Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/swapfile.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 5 | * Swap reorganised 29.12.95, Stephen Tweedie |
| 6 | */ |
| 7 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | #include <linux/mm.h> |
| 9 | #include <linux/hugetlb.h> |
| 10 | #include <linux/mman.h> |
| 11 | #include <linux/slab.h> |
| 12 | #include <linux/kernel_stat.h> |
| 13 | #include <linux/swap.h> |
| 14 | #include <linux/vmalloc.h> |
| 15 | #include <linux/pagemap.h> |
| 16 | #include <linux/namei.h> |
Hugh Dickins | 072441e | 2011-06-27 16:18:02 -0700 | [diff] [blame] | 17 | #include <linux/shmem_fs.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/blkdev.h> |
Hugh Dickins | 20137a4 | 2009-01-06 14:39:54 -0800 | [diff] [blame] | 19 | #include <linux/random.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/writeback.h> |
| 21 | #include <linux/proc_fs.h> |
| 22 | #include <linux/seq_file.h> |
| 23 | #include <linux/init.h> |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 24 | #include <linux/ksm.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/rmap.h> |
| 26 | #include <linux/security.h> |
| 27 | #include <linux/backing-dev.h> |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 28 | #include <linux/mutex.h> |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 29 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <linux/syscalls.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 31 | #include <linux/memcontrol.h> |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 32 | #include <linux/poll.h> |
David Rientjes | 72788c3 | 2011-05-24 17:11:40 -0700 | [diff] [blame] | 33 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | |
| 35 | #include <asm/pgtable.h> |
| 36 | #include <asm/tlbflush.h> |
| 37 | #include <linux/swapops.h> |
KAMEZAWA Hiroyuki | 27a7faa | 2009-01-07 18:07:58 -0800 | [diff] [blame] | 38 | #include <linux/page_cgroup.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 40 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
| 41 | unsigned char); |
| 42 | static void free_swap_count_continuations(struct swap_info_struct *); |
Lee Schermerhorn | d4906e1 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 43 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 44 | |
Adrian Bunk | 7c363b8 | 2008-07-25 19:46:24 -0700 | [diff] [blame] | 45 | static DEFINE_SPINLOCK(swap_lock); |
| 46 | static unsigned int nr_swapfiles; |
Hugh Dickins | b962716 | 2009-01-06 14:39:41 -0800 | [diff] [blame] | 47 | long nr_swap_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | long total_swap_pages; |
Hugh Dickins | 78ecba0 | 2008-07-23 21:28:23 -0700 | [diff] [blame] | 49 | static int least_priority; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | static const char Bad_file[] = "Bad swap file entry "; |
| 52 | static const char Unused_file[] = "Unused swap file entry "; |
| 53 | static const char Bad_offset[] = "Bad swap offset entry "; |
| 54 | static const char Unused_offset[] = "Unused swap offset entry "; |
| 55 | |
Adrian Bunk | 7c363b8 | 2008-07-25 19:46:24 -0700 | [diff] [blame] | 56 | static struct swap_list_t swap_list = {-1, -1}; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 58 | static struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 60 | static DEFINE_MUTEX(swapon_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 62 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
| 63 | /* Activity counter to indicate that a swapon or swapoff has occurred */ |
| 64 | static atomic_t proc_poll_event = ATOMIC_INIT(0); |
| 65 | |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 66 | static inline unsigned char swap_count(unsigned char ent) |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 67 | { |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 68 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 69 | } |
| 70 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 71 | /* returns 1 if swap entry is freed */ |
KAMEZAWA Hiroyuki | c9e4441 | 2009-06-16 15:32:54 -0700 | [diff] [blame] | 72 | static int |
| 73 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) |
| 74 | { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 75 | swp_entry_t entry = swp_entry(si->type, offset); |
KAMEZAWA Hiroyuki | c9e4441 | 2009-06-16 15:32:54 -0700 | [diff] [blame] | 76 | struct page *page; |
| 77 | int ret = 0; |
| 78 | |
| 79 | page = find_get_page(&swapper_space, entry.val); |
| 80 | if (!page) |
| 81 | return 0; |
| 82 | /* |
| 83 | * This function is called from scan_swap_map() and it's called |
| 84 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. |
| 85 | * We have to use trylock for avoiding deadlock. This is a special |
| 86 | * case and you should use try_to_free_swap() with explicit lock_page() |
| 87 | * in usual operations. |
| 88 | */ |
| 89 | if (trylock_page(page)) { |
| 90 | ret = try_to_free_swap(page); |
| 91 | unlock_page(page); |
| 92 | } |
| 93 | page_cache_release(page); |
| 94 | return ret; |
| 95 | } |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 96 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | /* |
Hugh Dickins | 6a6ba83 | 2009-01-06 14:39:51 -0800 | [diff] [blame] | 98 | * swapon tell device that all the old swap contents can be discarded, |
| 99 | * to allow the swap device to optimize its wear-levelling. |
| 100 | */ |
| 101 | static int discard_swap(struct swap_info_struct *si) |
| 102 | { |
| 103 | struct swap_extent *se; |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 104 | sector_t start_block; |
| 105 | sector_t nr_blocks; |
Hugh Dickins | 6a6ba83 | 2009-01-06 14:39:51 -0800 | [diff] [blame] | 106 | int err = 0; |
| 107 | |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 108 | /* Do not discard the swap header page! */ |
| 109 | se = &si->first_swap_extent; |
| 110 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
| 111 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); |
| 112 | if (nr_blocks) { |
| 113 | err = blkdev_issue_discard(si->bdev, start_block, |
Christoph Hellwig | dd3932e | 2010-09-16 20:51:46 +0200 | [diff] [blame] | 114 | nr_blocks, GFP_KERNEL, 0); |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 115 | if (err) |
| 116 | return err; |
| 117 | cond_resched(); |
| 118 | } |
Hugh Dickins | 6a6ba83 | 2009-01-06 14:39:51 -0800 | [diff] [blame] | 119 | |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 120 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
| 121 | start_block = se->start_block << (PAGE_SHIFT - 9); |
| 122 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); |
Hugh Dickins | 6a6ba83 | 2009-01-06 14:39:51 -0800 | [diff] [blame] | 123 | |
| 124 | err = blkdev_issue_discard(si->bdev, start_block, |
Christoph Hellwig | dd3932e | 2010-09-16 20:51:46 +0200 | [diff] [blame] | 125 | nr_blocks, GFP_KERNEL, 0); |
Hugh Dickins | 6a6ba83 | 2009-01-06 14:39:51 -0800 | [diff] [blame] | 126 | if (err) |
| 127 | break; |
| 128 | |
| 129 | cond_resched(); |
| 130 | } |
| 131 | return err; /* That will often be -EOPNOTSUPP */ |
| 132 | } |
| 133 | |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 134 | /* |
| 135 | * swap allocation tell device that a cluster of swap can now be discarded, |
| 136 | * to allow the swap device to optimize its wear-levelling. |
| 137 | */ |
| 138 | static void discard_swap_cluster(struct swap_info_struct *si, |
| 139 | pgoff_t start_page, pgoff_t nr_pages) |
| 140 | { |
| 141 | struct swap_extent *se = si->curr_swap_extent; |
| 142 | int found_extent = 0; |
| 143 | |
| 144 | while (nr_pages) { |
| 145 | struct list_head *lh; |
| 146 | |
| 147 | if (se->start_page <= start_page && |
| 148 | start_page < se->start_page + se->nr_pages) { |
| 149 | pgoff_t offset = start_page - se->start_page; |
| 150 | sector_t start_block = se->start_block + offset; |
Hugh Dickins | 858a2990 | 2009-01-06 14:39:56 -0800 | [diff] [blame] | 151 | sector_t nr_blocks = se->nr_pages - offset; |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 152 | |
| 153 | if (nr_blocks > nr_pages) |
| 154 | nr_blocks = nr_pages; |
| 155 | start_page += nr_blocks; |
| 156 | nr_pages -= nr_blocks; |
| 157 | |
| 158 | if (!found_extent++) |
| 159 | si->curr_swap_extent = se; |
| 160 | |
| 161 | start_block <<= PAGE_SHIFT - 9; |
| 162 | nr_blocks <<= PAGE_SHIFT - 9; |
| 163 | if (blkdev_issue_discard(si->bdev, start_block, |
Christoph Hellwig | dd3932e | 2010-09-16 20:51:46 +0200 | [diff] [blame] | 164 | nr_blocks, GFP_NOIO, 0)) |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 165 | break; |
| 166 | } |
| 167 | |
| 168 | lh = se->list.next; |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 169 | se = list_entry(lh, struct swap_extent, list); |
| 170 | } |
| 171 | } |
| 172 | |
| 173 | static int wait_for_discard(void *word) |
| 174 | { |
| 175 | schedule(); |
| 176 | return 0; |
| 177 | } |
| 178 | |
Hugh Dickins | 048c27f | 2005-09-03 15:54:40 -0700 | [diff] [blame] | 179 | #define SWAPFILE_CLUSTER 256 |
| 180 | #define LATENCY_LIMIT 256 |
| 181 | |
Cesar Eduardo Barros | 24b8ff7 | 2011-03-22 16:33:38 -0700 | [diff] [blame] | 182 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
| 183 | unsigned char usage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | { |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 185 | unsigned long offset; |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 186 | unsigned long scan_base; |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 187 | unsigned long last_in_cluster = 0; |
Hugh Dickins | 048c27f | 2005-09-03 15:54:40 -0700 | [diff] [blame] | 188 | int latency_ration = LATENCY_LIMIT; |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 189 | int found_free_cluster = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
Hugh Dickins | 886bb7e | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 191 | /* |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 192 | * We try to cluster swap pages by allocating them sequentially |
| 193 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this |
| 194 | * way, however, we resort to first-free allocation, starting |
| 195 | * a new cluster. This prevents us from scattering swap pages |
| 196 | * all over the entire swap partition, so that we reduce |
| 197 | * overall disk seek times between swap pages. -- sct |
| 198 | * But we do now try to find an empty cluster. -Andrea |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 199 | * And we let swap pages go all over an SSD partition. Hugh |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 200 | */ |
| 201 | |
Hugh Dickins | 52b7efdb | 2005-09-03 15:54:39 -0700 | [diff] [blame] | 202 | si->flags += SWP_SCANNING; |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 203 | scan_base = offset = si->cluster_next; |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 204 | |
| 205 | if (unlikely(!si->cluster_nr--)) { |
| 206 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
| 207 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
| 208 | goto checks; |
| 209 | } |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 210 | if (si->flags & SWP_DISCARDABLE) { |
| 211 | /* |
| 212 | * Start range check on racing allocations, in case |
| 213 | * they overlap the cluster we eventually decide on |
| 214 | * (we scan without swap_lock to allow preemption). |
| 215 | * It's hardly conceivable that cluster_nr could be |
| 216 | * wrapped during our scan, but don't depend on it. |
| 217 | */ |
| 218 | if (si->lowest_alloc) |
| 219 | goto checks; |
| 220 | si->lowest_alloc = si->max; |
| 221 | si->highest_alloc = 0; |
| 222 | } |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 223 | spin_unlock(&swap_lock); |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 224 | |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 225 | /* |
| 226 | * If seek is expensive, start searching for new cluster from |
| 227 | * start of partition, to minimize the span of allocated swap. |
| 228 | * But if seek is cheap, search from our current position, so |
| 229 | * that swap is allocated from all over the partition: if the |
| 230 | * Flash Translation Layer only remaps within limited zones, |
| 231 | * we don't want to wear out the first zone too quickly. |
| 232 | */ |
| 233 | if (!(si->flags & SWP_SOLIDSTATE)) |
| 234 | scan_base = offset = si->lowest_bit; |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 235 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
| 236 | |
| 237 | /* Locate the first empty (unaligned) cluster */ |
| 238 | for (; last_in_cluster <= si->highest_bit; offset++) { |
| 239 | if (si->swap_map[offset]) |
| 240 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
| 241 | else if (offset == last_in_cluster) { |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 242 | spin_lock(&swap_lock); |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 243 | offset -= SWAPFILE_CLUSTER - 1; |
| 244 | si->cluster_next = offset; |
| 245 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 246 | found_free_cluster = 1; |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 247 | goto checks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | } |
Hugh Dickins | 048c27f | 2005-09-03 15:54:40 -0700 | [diff] [blame] | 249 | if (unlikely(--latency_ration < 0)) { |
| 250 | cond_resched(); |
| 251 | latency_ration = LATENCY_LIMIT; |
| 252 | } |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 253 | } |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 254 | |
| 255 | offset = si->lowest_bit; |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 256 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
| 257 | |
| 258 | /* Locate the first empty (unaligned) cluster */ |
| 259 | for (; last_in_cluster < scan_base; offset++) { |
| 260 | if (si->swap_map[offset]) |
| 261 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
| 262 | else if (offset == last_in_cluster) { |
| 263 | spin_lock(&swap_lock); |
| 264 | offset -= SWAPFILE_CLUSTER - 1; |
| 265 | si->cluster_next = offset; |
| 266 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
| 267 | found_free_cluster = 1; |
| 268 | goto checks; |
| 269 | } |
| 270 | if (unlikely(--latency_ration < 0)) { |
| 271 | cond_resched(); |
| 272 | latency_ration = LATENCY_LIMIT; |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | offset = scan_base; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 277 | spin_lock(&swap_lock); |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 278 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 279 | si->lowest_alloc = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | } |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 281 | |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 282 | checks: |
| 283 | if (!(si->flags & SWP_WRITEOK)) |
Hugh Dickins | 52b7efdb | 2005-09-03 15:54:39 -0700 | [diff] [blame] | 284 | goto no_page; |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 285 | if (!si->highest_bit) |
| 286 | goto no_page; |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 287 | if (offset > si->highest_bit) |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 288 | scan_base = offset = si->lowest_bit; |
KAMEZAWA Hiroyuki | c9e4441 | 2009-06-16 15:32:54 -0700 | [diff] [blame] | 289 | |
Hugh Dickins | b73d7fc | 2010-09-09 16:38:09 -0700 | [diff] [blame] | 290 | /* reuse swap entry of cache-only swap if not busy. */ |
| 291 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
KAMEZAWA Hiroyuki | c9e4441 | 2009-06-16 15:32:54 -0700 | [diff] [blame] | 292 | int swap_was_freed; |
| 293 | spin_unlock(&swap_lock); |
| 294 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
| 295 | spin_lock(&swap_lock); |
| 296 | /* entry was freed successfully, try to use this again */ |
| 297 | if (swap_was_freed) |
| 298 | goto checks; |
| 299 | goto scan; /* check next one */ |
| 300 | } |
| 301 | |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 302 | if (si->swap_map[offset]) |
| 303 | goto scan; |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 304 | |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 305 | if (offset == si->lowest_bit) |
| 306 | si->lowest_bit++; |
| 307 | if (offset == si->highest_bit) |
| 308 | si->highest_bit--; |
