(vma, pvma)) |
| 223 | return -ENOMEM; |
| 224 | |
| 225 | /* Then add our own anon_vma. */ |
| 226 | anon_vma = anon_vma_alloc(); |
| 227 | if (!anon_vma) |
| 228 | goto out_error; |
| 229 | avc = anon_vma_chain_alloc(); |
| 230 | if (!avc) |
| 231 | goto out_error_free_anon_vma; |
Rik van Riel | 5c341ee | 2010-08-09 17:18:39 -0700 | [diff] [blame] | 232 | |
| 233 | /* |
| 234 | * The root anon_vma's spinlock is the lock actually used when we |
| 235 | * lock any of the anon_vmas in this anon_vma tree. |
| 236 | */ |
| 237 | anon_vma->root = pvma->anon_vma->root; |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 238 | /* |
| 239 | * With KSM refcounts, an anon_vma can stay around longer than the |
| 240 | * process it belongs to. The root anon_vma needs to be pinned |
| 241 | * until this anon_vma is freed, because the lock lives in the root. |
| 242 | */ |
| 243 | get_anon_vma(anon_vma->root); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 244 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
| 245 | vma->anon_vma = anon_vma; |
Rik van Riel | 5c341ee | 2010-08-09 17:18:39 -0700 | [diff] [blame] | 246 | anon_vma_chain_link(vma, avc, anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 247 | |
| 248 | return 0; |
| 249 | |
| 250 | out_error_free_anon_vma: |
| 251 | anon_vma_free(anon_vma); |
| 252 | out_error: |
Rik van Riel | 4946d54 | 2010-04-05 12:13:33 -0400 | [diff] [blame] | 253 | unlink_anon_vmas(vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 254 | return -ENOMEM; |
| 255 | } |
| 256 | |
| 257 | static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) |
| 258 | { |
| 259 | struct anon_vma *anon_vma = anon_vma_chain->anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | int empty; |
| 261 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 262 | /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 263 | if (!anon_vma) |
| 264 | return; |
| 265 | |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 266 | anon_vma_lock(anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 267 | list_del(&anon_vma_chain->same_anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | |
| 269 | /* We must garbage collect the anon_vma if it's empty */ |
Mel Gorman | 7f60c21 | 2010-05-24 14:32:18 -0700 | [diff] [blame] | 270 | empty = list_empty(&anon_vma->head) && !anonvma_external_refcount(anon_vma); |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 271 | anon_vma_unlock(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 273 | if (empty) { |
| 274 | /* We no longer need the root anon_vma */ |
| 275 | if (anon_vma->root != anon_vma) |
| 276 | drop_anon_vma(anon_vma->root); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | anon_vma_free(anon_vma); |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 278 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 279 | } |
| 280 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 281 | void unlink_anon_vmas(struct vm_area_struct *vma) |
| 282 | { |
| 283 | struct anon_vma_chain *avc, *next; |
| 284 | |
Rik van Riel | 5c341ee | 2010-08-09 17:18:39 -0700 | [diff] [blame] | 285 | /* |
| 286 | * Unlink each anon_vma chained to the VMA. This list is ordered |
| 287 | * from newest to oldest, ensuring the root anon_vma gets freed last. |
| 288 | */ |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 289 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
| 290 | anon_vma_unlink(avc); |
| 291 | list_del(&avc->same_vma); |
| 292 | anon_vma_chain_free(avc); |
| 293 | } |
| 294 | } |
| 295 | |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 296 | static void anon_vma_ctor(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | { |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 298 | struct anon_vma *anon_vma = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 300 | spin_lock_init(&anon_vma->lock); |
Mel Gorman | 7f60c21 | 2010-05-24 14:32:18 -0700 | [diff] [blame] | 301 | anonvma_external_refcount_init(anon_vma); |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 302 | INIT_LIST_HEAD(&anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | } |
| 304 | |
| 305 | void __init anon_vma_init(void) |
| 306 | { |
| 307 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 308 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 309 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | } |
| 311 | |
| 312 | /* |
| 313 | * Getting a lock on a stable anon_vma from a page off the LRU is |
| 314 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. |
| 315 | */ |
Andi Kleen | 10be22d | 2009-09-16 11:50:04 +0200 | [diff] [blame] | 316 | struct anon_vma *page_lock_anon_vma(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | { |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 318 | struct anon_vma *anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | unsigned long anon_mapping; |
| 320 | |
| 321 | rcu_read_lock(); |
Hugh Dickins | 80e1482 | 2009-12-14 17:59:29 -0800 | [diff] [blame] | 322 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
Hugh Dickins | 3ca7b3c | 2009-12-14 17:58:57 -0800 | [diff] [blame] | 323 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 324 | goto out; |
| 325 | if (!page_mapped(page)) |
| 326 | goto out; |
| 327 | |
| 328 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 329 | anon_vma_lock(anon_vma); |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 330 | return anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | out: |
| 332 | rcu_read_unlock(); |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 333 | return NULL; |
| 334 | } |
| 335 | |
Andi Kleen | 10be22d | 2009-09-16 11:50:04 +0200 | [diff] [blame] | 336 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 337 | { |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 338 | anon_vma_unlock(anon_vma); |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 339 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 340 | } |
| 341 | |
| 342 | /* |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 343 | * At what user virtual address is page expected in @vma? |
| 344 | * Returns virtual address or -EFAULT if page's index/offset is not |
| 345 | * within the range mapped the @vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | */ |
| 347 | static inline unsigned long |
| 348 | vma_address(struct page *page, struct vm_area_struct *vma) |
| 349 | { |
| 350 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 351 | unsigned long address; |
| 352 | |
| 353 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
| 354 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 355 | /* page should be within @vma mapping range */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | return -EFAULT; |
| 357 | } |
| 358 | return address; |
| 359 | } |
| 360 | |
| 361 | /* |
Huang Shijie | bf89c8c | 2009-10-01 15:44:04 -0700 | [diff] [blame] | 362 | * At what user virtual address is page expected in vma? |
Naoya Horiguchi | ab941e0 | 2010-05-11 14:06:55 -0700 | [diff] [blame] | 363 | * Caller should check the page is actually part of the vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | */ |
| 365 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) |
| 366 | { |
Naoya Horiguchi | ab941e0 | 2010-05-11 14:06:55 -0700 | [diff] [blame] | 367 | if (PageAnon(page)) |
| 368 | ; |
| 369 | else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 370 | if (!vma->vm_file || |
| 371 | vma->vm_file->f_mapping != page->mapping) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | return -EFAULT; |
| 373 | } else |
| 374 | return -EFAULT; |
| 375 | return vma_address(page, vma); |
| 376 | } |
| 377 | |
| 378 | /* |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 379 | * Check that @page is mapped at @address into @mm. |
| 380 | * |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 381 | * If @sync is false, page_check_address may perform a racy check to avoid |
| 382 | * the page table lock when the pte is not present (helpful when reclaiming |
| 383 | * highly shared pages). |
| 384 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 385 | * On success returns with pte mapped and locked. |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 386 | */ |
Carsten Otte | ceffc07 | 2005-06-23 22:05:25 -0700 | [diff] [blame] | 387 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 388 | unsigned long address, spinlock_t **ptlp, int sync) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 389 | { |
| 390 | pgd_t *pgd; |
| 391 | pud_t *pud; |
| 392 | pmd_t *pmd; |
| 393 | pte_t *pte; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 394 | spinlock_t *ptl; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 395 | |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 396 | pgd = pgd_offset(mm, address); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 397 | if (!pgd_present(*pgd)) |
| 398 | return NULL; |
| 399 | |
| 400 | pud = pud_offset(pgd, address); |
| 401 | if (!pud_present(*pud)) |
| 402 | return NULL; |
| 403 | |
| 404 | pmd = pmd_offset(pud, address); |
| 405 | if (!pmd_present(*pmd)) |
| 406 | return NULL; |
| 407 | |
| 408 | pte = pte_offset_map(pmd, address); |
| 409 | /* Make a quick check before getting the lock */ |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 410 | if (!sync && !pte_present(*pte)) { |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 411 | pte_unmap(pte); |
| 412 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 413 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 414 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 415 | ptl = pte_lockptr(mm, pmd); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 416 | spin_lock(ptl); |
| 417 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { |
| 418 | *ptlp = ptl; |
| 419 | return pte; |
| 420 | } |
| 421 | pte_unmap_unlock(pte, ptl); |
| 422 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 423 | } |
| 424 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 425 | /** |
| 426 | * page_mapped_in_vma - check whether a page is really mapped in a VMA |
| 427 | * @page: the page to test |
| 428 | * @vma: the VMA to test |
| 429 | * |
| 430 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 |
| 431 | * if the page is not mapped into the page tables of this VMA. Only |
| 432 | * valid for normal file or anonymous VMAs. |
| 433 | */ |
Andi Kleen | 6a46079 | 2009-09-16 11:50:15 +0200 | [diff] [blame] | 434 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 435 | { |
| 436 | unsigned long address; |
| 437 | pte_t *pte; |
| 438 | spinlock_t *ptl; |
| 439 | |
| 440 | address = vma_address(page, vma); |
| 441 | if (address == -EFAULT) /* out of vma range */ |
| 442 | return 0; |
| 443 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); |
| 444 | if (!pte) /* the page is not in this mm */ |
| 445 | return 0; |
| 446 | pte_unmap_unlock(pte, ptl); |
| 447 | |
| 448 | return 1; |
| 449 | } |
| 450 | |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 451 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 452 | * Subfunctions of page_referenced: page_referenced_one called |
| 453 | * repeatedly from either page_referenced_anon or page_referenced_file. |
| 454 | */ |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 455 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
| 456 | unsigned long address, unsigned int *mapcount, |
| 457 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | { |
| 459 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | pte_t *pte; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 461 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | int referenced = 0; |
| 463 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 464 | pte = page_check_address(page, mm, address, &ptl, 0); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 465 | if (!pte) |
| 466 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 468 | /* |
| 469 | * Don't want to elevate referenced for mlocked page that gets this far, |
| 470 | * in order that it progresses to try_to_unmap and is moved to the |
| 471 | * unevictable list. |
| 472 | */ |
Hugh Dickins | 5a9bbdc | 2008-02-04 22:29:23 -0800 | [diff] [blame] | 473 | if (vma->vm_flags & VM_LOCKED) { |
Hugh Dickins | 5a9bbdc | 2008-02-04 22:29:23 -0800 | [diff] [blame] | 474 | *mapcount = 1; /* break early from loop */ |
Minchan Kim | 03ef83a | 2009-08-26 14:29:23 -0700 | [diff] [blame] | 475 | *vm_flags |= VM_LOCKED; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 476 | goto out_unmap; |
| 477 | } |
| 478 | |
Johannes Weiner | 4917e5d | 2009-01-06 14:39:17 -0800 | [diff] [blame] | 479 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
| 480 | /* |
| 481 | * Don't treat a reference through a sequentially read |
| 482 | * mapping as such. If the page has been used in |
| 483 | * another mapping, we will catch it; if this other |
| 484 | * mapping is already gone, the unmap path will have |
| 485 | * set PG_referenced or activated the page. |
| 486 | */ |
| 487 | if (likely(!VM_SequentialReadHint(vma))) |
| 488 | referenced++; |
| 489 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 491 | /* Pretend the page is referenced if the task has the |
| 492 | swap token and is in the middle of a page fault. */ |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 493 | if (mm != current->mm && has_swap_token(mm) && |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 494 | rwsem_is_locked(&mm->mmap_sem)) |
| 495 | referenced++; |
| 496 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 497 | out_unmap: |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 498 | (*mapcount)--; |
| 499 | pte_unmap_unlock(pte, ptl); |
Huang Shijie | 273f047 | 2009-12-14 17:58:51 -0800 | [diff] [blame] | 500 | |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 501 | if (referenced) |
| 502 | *vm_flags |= vma->vm_flags; |
Huang Shijie | 273f047 | 2009-12-14 17:58:51 -0800 | [diff] [blame] | 503 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 504 | return referenced; |
| 505 | } |
| 506 | |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 507 | static int page_referenced_anon(struct page *page, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 508 | struct mem_cgroup *mem_cont, |
| 509 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | { |
| 511 | unsigned int mapcount; |
| 512 | struct anon_vma *anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 513 | struct anon_vma_chain *avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | int referenced = 0; |
| 515 | |
| 516 | anon_vma = page_lock_anon_vma(page); |
| 517 | if (!anon_vma) |
| 518 | return referenced; |
| 519 | |
| 520 | mapcount = page_mapcount(page); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 521 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
| 522 | struct vm_area_struct *vma = avc->vma; |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 523 | unsigned long address = vma_address(page, vma); |
| 524 | if (address == -EFAULT) |
| 525 | continue; |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 526 | /* |
| 527 | * If we are reclaiming on behalf of a cgroup, skip |
| 528 | * counting on behalf of references from different |
| 529 | * cgroups |
| 530 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 531 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 532 | continue; |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 533 | referenced += page_referenced_one(page, vma, address, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 534 | &mapcount, vm_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | if (!mapcount) |
| 536 | break; |
| 537 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 538 | |
| 539 | page_unlock_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | return referenced; |
| 541 | } |
| 542 | |
| 543 | /** |
| 544 | * page_referenced_file - referenced check for object-based rmap |