| 309 | si->inuse_pages++; |
| 310 | if (si->inuse_pages == si->pages) { |
| 311 | si->lowest_bit = si->max; |
| 312 | si->highest_bit = 0; |
| 313 | } |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 314 | si->swap_map[offset] = usage; |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 315 | si->cluster_next = offset + 1; |
| 316 | si->flags -= SWP_SCANNING; |
Hugh Dickins | 7992fde | 2009-01-06 14:39:53 -0800 | [diff] [blame] | 317 | |
| 318 | if (si->lowest_alloc) { |
| 319 | /* |
| 320 | * Only set when SWP_DISCARDABLE, and there's a scan |
| 321 | * for a free cluster in progress or just completed. |
| 322 | */ |
| 323 | if (found_free_cluster) { |
| 324 | /* |
| 325 | * To optimize wear-levelling, discard the |
| 326 | * old data of the cluster, taking care not to |
| 327 | * discard any of its pages that have already |
| 328 | * been allocated by racing tasks (offset has |
| 329 | * already stepped over any at the beginning). |
| 330 | */ |
| 331 | if (offset < si->highest_alloc && |
| 332 | si->lowest_alloc <= last_in_cluster) |
| 333 | last_in_cluster = si->lowest_alloc - 1; |
| 334 | si->flags |= SWP_DISCARDING; |
| 335 | spin_unlock(&swap_lock); |
| 336 | |
| 337 | if (offset < last_in_cluster) |
| 338 | discard_swap_cluster(si, offset, |
| 339 | last_in_cluster - offset + 1); |
| 340 | |
| 341 | spin_lock(&swap_lock); |
| 342 | si->lowest_alloc = 0; |
| 343 | si->flags &= ~SWP_DISCARDING; |
| 344 | |
| 345 | smp_mb(); /* wake_up_bit advises this */ |
| 346 | wake_up_bit(&si->flags, ilog2(SWP_DISCARDING)); |
| 347 | |
| 348 | } else if (si->flags & SWP_DISCARDING) { |
| 349 | /* |
| 350 | * Delay using pages allocated by racing tasks |
| 351 | * until the whole discard has been issued. We |
| 352 | * could defer that delay until swap_writepage, |
| 353 | * but it's easier to keep this self-contained. |
| 354 | */ |
| 355 | spin_unlock(&swap_lock); |
| 356 | wait_on_bit(&si->flags, ilog2(SWP_DISCARDING), |
| 357 | wait_for_discard, TASK_UNINTERRUPTIBLE); |
| 358 | spin_lock(&swap_lock); |
| 359 | } else { |
| 360 | /* |
| 361 | * Note pages allocated by racing tasks while |
| 362 | * scan for a free cluster is in progress, so |
| 363 | * that its final discard can exclude them. |
| 364 | */ |
| 365 | if (offset < si->lowest_alloc) |
| 366 | si->lowest_alloc = offset; |
| 367 | if (offset > si->highest_alloc) |
| 368 | si->highest_alloc = offset; |
| 369 | } |
| 370 | } |
Hugh Dickins | ebebbbe | 2009-01-06 14:39:50 -0800 | [diff] [blame] | 371 | return offset; |
| 372 | |
| 373 | scan: |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 374 | spin_unlock(&swap_lock); |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 375 | while (++offset <= si->highest_bit) { |
Hugh Dickins | 52b7efdb | 2005-09-03 15:54:39 -0700 | [diff] [blame] | 376 | if (!si->swap_map[offset]) { |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 377 | spin_lock(&swap_lock); |
Hugh Dickins | 52b7efdb | 2005-09-03 15:54:39 -0700 | [diff] [blame] | 378 | goto checks; |
| 379 | } |
KAMEZAWA Hiroyuki | c9e4441 | 2009-06-16 15:32:54 -0700 | [diff] [blame] | 380 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
| 381 | spin_lock(&swap_lock); |
| 382 | goto checks; |
| 383 | } |
Hugh Dickins | 048c27f | 2005-09-03 15:54:40 -0700 | [diff] [blame] | 384 | if (unlikely(--latency_ration < 0)) { |
| 385 | cond_resched(); |
| 386 | latency_ration = LATENCY_LIMIT; |
| 387 | } |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 388 | } |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 389 | offset = si->lowest_bit; |
| 390 | while (++offset < scan_base) { |
| 391 | if (!si->swap_map[offset]) { |
| 392 | spin_lock(&swap_lock); |
| 393 | goto checks; |
| 394 | } |
KAMEZAWA Hiroyuki | c9e4441 | 2009-06-16 15:32:54 -0700 | [diff] [blame] | 395 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
| 396 | spin_lock(&swap_lock); |
| 397 | goto checks; |
| 398 | } |
Hugh Dickins | c60aa17 | 2009-01-06 14:39:55 -0800 | [diff] [blame] | 399 | if (unlikely(--latency_ration < 0)) { |
| 400 | cond_resched(); |
| 401 | latency_ration = LATENCY_LIMIT; |
| 402 | } |
| 403 | } |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 404 | spin_lock(&swap_lock); |
Hugh Dickins | 7dfad41 | 2005-09-03 15:54:38 -0700 | [diff] [blame] | 405 | |
| 406 | no_page: |
Hugh Dickins | 52b7efdb | 2005-09-03 15:54:39 -0700 | [diff] [blame] | 407 | si->flags -= SWP_SCANNING; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | return 0; |
| 409 | } |
| 410 | |
| 411 | swp_entry_t get_swap_page(void) |
| 412 | { |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 413 | struct swap_info_struct *si; |
| 414 | pgoff_t offset; |
| 415 | int type, next; |
| 416 | int wrapped = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 418 | spin_lock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 419 | if (nr_swap_pages <= 0) |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 420 | goto noswap; |
| 421 | nr_swap_pages--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 423 | for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 424 | si = swap_info[type]; |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 425 | next = si->next; |
| 426 | if (next < 0 || |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 427 | (!wrapped && si->prio != swap_info[next]->prio)) { |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 428 | next = swap_list.head; |
| 429 | wrapped++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | } |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 431 | |
| 432 | if (!si->highest_bit) |
| 433 | continue; |
| 434 | if (!(si->flags & SWP_WRITEOK)) |
| 435 | continue; |
| 436 | |
| 437 | swap_list.next = next; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 438 | /* This is called for allocating swap entry for cache */ |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 439 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 440 | if (offset) { |
| 441 | spin_unlock(&swap_lock); |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 442 | return swp_entry(type, offset); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 443 | } |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 444 | next = swap_list.next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | } |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 446 | |
| 447 | nr_swap_pages++; |
| 448 | noswap: |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 449 | spin_unlock(&swap_lock); |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 450 | return (swp_entry_t) {0}; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | } |
| 452 | |
Hugh Dickins | 910321e | 2010-09-09 16:38:07 -0700 | [diff] [blame] | 453 | /* The only caller of this function is now susupend routine */ |
| 454 | swp_entry_t get_swap_page_of_type(int type) |
| 455 | { |
| 456 | struct swap_info_struct *si; |
| 457 | pgoff_t offset; |
| 458 | |
| 459 | spin_lock(&swap_lock); |
| 460 | si = swap_info[type]; |
| 461 | if (si && (si->flags & SWP_WRITEOK)) { |
| 462 | nr_swap_pages--; |
| 463 | /* This is called for allocating swap entry, not cache */ |
| 464 | offset = scan_swap_map(si, 1); |
| 465 | if (offset) { |
| 466 | spin_unlock(&swap_lock); |
| 467 | return swp_entry(type, offset); |
| 468 | } |
| 469 | nr_swap_pages++; |
| 470 | } |
| 471 | spin_unlock(&swap_lock); |
| 472 | return (swp_entry_t) {0}; |
| 473 | } |
| 474 | |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 475 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | { |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 477 | struct swap_info_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | unsigned long offset, type; |
| 479 | |
| 480 | if (!entry.val) |
| 481 | goto out; |
| 482 | type = swp_type(entry); |
| 483 | if (type >= nr_swapfiles) |
| 484 | goto bad_nofile; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 485 | p = swap_info[type]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | if (!(p->flags & SWP_USED)) |
| 487 | goto bad_device; |
| 488 | offset = swp_offset(entry); |
| 489 | if (offset >= p->max) |
| 490 | goto bad_offset; |
| 491 | if (!p->swap_map[offset]) |
| 492 | goto bad_free; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 493 | spin_lock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | return p; |
| 495 | |
| 496 | bad_free: |
| 497 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val); |
| 498 | goto out; |
| 499 | bad_offset: |
| 500 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val); |
| 501 | goto out; |
| 502 | bad_device: |
| 503 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val); |
| 504 | goto out; |
| 505 | bad_nofile: |
| 506 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val); |
| 507 | out: |
| 508 | return NULL; |
Hugh Dickins | 886bb7e | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 509 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 511 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
| 512 | swp_entry_t entry, unsigned char usage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | { |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 514 | unsigned long offset = swp_offset(entry); |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 515 | unsigned char count; |
| 516 | unsigned char has_cache; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 518 | count = p->swap_map[offset]; |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 519 | has_cache = count & SWAP_HAS_CACHE; |
| 520 | count &= ~SWAP_HAS_CACHE; |
| 521 | |
| 522 | if (usage == SWAP_HAS_CACHE) { |
| 523 | VM_BUG_ON(!has_cache); |
| 524 | has_cache = 0; |
Hugh Dickins | aaa4686 | 2009-12-14 17:58:47 -0800 | [diff] [blame] | 525 | } else if (count == SWAP_MAP_SHMEM) { |
| 526 | /* |
| 527 | * Or we could insist on shmem.c using a special |
| 528 | * swap_shmem_free() and free_shmem_swap_and_cache()... |
| 529 | */ |
| 530 | count = 0; |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 531 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
| 532 | if (count == COUNT_CONTINUED) { |
| 533 | if (swap_count_continued(p, offset, count)) |
| 534 | count = SWAP_MAP_MAX | COUNT_CONTINUED; |
| 535 | else |
| 536 | count = SWAP_MAP_MAX; |
| 537 | } else |
| 538 | count--; |
| 539 | } |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 540 | |
| 541 | if (!count) |
| 542 | mem_cgroup_uncharge_swap(entry); |
| 543 | |
| 544 | usage = count | has_cache; |
| 545 | p->swap_map[offset] = usage; |
| 546 | |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 547 | /* free if no reference */ |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 548 | if (!usage) { |
Nitin Gupta | b3a27d0 | 2010-05-17 11:02:43 +0530 | [diff] [blame] | 549 | struct gendisk *disk = p->bdev->bd_disk; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 550 | if (offset < p->lowest_bit) |
| 551 | p->lowest_bit = offset; |
| 552 | if (offset > p->highest_bit) |
| 553 | p->highest_bit = offset; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 554 | if (swap_list.next >= 0 && |
| 555 | p->prio > swap_info[swap_list.next]->prio) |
| 556 | swap_list.next = p->type; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 557 | nr_swap_pages++; |
| 558 | p->inuse_pages--; |
Nitin Gupta | b3a27d0 | 2010-05-17 11:02:43 +0530 | [diff] [blame] | 559 | if ((p->flags & SWP_BLKDEV) && |
| 560 | disk->fops->swap_slot_free_notify) |
| 561 | disk->fops->swap_slot_free_notify(p->bdev, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | } |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 563 | |
| 564 | return usage; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | } |
| 566 | |
| 567 | /* |
| 568 | * Caller has made sure that the swapdevice corresponding to entry |
| 569 | * is still around or has not been recycled. |
| 570 | */ |
| 571 | void swap_free(swp_entry_t entry) |
| 572 | { |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 573 | struct swap_info_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | |
| 575 | p = swap_info_get(entry); |
| 576 | if (p) { |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 577 | swap_entry_free(p, entry, 1); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 578 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | } |
| 580 | } |
| 581 | |
| 582 | /* |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 583 | * Called after dropping swapcache to decrease refcnt to swap entries. |
| 584 | */ |
| 585 | void swapcache_free(swp_entry_t entry, struct page *page) |
| 586 | { |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 587 | struct swap_info_struct *p; |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 588 | unsigned char count; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 589 | |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 590 | p = swap_info_get(entry); |
| 591 | if (p) { |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 592 | count = swap_entry_free(p, entry, SWAP_HAS_CACHE); |
| 593 | if (page) |
| 594 | mem_cgroup_uncharge_swapcache(page, entry, count != 0); |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 595 | spin_unlock(&swap_lock); |
| 596 | } |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 597 | } |
| 598 | |
| 599 | /* |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 600 | * How many references to page are currently swapped out? |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 601 | * This does not give an exact answer when swap count is continued, |
| 602 | * but does include the high COUNT_CONTINUED flag to allow for that. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | */ |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 604 | static inline int page_swapcount(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 605 | { |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 606 | int count = 0; |
| 607 | struct swap_info_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | swp_entry_t entry; |
| 609 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 610 | entry.val = page_private(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | p = swap_info_get(entry); |
| 612 | if (p) { |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 613 | count = swap_count(p->swap_map[swp_offset(entry)]); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 614 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 615 | } |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 616 | return count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | } |
| 618 | |
| 619 | /* |
Hugh Dickins | 7b1fe59 | 2009-01-06 14:39:34 -0800 | [diff] [blame] | 620 | * We can write to an anon page without COW if there are no other references |
| 621 | * to it. And as a side-effect, free up its swap: because the old content |
| 622 | * on disk will never be read, and seeking back there to write new content |
| 623 | * later would only waste time away from clustering. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 624 | */ |
Hugh Dickins | 7b1fe59 | 2009-01-06 14:39:34 -0800 | [diff] [blame] | 625 | int reuse_swap_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | { |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 627 | int count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | |
Hugh Dickins | 51726b1 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 629 | VM_BUG_ON(!PageLocked(page)); |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 630 | if (unlikely(PageKsm(page))) |
| 631 | return 0; |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 632 | count = page_mapcount(page); |
Hugh Dickins | 7b1fe59 | 2009-01-06 14:39:34 -0800 | [diff] [blame] | 633 | if (count <= 1 && PageSwapCache(page)) { |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 634 | count += page_swapcount(page); |
Hugh Dickins | 7b1fe59 | 2009-01-06 14:39:34 -0800 | [diff] [blame] | 635 | if (count == 1 && !PageWriteback(page)) { |
| 636 | delete_from_swap_cache(page); |
| 637 | SetPageDirty(page); |
| 638 | } |
| 639 | } |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 640 | return count <= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | } |
| 642 | |