| 545 | * @page: the page we're checking references on. |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 546 | * @mem_cont: target memory controller |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 547 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | * |
| 549 | * For an object-based mapped page, find all the places it is mapped and |
| 550 | * check/clear the referenced flag. This is done by following the page->mapping |
| 551 | * pointer, then walking the chain of vmas it holds. It returns the number |
| 552 | * of references it found. |
| 553 | * |
| 554 | * This function is only called from page_referenced for object-based pages. |
| 555 | */ |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 556 | static int page_referenced_file(struct page *page, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 557 | struct mem_cgroup *mem_cont, |
| 558 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | { |
| 560 | unsigned int mapcount; |
| 561 | struct address_space *mapping = page->mapping; |
| 562 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 563 | struct vm_area_struct *vma; |
| 564 | struct prio_tree_iter iter; |
| 565 | int referenced = 0; |
| 566 | |
| 567 | /* |
| 568 | * The caller's checks on page->mapping and !PageAnon have made |
| 569 | * sure that this is a file page: the check for page->mapping |
| 570 | * excludes the case just before it gets set on an anon page. |
| 571 | */ |
| 572 | BUG_ON(PageAnon(page)); |
| 573 | |
| 574 | /* |
| 575 | * The page lock not only makes sure that page->mapping cannot |
| 576 | * suddenly be NULLified by truncation, it makes sure that the |
| 577 | * structure at mapping cannot be freed and reused yet, |
| 578 | * so we can safely take mapping->i_mmap_lock. |
| 579 | */ |
| 580 | BUG_ON(!PageLocked(page)); |
| 581 | |
| 582 | spin_lock(&mapping->i_mmap_lock); |
| 583 | |
| 584 | /* |
| 585 | * i_mmap_lock does not stabilize mapcount at all, but mapcount |
| 586 | * is more likely to be accurate if we note it after spinning. |
| 587 | */ |
| 588 | mapcount = page_mapcount(page); |
| 589 | |
| 590 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 591 | unsigned long address = vma_address(page, vma); |
| 592 | if (address == -EFAULT) |
| 593 | continue; |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 594 | /* |
| 595 | * If we are reclaiming on behalf of a cgroup, skip |
| 596 | * counting on behalf of references from different |
| 597 | * cgroups |
| 598 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 599 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 600 | continue; |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 601 | referenced += page_referenced_one(page, vma, address, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 602 | &mapcount, vm_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | if (!mapcount) |
| 604 | break; |
| 605 | } |
| 606 | |
| 607 | spin_unlock(&mapping->i_mmap_lock); |
| 608 | return referenced; |
| 609 | } |
| 610 | |
| 611 | /** |
| 612 | * page_referenced - test if the page was referenced |
| 613 | * @page: the page to test |
| 614 | * @is_locked: caller holds lock on the page |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 615 | * @mem_cont: target memory controller |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 616 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | * |
| 618 | * Quick test_and_clear_referenced for all mappings to a page, |
| 619 | * returns the number of ptes which referenced the page. |
| 620 | */ |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 621 | int page_referenced(struct page *page, |
| 622 | int is_locked, |
| 623 | struct mem_cgroup *mem_cont, |
| 624 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | { |
| 626 | int referenced = 0; |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 627 | int we_locked = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 629 | *vm_flags = 0; |
Hugh Dickins | 3ca7b3c | 2009-12-14 17:58:57 -0800 | [diff] [blame] | 630 | if (page_mapped(page) && page_rmapping(page)) { |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 631 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
| 632 | we_locked = trylock_page(page); |
| 633 | if (!we_locked) { |
| 634 | referenced++; |
| 635 | goto out; |
| 636 | } |
| 637 | } |
| 638 | if (unlikely(PageKsm(page))) |
| 639 | referenced += page_referenced_ksm(page, mem_cont, |
| 640 | vm_flags); |
| 641 | else if (PageAnon(page)) |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 642 | referenced += page_referenced_anon(page, mem_cont, |
| 643 | vm_flags); |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 644 | else if (page->mapping) |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 645 | referenced += page_referenced_file(page, mem_cont, |
| 646 | vm_flags); |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 647 | if (we_locked) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | unlock_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | } |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 650 | out: |
Christian Borntraeger | 5b7baf0 | 2008-03-25 18:47:12 +0100 | [diff] [blame] | 651 | if (page_test_and_clear_young(page)) |
| 652 | referenced++; |
| 653 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | return referenced; |
| 655 | } |
| 656 | |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 657 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
| 658 | unsigned long address) |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 659 | { |
| 660 | struct mm_struct *mm = vma->vm_mm; |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 661 | pte_t *pte; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 662 | spinlock_t *ptl; |
| 663 | int ret = 0; |
| 664 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 665 | pte = page_check_address(page, mm, address, &ptl, 1); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 666 | if (!pte) |
| 667 | goto out; |
| 668 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 669 | if (pte_dirty(*pte) || pte_write(*pte)) { |
| 670 | pte_t entry; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 671 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 672 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 673 | entry = ptep_clear_flush_notify(vma, address, pte); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 674 | entry = pte_wrprotect(entry); |
| 675 | entry = pte_mkclean(entry); |
Al Viro | d6e88e6 | 2006-12-29 16:48:35 -0800 | [diff] [blame] | 676 | set_pte_at(mm, address, pte, entry); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 677 | ret = 1; |
| 678 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 679 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 680 | pte_unmap_unlock(pte, ptl); |
| 681 | out: |
| 682 | return ret; |
| 683 | } |
| 684 | |
| 685 | static int page_mkclean_file(struct address_space *mapping, struct page *page) |
| 686 | { |
| 687 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 688 | struct vm_area_struct *vma; |
| 689 | struct prio_tree_iter iter; |
| 690 | int ret = 0; |
| 691 | |
| 692 | BUG_ON(PageAnon(page)); |
| 693 | |
| 694 | spin_lock(&mapping->i_mmap_lock); |
| 695 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 696 | if (vma->vm_flags & VM_SHARED) { |
| 697 | unsigned long address = vma_address(page, vma); |
| 698 | if (address == -EFAULT) |
| 699 | continue; |
| 700 | ret += page_mkclean_one(page, vma, address); |
| 701 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 702 | } |
| 703 | spin_unlock(&mapping->i_mmap_lock); |
| 704 | return ret; |
| 705 | } |
| 706 | |
| 707 | int page_mkclean(struct page *page) |
| 708 | { |
| 709 | int ret = 0; |
| 710 | |
| 711 | BUG_ON(!PageLocked(page)); |
| 712 | |
| 713 | if (page_mapped(page)) { |
| 714 | struct address_space *mapping = page_mapping(page); |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 715 | if (mapping) { |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 716 | ret = page_mkclean_file(mapping, page); |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 717 | if (page_test_dirty(page)) { |
| 718 | page_clear_dirty(page); |
| 719 | ret = 1; |
| 720 | } |
Martin Schwidefsky | 6c21048 | 2007-04-27 16:01:57 +0200 | [diff] [blame] | 721 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 722 | } |
| 723 | |
| 724 | return ret; |
| 725 | } |
Jaya Kumar | 60b59be | 2007-05-08 00:37:37 -0700 | [diff] [blame] | 726 | EXPORT_SYMBOL_GPL(page_mkclean); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 727 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | /** |
Rik van Riel | c44b674 | 2010-03-05 13:42:09 -0800 | [diff] [blame] | 729 | * page_move_anon_rmap - move a page to our anon_vma |
| 730 | * @page: the page to move to our anon_vma |
| 731 | * @vma: the vma the page belongs to |
| 732 | * @address: the user virtual address mapped |
| 733 | * |
| 734 | * When a page belongs exclusively to one process after a COW event, |
| 735 | * that page can be moved into the anon_vma that belongs to just that |
| 736 | * process, so the rmap code will not search the parent or sibling |
| 737 | * processes. |
| 738 | */ |
| 739 | void page_move_anon_rmap(struct page *page, |
| 740 | struct vm_area_struct *vma, unsigned long address) |
| 741 | { |
| 742 | struct anon_vma *anon_vma = vma->anon_vma; |
| 743 | |
| 744 | VM_BUG_ON(!PageLocked(page)); |
| 745 | VM_BUG_ON(!anon_vma); |
| 746 | VM_BUG_ON(page->index != linear_page_index(vma, address)); |
| 747 | |
| 748 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 749 | page->mapping = (struct address_space *) anon_vma; |
| 750 | } |
| 751 | |
| 752 | /** |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 753 | * __page_set_anon_rmap - setup new anonymous rmap |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 754 | * @page: the page to add the mapping to |
| 755 | * @vma: the vm area in which the mapping is added |
| 756 | * @address: the user virtual address mapped |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 757 | * @exclusive: the page is exclusively owned by the current process |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 758 | */ |
| 759 | static void __page_set_anon_rmap(struct page *page, |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 760 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 761 | { |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 762 | struct anon_vma *anon_vma = vma->anon_vma; |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 763 | |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 764 | BUG_ON(!anon_vma); |
Linus Torvalds | ea90002 | 2010-04-12 12:44:29 -0700 | [diff] [blame] | 765 | |
| 766 | /* |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 767 | * If the page isn't exclusively mapped into this vma, |
| 768 | * we must use the _oldest_ possible anon_vma for the |
| 769 | * page mapping! |
Linus Torvalds | ea90002 | 2010-04-12 12:44:29 -0700 | [diff] [blame] | 770 | */ |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 771 | if (!exclusive) { |
Andrea Arcangeli | 288468c | 2010-08-09 17:19:09 -0700 | [diff] [blame^] | 772 | if (PageAnon(page)) |
| 773 | return; |
| 774 | anon_vma = anon_vma->root; |
| 775 | } else { |
| 776 | /* |
| 777 | * In this case, swapped-out-but-not-discarded swap-cache |
| 778 | * is remapped. So, no need to update page->mapping here. |
| 779 | * We convice anon_vma poitned by page->mapping is not obsolete |
| 780 | * because vma->anon_vma is necessary to be a family of it. |
| 781 | */ |
| 782 | if (PageAnon(page)) |
| 783 | return; |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 784 | } |
Linus Torvalds | ea90002 | 2010-04-12 12:44:29 -0700 | [diff] [blame] | 785 | |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 786 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 787 | page->mapping = (struct address_space *) anon_vma; |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 788 | page->index = linear_page_index(vma, address); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 789 | } |
| 790 | |
| 791 | /** |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 792 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 793 | * @page: the page to add the mapping to |
| 794 | * @vma: the vm area in which the mapping is added |
| 795 | * @address: the user virtual address mapped |
| 796 | */ |
| 797 | static void __page_check_anon_rmap(struct page *page, |
| 798 | struct vm_area_struct *vma, unsigned long address) |
| 799 | { |
| 800 | #ifdef CONFIG_DEBUG_VM |
| 801 | /* |
| 802 | * The page's anon-rmap details (mapping and index) are guaranteed to |
| 803 | * be set up correctly at this point. |
| 804 | * |
| 805 | * We have exclusion against page_add_anon_rmap because the caller |
| 806 | * always holds the page locked, except if called from page_dup_rmap, |
| 807 | * in which case the page is already known to be setup. |
| 808 | * |
| 809 | * We have exclusion against page_add_new_anon_rmap because those pages |
| 810 | * are initially only visible via the pagetables, and the pte is locked |
| 811 | * over the call to page_add_new_anon_rmap. |
| 812 | */ |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 813 | BUG_ON(page->index != linear_page_index(vma, address)); |
| 814 | #endif |
| 815 | } |
| 816 | |
| 817 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 818 | * page_add_anon_rmap - add pte mapping to an anonymous page |
| 819 | * @page: the page to add the mapping to |
| 820 | * @vma: the vm area in which the mapping is added |
| 821 | * @address: the user virtual address mapped |
| 822 | * |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 823 | * The caller needs to hold the pte lock, and the page must be locked in |
Hugh Dickins | 80e1482 | 2009-12-14 17:59:29 -0800 | [diff] [blame] | 824 | * the anon_vma case: to serialize mapping,index checking after setting, |
| 825 | * and to ensure that PageAnon is not being upgraded racily to PageKsm |
| 826 | * (but PageKsm is never downgraded to PageAnon). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | */ |
| 828 | void page_add_anon_rmap(struct page *page, |
| 829 | struct vm_area_struct *vma, unsigned long address) |
| 830 | { |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 831 | int first = atomic_inc_and_test(&page->_mapcount); |
| 832 | if (first) |
| 833 | __inc_zone_page_state(page, NR_ANON_PAGES); |
| 834 | if (unlikely(PageKsm(page))) |
| 835 | return; |
| 836 | |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 837 | VM_BUG_ON(!PageLocked(page)); |
| 838 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 839 | if (first) |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 840 | __page_set_anon_rmap(page, vma, address, 0); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 841 | else |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 842 | __page_check_anon_rmap(page, vma, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 843 | } |
| 844 | |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 845 | /** |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 846 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
| 847 | * @page: the page to add the mapping to |
| 848 | * @vma: the vm area in which the mapping is added |
| 849 | * @address: the user virtual address mapped |
| 850 | * |
| 851 | * Same as page_add_anon_rmap but must only be called on *new* pages. |
| 852 | * This means the inc-and-test can be bypassed. |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 853 | * Page does not have to be locked. |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 854 | */ |
| 855 | void page_add_new_anon_rmap(struct page *page, |
| 856 | struct vm_area_struct *vma, unsigned long address) |
| 857 | { |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 858 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
Hugh Dickins | cbf84b7 | 2009-01-06 14:39:27 -0800 | [diff] [blame] | 859 | SetPageSwapBacked(page); |