| 643 | /* |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 644 | * If swap is getting full, or if there are no more mappings of this page, |
| 645 | * then try_to_free_swap is called to free its swap space. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | */ |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 647 | int try_to_free_swap(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | { |
Hugh Dickins | 51726b1 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 649 | VM_BUG_ON(!PageLocked(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | |
| 651 | if (!PageSwapCache(page)) |
| 652 | return 0; |
| 653 | if (PageWriteback(page)) |
| 654 | return 0; |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 655 | if (page_swapcount(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 656 | return 0; |
| 657 | |
Hugh Dickins | b73d7fc | 2010-09-09 16:38:09 -0700 | [diff] [blame] | 658 | /* |
| 659 | * Once hibernation has begun to create its image of memory, |
| 660 | * there's a danger that one of the calls to try_to_free_swap() |
| 661 | * - most probably a call from __try_to_reclaim_swap() while |
| 662 | * hibernation is allocating its own swap pages for the image, |
| 663 | * but conceivably even a call from memory reclaim - will free |
| 664 | * the swap from a page which has already been recorded in the |
| 665 | * image as a clean swapcache page, and then reuse its swap for |
| 666 | * another page of the image. On waking from hibernation, the |
| 667 | * original page might be freed under memory pressure, then |
| 668 | * later read back in from swap, now with the wrong data. |
| 669 | * |
| 670 | * Hibernation clears bits from gfp_allowed_mask to prevent |
| 671 | * memory reclaim from writing to disk, so check that here. |
| 672 | */ |
| 673 | if (!(gfp_allowed_mask & __GFP_IO)) |
| 674 | return 0; |
| 675 | |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 676 | delete_from_swap_cache(page); |
| 677 | SetPageDirty(page); |
| 678 | return 1; |
Rik van Riel | 68a22394 | 2008-10-18 20:26:23 -0700 | [diff] [blame] | 679 | } |
| 680 | |
| 681 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | * Free the swap entry like above, but also try to |
| 683 | * free the page cache entry if it is the last user. |
| 684 | */ |
Hugh Dickins | 2509ef2 | 2009-01-06 14:40:10 -0800 | [diff] [blame] | 685 | int free_swap_and_cache(swp_entry_t entry) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | { |
Hugh Dickins | 2509ef2 | 2009-01-06 14:40:10 -0800 | [diff] [blame] | 687 | struct swap_info_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | struct page *page = NULL; |
| 689 | |
Andi Kleen | a7420aa | 2009-09-16 11:50:05 +0200 | [diff] [blame] | 690 | if (non_swap_entry(entry)) |
Hugh Dickins | 2509ef2 | 2009-01-06 14:40:10 -0800 | [diff] [blame] | 691 | return 1; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 692 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | p = swap_info_get(entry); |
| 694 | if (p) { |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 695 | if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) { |
Nick Piggin | 93fac70 | 2006-03-31 02:29:56 -0800 | [diff] [blame] | 696 | page = find_get_page(&swapper_space, entry.val); |
Nick Piggin | 8413ac9 | 2008-10-18 20:26:59 -0700 | [diff] [blame] | 697 | if (page && !trylock_page(page)) { |
Nick Piggin | 93fac70 | 2006-03-31 02:29:56 -0800 | [diff] [blame] | 698 | page_cache_release(page); |
| 699 | page = NULL; |
| 700 | } |
| 701 | } |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 702 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | } |
| 704 | if (page) { |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 705 | /* |
| 706 | * Not mapped elsewhere, or swap space full? Free it! |
| 707 | * Also recheck PageSwapCache now page is locked (above). |
| 708 | */ |
Nick Piggin | 93fac70 | 2006-03-31 02:29:56 -0800 | [diff] [blame] | 709 | if (PageSwapCache(page) && !PageWriteback(page) && |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 710 | (!page_mapped(page) || vm_swap_full())) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 711 | delete_from_swap_cache(page); |
| 712 | SetPageDirty(page); |
| 713 | } |
| 714 | unlock_page(page); |
| 715 | page_cache_release(page); |
| 716 | } |
Hugh Dickins | 2509ef2 | 2009-01-06 14:40:10 -0800 | [diff] [blame] | 717 | return p != NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | } |
| 719 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 720 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
| 721 | /** |
| 722 | * mem_cgroup_count_swap_user - count the user of a swap entry |
| 723 | * @ent: the swap entry to be checked |
| 724 | * @pagep: the pointer for the swap cache page of the entry to be stored |
| 725 | * |
| 726 | * Returns the number of the user of the swap entry. The number is valid only |
| 727 | * for swaps of anonymous pages. |
| 728 | * If the entry is found on swap cache, the page is stored to pagep with |
| 729 | * refcount of it being incremented. |
| 730 | */ |
| 731 | int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep) |
| 732 | { |
| 733 | struct page *page; |
| 734 | struct swap_info_struct *p; |
| 735 | int count = 0; |
| 736 | |
| 737 | page = find_get_page(&swapper_space, ent.val); |
| 738 | if (page) |
| 739 | count += page_mapcount(page); |
| 740 | p = swap_info_get(ent); |
| 741 | if (p) { |
| 742 | count += swap_count(p->swap_map[swp_offset(ent)]); |
| 743 | spin_unlock(&swap_lock); |
| 744 | } |
| 745 | |
| 746 | *pagep = page; |
| 747 | return count; |
| 748 | } |
| 749 | #endif |
| 750 | |
Rafael J. Wysocki | b0cb1a1 | 2007-07-29 23:24:36 +0200 | [diff] [blame] | 751 | #ifdef CONFIG_HIBERNATION |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 752 | /* |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 753 | * Find the swap type that corresponds to given device (if any). |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 754 | * |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 755 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
| 756 | * from 0, in which the swap header is expected to be located. |
| 757 | * |
| 758 | * This is needed for the suspend to disk (aka swsusp). |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 759 | */ |
Rafael J. Wysocki | 7bf2368 | 2007-01-05 16:36:28 -0800 | [diff] [blame] | 760 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 761 | { |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 762 | struct block_device *bdev = NULL; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 763 | int type; |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 764 | |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 765 | if (device) |
| 766 | bdev = bdget(device); |
| 767 | |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 768 | spin_lock(&swap_lock); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 769 | for (type = 0; type < nr_swapfiles; type++) { |
| 770 | struct swap_info_struct *sis = swap_info[type]; |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 771 | |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 772 | if (!(sis->flags & SWP_WRITEOK)) |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 773 | continue; |
Rafael J. Wysocki | b6b5bce | 2006-08-27 01:23:25 -0700 | [diff] [blame] | 774 | |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 775 | if (!bdev) { |
Rafael J. Wysocki | 7bf2368 | 2007-01-05 16:36:28 -0800 | [diff] [blame] | 776 | if (bdev_p) |
Alan Jenkins | dddac6a | 2009-07-29 21:07:55 +0200 | [diff] [blame] | 777 | *bdev_p = bdgrab(sis->bdev); |
Rafael J. Wysocki | 7bf2368 | 2007-01-05 16:36:28 -0800 | [diff] [blame] | 778 | |
Rafael J. Wysocki | 6e1819d | 2006-03-23 03:00:03 -0800 | [diff] [blame] | 779 | spin_unlock(&swap_lock); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 780 | return type; |
Rafael J. Wysocki | 6e1819d | 2006-03-23 03:00:03 -0800 | [diff] [blame] | 781 | } |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 782 | if (bdev == sis->bdev) { |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 783 | struct swap_extent *se = &sis->first_swap_extent; |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 784 | |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 785 | if (se->start_block == offset) { |
Rafael J. Wysocki | 7bf2368 | 2007-01-05 16:36:28 -0800 | [diff] [blame] | 786 | if (bdev_p) |
Alan Jenkins | dddac6a | 2009-07-29 21:07:55 +0200 | [diff] [blame] | 787 | *bdev_p = bdgrab(sis->bdev); |
Rafael J. Wysocki | 7bf2368 | 2007-01-05 16:36:28 -0800 | [diff] [blame] | 788 | |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 789 | spin_unlock(&swap_lock); |
| 790 | bdput(bdev); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 791 | return type; |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 792 | } |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 793 | } |
| 794 | } |
| 795 | spin_unlock(&swap_lock); |
Rafael J. Wysocki | 915bae9 | 2006-12-06 20:34:07 -0800 | [diff] [blame] | 796 | if (bdev) |
| 797 | bdput(bdev); |
| 798 | |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 799 | return -ENODEV; |
| 800 | } |
| 801 | |
| 802 | /* |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 803 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev |
| 804 | * corresponding to given index in swap_info (swap type). |
| 805 | */ |
| 806 | sector_t swapdev_block(int type, pgoff_t offset) |
| 807 | { |
| 808 | struct block_device *bdev; |
| 809 | |
| 810 | if ((unsigned int)type >= nr_swapfiles) |
| 811 | return 0; |
| 812 | if (!(swap_info[type]->flags & SWP_WRITEOK)) |
| 813 | return 0; |
Lee Schermerhorn | d4906e1 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 814 | return map_swap_entry(swp_entry(type, offset), &bdev); |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 815 | } |
| 816 | |
| 817 | /* |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 818 | * Return either the total number of swap pages of given type, or the number |
| 819 | * of free pages of that type (depending on @free) |
| 820 | * |
| 821 | * This is needed for software suspend |
| 822 | */ |
| 823 | unsigned int count_swap_pages(int type, int free) |
| 824 | { |
| 825 | unsigned int n = 0; |
| 826 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 827 | spin_lock(&swap_lock); |
| 828 | if ((unsigned int)type < nr_swapfiles) { |
| 829 | struct swap_info_struct *sis = swap_info[type]; |
| 830 | |
| 831 | if (sis->flags & SWP_WRITEOK) { |
| 832 | n = sis->pages; |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 833 | if (free) |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 834 | n -= sis->inuse_pages; |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 835 | } |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 836 | } |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 837 | spin_unlock(&swap_lock); |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 838 | return n; |
| 839 | } |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 840 | #endif /* CONFIG_HIBERNATION */ |
Rafael J. Wysocki | f577eb3 | 2006-03-23 02:59:59 -0800 | [diff] [blame] | 841 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 842 | /* |
Hugh Dickins | 72866f6 | 2005-10-29 18:15:55 -0700 | [diff] [blame] | 843 | * No need to decide whether this PTE shares the swap entry with others, |
| 844 | * just let do_wp_page work it out if a write is requested later - to |
| 845 | * force COW, vm_page_prot omits write permission from any private vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | */ |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 847 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | unsigned long addr, swp_entry_t entry, struct page *page) |
| 849 | { |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 850 | struct mem_cgroup *ptr; |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 851 | spinlock_t *ptl; |
| 852 | pte_t *pte; |
| 853 | int ret = 1; |
| 854 | |
KAMEZAWA Hiroyuki | 85d9fc8 | 2009-01-29 14:25:13 -0800 | [diff] [blame] | 855 | if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) { |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 856 | ret = -ENOMEM; |
KAMEZAWA Hiroyuki | 85d9fc8 | 2009-01-29 14:25:13 -0800 | [diff] [blame] | 857 | goto out_nolock; |
| 858 | } |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 859 | |
| 860 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 861 | if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) { |
| 862 | if (ret > 0) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 863 | mem_cgroup_cancel_charge_swapin(ptr); |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 864 | ret = 0; |
| 865 | goto out; |
| 866 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 867 | |
KAMEZAWA Hiroyuki | b084d43 | 2010-03-05 13:41:42 -0800 | [diff] [blame] | 868 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 869 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 870 | get_page(page); |
| 871 | set_pte_at(vma->vm_mm, addr, pte, |
| 872 | pte_mkold(mk_pte(page, vma->vm_page_prot))); |
| 873 | page_add_anon_rmap(page, vma, addr); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 874 | mem_cgroup_commit_charge_swapin(page, ptr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | swap_free(entry); |
| 876 | /* |
| 877 | * Move the page to the active list so it is not |
| 878 | * immediately swapped out again after swapon. |
| 879 | */ |
| 880 | activate_page(page); |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 881 | out: |
| 882 | pte_unmap_unlock(pte, ptl); |
KAMEZAWA Hiroyuki | 85d9fc8 | 2009-01-29 14:25:13 -0800 | [diff] [blame] | 883 | out_nolock: |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 884 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | } |
| 886 | |
| 887 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
| 888 | unsigned long addr, unsigned long end, |
| 889 | swp_entry_t entry, struct page *page) |
| 890 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 891 | pte_t swp_pte = swp_entry_to_pte(entry); |
Hugh Dickins | 705e87c | 2005-10-29 18:16:27 -0700 | [diff] [blame] | 892 | pte_t *pte; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 893 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 895 | /* |
| 896 | * We don't actually need pte lock while scanning for swp_pte: since |
| 897 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the |
| 898 | * page table while we're scanning; though it could get zapped, and on |
| 899 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse |
| 900 | * of unmatched parts which look like swp_pte, so unuse_pte must |
| 901 | * recheck under pte lock. Scanning without pte lock lets it be |
| 902 | * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
| 903 | */ |
| 904 | pte = pte_offset_map(pmd, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | do { |
| 906 | /* |
| 907 | * swapoff spends a _lot_ of time in this loop! |
| 908 | * Test inline before going to call unuse_pte. |
| 909 | */ |
| 910 | if (unlikely(pte_same(*pte, swp_pte))) { |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 911 | pte_unmap(pte); |
| 912 | ret = unuse_pte(vma, pmd, addr, entry, page); |
| 913 | if (ret) |
| 914 | goto out; |
| 915 | pte = pte_offset_map(pmd, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | } |
| 917 | } while (pte++, addr += PAGE_SIZE, addr != end); |
Hugh Dickins | 044d66c | 2008-02-07 00:14:04 -0800 | [diff] [blame] | 918 | pte_unmap(pte - 1); |
| 919 | out: |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 920 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | } |
| 922 | |
| 923 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
| 924 | unsigned long addr, unsigned long end, |
| 925 | swp_entry_t entry, struct page *page) |
| 926 | { |
| 927 | pmd_t *pmd; |
| 928 | unsigned long next; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 929 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | |
| 931 | pmd = pmd_offset(pud, addr); |
| 932 | do { |
| 933 | next = pmd_addr_end(addr, end); |
Andrea Arcangeli | 3f04f62 | 2011-01-13 15:46:47 -0800 | [diff] [blame] | 934 | if (unlikely(pmd_trans_huge(*pmd))) |
| 935 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 936 | if (pmd_none_or_clear_bad(pmd)) |