| 860 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 861 | __inc_zone_page_state(page, NR_ANON_PAGES); |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 862 | __page_set_anon_rmap(page, vma, address, 1); |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 863 | if (page_evictable(page, vma)) |
Hugh Dickins | cbf84b7 | 2009-01-06 14:39:27 -0800 | [diff] [blame] | 864 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 865 | else |
| 866 | add_page_to_unevictable_list(page); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 867 | } |
| 868 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 869 | /** |
| 870 | * page_add_file_rmap - add pte mapping to a file page |
| 871 | * @page: the page to add the mapping to |
| 872 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 873 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 874 | */ |
| 875 | void page_add_file_rmap(struct page *page) |
| 876 | { |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 877 | if (atomic_inc_and_test(&page->_mapcount)) { |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 878 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 879 | mem_cgroup_update_file_mapped(page, 1); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 880 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 881 | } |
| 882 | |
| 883 | /** |
| 884 | * page_remove_rmap - take down pte mapping from a page |
| 885 | * @page: page to remove mapping from |
| 886 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 887 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | */ |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 889 | void page_remove_rmap(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | { |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 891 | /* page still mapped by someone else? */ |
| 892 | if (!atomic_add_negative(-1, &page->_mapcount)) |
| 893 | return; |
| 894 | |
| 895 | /* |
| 896 | * Now that the last pte has gone, s390 must transfer dirty |
| 897 | * flag from storage key to struct page. We can usually skip |
| 898 | * this if the page is anon, so about to be freed; but perhaps |
| 899 | * not if it's in swapcache - there might be another pte slot |
| 900 | * containing the swap entry, but page not yet written to swap. |
| 901 | */ |
| 902 | if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { |
| 903 | page_clear_dirty(page); |
| 904 | set_page_dirty(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | } |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 906 | if (PageAnon(page)) { |
| 907 | mem_cgroup_uncharge_page(page); |
| 908 | __dec_zone_page_state(page, NR_ANON_PAGES); |
| 909 | } else { |
| 910 | __dec_zone_page_state(page, NR_FILE_MAPPED); |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 911 | mem_cgroup_update_file_mapped(page, -1); |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 912 | } |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 913 | /* |
| 914 | * It would be tidy to reset the PageAnon mapping here, |
| 915 | * but that might overwrite a racing page_add_anon_rmap |
| 916 | * which increments mapcount after us but sets mapping |
| 917 | * before us: so leave the reset to free_hot_cold_page, |
| 918 | * and remember that it's only reliable while mapped. |
| 919 | * Leaving it set also helps swapoff to reinstate ptes |
| 920 | * faster for those pages still in swapcache. |
| 921 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | } |
| 923 | |
| 924 | /* |
| 925 | * Subfunctions of try_to_unmap: try_to_unmap_one called |
| 926 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. |
| 927 | */ |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 928 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
| 929 | unsigned long address, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | { |
| 931 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | pte_t *pte; |
| 933 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 934 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | int ret = SWAP_AGAIN; |
| 936 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 937 | pte = page_check_address(page, mm, address, &ptl, 0); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 938 | if (!pte) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 939 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 940 | |
| 941 | /* |
| 942 | * If the page is mlock()d, we cannot swap it out. |
| 943 | * If it's recently referenced (perhaps page_referenced |
| 944 | * skipped over this mm) then we should reactivate it. |
| 945 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 946 | if (!(flags & TTU_IGNORE_MLOCK)) { |
KOSAKI Motohiro | caed0f4 | 2009-12-14 17:59:45 -0800 | [diff] [blame] | 947 | if (vma->vm_flags & VM_LOCKED) |
| 948 | goto out_mlock; |
| 949 | |
Hugh Dickins | af8e335 | 2009-12-14 17:58:59 -0800 | [diff] [blame] | 950 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 951 | goto out_unmap; |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 952 | } |
| 953 | if (!(flags & TTU_IGNORE_ACCESS)) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 954 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
| 955 | ret = SWAP_FAIL; |
| 956 | goto out_unmap; |
| 957 | } |
| 958 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 960 | /* Nuke the page table entry. */ |
| 961 | flush_cache_page(vma, address, page_to_pfn(page)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 962 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 963 | |
| 964 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 965 | if (pte_dirty(pteval)) |
| 966 | set_page_dirty(page); |
| 967 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 968 | /* Update high watermark before we lower rss */ |
| 969 | update_hiwater_rss(mm); |
| 970 | |
Andi Kleen | 888b9f7 | 2009-09-16 11:50:11 +0200 | [diff] [blame] | 971 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
| 972 | if (PageAnon(page)) |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 973 | dec_mm_counter(mm, MM_ANONPAGES); |
Andi Kleen | 888b9f7 | 2009-09-16 11:50:11 +0200 | [diff] [blame] | 974 | else |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 975 | dec_mm_counter(mm, MM_FILEPAGES); |
Andi Kleen | 888b9f7 | 2009-09-16 11:50:11 +0200 | [diff] [blame] | 976 | set_pte_at(mm, address, pte, |
| 977 | swp_entry_to_pte(make_hwpoison_entry(page))); |
| 978 | } else if (PageAnon(page)) { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 979 | swp_entry_t entry = { .val = page_private(page) }; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 980 | |
| 981 | if (PageSwapCache(page)) { |
| 982 | /* |
| 983 | * Store the swap location in the pte. |
| 984 | * See handle_pte_fault() ... |
| 985 | */ |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 986 | if (swap_duplicate(entry) < 0) { |
| 987 | set_pte_at(mm, address, pte, pteval); |
| 988 | ret = SWAP_FAIL; |
| 989 | goto out_unmap; |
| 990 | } |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 991 | if (list_empty(&mm->mmlist)) { |
| 992 | spin_lock(&mmlist_lock); |
| 993 | if (list_empty(&mm->mmlist)) |
| 994 | list_add(&mm->mmlist, &init_mm.mmlist); |
| 995 | spin_unlock(&mmlist_lock); |
| 996 | } |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 997 | dec_mm_counter(mm, MM_ANONPAGES); |
KAMEZAWA Hiroyuki | b084d43 | 2010-03-05 13:41:42 -0800 | [diff] [blame] | 998 | inc_mm_counter(mm, MM_SWAPENTS); |
KOSAKI Motohiro | 64cdd54 | 2009-01-06 14:39:16 -0800 | [diff] [blame] | 999 | } else if (PAGE_MIGRATION) { |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 1000 | /* |
| 1001 | * Store the pfn of the page in a special migration |
| 1002 | * pte. do_swap_page() will wait until the migration |
| 1003 | * pte is removed and then restart fault handling. |
| 1004 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1005 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 1006 | entry = make_migration_entry(page, pte_write(pteval)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | } |
| 1008 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 1009 | BUG_ON(pte_file(*pte)); |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1010 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 1011 | /* Establish migration entry for a file page */ |
| 1012 | swp_entry_t entry; |
| 1013 | entry = make_migration_entry(page, pte_write(pteval)); |
| 1014 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 1015 | } else |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1016 | dec_mm_counter(mm, MM_FILEPAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 1018 | page_remove_rmap(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1019 | page_cache_release(page); |
| 1020 | |
| 1021 | out_unmap: |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1022 | pte_unmap_unlock(pte, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1023 | out: |
| 1024 | return ret; |
KOSAKI Motohiro | caed0f4 | 2009-12-14 17:59:45 -0800 | [diff] [blame] | 1025 | |
| 1026 | out_mlock: |
| 1027 | pte_unmap_unlock(pte, ptl); |
| 1028 | |
| 1029 | |
| 1030 | /* |
| 1031 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes |
| 1032 | * unstable result and race. Plus, We can't wait here because |
| 1033 | * we now hold anon_vma->lock or mapping->i_mmap_lock. |
| 1034 | * if trylock failed, the page remain in evictable lru and later |
| 1035 | * vmscan could retry to move the page to unevictable lru if the |
| 1036 | * page is actually mlocked. |
| 1037 | */ |
| 1038 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
| 1039 | if (vma->vm_flags & VM_LOCKED) { |
| 1040 | mlock_vma_page(page); |
| 1041 | ret = SWAP_MLOCK; |
| 1042 | } |
| 1043 | up_read(&vma->vm_mm->mmap_sem); |
| 1044 | } |
| 1045 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | } |
| 1047 | |
| 1048 | /* |
| 1049 | * objrmap doesn't work for nonlinear VMAs because the assumption that |
| 1050 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. |
| 1051 | * Consequently, given a particular page and its ->index, we cannot locate the |
| 1052 | * ptes which are mapping that page without an exhaustive linear search. |
| 1053 | * |
| 1054 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which |
| 1055 | * maps the file to which the target page belongs. The ->vm_private_data field |
| 1056 | * holds the current cursor into that scan. Successive searches will circulate |
| 1057 | * around the vma's virtual address space. |
| 1058 | * |
| 1059 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, |
| 1060 | * more scanning pressure is placed against them as well. Eventually pages |
| 1061 | * will become fully unmapped and are eligible for eviction. |
| 1062 | * |
| 1063 | * For very sparsely populated VMAs this is a little inefficient - chances are |
| 1064 | * there there won't be many ptes located within the scan cluster. In this case |
| 1065 | * maybe we could scan further - to the end of the pte page, perhaps. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1066 | * |
| 1067 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can |
| 1068 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, |
| 1069 | * rather than unmapping them. If we encounter the "check_page" that vmscan is |
| 1070 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1071 | */ |
| 1072 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) |
| 1073 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) |
| 1074 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1075 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
| 1076 | struct vm_area_struct *vma, struct page *check_page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | { |
| 1078 | struct mm_struct *mm = vma->vm_mm; |
| 1079 | pgd_t *pgd; |
| 1080 | pud_t *pud; |
| 1081 | pmd_t *pmd; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1082 | pte_t *pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1083 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1084 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | struct page *page; |
| 1086 | unsigned long address; |
| 1087 | unsigned long end; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1088 | int ret = SWAP_AGAIN; |
| 1089 | int locked_vma = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1091 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
| 1092 | end = address + CLUSTER_SIZE; |
| 1093 | if (address < vma->vm_start) |
| 1094 | address = vma->vm_start; |
| 1095 | if (end > vma->vm_end) |
| 1096 | end = vma->vm_end; |
| 1097 | |
| 1098 | pgd = pgd_offset(mm, address); |
| 1099 | if (!pgd_present(*pgd)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1100 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1101 | |
| 1102 | pud = pud_offset(pgd, address); |
| 1103 | if (!pud_present(*pud)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1104 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | |
| 1106 | pmd = pmd_offset(pud, address); |
| 1107 | if (!pmd_present(*pmd)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1108 | return ret; |
| 1109 | |
| 1110 | /* |
Hugh Dickins | af8e335 | 2009-12-14 17:58:59 -0800 | [diff] [blame] | 1111 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1112 | * keep the sem while scanning the cluster for mlocking pages. |
| 1113 | */ |
Hugh Dickins | af8e335 | 2009-12-14 17:58:59 -0800 | [diff] [blame] | 1114 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1115 | locked_vma = (vma->vm_flags & VM_LOCKED); |
| 1116 | if (!locked_vma) |
| 1117 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ |
| 1118 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1119 | |
| 1120 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 1122 | /* Update high watermark before we lower rss */ |
| 1123 | update_hiwater_rss(mm); |
| 1124 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1125 | for (; address < end; pte++, address += PAGE_SIZE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | if (!pte_present(*pte)) |
| 1127 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1128 | page = vm_normal_page(vma, address, *pte); |
| 1129 | BUG_ON(!page || PageAnon(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1131 | if (locked_vma) { |
| 1132 | mlock_vma_page(page); /* no-op if already mlocked */ |
| 1133 | if (page == check_page) |
| 1134 | ret = SWAP_MLOCK; |
| 1135 | continue; /* don't unmap */ |
| 1136 | } |
| 1137 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1138 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | continue; |
| 1140 | |
| 1141 | /* Nuke the page table entry. */ |
Ben Collins | eca3513 | 2005-11-29 11:45:26 -0800 | [diff] [blame] | 1142 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1143 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1144 | |
| 1145 | /* If nonlinear, store the file page offset in the pte. */ |
| 1146 | if (page->index != linear_page_index(vma, address)) |
| 1147 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); |
| 1148 | |
| 1149 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 1150 | if (pte_dirty(pteval)) |
| 1151 | set_page_dirty(page); |
| 1152 | |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 1153 | page_remove_rmap(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | page_cache_release(page); |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1155 | dec_mm_counter(mm, MM_FILEPAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1156 | (*mapcount)--; |
| 1157 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1158 | pte_unmap_unlock(pte - 1, ptl); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1159 | if (locked_vma) |
| 1160 | up_read(&vma->vm_mm->mmap_sem); |
| 1161 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | } |
| 1163 | |
Mel Gorman | a8bef8ff | 2010-05-24 14:32:24 -0700 | [diff] [blame] | 1164 | static bool is_vma_temporary_stack(struct vm_area_struct *vma) |
| 1165 | { |
| 1166 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); |
| 1167 | |
| 1168 | if (!maybe_stack) |
| 1169 | return false; |
| 1170 | |
| 1171 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == |
| 1172 | VM_STACK_INCOMPLETE_SETUP) |