| 937 | continue; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 938 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
| 939 | if (ret) |
| 940 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 941 | } while (pmd++, addr = next, addr != end); |
| 942 | return 0; |
| 943 | } |
| 944 | |
| 945 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
| 946 | unsigned long addr, unsigned long end, |
| 947 | swp_entry_t entry, struct page *page) |
| 948 | { |
| 949 | pud_t *pud; |
| 950 | unsigned long next; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 951 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 952 | |
| 953 | pud = pud_offset(pgd, addr); |
| 954 | do { |
| 955 | next = pud_addr_end(addr, end); |
| 956 | if (pud_none_or_clear_bad(pud)) |
| 957 | continue; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 958 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
| 959 | if (ret) |
| 960 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | } while (pud++, addr = next, addr != end); |
| 962 | return 0; |
| 963 | } |
| 964 | |
| 965 | static int unuse_vma(struct vm_area_struct *vma, |
| 966 | swp_entry_t entry, struct page *page) |
| 967 | { |
| 968 | pgd_t *pgd; |
| 969 | unsigned long addr, end, next; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 970 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | |
Hugh Dickins | 3ca7b3c | 2009-12-14 17:58:57 -0800 | [diff] [blame] | 972 | if (page_anon_vma(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | addr = page_address_in_vma(page, vma); |
| 974 | if (addr == -EFAULT) |
| 975 | return 0; |
| 976 | else |
| 977 | end = addr + PAGE_SIZE; |
| 978 | } else { |
| 979 | addr = vma->vm_start; |
| 980 | end = vma->vm_end; |
| 981 | } |
| 982 | |
| 983 | pgd = pgd_offset(vma->vm_mm, addr); |
| 984 | do { |
| 985 | next = pgd_addr_end(addr, end); |
| 986 | if (pgd_none_or_clear_bad(pgd)) |
| 987 | continue; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 988 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
| 989 | if (ret) |
| 990 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 991 | } while (pgd++, addr = next, addr != end); |
| 992 | return 0; |
| 993 | } |
| 994 | |
| 995 | static int unuse_mm(struct mm_struct *mm, |
| 996 | swp_entry_t entry, struct page *page) |
| 997 | { |
| 998 | struct vm_area_struct *vma; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 999 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | |
| 1001 | if (!down_read_trylock(&mm->mmap_sem)) { |
| 1002 | /* |
Fernando Luis Vazquez Cao | 7d03431 | 2008-07-29 22:33:41 -0700 | [diff] [blame] | 1003 | * Activate page so shrink_inactive_list is unlikely to unmap |
| 1004 | * its ptes while lock is dropped, so swapoff can make progress. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | */ |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 1006 | activate_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | unlock_page(page); |
| 1008 | down_read(&mm->mmap_sem); |
| 1009 | lock_page(page); |
| 1010 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1011 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1012 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1013 | break; |
| 1014 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | up_read(&mm->mmap_sem); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1016 | return (ret < 0)? ret: 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | } |
| 1018 | |
| 1019 | /* |
| 1020 | * Scan swap_map from current position to next entry still in use. |
| 1021 | * Recycle to start on reaching the end, returning 0 when empty. |
| 1022 | */ |
Hugh Dickins | 6eb396d | 2005-09-03 15:54:35 -0700 | [diff] [blame] | 1023 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
| 1024 | unsigned int prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1025 | { |
Hugh Dickins | 6eb396d | 2005-09-03 15:54:35 -0700 | [diff] [blame] | 1026 | unsigned int max = si->max; |
| 1027 | unsigned int i = prev; |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 1028 | unsigned char count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | |
| 1030 | /* |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1031 | * No need for swap_lock here: we're just looking |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | * for whether an entry is in use, not modifying it; false |
| 1033 | * hits are okay, and sys_swapoff() has already prevented new |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1034 | * allocations from this area (while holding swap_lock). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1035 | */ |
| 1036 | for (;;) { |
| 1037 | if (++i >= max) { |
| 1038 | if (!prev) { |
| 1039 | i = 0; |
| 1040 | break; |
| 1041 | } |
| 1042 | /* |
| 1043 | * No entries in use at top of swap_map, |
| 1044 | * loop back to start and recheck there. |
| 1045 | */ |
| 1046 | max = prev + 1; |
| 1047 | prev = 0; |
| 1048 | i = 1; |
| 1049 | } |
| 1050 | count = si->swap_map[i]; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 1051 | if (count && swap_count(count) != SWAP_MAP_BAD) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1052 | break; |
| 1053 | } |
| 1054 | return i; |
| 1055 | } |
| 1056 | |
| 1057 | /* |
| 1058 | * We completely avoid races by reading each swap page in advance, |
| 1059 | * and then search for the process using it. All the necessary |
| 1060 | * page table adjustments can then be made atomically. |
| 1061 | */ |
| 1062 | static int try_to_unuse(unsigned int type) |
| 1063 | { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1064 | struct swap_info_struct *si = swap_info[type]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | struct mm_struct *start_mm; |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 1066 | unsigned char *swap_map; |
| 1067 | unsigned char swcount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | struct page *page; |
| 1069 | swp_entry_t entry; |
Hugh Dickins | 6eb396d | 2005-09-03 15:54:35 -0700 | [diff] [blame] | 1070 | unsigned int i = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1071 | int retval = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | |
| 1073 | /* |
| 1074 | * When searching mms for an entry, a good strategy is to |
| 1075 | * start at the first mm we freed the previous entry from |
| 1076 | * (though actually we don't notice whether we or coincidence |
| 1077 | * freed the entry). Initialize this start_mm with a hold. |
| 1078 | * |
| 1079 | * A simpler strategy would be to start at the last mm we |
| 1080 | * freed the previous entry from; but that would take less |
| 1081 | * advantage of mmlist ordering, which clusters forked mms |
| 1082 | * together, child after parent. If we race with dup_mmap(), we |
| 1083 | * prefer to resolve parent before child, lest we miss entries |
| 1084 | * duplicated after we scanned child: using last mm would invert |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 1085 | * that. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | */ |
| 1087 | start_mm = &init_mm; |
| 1088 | atomic_inc(&init_mm.mm_users); |
| 1089 | |
| 1090 | /* |
| 1091 | * Keep on scanning until all entries have gone. Usually, |
| 1092 | * one pass through swap_map is enough, but not necessarily: |
| 1093 | * there are races when an instance of an entry might be missed. |
| 1094 | */ |
| 1095 | while ((i = find_next_to_unuse(si, i)) != 0) { |
| 1096 | if (signal_pending(current)) { |
| 1097 | retval = -EINTR; |
| 1098 | break; |
| 1099 | } |
| 1100 | |
Hugh Dickins | 886bb7e | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 1101 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1102 | * Get a page for the entry, using the existing swap |
| 1103 | * cache page if there is one. Otherwise, get a clean |
Hugh Dickins | 886bb7e | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 1104 | * page and read the swap into it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | */ |
| 1106 | swap_map = &si->swap_map[i]; |
| 1107 | entry = swp_entry(type, i); |
Hugh Dickins | 02098fe | 2008-02-04 22:28:42 -0800 | [diff] [blame] | 1108 | page = read_swap_cache_async(entry, |
| 1109 | GFP_HIGHUSER_MOVABLE, NULL, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | if (!page) { |
| 1111 | /* |
| 1112 | * Either swap_duplicate() failed because entry |
| 1113 | * has been freed independently, and will not be |
| 1114 | * reused since sys_swapoff() already disabled |
| 1115 | * allocation from here, or alloc_page() failed. |
| 1116 | */ |
| 1117 | if (!*swap_map) |
| 1118 | continue; |
| 1119 | retval = -ENOMEM; |
| 1120 | break; |
| 1121 | } |
| 1122 | |
| 1123 | /* |
| 1124 | * Don't hold on to start_mm if it looks like exiting. |
| 1125 | */ |
| 1126 | if (atomic_read(&start_mm->mm_users) == 1) { |
| 1127 | mmput(start_mm); |
| 1128 | start_mm = &init_mm; |
| 1129 | atomic_inc(&init_mm.mm_users); |
| 1130 | } |
| 1131 | |
| 1132 | /* |
| 1133 | * Wait for and lock page. When do_swap_page races with |
| 1134 | * try_to_unuse, do_swap_page can handle the fault much |
| 1135 | * faster than try_to_unuse can locate the entry. This |
| 1136 | * apparently redundant "wait_on_page_locked" lets try_to_unuse |
| 1137 | * defer to do_swap_page in such a case - in some tests, |
| 1138 | * do_swap_page and try_to_unuse repeatedly compete. |
| 1139 | */ |
| 1140 | wait_on_page_locked(page); |
| 1141 | wait_on_page_writeback(page); |
| 1142 | lock_page(page); |
| 1143 | wait_on_page_writeback(page); |
| 1144 | |
| 1145 | /* |
| 1146 | * Remove all references to entry. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1147 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | swcount = *swap_map; |
Hugh Dickins | aaa4686 | 2009-12-14 17:58:47 -0800 | [diff] [blame] | 1149 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
| 1150 | retval = shmem_unuse(entry, page); |
| 1151 | /* page has already been unlocked and released */ |
| 1152 | if (retval < 0) |
| 1153 | break; |
| 1154 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1155 | } |
Hugh Dickins | aaa4686 | 2009-12-14 17:58:47 -0800 | [diff] [blame] | 1156 | if (swap_count(swcount) && start_mm != &init_mm) |
| 1157 | retval = unuse_mm(start_mm, entry, page); |
| 1158 | |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 1159 | if (swap_count(*swap_map)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1160 | int set_start_mm = (*swap_map >= swcount); |
| 1161 | struct list_head *p = &start_mm->mmlist; |
| 1162 | struct mm_struct *new_start_mm = start_mm; |
| 1163 | struct mm_struct *prev_mm = start_mm; |
| 1164 | struct mm_struct *mm; |
| 1165 | |
| 1166 | atomic_inc(&new_start_mm->mm_users); |
| 1167 | atomic_inc(&prev_mm->mm_users); |
| 1168 | spin_lock(&mmlist_lock); |
Hugh Dickins | aaa4686 | 2009-12-14 17:58:47 -0800 | [diff] [blame] | 1169 | while (swap_count(*swap_map) && !retval && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | (p = p->next) != &start_mm->mmlist) { |
| 1171 | mm = list_entry(p, struct mm_struct, mmlist); |
Hugh Dickins | 70af7c5 | 2006-06-23 02:03:44 -0700 | [diff] [blame] | 1172 | if (!atomic_inc_not_zero(&mm->mm_users)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1173 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1174 | spin_unlock(&mmlist_lock); |
| 1175 | mmput(prev_mm); |
| 1176 | prev_mm = mm; |
| 1177 | |
| 1178 | cond_resched(); |
| 1179 | |
| 1180 | swcount = *swap_map; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 1181 | if (!swap_count(swcount)) /* any usage ? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1182 | ; |
Hugh Dickins | aaa4686 | 2009-12-14 17:58:47 -0800 | [diff] [blame] | 1183 | else if (mm == &init_mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1184 | set_start_mm = 1; |
Hugh Dickins | aaa4686 | 2009-12-14 17:58:47 -0800 | [diff] [blame] | 1185 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1186 | retval = unuse_mm(mm, entry, page); |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 1187 | |
Bo Liu | 32c5fc1 | 2009-11-02 16:50:33 +0000 | [diff] [blame] | 1188 | if (set_start_mm && *swap_map < swcount) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1189 | mmput(new_start_mm); |
| 1190 | atomic_inc(&mm->mm_users); |
| 1191 | new_start_mm = mm; |
| 1192 | set_start_mm = 0; |
| 1193 | } |
| 1194 | spin_lock(&mmlist_lock); |
| 1195 | } |
| 1196 | spin_unlock(&mmlist_lock); |
| 1197 | mmput(prev_mm); |
| 1198 | mmput(start_mm); |
| 1199 | start_mm = new_start_mm; |
| 1200 | } |
| 1201 | if (retval) { |
| 1202 | unlock_page(page); |
| 1203 | page_cache_release(page); |
| 1204 | break; |
| 1205 | } |
| 1206 | |
| 1207 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | * If a reference remains (rare), we would like to leave |
| 1209 | * the page in the swap cache; but try_to_unmap could |
| 1210 | * then re-duplicate the entry once we drop page lock, |
| 1211 | * so we might loop indefinitely; also, that page could |
| 1212 | * not be swapped out to other storage meanwhile. So: |
| 1213 | * delete from cache even if there's another reference, |
| 1214 | * after ensuring that the data has been saved to disk - |
| 1215 | * since if the reference remains (rarer), it will be |
| 1216 | * read from disk into another page. Splitting into two |
| 1217 | * pages would be incorrect if swap supported "shared |
| 1218 | * private" pages, but they are handled by tmpfs files. |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1219 | * |
| 1220 | * Given how unuse_vma() targets one particular offset |
| 1221 | * in an anon_vma, once the anon_vma has been determined, |
| 1222 | * this splitting happens to be just what is needed to |
| 1223 | * handle where KSM pages have been swapped out: re-reading |
| 1224 | * is unnecessarily slow, but we can fix that later on. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1225 | */ |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 1226 | if (swap_count(*swap_map) && |
| 1227 | PageDirty(page) && PageSwapCache(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1228 | struct writeback_control wbc = { |
| 1229 | .sync_mode = WB_SYNC_NONE, |
| 1230 | }; |
| 1231 | |
| 1232 | swap_writepage(page, &wbc); |
| 1233 | lock_page(page); |
| 1234 | wait_on_page_writeback(page); |
| 1235 | } |
Hugh Dickins | 68bdc8d | 2009-01-06 14:39:37 -0800 | [diff] [blame] | 1236 | |
| 1237 | /* |
| 1238 | * It is conceivable that a racing task removed this page from |
| 1239 | * swap cache just before we acquired the page lock at the top, |
| 1240 | * or while we dropped it in unuse_mm(). The page might even |
| 1241 | * be back in swap cache on another swap area: that we must not |
| 1242 | * delete, since it may not have been written out to swap yet. |
| 1243 | */ |
| 1244 | if (PageSwapCache(page) && |
| 1245 | likely(page_private(page) == entry.val)) |
Hugh Dickins | 2e0e26c | 2008-02-04 22:28:53 -0800 | [diff] [blame] | 1246 | delete_from_swap_cache(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1247 | |
| 1248 | /* |
| 1249 | * So we could skip searching mms once swap count went |
| 1250 | * to 1, we did not mark any present ptes as dirty: must |
Anderson Briglia | 2706a1b | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 1251 | * mark page dirty so shrink_page_list will preserve it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1252 | */ |
| 1253 | SetPageDirty(page); |
| 1254 | unlock_page(page); |
| 1255 | page_cache_release(page); |
| 1256 | |
| 1257 | /* |
| 1258 | * Make sure that we aren't completely killing |
| 1259 | * interactive performance. |
| 1260 | */ |
| 1261 | cond_resched(); |
| 1262 | } |
| 1263 | |
| 1264 | mmput(start_mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | return retval; |
| 1266 | } |
| 1267 | |
| 1268 | /* |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1269 | * After a successful try_to_unuse, if no swap is now in use, we know |
| 1270 | * we can empty the mmlist. swap_lock must be held on entry and exit. |
| 1271 | * Note that mmlist_lock nests inside swap_lock, and an mm must be |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1272 | * added to the mmlist just after page_duplicate - before would be racy. |