| 1173 | return true; |
| 1174 | |
| 1175 | return false; |
| 1176 | } |
| 1177 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1178 | /** |
| 1179 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based |
| 1180 | * rmap method |
| 1181 | * @page: the page to unmap/unlock |
Huang Shijie | 8051be5 | 2009-12-14 17:58:50 -0800 | [diff] [blame] | 1182 | * @flags: action and flags |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1183 | * |
| 1184 | * Find all the mappings of a page using the mapping pointer and the vma chains |
| 1185 | * contained in the anon_vma struct it points to. |
| 1186 | * |
| 1187 | * This function is only called from try_to_unmap/try_to_munlock for |
| 1188 | * anonymous pages. |
| 1189 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
| 1190 | * where the page was found will be held for write. So, we won't recheck |
| 1191 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
| 1192 | * 'LOCKED. |
| 1193 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1194 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | { |
| 1196 | struct anon_vma *anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1197 | struct anon_vma_chain *avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | int ret = SWAP_AGAIN; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1199 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1200 | anon_vma = page_lock_anon_vma(page); |
| 1201 | if (!anon_vma) |
| 1202 | return ret; |
| 1203 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1204 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
| 1205 | struct vm_area_struct *vma = avc->vma; |
Mel Gorman | a8bef8ff | 2010-05-24 14:32:24 -0700 | [diff] [blame] | 1206 | unsigned long address; |
| 1207 | |
| 1208 | /* |
| 1209 | * During exec, a temporary VMA is setup and later moved. |
| 1210 | * The VMA is moved under the anon_vma lock but not the |
| 1211 | * page tables leading to a race where migration cannot |
| 1212 | * find the migration ptes. Rather than increasing the |
| 1213 | * locking requirements of exec(), migration skips |
| 1214 | * temporary VMAs until after exec() completes. |
| 1215 | */ |
| 1216 | if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && |
| 1217 | is_vma_temporary_stack(vma)) |
| 1218 | continue; |
| 1219 | |
| 1220 | address = vma_address(page, vma); |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 1221 | if (address == -EFAULT) |
| 1222 | continue; |
| 1223 | ret = try_to_unmap_one(page, vma, address, flags); |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1224 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
| 1225 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1226 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 1227 | |
| 1228 | page_unlock_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | return ret; |
| 1230 | } |
| 1231 | |
| 1232 | /** |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1233 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
| 1234 | * @page: the page to unmap/unlock |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1235 | * @flags: action and flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | * |
| 1237 | * Find all the mappings of a page using the mapping pointer and the vma chains |
| 1238 | * contained in the address_space struct it points to. |
| 1239 | * |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1240 | * This function is only called from try_to_unmap/try_to_munlock for |
| 1241 | * object-based pages. |
| 1242 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
| 1243 | * where the page was found will be held for write. So, we won't recheck |
| 1244 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
| 1245 | * 'LOCKED. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1246 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1247 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1248 | { |
| 1249 | struct address_space *mapping = page->mapping; |
| 1250 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 1251 | struct vm_area_struct *vma; |
| 1252 | struct prio_tree_iter iter; |
| 1253 | int ret = SWAP_AGAIN; |
| 1254 | unsigned long cursor; |
| 1255 | unsigned long max_nl_cursor = 0; |
| 1256 | unsigned long max_nl_size = 0; |
| 1257 | unsigned int mapcount; |
| 1258 | |
| 1259 | spin_lock(&mapping->i_mmap_lock); |
| 1260 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 1261 | unsigned long address = vma_address(page, vma); |
| 1262 | if (address == -EFAULT) |
| 1263 | continue; |
| 1264 | ret = try_to_unmap_one(page, vma, address, flags); |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1265 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
| 1266 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | } |
| 1268 | |
| 1269 | if (list_empty(&mapping->i_mmap_nonlinear)) |
| 1270 | goto out; |
| 1271 | |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1272 | /* |
| 1273 | * We don't bother to try to find the munlocked page in nonlinears. |
| 1274 | * It's costly. Instead, later, page reclaim logic may call |
| 1275 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. |
| 1276 | */ |
| 1277 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
| 1278 | goto out; |
| 1279 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 1281 | shared.vm_set.list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | cursor = (unsigned long) vma->vm_private_data; |
| 1283 | if (cursor > max_nl_cursor) |
| 1284 | max_nl_cursor = cursor; |
| 1285 | cursor = vma->vm_end - vma->vm_start; |
| 1286 | if (cursor > max_nl_size) |
| 1287 | max_nl_size = cursor; |
| 1288 | } |
| 1289 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1290 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1291 | ret = SWAP_FAIL; |
| 1292 | goto out; |
| 1293 | } |
| 1294 | |
| 1295 | /* |
| 1296 | * We don't try to search for this page in the nonlinear vmas, |
| 1297 | * and page_referenced wouldn't have found it anyway. Instead |
| 1298 | * just walk the nonlinear vmas trying to age and unmap some. |
| 1299 | * The mapcount of the page we came in with is irrelevant, |
| 1300 | * but even so use it as a guide to how hard we should try? |
| 1301 | */ |
| 1302 | mapcount = page_mapcount(page); |
| 1303 | if (!mapcount) |
| 1304 | goto out; |
| 1305 | cond_resched_lock(&mapping->i_mmap_lock); |
| 1306 | |
| 1307 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; |
| 1308 | if (max_nl_cursor == 0) |
| 1309 | max_nl_cursor = CLUSTER_SIZE; |
| 1310 | |
| 1311 | do { |
| 1312 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 1313 | shared.vm_set.list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1314 | cursor = (unsigned long) vma->vm_private_data; |
Hugh Dickins | 839b968 | 2005-09-03 15:54:43 -0700 | [diff] [blame] | 1315 | while ( cursor < max_nl_cursor && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | cursor < vma->vm_end - vma->vm_start) { |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1317 | if (try_to_unmap_cluster(cursor, &mapcount, |
| 1318 | vma, page) == SWAP_MLOCK) |
| 1319 | ret = SWAP_MLOCK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | cursor += CLUSTER_SIZE; |
| 1321 | vma->vm_private_data = (void *) cursor; |
| 1322 | if ((int)mapcount <= 0) |
| 1323 | goto out; |
| 1324 | } |
| 1325 | vma->vm_private_data = (void *) max_nl_cursor; |
| 1326 | } |
| 1327 | cond_resched_lock(&mapping->i_mmap_lock); |
| 1328 | max_nl_cursor += CLUSTER_SIZE; |
| 1329 | } while (max_nl_cursor <= max_nl_size); |
| 1330 | |
| 1331 | /* |
| 1332 | * Don't loop forever (perhaps all the remaining pages are |
| 1333 | * in locked vmas). Reset cursor on all unreserved nonlinear |
| 1334 | * vmas, now forgetting on which ones it had fallen behind. |
| 1335 | */ |
Hugh Dickins | 101d2be | 2005-11-21 21:32:16 -0800 | [diff] [blame] | 1336 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
| 1337 | vma->vm_private_data = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | out: |
| 1339 | spin_unlock(&mapping->i_mmap_lock); |
| 1340 | return ret; |
| 1341 | } |
| 1342 | |
| 1343 | /** |
| 1344 | * try_to_unmap - try to remove all page table mappings to a page |
| 1345 | * @page: the page to get unmapped |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1346 | * @flags: action and flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1347 | * |
| 1348 | * Tries to remove all the page table entries which are mapping this |
| 1349 | * page, used in the pageout path. Caller must hold the page lock. |
| 1350 | * Return values are: |
| 1351 | * |
| 1352 | * SWAP_SUCCESS - we succeeded in removing all mappings |
| 1353 | * SWAP_AGAIN - we missed a mapping, try again later |
| 1354 | * SWAP_FAIL - the page is unswappable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1355 | * SWAP_MLOCK - page is mlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1357 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1358 | { |
| 1359 | int ret; |
| 1360 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | BUG_ON(!PageLocked(page)); |
| 1362 | |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1363 | if (unlikely(PageKsm(page))) |
| 1364 | ret = try_to_unmap_ksm(page, flags); |
| 1365 | else if (PageAnon(page)) |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1366 | ret = try_to_unmap_anon(page, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1367 | else |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1368 | ret = try_to_unmap_file(page, flags); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1369 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | ret = SWAP_SUCCESS; |
| 1371 | return ret; |
| 1372 | } |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 1373 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1374 | /** |
| 1375 | * try_to_munlock - try to munlock a page |
| 1376 | * @page: the page to be munlocked |
| 1377 | * |
| 1378 | * Called from munlock code. Checks all of the VMAs mapping the page |
| 1379 | * to make sure nobody else has this page mlocked. The page will be |
| 1380 | * returned with PG_mlocked cleared if no other vmas have it mlocked. |
| 1381 | * |
| 1382 | * Return values are: |
| 1383 | * |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1384 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1385 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1386 | * SWAP_FAIL - page cannot be located at present |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1387 | * SWAP_MLOCK - page is now mlocked. |
| 1388 | */ |
| 1389 | int try_to_munlock(struct page *page) |
| 1390 | { |
| 1391 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); |
| 1392 | |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1393 | if (unlikely(PageKsm(page))) |
| 1394 | return try_to_unmap_ksm(page, TTU_MUNLOCK); |
| 1395 | else if (PageAnon(page)) |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1396 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1397 | else |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1398 | return try_to_unmap_file(page, TTU_MUNLOCK); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1399 | } |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1400 | |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 1401 | #if defined(CONFIG_KSM) || defined(CONFIG_MIGRATION) |
| 1402 | /* |
| 1403 | * Drop an anon_vma refcount, freeing the anon_vma and anon_vma->root |
| 1404 | * if necessary. Be careful to do all the tests under the lock. Once |
| 1405 | * we know we are the last user, nobody else can get a reference and we |
| 1406 | * can do the freeing without the lock. |
| 1407 | */ |
| 1408 | void drop_anon_vma(struct anon_vma *anon_vma) |
| 1409 | { |
| 1410 | if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) { |
| 1411 | struct anon_vma *root = anon_vma->root; |
| 1412 | int empty = list_empty(&anon_vma->head); |
| 1413 | int last_root_user = 0; |
| 1414 | int root_empty = 0; |
| 1415 | |
| 1416 | /* |
| 1417 | * The refcount on a non-root anon_vma got dropped. Drop |
| 1418 | * the refcount on the root and check if we need to free it. |
| 1419 | */ |
| 1420 | if (empty && anon_vma != root) { |
| 1421 | last_root_user = atomic_dec_and_test(&root->external_refcount); |
| 1422 | root_empty = list_empty(&root->head); |
| 1423 | } |
| 1424 | anon_vma_unlock(anon_vma); |
| 1425 | |
| 1426 | if (empty) { |
| 1427 | anon_vma_free(anon_vma); |
| 1428 | if (root_empty && last_root_user) |
| 1429 | anon_vma_free(root); |
| 1430 | } |
| 1431 | } |
| 1432 | } |
| 1433 | #endif |
| 1434 | |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1435 | #ifdef CONFIG_MIGRATION |
| 1436 | /* |
| 1437 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): |
| 1438 | * Called by migrate.c to remove migration ptes, but might be used more later. |
| 1439 | */ |
| 1440 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, |
| 1441 | struct vm_area_struct *, unsigned long, void *), void *arg) |
| 1442 | { |
| 1443 | struct anon_vma *anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1444 | struct anon_vma_chain *avc; |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1445 | int ret = SWAP_AGAIN; |
| 1446 | |
| 1447 | /* |
| 1448 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma() |
| 1449 | * because that depends on page_mapped(); but not all its usages |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 1450 | * are holding mmap_sem. Users without mmap_sem are required to |
| 1451 | * take a reference count to prevent the anon_vma disappearing |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1452 | */ |
| 1453 | anon_vma = page_anon_vma(page); |
| 1454 | if (!anon_vma) |
| 1455 | return ret; |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 1456 | anon_vma_lock(anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1457 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
| 1458 | struct vm_area_struct *vma = avc->vma; |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1459 | unsigned long address = vma_address(page, vma); |
| 1460 | if (address == -EFAULT) |
| 1461 | continue; |
| 1462 | ret = rmap_one(page, vma, address, arg); |
| 1463 | if (ret != SWAP_AGAIN) |
| 1464 | break; |
| 1465 | } |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 1466 | anon_vma_unlock(anon_vma); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1467 | return ret; |
| 1468 | } |
| 1469 | |
| 1470 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, |
| 1471 | struct vm_area_struct *, unsigned long, void *), void *arg) |
| 1472 | { |
| 1473 | struct address_space *mapping = page->mapping; |
| 1474 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 1475 | struct vm_area_struct *vma; |
| 1476 | struct prio_tree_iter iter; |
| 1477 | int ret = SWAP_AGAIN; |
| 1478 | |
| 1479 | if (!mapping) |
| 1480 | return ret; |
| 1481 | spin_lock(&mapping->i_mmap_lock); |
| 1482 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
| 1483 | unsigned long address = vma_address(page, vma); |
| 1484 | if (address == -EFAULT) |
| 1485 | continue; |
| 1486 | ret = rmap_one(page, vma, address, arg); |
| 1487 | if (ret != SWAP_AGAIN) |
| 1488 | break; |
| 1489 | } |
| 1490 | /* |
| 1491 | * No nonlinear handling: being always shared, nonlinear vmas |
| 1492 | * never contain migration ptes. Decide what to do about this |
| 1493 | * limitation to linear when we need rmap_walk() on nonlinear. |
| 1494 | */ |
| 1495 | spin_unlock(&mapping->i_mmap_lock); |
| 1496 | return ret; |
| 1497 | } |
| 1498 | |
| 1499 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, |
| 1500 | struct vm_area_struct *, unsigned long, void *), void *arg) |
| 1501 | { |
| 1502 | VM_BUG_ON(!PageLocked(page)); |
| 1503 | |
| 1504 | if (unlikely(PageKsm(page))) |
| 1505 | return rmap_walk_ksm(page, rmap_one, arg); |
| 1506 | else if (PageAnon(page)) |
| 1507 | return rmap_walk_anon(page, rmap_one, arg); |
| 1508 | else |
| 1509 | return rmap_walk_file(page, rmap_one, arg); |
| 1510 | } |
| 1511 | #endif /* CONFIG_MIGRATION */ |