| 1273 | */ |
| 1274 | static void drain_mmlist(void) |
| 1275 | { |
| 1276 | struct list_head *p, *next; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1277 | unsigned int type; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1278 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1279 | for (type = 0; type < nr_swapfiles; type++) |
| 1280 | if (swap_info[type]->inuse_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1281 | return; |
| 1282 | spin_lock(&mmlist_lock); |
| 1283 | list_for_each_safe(p, next, &init_mm.mmlist) |
| 1284 | list_del_init(p); |
| 1285 | spin_unlock(&mmlist_lock); |
| 1286 | } |
| 1287 | |
| 1288 | /* |
| 1289 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which |
Lee Schermerhorn | d4906e1 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 1290 | * corresponds to page offset for the specified swap entry. |
| 1291 | * Note that the type of this function is sector_t, but it returns page offset |
| 1292 | * into the bdev, not sector offset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | */ |
Lee Schermerhorn | d4906e1 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 1294 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1295 | { |
Hugh Dickins | f29ad6a | 2009-12-14 17:58:40 -0800 | [diff] [blame] | 1296 | struct swap_info_struct *sis; |
| 1297 | struct swap_extent *start_se; |
| 1298 | struct swap_extent *se; |
| 1299 | pgoff_t offset; |
| 1300 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1301 | sis = swap_info[swp_type(entry)]; |
Hugh Dickins | f29ad6a | 2009-12-14 17:58:40 -0800 | [diff] [blame] | 1302 | *bdev = sis->bdev; |
| 1303 | |
| 1304 | offset = swp_offset(entry); |
| 1305 | start_se = sis->curr_swap_extent; |
| 1306 | se = start_se; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1307 | |
| 1308 | for ( ; ; ) { |
| 1309 | struct list_head *lh; |
| 1310 | |
| 1311 | if (se->start_page <= offset && |
| 1312 | offset < (se->start_page + se->nr_pages)) { |
| 1313 | return se->start_block + (offset - se->start_page); |
| 1314 | } |
Hugh Dickins | 11d3188 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1315 | lh = se->list.next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | se = list_entry(lh, struct swap_extent, list); |
| 1317 | sis->curr_swap_extent = se; |
| 1318 | BUG_ON(se == start_se); /* It *must* be present */ |
| 1319 | } |
| 1320 | } |
| 1321 | |
| 1322 | /* |
Lee Schermerhorn | d4906e1 | 2009-12-14 17:58:49 -0800 | [diff] [blame] | 1323 | * Returns the page offset into bdev for the specified page's swap entry. |
| 1324 | */ |
| 1325 | sector_t map_swap_page(struct page *page, struct block_device **bdev) |
| 1326 | { |
| 1327 | swp_entry_t entry; |
| 1328 | entry.val = page_private(page); |
| 1329 | return map_swap_entry(entry, bdev); |
| 1330 | } |
| 1331 | |
| 1332 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | * Free all of a swapdev's extent information |
| 1334 | */ |
| 1335 | static void destroy_swap_extents(struct swap_info_struct *sis) |
| 1336 | { |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1337 | while (!list_empty(&sis->first_swap_extent.list)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | struct swap_extent *se; |
| 1339 | |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1340 | se = list_entry(sis->first_swap_extent.list.next, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | struct swap_extent, list); |
| 1342 | list_del(&se->list); |
| 1343 | kfree(se); |
| 1344 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1345 | } |
| 1346 | |
| 1347 | /* |
| 1348 | * Add a block range (and the corresponding page range) into this swapdev's |
Hugh Dickins | 11d3188 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1349 | * extent list. The extent list is kept sorted in page order. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | * |
Hugh Dickins | 11d3188 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1351 | * This function rather assumes that it is called in ascending page order. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | */ |
| 1353 | static int |
| 1354 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
| 1355 | unsigned long nr_pages, sector_t start_block) |
| 1356 | { |
| 1357 | struct swap_extent *se; |
| 1358 | struct swap_extent *new_se; |
| 1359 | struct list_head *lh; |
| 1360 | |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1361 | if (start_page == 0) { |
| 1362 | se = &sis->first_swap_extent; |
| 1363 | sis->curr_swap_extent = se; |
| 1364 | se->start_page = 0; |
| 1365 | se->nr_pages = nr_pages; |
| 1366 | se->start_block = start_block; |
| 1367 | return 1; |
| 1368 | } else { |
| 1369 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | se = list_entry(lh, struct swap_extent, list); |
Hugh Dickins | 11d3188 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1371 | BUG_ON(se->start_page + se->nr_pages != start_page); |
| 1372 | if (se->start_block + se->nr_pages == start_block) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1373 | /* Merge it */ |
| 1374 | se->nr_pages += nr_pages; |
| 1375 | return 0; |
| 1376 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1377 | } |
| 1378 | |
| 1379 | /* |
| 1380 | * No merge. Insert a new extent, preserving ordering. |
| 1381 | */ |
| 1382 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
| 1383 | if (new_se == NULL) |
| 1384 | return -ENOMEM; |
| 1385 | new_se->start_page = start_page; |
| 1386 | new_se->nr_pages = nr_pages; |
| 1387 | new_se->start_block = start_block; |
| 1388 | |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1389 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1390 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | } |
| 1392 | |
| 1393 | /* |
| 1394 | * A `swap extent' is a simple thing which maps a contiguous range of pages |
| 1395 | * onto a contiguous range of disk blocks. An ordered list of swap extents |
| 1396 | * is built at swapon time and is then used at swap_writepage/swap_readpage |
| 1397 | * time for locating where on disk a page belongs. |
| 1398 | * |
| 1399 | * If the swapfile is an S_ISBLK block device, a single extent is installed. |
| 1400 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG |
| 1401 | * swap files identically. |
| 1402 | * |
| 1403 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap |
| 1404 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
| 1405 | * swapfiles are handled *identically* after swapon time. |
| 1406 | * |
| 1407 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks |
| 1408 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If |
| 1409 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment |
| 1410 | * requirements, they are simply tossed out - we will never use those blocks |
| 1411 | * for swapping. |
| 1412 | * |
Hugh Dickins | b0d9bcd | 2005-09-03 15:54:31 -0700 | [diff] [blame] | 1413 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1414 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
| 1415 | * which will scribble on the fs. |
| 1416 | * |
| 1417 | * The amount of disk space which a single swap extent represents varies. |
| 1418 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of |
| 1419 | * extents in the list. To avoid much list walking, we cache the previous |
| 1420 | * search location in `curr_swap_extent', and start new searches from there. |
| 1421 | * This is extremely effective. The average number of iterations in |
| 1422 | * map_swap_page() has been measured at about 0.3 per page. - akpm. |
| 1423 | */ |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1424 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1425 | { |
| 1426 | struct inode *inode; |
| 1427 | unsigned blocks_per_page; |
| 1428 | unsigned long page_no; |
| 1429 | unsigned blkbits; |
| 1430 | sector_t probe_block; |
| 1431 | sector_t last_block; |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1432 | sector_t lowest_block = -1; |
| 1433 | sector_t highest_block = 0; |
| 1434 | int nr_extents = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1435 | int ret; |
| 1436 | |
| 1437 | inode = sis->swap_file->f_mapping->host; |
| 1438 | if (S_ISBLK(inode->i_mode)) { |
| 1439 | ret = add_swap_extent(sis, 0, sis->max, 0); |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1440 | *span = sis->pages; |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1441 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1442 | } |
| 1443 | |
| 1444 | blkbits = inode->i_blkbits; |
| 1445 | blocks_per_page = PAGE_SIZE >> blkbits; |
| 1446 | |
| 1447 | /* |
| 1448 | * Map all the blocks into the extent list. This code doesn't try |
| 1449 | * to be very smart. |
| 1450 | */ |
| 1451 | probe_block = 0; |
| 1452 | page_no = 0; |
| 1453 | last_block = i_size_read(inode) >> blkbits; |
| 1454 | while ((probe_block + blocks_per_page) <= last_block && |
| 1455 | page_no < sis->max) { |
| 1456 | unsigned block_in_page; |
| 1457 | sector_t first_block; |
| 1458 | |
| 1459 | first_block = bmap(inode, probe_block); |
| 1460 | if (first_block == 0) |
| 1461 | goto bad_bmap; |
| 1462 | |
| 1463 | /* |
| 1464 | * It must be PAGE_SIZE aligned on-disk |
| 1465 | */ |
| 1466 | if (first_block & (blocks_per_page - 1)) { |
| 1467 | probe_block++; |
| 1468 | goto reprobe; |
| 1469 | } |
| 1470 | |
| 1471 | for (block_in_page = 1; block_in_page < blocks_per_page; |
| 1472 | block_in_page++) { |
| 1473 | sector_t block; |
| 1474 | |
| 1475 | block = bmap(inode, probe_block + block_in_page); |
| 1476 | if (block == 0) |
| 1477 | goto bad_bmap; |
| 1478 | if (block != first_block + block_in_page) { |
| 1479 | /* Discontiguity */ |
| 1480 | probe_block++; |
| 1481 | goto reprobe; |
| 1482 | } |
| 1483 | } |
| 1484 | |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1485 | first_block >>= (PAGE_SHIFT - blkbits); |
| 1486 | if (page_no) { /* exclude the header page */ |
| 1487 | if (first_block < lowest_block) |
| 1488 | lowest_block = first_block; |
| 1489 | if (first_block > highest_block) |
| 1490 | highest_block = first_block; |
| 1491 | } |
| 1492 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1493 | /* |
| 1494 | * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks |
| 1495 | */ |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1496 | ret = add_swap_extent(sis, page_no, 1, first_block); |
| 1497 | if (ret < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | goto out; |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1499 | nr_extents += ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | page_no++; |
| 1501 | probe_block += blocks_per_page; |
| 1502 | reprobe: |
| 1503 | continue; |
| 1504 | } |
Hugh Dickins | 53092a7 | 2005-09-03 15:54:34 -0700 | [diff] [blame] | 1505 | ret = nr_extents; |
| 1506 | *span = 1 + highest_block - lowest_block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1507 | if (page_no == 0) |
Hugh Dickins | e2244ec | 2005-09-03 15:54:32 -0700 | [diff] [blame] | 1508 | page_no = 1; /* force Empty message */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1509 | sis->max = page_no; |
Hugh Dickins | e2244ec | 2005-09-03 15:54:32 -0700 | [diff] [blame] | 1510 | sis->pages = page_no - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | sis->highest_bit = page_no - 1; |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1512 | out: |
| 1513 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | bad_bmap: |
| 1515 | printk(KERN_ERR "swapon: swapfile has holes\n"); |
| 1516 | ret = -EINVAL; |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1517 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1518 | } |
| 1519 | |
Cesar Eduardo Barros | 4053154 | 2011-03-22 16:33:37 -0700 | [diff] [blame] | 1520 | static void enable_swap_info(struct swap_info_struct *p, int prio, |
| 1521 | unsigned char *swap_map) |
| 1522 | { |
| 1523 | int i, prev; |
| 1524 | |
| 1525 | spin_lock(&swap_lock); |
| 1526 | if (prio >= 0) |
| 1527 | p->prio = prio; |
| 1528 | else |
| 1529 | p->prio = --least_priority; |
| 1530 | p->swap_map = swap_map; |
| 1531 | p->flags |= SWP_WRITEOK; |
| 1532 | nr_swap_pages += p->pages; |
| 1533 | total_swap_pages += p->pages; |
| 1534 | |
| 1535 | /* insert swap space into swap_list: */ |
| 1536 | prev = -1; |
| 1537 | for (i = swap_list.head; i >= 0; i = swap_info[i]->next) { |
| 1538 | if (p->prio >= swap_info[i]->prio) |
| 1539 | break; |
| 1540 | prev = i; |
| 1541 | } |
| 1542 | p->next = i; |
| 1543 | if (prev < 0) |
| 1544 | swap_list.head = swap_list.next = p->type; |
| 1545 | else |
| 1546 | swap_info[prev]->next = p->type; |
| 1547 | spin_unlock(&swap_lock); |
| 1548 | } |
| 1549 | |
Heiko Carstens | c4ea37c | 2009-01-14 14:14:28 +0100 | [diff] [blame] | 1550 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1551 | { |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 1552 | struct swap_info_struct *p = NULL; |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 1553 | unsigned char *swap_map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1554 | struct file *swap_file, *victim; |
| 1555 | struct address_space *mapping; |
| 1556 | struct inode *inode; |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 1557 | char *pathname; |
David Rientjes | 72788c3 | 2011-05-24 17:11:40 -0700 | [diff] [blame] | 1558 | int oom_score_adj; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1559 | int i, type, prev; |
| 1560 | int err; |
Hugh Dickins | 886bb7e | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 1561 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1562 | if (!capable(CAP_SYS_ADMIN)) |
| 1563 | return -EPERM; |
| 1564 | |
| 1565 | pathname = getname(specialfile); |
| 1566 | err = PTR_ERR(pathname); |
| 1567 | if (IS_ERR(pathname)) |
| 1568 | goto out; |
| 1569 | |
| 1570 | victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0); |
| 1571 | putname(pathname); |
| 1572 | err = PTR_ERR(victim); |
| 1573 | if (IS_ERR(victim)) |
| 1574 | goto out; |
| 1575 | |
| 1576 | mapping = victim->f_mapping; |
| 1577 | prev = -1; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1578 | spin_lock(&swap_lock); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1579 | for (type = swap_list.head; type >= 0; type = swap_info[type]->next) { |
| 1580 | p = swap_info[type]; |
Hugh Dickins | 22c6f8f | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 1581 | if (p->flags & SWP_WRITEOK) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1582 | if (p->swap_file->f_mapping == mapping) |
| 1583 | break; |
| 1584 | } |
| 1585 | prev = type; |
| 1586 | } |
| 1587 | if (type < 0) { |
| 1588 | err = -EINVAL; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1589 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1590 | goto out_dput; |
| 1591 | } |
| 1592 | if (!security_vm_enough_memory(p->pages)) |
| 1593 | vm_unacct_memory(p->pages); |
| 1594 | else { |
| 1595 | err = -ENOMEM; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1596 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1597 | goto out_dput; |
| 1598 | } |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1599 | if (prev < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1600 | swap_list.head = p->next; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1601 | else |
| 1602 | swap_info[prev]->next = p->next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1603 | if (type == swap_list.next) { |
| 1604 | /* just pick something that's safe... */ |
| 1605 | swap_list.next = swap_list.head; |
| 1606 | } |
Hugh Dickins | 78ecba0 | 2008-07-23 21:28:23 -0700 | [diff] [blame] | 1607 | if (p->prio < 0) { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1608 | for (i = p->next; i >= 0; i = swap_info[i]->next) |
| 1609 | swap_info[i]->prio = p->prio--; |
Hugh Dickins | 78ecba0 | 2008-07-23 21:28:23 -0700 | [diff] [blame] | 1610 | least_priority++; |
| 1611 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1612 | nr_swap_pages -= p->pages; |
| 1613 | total_swap_pages -= p->pages; |
| 1614 | p->flags &= ~SWP_WRITEOK; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1615 | spin_unlock(&swap_lock); |
Hugh Dickins | fb4f88d | 2005-09-03 15:54:37 -0700 | [diff] [blame] | 1616 | |
David Rientjes | 72788c3 | 2011-05-24 17:11:40 -0700 | [diff] [blame] | 1617 | oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1618 | err = try_to_unuse(type); |
David Rientjes | 43362a4 | 2011-10-31 17:07:18 -0700 | [diff] [blame] | 1619 | compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX, oom_score_adj); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1620 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1621 | if (err) { |
Cesar Eduardo Barros | 4053154 | 2011-03-22 16:33:37 -0700 | [diff] [blame] | 1622 | /* |
| 1623 | * reading p->prio and p->swap_map outside the lock is |
| 1624 | * safe here because only sys_swapon and sys_swapoff |
| 1625 | * change them, and there can be no other sys_swapon or |
| 1626 | * sys_swapoff for this swap_info_struct at this point. |
| 1627 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | /* re-insert swap space back into swap_list */ |
Cesar Eduardo Barros | 4053154 | 2011-03-22 16:33:37 -0700 | [diff] [blame] | 1629 | enable_swap_info(p, p->prio, p->swap_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | goto out_dput; |
| 1631 | } |
Hugh Dickins | 52b7efdb | 2005-09-03 15:54:39 -0700 | [diff] [blame] | 1632 | |
Hugh Dickins | 4cd3bb1 | 2005-09-03 15:54:33 -0700 | [diff] [blame] | 1633 | destroy_swap_extents(p); |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 1634 | if (p->flags & SWP_CONTINUED) |
| 1635 | free_swap_count_continuations(p); |
| 1636 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1637 | mutex_lock(&swapon_mutex); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1638 | spin_lock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1639 | drain_mmlist(); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1640 | |
| 1641 | /* wait for anyone still in scan_swap_map */ |
| 1642 | p->highest_bit = 0; /* cuts scans short */ |
| 1643 | while (p->flags >= SWP_SCANNING) { |
| 1644 | spin_unlock(&swap_lock); |
Nishanth Aravamudan | 13e4b57 | 2005-09-10 00:27:25 -0700 | [diff] [blame] | 1645 | schedule_timeout_uninterruptible(1); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1646 | spin_lock(&swap_lock); |
| 1647 | } |
| 1648 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1649 | swap_file = p->swap_file; |
| 1650 | p->swap_file = NULL; |
| 1651 | p->max = 0; |
| 1652 | swap_map = p->swap_map; |
| 1653 | p->swap_map = NULL; |
| 1654 | p->flags = 0; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1655 | spin_unlock(&swap_lock); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1656 | mutex_unlock(&swapon_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | vfree(swap_map); |
KAMEZAWA Hiroyuki | 27a7faa | 2009-01-07 18:07:58 -0800 | [diff] [blame] | 1658 | /* Destroy swap account informatin */ |
| 1659 | swap_cgroup_swapoff(type); |
| 1660 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1661 | inode = mapping->host; |
| 1662 | if (S_ISBLK(inode->i_mode)) { |
| 1663 | struct block_device *bdev = I_BDEV(inode); |
| 1664 | set_blocksize(bdev, p->old_block_size); |
Tejun Heo | e525fd8 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1665 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1666 | } else { |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1667 | mutex_lock(&inode->i_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1668 | inode->i_flags &= ~S_SWAPFILE; |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1669 | mutex_unlock(&inode->i_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1670 | } |
| 1671 | filp_close(swap_file, NULL); |
| 1672 | err = 0; |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1673 | atomic_inc(&proc_poll_event); |
| 1674 | wake_up_interruptible(&proc_poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1675 | |
| 1676 | out_dput: |
| 1677 | filp_close(victim, NULL); |
| 1678 | out: |
| 1679 | return err; |
| 1680 | } |
| 1681 | |
| 1682 | #ifdef CONFIG_PROC_FS |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1683 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
| 1684 | { |
Kay Sievers | f151463 | 2011-07-12 20:48:39 +0200 | [diff] [blame] | 1685 | struct seq_file *seq = file->private_data; |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1686 | |
| 1687 | poll_wait(file, &proc_poll_wait, wait); |
| 1688 | |
Kay Sievers | f151463 | 2011-07-12 20:48:39 +0200 | [diff] [blame] | 1689 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
| 1690 | seq->poll_event = atomic_read(&proc_poll_event); |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1691 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
| 1692 | } |
| 1693 | |
| 1694 | return POLLIN | POLLRDNORM; |
| 1695 | } |
| 1696 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1697 | /* iterator */ |
| 1698 | static void *swap_start(struct seq_file *swap, loff_t *pos) |
| 1699 | { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1700 | struct swap_info_struct *si; |
| 1701 | int type; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1702 | loff_t l = *pos; |
| 1703 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1704 | mutex_lock(&swapon_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1705 | |
Suleiman Souhlal | 881e4aa | 2006-12-06 20:32:28 -0800 | [diff] [blame] | 1706 | if (!l) |
| 1707 | return SEQ_START_TOKEN; |
| 1708 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1709 | for (type = 0; type < nr_swapfiles; type++) { |
| 1710 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ |
| 1711 | si = swap_info[type]; |
| 1712 | if (!(si->flags & SWP_USED) || !si->swap_map) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1713 | continue; |
Suleiman Souhlal | 881e4aa | 2006-12-06 20:32:28 -0800 | [diff] [blame] | 1714 | if (!--l) |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1715 | return si; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1716 | } |
| 1717 | |
| 1718 | return NULL; |
| 1719 | } |
| 1720 | |
| 1721 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) |
| 1722 | { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1723 | struct swap_info_struct *si = v; |
| 1724 | int type; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1725 | |
Suleiman Souhlal | 881e4aa | 2006-12-06 20:32:28 -0800 | [diff] [blame] | 1726 | if (v == SEQ_START_TOKEN) |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1727 | type = 0; |
| 1728 | else |
| 1729 | type = si->type + 1; |
Suleiman Souhlal | 881e4aa | 2006-12-06 20:32:28 -0800 | [diff] [blame] | 1730 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1731 | for (; type < nr_swapfiles; type++) { |
| 1732 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ |
| 1733 | si = swap_info[type]; |
| 1734 | if (!(si->flags & SWP_USED) || !si->swap_map) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1735 | continue; |
| 1736 | ++*pos; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1737 | return si; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | } |
| 1739 | |
| 1740 | return NULL; |
| 1741 | } |
| 1742 | |
| 1743 | static void swap_stop(struct seq_file *swap, void *v) |
| 1744 | { |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1745 | mutex_unlock(&swapon_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1746 | } |
| 1747 | |
| 1748 | static int swap_show(struct seq_file *swap, void *v) |
| 1749 | { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1750 | struct swap_info_struct *si = v; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1751 | struct file *file; |
| 1752 | int len; |
| 1753 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1754 | if (si == SEQ_START_TOKEN) { |
Suleiman Souhlal | 881e4aa | 2006-12-06 20:32:28 -0800 | [diff] [blame] | 1755 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
| 1756 | return 0; |
| 1757 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1758 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1759 | file = si->swap_file; |
Jan Blunck | c32c2f6 | 2008-02-14 19:38:43 -0800 | [diff] [blame] | 1760 | len = seq_path(swap, &file->f_path, " \t\n\\"); |
Hugh Dickins | 6eb396d | 2005-09-03 15:54:35 -0700 | [diff] [blame] | 1761 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
Hugh Dickins | 886bb7e | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 1762 | len < 40 ? 40 - len : 1, " ", |
| 1763 | S_ISBLK(file->f_path.dentry->d_inode->i_mode) ? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1764 | "partition" : "file\t", |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1765 | si->pages << (PAGE_SHIFT - 10), |
| 1766 | si->inuse_pages << (PAGE_SHIFT - 10), |
| 1767 | si->prio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1768 | return 0; |
| 1769 | } |
| 1770 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 1771 | static const struct seq_operations swaps_op = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1772 | .start = swap_start, |
| 1773 | .next = swap_next, |
| 1774 | .stop = swap_stop, |
| 1775 | .show = swap_show |
| 1776 | }; |
| 1777 | |
| 1778 | static int swaps_open(struct inode *inode, struct file *file) |
| 1779 | { |
Kay Sievers | f151463 | 2011-07-12 20:48:39 +0200 | [diff] [blame] | 1780 | struct seq_file *seq; |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1781 | int ret; |
| 1782 | |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1783 | ret = seq_open(file, &swaps_op); |
Kay Sievers | f151463 | 2011-07-12 20:48:39 +0200 | [diff] [blame] | 1784 | if (ret) |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1785 | return ret; |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1786 | |
Kay Sievers | f151463 | 2011-07-12 20:48:39 +0200 | [diff] [blame] | 1787 | seq = file->private_data; |
| 1788 | seq->poll_event = atomic_read(&proc_poll_event); |
| 1789 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1790 | } |
| 1791 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 1792 | static const struct file_operations proc_swaps_operations = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1793 | .open = swaps_open, |
| 1794 | .read = seq_read, |
| 1795 | .llseek = seq_lseek, |
| 1796 | .release = seq_release, |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1797 | .poll = swaps_poll, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1798 | }; |
| 1799 | |
| 1800 | static int __init procswaps_init(void) |
| 1801 | { |
Denis V. Lunev | 3d71f86 | 2008-04-29 01:02:13 -0700 | [diff] [blame] | 1802 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1803 | return 0; |
| 1804 | } |
| 1805 | __initcall(procswaps_init); |
| 1806 | #endif /* CONFIG_PROC_FS */ |
| 1807 | |
Jan Beulich | 1796316 | 2008-12-16 11:35:24 +0000 | [diff] [blame] | 1808 | #ifdef MAX_SWAPFILES_CHECK |
| 1809 | static int __init max_swapfiles_check(void) |
| 1810 | { |
| 1811 | MAX_SWAPFILES_CHECK(); |
| 1812 | return 0; |
| 1813 | } |
| 1814 | late_initcall(max_swapfiles_check); |
| 1815 | #endif |
| 1816 | |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 1817 | static struct swap_info_struct *alloc_swap_info(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1818 | { |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 1819 | struct swap_info_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1820 | unsigned int type; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1821 | |
| 1822 | p = kzalloc(sizeof(*p), GFP_KERNEL); |
| 1823 | if (!p) |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 1824 | return ERR_PTR(-ENOMEM); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1825 | |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1826 | spin_lock(&swap_lock); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1827 | for (type = 0; type < nr_swapfiles; type++) { |
| 1828 | if (!(swap_info[type]->flags & SWP_USED)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1829 | break; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1830 | } |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 1831 | if (type >= MAX_SWAPFILES) { |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1832 | spin_unlock(&swap_lock); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1833 | kfree(p); |
Cesar Eduardo Barros | 730c058 | 2011-03-22 16:33:19 -0700 | [diff] [blame] | 1834 | return ERR_PTR(-EPERM); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1835 | } |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1836 | if (type >= nr_swapfiles) { |
| 1837 | p->type = type; |
| 1838 | swap_info[type] = p; |
| 1839 | /* |
| 1840 | * Write swap_info[type] before nr_swapfiles, in case a |
| 1841 | * racing procfs swap_start() or swap_next() is reading them. |
| 1842 | * (We never shrink nr_swapfiles, we never free this entry.) |
| 1843 | */ |
| 1844 | smp_wmb(); |
| 1845 | nr_swapfiles++; |
| 1846 | } else { |
| 1847 | kfree(p); |
| 1848 | p = swap_info[type]; |
| 1849 | /* |
| 1850 | * Do not memset this entry: a racing procfs swap_next() |
| 1851 | * would be relying on p->type to remain valid. |
| 1852 | */ |
| 1853 | } |
Hugh Dickins | 9625a5f | 2009-12-14 17:58:42 -0800 | [diff] [blame] | 1854 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1855 | p->flags = SWP_USED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1856 | p->next = -1; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 1857 | spin_unlock(&swap_lock); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 1858 | |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 1859 | return p; |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 1860 | } |
| 1861 | |
Cesar Eduardo Barros | 4d0e1e1 | 2011-03-22 16:33:26 -0700 | [diff] [blame] | 1862 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
| 1863 | { |
| 1864 | int error; |
| 1865 | |
| 1866 | if (S_ISBLK(inode->i_mode)) { |
| 1867 | p->bdev = bdgrab(I_BDEV(inode)); |
| 1868 | error = blkdev_get(p->bdev, |
| 1869 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, |
| 1870 | sys_swapon); |
| 1871 | if (error < 0) { |
| 1872 | p->bdev = NULL; |
Cesar Eduardo Barros | 87ade72 | 2011-03-22 16:33:27 -0700 | [diff] [blame] | 1873 | return -EINVAL; |
Cesar Eduardo Barros | 4d0e1e1 | 2011-03-22 16:33:26 -0700 | [diff] [blame] | 1874 | } |
| 1875 | p->old_block_size = block_size(p->bdev); |
| 1876 | error = set_blocksize(p->bdev, PAGE_SIZE); |
| 1877 | if (error < 0) |
Cesar Eduardo Barros | 87ade72 | 2011-03-22 16:33:27 -0700 | [diff] [blame] | 1878 | return error; |
Cesar Eduardo Barros | 4d0e1e1 | 2011-03-22 16:33:26 -0700 | [diff] [blame] | 1879 | p->flags |= SWP_BLKDEV; |
| 1880 | } else if (S_ISREG(inode->i_mode)) { |
| 1881 | p->bdev = inode->i_sb->s_bdev; |
| 1882 | mutex_lock(&inode->i_mutex); |
Cesar Eduardo Barros | 87ade72 | 2011-03-22 16:33:27 -0700 | [diff] [blame] | 1883 | if (IS_SWAPFILE(inode)) |
| 1884 | return -EBUSY; |
| 1885 | } else |
| 1886 | return -EINVAL; |
Cesar Eduardo Barros | 4d0e1e1 | 2011-03-22 16:33:26 -0700 | [diff] [blame] | 1887 | |
| 1888 | return 0; |
Cesar Eduardo Barros | 4d0e1e1 | 2011-03-22 16:33:26 -0700 | [diff] [blame] | 1889 | } |
| 1890 | |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1891 | static unsigned long read_swap_header(struct swap_info_struct *p, |
| 1892 | union swap_header *swap_header, |
| 1893 | struct inode *inode) |
| 1894 | { |
| 1895 | int i; |
| 1896 | unsigned long maxpages; |
| 1897 | unsigned long swapfilepages; |
| 1898 | |
| 1899 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { |
| 1900 | printk(KERN_ERR "Unable to find swap-space signature\n"); |
Cesar Eduardo Barros | 3871902 | 2011-03-22 16:33:30 -0700 | [diff] [blame] | 1901 | return 0; |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1902 | } |
| 1903 | |
| 1904 | /* swap partition endianess hack... */ |
| 1905 | if (swab32(swap_header->info.version) == 1) { |
| 1906 | swab32s(&swap_header->info.version); |
| 1907 | swab32s(&swap_header->info.last_page); |
| 1908 | swab32s(&swap_header->info.nr_badpages); |
| 1909 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
| 1910 | swab32s(&swap_header->info.badpages[i]); |
| 1911 | } |
| 1912 | /* Check the swap header's sub-version */ |
| 1913 | if (swap_header->info.version != 1) { |
| 1914 | printk(KERN_WARNING |
| 1915 | "Unable to handle swap header version %d\n", |
| 1916 | swap_header->info.version); |
Cesar Eduardo Barros | 3871902 | 2011-03-22 16:33:30 -0700 | [diff] [blame] | 1917 | return 0; |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1918 | } |
| 1919 | |
| 1920 | p->lowest_bit = 1; |
| 1921 | p->cluster_next = 1; |
| 1922 | p->cluster_nr = 0; |
| 1923 | |
| 1924 | /* |
| 1925 | * Find out how many pages are allowed for a single swap |
Hugh Dickins | a2c16d6 | 2011-08-03 16:21:19 -0700 | [diff] [blame] | 1926 | * device. There are three limiting factors: 1) the number |
| 1927 | * of bits for the swap offset in the swp_entry_t type, and |
| 1928 | * 2) the number of bits in the swap pte as defined by the |
| 1929 | * the different architectures, and 3) the number of free bits |
| 1930 | * in an exceptional radix_tree entry. In order to find the |
| 1931 | * largest possible bit mask, a swap entry with swap type 0 |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1932 | * and swap offset ~0UL is created, encoded to a swap pte, |
Hugh Dickins | a2c16d6 | 2011-08-03 16:21:19 -0700 | [diff] [blame] | 1933 | * decoded to a swp_entry_t again, and finally the swap |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1934 | * offset is extracted. This will mask all the bits from |
| 1935 | * the initial ~0UL mask that can't be encoded in either |
| 1936 | * the swp_entry_t or the architecture definition of a |
Hugh Dickins | a2c16d6 | 2011-08-03 16:21:19 -0700 | [diff] [blame] | 1937 | * swap pte. Then the same is done for a radix_tree entry. |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1938 | */ |
| 1939 | maxpages = swp_offset(pte_to_swp_entry( |
Hugh Dickins | a2c16d6 | 2011-08-03 16:21:19 -0700 | [diff] [blame] | 1940 | swp_entry_to_pte(swp_entry(0, ~0UL)))); |
| 1941 | maxpages = swp_offset(radix_to_swp_entry( |
| 1942 | swp_to_radix_entry(swp_entry(0, maxpages)))) + 1; |
| 1943 | |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1944 | if (maxpages > swap_header->info.last_page) { |
| 1945 | maxpages = swap_header->info.last_page + 1; |
| 1946 | /* p->max is an unsigned int: don't overflow it */ |
| 1947 | if ((unsigned int)maxpages == 0) |
| 1948 | maxpages = UINT_MAX; |
| 1949 | } |
| 1950 | p->highest_bit = maxpages - 1; |
| 1951 | |
| 1952 | if (!maxpages) |
Cesar Eduardo Barros | 3871902 | 2011-03-22 16:33:30 -0700 | [diff] [blame] | 1953 | return 0; |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1954 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
| 1955 | if (swapfilepages && maxpages > swapfilepages) { |
| 1956 | printk(KERN_WARNING |
| 1957 | "Swap area shorter than signature indicates\n"); |
Cesar Eduardo Barros | 3871902 | 2011-03-22 16:33:30 -0700 | [diff] [blame] | 1958 | return 0; |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1959 | } |
| 1960 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) |
Cesar Eduardo Barros | 3871902 | 2011-03-22 16:33:30 -0700 | [diff] [blame] | 1961 | return 0; |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1962 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
Cesar Eduardo Barros | 3871902 | 2011-03-22 16:33:30 -0700 | [diff] [blame] | 1963 | return 0; |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1964 | |
| 1965 | return maxpages; |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 1966 | } |
| 1967 | |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 1968 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
| 1969 | union swap_header *swap_header, |
| 1970 | unsigned char *swap_map, |
| 1971 | unsigned long maxpages, |
| 1972 | sector_t *span) |
| 1973 | { |
| 1974 | int i; |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 1975 | unsigned int nr_good_pages; |
| 1976 | int nr_extents; |
| 1977 | |
| 1978 | nr_good_pages = maxpages - 1; /* omit header page */ |
| 1979 | |
| 1980 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
| 1981 | unsigned int page_nr = swap_header->info.badpages[i]; |
Cesar Eduardo Barros | bdb8e3f | 2011-03-22 16:33:33 -0700 | [diff] [blame] | 1982 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
| 1983 | return -EINVAL; |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 1984 | if (page_nr < maxpages) { |
| 1985 | swap_map[page_nr] = SWAP_MAP_BAD; |
| 1986 | nr_good_pages--; |
| 1987 | } |
| 1988 | } |
| 1989 | |
| 1990 | if (nr_good_pages) { |
| 1991 | swap_map[0] = SWAP_MAP_BAD; |
| 1992 | p->max = maxpages; |
| 1993 | p->pages = nr_good_pages; |
| 1994 | nr_extents = setup_swap_extents(p, span); |
Cesar Eduardo Barros | bdb8e3f | 2011-03-22 16:33:33 -0700 | [diff] [blame] | 1995 | if (nr_extents < 0) |
| 1996 | return nr_extents; |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 1997 | nr_good_pages = p->pages; |
| 1998 | } |
| 1999 | if (!nr_good_pages) { |
| 2000 | printk(KERN_WARNING "Empty swap-file\n"); |
Cesar Eduardo Barros | bdb8e3f | 2011-03-22 16:33:33 -0700 | [diff] [blame] | 2001 | return -EINVAL; |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 2002 | } |
| 2003 | |
| 2004 | return nr_extents; |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 2005 | } |
| 2006 | |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 2007 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
| 2008 | { |
| 2009 | struct swap_info_struct *p; |
Cesar Eduardo Barros | 28b36bd | 2011-03-22 16:33:20 -0700 | [diff] [blame] | 2010 | char *name; |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 2011 | struct file *swap_file = NULL; |
| 2012 | struct address_space *mapping; |
Cesar Eduardo Barros | 4053154 | 2011-03-22 16:33:37 -0700 | [diff] [blame] | 2013 | int i; |
| 2014 | int prio; |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 2015 | int error; |
| 2016 | union swap_header *swap_header; |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 2017 | int nr_extents; |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 2018 | sector_t span; |
| 2019 | unsigned long maxpages; |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 2020 | unsigned char *swap_map = NULL; |
| 2021 | struct page *page = NULL; |
| 2022 | struct inode *inode = NULL; |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 2023 | |
| 2024 | if (!capable(CAP_SYS_ADMIN)) |
| 2025 | return -EPERM; |
| 2026 | |
| 2027 | p = alloc_swap_info(); |
Cesar Eduardo Barros | 2542e513 | 2011-03-22 16:33:18 -0700 | [diff] [blame] | 2028 | if (IS_ERR(p)) |
| 2029 | return PTR_ERR(p); |
Cesar Eduardo Barros | 53cbb24 | 2011-03-22 16:33:17 -0700 | [diff] [blame] | 2030 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2031 | name = getname(specialfile); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | if (IS_ERR(name)) { |
Cesar Eduardo Barros | 7de7fb6 | 2011-03-22 16:33:22 -0700 | [diff] [blame] | 2033 | error = PTR_ERR(name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2034 | name = NULL; |
Cesar Eduardo Barros | bd69010 | 2011-03-22 16:33:25 -0700 | [diff] [blame] | 2035 | goto bad_swap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2036 | } |
| 2037 | swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2038 | if (IS_ERR(swap_file)) { |
Cesar Eduardo Barros | 7de7fb6 | 2011-03-22 16:33:22 -0700 | [diff] [blame] | 2039 | error = PTR_ERR(swap_file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2040 | swap_file = NULL; |
Cesar Eduardo Barros | bd69010 | 2011-03-22 16:33:25 -0700 | [diff] [blame] | 2041 | goto bad_swap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | } |
| 2043 | |
| 2044 | p->swap_file = swap_file; |
| 2045 | mapping = swap_file->f_mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2046 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2047 | for (i = 0; i < nr_swapfiles; i++) { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 2048 | struct swap_info_struct *q = swap_info[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2049 | |
Cesar Eduardo Barros | e8e6c2e | 2011-03-22 16:33:16 -0700 | [diff] [blame] | 2050 | if (q == p || !q->swap_file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2051 | continue; |
Cesar Eduardo Barros | 7de7fb6 | 2011-03-22 16:33:22 -0700 | [diff] [blame] | 2052 | if (mapping == q->swap_file->f_mapping) { |
| 2053 | error = -EBUSY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2054 | goto bad_swap; |
Cesar Eduardo Barros | 7de7fb6 | 2011-03-22 16:33:22 -0700 | [diff] [blame] | 2055 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2056 | } |
| 2057 | |
Cesar Eduardo Barros | 2130781 | 2011-03-22 23:03:13 -0300 | [diff] [blame] | 2058 | inode = mapping->host; |
| 2059 | /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */ |
Cesar Eduardo Barros | 4d0e1e1 | 2011-03-22 16:33:26 -0700 | [diff] [blame] | 2060 | error = claim_swapfile(p, inode); |
| 2061 | if (unlikely(error)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2062 | goto bad_swap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2063 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2064 | /* |
| 2065 | * Read the swap header. |
| 2066 | */ |
| 2067 | if (!mapping->a_ops->readpage) { |
| 2068 | error = -EINVAL; |
| 2069 | goto bad_swap; |
| 2070 | } |
Pekka Enberg | 090d2b1 | 2006-06-23 02:05:08 -0700 | [diff] [blame] | 2071 | page = read_mapping_page(mapping, 0, swap_file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2072 | if (IS_ERR(page)) { |
| 2073 | error = PTR_ERR(page); |
| 2074 | goto bad_swap; |
| 2075 | } |
Hugh Dickins | 81e3397 | 2009-01-06 14:39:49 -0800 | [diff] [blame] | 2076 | swap_header = kmap(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2077 | |
Cesar Eduardo Barros | ca8bd38 | 2011-03-22 16:33:29 -0700 | [diff] [blame] | 2078 | maxpages = read_swap_header(p, swap_header, inode); |
| 2079 | if (unlikely(!maxpages)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2080 | error = -EINVAL; |
| 2081 | goto bad_swap; |
| 2082 | } |
Hugh Dickins | 886bb7e | 2009-01-06 14:39:48 -0800 | [diff] [blame] | 2083 | |
Hugh Dickins | 81e3397 | 2009-01-06 14:39:49 -0800 | [diff] [blame] | 2084 | /* OK, set up the swap map and apply the bad block list */ |
Cesar Eduardo Barros | 803d0c83 | 2011-03-22 16:33:14 -0700 | [diff] [blame] | 2085 | swap_map = vzalloc(maxpages); |
Hugh Dickins | 81e3397 | 2009-01-06 14:39:49 -0800 | [diff] [blame] | 2086 | if (!swap_map) { |
| 2087 | error = -ENOMEM; |
| 2088 | goto bad_swap; |
| 2089 | } |
| 2090 | |
Cesar Eduardo Barros | 1421ef3 | 2011-03-22 16:33:31 -0700 | [diff] [blame] | 2091 | error = swap_cgroup_swapon(p->type, maxpages); |
| 2092 | if (error) |
| 2093 | goto bad_swap; |
| 2094 | |
Cesar Eduardo Barros | 915d4d7 | 2011-03-22 16:33:32 -0700 | [diff] [blame] | 2095 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
| 2096 | maxpages, &span); |
| 2097 | if (unlikely(nr_extents < 0)) { |
| 2098 | error = nr_extents; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2099 | goto bad_swap; |
| 2100 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2101 | |
Suresh Jayaraman | 3bd0f0c | 2009-09-30 10:53:48 +0200 | [diff] [blame] | 2102 | if (p->bdev) { |
| 2103 | if (blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
| 2104 | p->flags |= SWP_SOLIDSTATE; |
| 2105 | p->cluster_next = 1 + (random32() % p->highest_bit); |
| 2106 | } |
Hugh Dickins | 3399446 | 2010-09-09 16:38:11 -0700 | [diff] [blame] | 2107 | if (discard_swap(p) == 0 && (swap_flags & SWAP_FLAG_DISCARD)) |
Suresh Jayaraman | 3bd0f0c | 2009-09-30 10:53:48 +0200 | [diff] [blame] | 2108 | p->flags |= SWP_DISCARDABLE; |
Hugh Dickins | 20137a4 | 2009-01-06 14:39:54 -0800 | [diff] [blame] | 2109 | } |
Hugh Dickins | 6a6ba83 | 2009-01-06 14:39:51 -0800 | [diff] [blame] | 2110 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2111 | mutex_lock(&swapon_mutex); |
Cesar Eduardo Barros | 4053154 | 2011-03-22 16:33:37 -0700 | [diff] [blame] | 2112 | prio = -1; |
Hugh Dickins | 78ecba0 | 2008-07-23 21:28:23 -0700 | [diff] [blame] | 2113 | if (swap_flags & SWAP_FLAG_PREFER) |
Cesar Eduardo Barros | 4053154 | 2011-03-22 16:33:37 -0700 | [diff] [blame] | 2114 | prio = |
Hugh Dickins | 78ecba0 | 2008-07-23 21:28:23 -0700 | [diff] [blame] | 2115 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
Cesar Eduardo Barros | 4053154 | 2011-03-22 16:33:37 -0700 | [diff] [blame] | 2116 | enable_swap_info(p, prio, swap_map); |
Cesar Eduardo Barros | c69dbfb | 2011-03-22 16:33:35 -0700 | [diff] [blame] | 2117 | |
| 2118 | printk(KERN_INFO "Adding %uk swap on %s. " |
| 2119 | "Priority:%d extents:%d across:%lluk %s%s\n", |
| 2120 | p->pages<<(PAGE_SHIFT-10), name, p->prio, |
| 2121 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
| 2122 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", |
| 2123 | (p->flags & SWP_DISCARDABLE) ? "D" : ""); |
| 2124 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2125 | mutex_unlock(&swapon_mutex); |
Kay Sievers | 66d7dd5 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 2126 | atomic_inc(&proc_poll_event); |
| 2127 | wake_up_interruptible(&proc_poll_wait); |
| 2128 | |
Cesar Eduardo Barros | 9b01c35 | 2011-03-22 16:33:24 -0700 | [diff] [blame] | 2129 | if (S_ISREG(inode->i_mode)) |
| 2130 | inode->i_flags |= S_SWAPFILE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | error = 0; |
| 2132 | goto out; |
| 2133 | bad_swap: |
Cesar Eduardo Barros | bd69010 | 2011-03-22 16:33:25 -0700 | [diff] [blame] | 2134 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
Cesar Eduardo Barros | f2090d2 | 2011-03-22 16:33:23 -0700 | [diff] [blame] | 2135 | set_blocksize(p->bdev, p->old_block_size); |
| 2136 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2137 | } |
Hugh Dickins | 4cd3bb1 | 2005-09-03 15:54:33 -0700 | [diff] [blame] | 2138 | destroy_swap_extents(p); |
Cesar Eduardo Barros | e8e6c2e | 2011-03-22 16:33:16 -0700 | [diff] [blame] | 2139 | swap_cgroup_swapoff(p->type); |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2140 | spin_lock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2141 | p->swap_file = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2142 | p->flags = 0; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2143 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | vfree(swap_map); |
Mel Gorman | 52c5056 | 2011-03-22 16:30:08 -0700 | [diff] [blame] | 2145 | if (swap_file) { |
Cesar Eduardo Barros | 2130781 | 2011-03-22 23:03:13 -0300 | [diff] [blame] | 2146 | if (inode && S_ISREG(inode->i_mode)) { |
Mel Gorman | 52c5056 | 2011-03-22 16:30:08 -0700 | [diff] [blame] | 2147 | mutex_unlock(&inode->i_mutex); |
Cesar Eduardo Barros | 2130781 | 2011-03-22 23:03:13 -0300 | [diff] [blame] | 2148 | inode = NULL; |
| 2149 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2150 | filp_close(swap_file, NULL); |
Mel Gorman | 52c5056 | 2011-03-22 16:30:08 -0700 | [diff] [blame] | 2151 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | out: |
| 2153 | if (page && !IS_ERR(page)) { |
| 2154 | kunmap(page); |
| 2155 | page_cache_release(page); |
| 2156 | } |
| 2157 | if (name) |
| 2158 | putname(name); |
Cesar Eduardo Barros | 9b01c35 | 2011-03-22 16:33:24 -0700 | [diff] [blame] | 2159 | if (inode && S_ISREG(inode->i_mode)) |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2160 | mutex_unlock(&inode->i_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2161 | return error; |
| 2162 | } |
| 2163 | |
| 2164 | void si_swapinfo(struct sysinfo *val) |
| 2165 | { |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 2166 | unsigned int type; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2167 | unsigned long nr_to_be_unused = 0; |
| 2168 | |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2169 | spin_lock(&swap_lock); |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 2170 | for (type = 0; type < nr_swapfiles; type++) { |
| 2171 | struct swap_info_struct *si = swap_info[type]; |
| 2172 | |
| 2173 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) |
| 2174 | nr_to_be_unused += si->inuse_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | } |
| 2176 | val->freeswap = nr_swap_pages + nr_to_be_unused; |
| 2177 | val->totalswap = total_swap_pages + nr_to_be_unused; |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2178 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2179 | } |
| 2180 | |
| 2181 | /* |
| 2182 | * Verify that a swap entry is valid and increment its swap map count. |
| 2183 | * |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2184 | * Returns error code in following case. |
| 2185 | * - success -> 0 |
| 2186 | * - swp_entry is invalid -> EINVAL |
| 2187 | * - swp_entry is migration entry -> EINVAL |
| 2188 | * - swap-cache reference is requested but there is already one. -> EEXIST |
| 2189 | * - swap-cache reference is requested but the entry is not used. -> ENOENT |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 2190 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2191 | */ |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 2192 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2193 | { |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 2194 | struct swap_info_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2195 | unsigned long offset, type; |
Hugh Dickins | 8d69aae | 2009-12-14 17:58:45 -0800 | [diff] [blame] | 2196 | unsigned char count; |
| 2197 | unsigned char has_cache; |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2198 | int err = -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2199 | |
Andi Kleen | a7420aa | 2009-09-16 11:50:05 +0200 | [diff] [blame] | 2200 | if (non_swap_entry(entry)) |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2201 | goto out; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 2202 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2203 | type = swp_type(entry); |
| 2204 | if (type >= nr_swapfiles) |
| 2205 | goto bad_file; |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 2206 | p = swap_info[type]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2207 | offset = swp_offset(entry); |
| 2208 | |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2209 | spin_lock(&swap_lock); |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2210 | if (unlikely(offset >= p->max)) |
| 2211 | goto unlock_out; |
| 2212 | |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2213 | count = p->swap_map[offset]; |
| 2214 | has_cache = count & SWAP_HAS_CACHE; |
| 2215 | count &= ~SWAP_HAS_CACHE; |
| 2216 | err = 0; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2217 | |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2218 | if (usage == SWAP_HAS_CACHE) { |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2219 | |
| 2220 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2221 | if (!has_cache && count) |
| 2222 | has_cache = SWAP_HAS_CACHE; |
| 2223 | else if (has_cache) /* someone else added cache */ |
| 2224 | err = -EEXIST; |
| 2225 | else /* no users remaining */ |
| 2226 | err = -ENOENT; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2227 | |
| 2228 | } else if (count || has_cache) { |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2229 | |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 2230 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
| 2231 | count += usage; |
| 2232 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2233 | err = -EINVAL; |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 2234 | else if (swap_count_continued(p, offset, count)) |
| 2235 | count = COUNT_CONTINUED; |
| 2236 | else |
| 2237 | err = -ENOMEM; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2238 | } else |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2239 | err = -ENOENT; /* unused swap entry */ |
| 2240 | |
| 2241 | p->swap_map[offset] = count | has_cache; |
| 2242 | |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2243 | unlock_out: |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2244 | spin_unlock(&swap_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2245 | out: |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2246 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2247 | |
| 2248 | bad_file: |
| 2249 | printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val); |
| 2250 | goto out; |
| 2251 | } |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2252 | |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2253 | /* |
Hugh Dickins | aaa4686 | 2009-12-14 17:58:47 -0800 | [diff] [blame] | 2254 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs |
| 2255 | * (in which case its reference count is never incremented). |
| 2256 | */ |
| 2257 | void swap_shmem_alloc(swp_entry_t entry) |
| 2258 | { |
| 2259 | __swap_duplicate(entry, SWAP_MAP_SHMEM); |
| 2260 | } |
| 2261 | |
| 2262 | /* |
Hugh Dickins | 08259d5 | 2010-03-05 13:42:25 -0800 | [diff] [blame] | 2263 | * Increase reference count of swap entry by 1. |
| 2264 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required |
| 2265 | * but could not be atomically allocated. Returns 0, just as if it succeeded, |
| 2266 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which |
| 2267 | * might occur if a page table entry has got corrupted. |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2268 | */ |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 2269 | int swap_duplicate(swp_entry_t entry) |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2270 | { |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 2271 | int err = 0; |
| 2272 | |
| 2273 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) |
| 2274 | err = add_swap_count_continuation(entry, GFP_ATOMIC); |
| 2275 | return err; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2276 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2277 | |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 2278 | /* |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2279 | * @entry: swap entry for which we allocate swap cache. |
| 2280 | * |
Hugh Dickins | 73c34b6 | 2009-12-14 17:58:43 -0800 | [diff] [blame] | 2281 | * Called when allocating swap cache for existing swap entry, |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2282 | * This can return error codes. Returns 0 at success. |
| 2283 | * -EBUSY means there is a swap cache. |
| 2284 | * Note: return code is different from swap_duplicate(). |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 2285 | */ |
| 2286 | int swapcache_prepare(swp_entry_t entry) |
| 2287 | { |
Hugh Dickins | 253d553 | 2009-12-14 17:58:44 -0800 | [diff] [blame] | 2288 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
KAMEZAWA Hiroyuki | cb4b86b | 2009-06-16 15:32:52 -0700 | [diff] [blame] | 2289 | } |
| 2290 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2291 | /* |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2292 | * swap_lock prevents swap_map being freed. Don't grab an extra |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2293 | * reference on the swaphandle, it doesn't matter if it becomes unused. |
| 2294 | */ |
| 2295 | int valid_swaphandles(swp_entry_t entry, unsigned long *offset) |
| 2296 | { |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2297 | struct swap_info_struct *si; |
Hugh Dickins | 3f9e794 | 2006-09-29 02:01:26 -0700 | [diff] [blame] | 2298 | int our_page_cluster = page_cluster; |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2299 | pgoff_t target, toff; |
| 2300 | pgoff_t base, end; |
| 2301 | int nr_pages = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2302 | |
Hugh Dickins | 3f9e794 | 2006-09-29 02:01:26 -0700 | [diff] [blame] | 2303 | if (!our_page_cluster) /* no readahead */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2304 | return 0; |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2305 | |
Hugh Dickins | efa90a9 | 2009-12-14 17:58:41 -0800 | [diff] [blame] | 2306 | si = swap_info[swp_type(entry)]; |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2307 | target = swp_offset(entry); |
| 2308 | base = (target >> our_page_cluster) << our_page_cluster; |
| 2309 | end = base + (1 << our_page_cluster); |
| 2310 | if (!base) /* first page is swap header */ |
| 2311 | base++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2313 | spin_lock(&swap_lock); |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2314 | if (end > si->max) /* don't go beyond end of map */ |
| 2315 | end = si->max; |
| 2316 | |
| 2317 | /* Count contiguous allocated slots above our target */ |
| 2318 | for (toff = target; ++toff < end; nr_pages++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2319 | /* Don't read in free or bad pages */ |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2320 | if (!si->swap_map[toff]) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2321 | break; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2322 | if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2323 | break; |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2324 | } |
| 2325 | /* Count contiguous allocated slots below our target */ |
| 2326 | for (toff = target; --toff >= base; nr_pages++) { |
| 2327 | /* Don't read in free or bad pages */ |
| 2328 | if (!si->swap_map[toff]) |
| 2329 | break; |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 2330 | if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD) |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2331 | break; |
| 2332 | } |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 2333 | spin_unlock(&swap_lock); |
Hugh Dickins | 8952898 | 2008-02-04 22:28:45 -0800 | [diff] [blame] | 2334 | |
| 2335 | /* |
| 2336 | * Indicate starting offset, and return number of pages to get: |
| 2337 | * if only 1, say 0, since there's then no readahead to be done. |
| 2338 | */ |
| 2339 | *offset = ++toff; |
| 2340 | return nr_pages? ++nr_pages: 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2341 | } |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 2342 | |
| 2343 | /* |
| 2344 | * add_swap_count_continuation - called when a swap count is duplicated |
| 2345 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's |
| 2346 | * page of the original vmalloc'ed swap_map, to hold the continuation count |
| 2347 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called |
| 2348 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. |
| 2349 | * |
| 2350 | * These continuation pages are seldom referenced: the common paths all work |
| 2351 | * on the original swap_map, only referring to a continuation page when the |
| 2352 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. |
| 2353 | * |
| 2354 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding |
| 2355 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) |
| 2356 | * can be called after dropping locks. |
| 2357 | */ |
| 2358 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) |
| 2359 | { |
| 2360 | struct swap_info_struct *si; |
| 2361 | struct page *head; |
| 2362 | struct page *page; |
| 2363 | struct page *list_page; |
| 2364 | pgoff_t offset; |
| 2365 | unsigned char count; |
| 2366 | |
| 2367 | /* |
| 2368 | * When debugging, it's easier to use __GFP_ZERO here; but it's better |
| 2369 | * for latency not to zero a page while GFP_ATOMIC and holding locks. |
| 2370 | */ |
| 2371 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); |
| 2372 | |
| 2373 | si = swap_info_get(entry); |
| 2374 | if (!si) { |
| 2375 | /* |
| 2376 | * An acceptable race has occurred since the failing |
| 2377 | * __swap_duplicate(): the swap entry has been freed, |
| 2378 | * perhaps even the whole swap_map cleared for swapoff. |
| 2379 | */ |
| 2380 | goto outer; |
| 2381 | } |
| 2382 | |
| 2383 | offset = swp_offset(entry); |
| 2384 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
| 2385 | |
| 2386 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { |
| 2387 | /* |
| 2388 | * The higher the swap count, the more likely it is that tasks |
| 2389 | * will race to add swap count continuation: we need to avoid |
| 2390 | * over-provisioning. |
| 2391 | */ |
| 2392 | goto out; |
| 2393 | } |
| 2394 | |
| 2395 | if (!page) { |
| 2396 | spin_unlock(&swap_lock); |
| 2397 | return -ENOMEM; |
| 2398 | } |
| 2399 | |
| 2400 | /* |
| 2401 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, |
| 2402 | * no architecture is using highmem pages for kernel pagetables: so it |
| 2403 | * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps. |
| 2404 | */ |
| 2405 | head = vmalloc_to_page(si->swap_map + offset); |
| 2406 | offset &= ~PAGE_MASK; |
| 2407 | |
| 2408 | /* |
| 2409 | * Page allocation does not initialize the page's lru field, |
| 2410 | * but it does always reset its private field. |
| 2411 | */ |
| 2412 | if (!page_private(head)) { |
| 2413 | BUG_ON(count & COUNT_CONTINUED); |
| 2414 | INIT_LIST_HEAD(&head->lru); |
| 2415 | set_page_private(head, SWP_CONTINUED); |
| 2416 | si->flags |= SWP_CONTINUED; |
| 2417 | } |
| 2418 | |
| 2419 | list_for_each_entry(list_page, &head->lru, lru) { |
| 2420 | unsigned char *map; |
| 2421 | |
| 2422 | /* |
| 2423 | * If the previous map said no continuation, but we've found |
| 2424 | * a continuation page, free our allocation and use this one. |
| 2425 | */ |
| 2426 | if (!(count & COUNT_CONTINUED)) |
| 2427 | goto out; |
| 2428 | |
| 2429 | map = kmap_atomic(list_page, KM_USER0) + offset; |
| 2430 | count = *map; |
| 2431 | kunmap_atomic(map, KM_USER0); |
| 2432 | |
| 2433 | /* |
| 2434 | * If this continuation count now has some space in it, |
| 2435 | * free our allocation and use this one. |
| 2436 | */ |
| 2437 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) |
| 2438 | goto out; |
| 2439 | } |
| 2440 | |
| 2441 | list_add_tail(&page->lru, &head->lru); |
| 2442 | page = NULL; /* now it's attached, don't free it */ |
| 2443 | out: |
| 2444 | spin_unlock(&swap_lock); |
| 2445 | outer: |
| 2446 | if (page) |
| 2447 | __free_page(page); |
| 2448 | return 0; |
| 2449 | } |
| 2450 | |
| 2451 | /* |
| 2452 | * swap_count_continued - when the original swap_map count is incremented |
| 2453 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry |
| 2454 | * into, carry if so, or else fail until a new continuation page is allocated; |
| 2455 | * when the original swap_map count is decremented from 0 with continuation, |
| 2456 | * borrow from the continuation and report whether it still holds more. |
| 2457 | * Called while __swap_duplicate() or swap_entry_free() holds swap_lock. |
| 2458 | */ |
| 2459 | static bool swap_count_continued(struct swap_info_struct *si, |
| 2460 | pgoff_t offset, unsigned char count) |
| 2461 | { |
| 2462 | struct page *head; |
| 2463 | struct page *page; |
| 2464 | unsigned char *map; |
| 2465 | |
| 2466 | head = vmalloc_to_page(si->swap_map + offset); |
| 2467 | if (page_private(head) != SWP_CONTINUED) { |
| 2468 | BUG_ON(count & COUNT_CONTINUED); |
| 2469 | return false; /* need to add count continuation */ |
| 2470 | } |
| 2471 | |
| 2472 | offset &= ~PAGE_MASK; |
| 2473 | page = list_entry(head->lru.next, struct page, lru); |
| 2474 | map = kmap_atomic(page, KM_USER0) + offset; |
| 2475 | |
| 2476 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ |
| 2477 | goto init_map; /* jump over SWAP_CONT_MAX checks */ |
| 2478 | |
| 2479 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ |
| 2480 | /* |
| 2481 | * Think of how you add 1 to 999 |
| 2482 | */ |
| 2483 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { |
| 2484 | kunmap_atomic(map, KM_USER0); |
| 2485 | page = list_entry(page->lru.next, struct page, lru); |
| 2486 | BUG_ON(page == head); |
| 2487 | map = kmap_atomic(page, KM_USER0) + offset; |
| 2488 | } |
| 2489 | if (*map == SWAP_CONT_MAX) { |
| 2490 | kunmap_atomic(map, KM_USER0); |
| 2491 | page = list_entry(page->lru.next, struct page, lru); |
| 2492 | if (page == head) |
| 2493 | return false; /* add count continuation */ |
| 2494 | map = kmap_atomic(page, KM_USER0) + offset; |
| 2495 | init_map: *map = 0; /* we didn't zero the page */ |
| 2496 | } |
| 2497 | *map += 1; |
| 2498 | kunmap_atomic(map, KM_USER0); |
| 2499 | page = list_entry(page->lru.prev, struct page, lru); |
| 2500 | while (page != head) { |
| 2501 | map = kmap_atomic(page, KM_USER0) + offset; |
| 2502 | *map = COUNT_CONTINUED; |
| 2503 | kunmap_atomic(map, KM_USER0); |
| 2504 | page = list_entry(page->lru.prev, struct page, lru); |
| 2505 | } |
| 2506 | return true; /* incremented */ |
| 2507 | |
| 2508 | } else { /* decrementing */ |
| 2509 | /* |
| 2510 | * Think of how you subtract 1 from 1000 |
| 2511 | */ |
| 2512 | BUG_ON(count != COUNT_CONTINUED); |
| 2513 | while (*map == COUNT_CONTINUED) { |
| 2514 | kunmap_atomic(map, KM_USER0); |
| 2515 | page = list_entry(page->lru.next, struct page, lru); |
| 2516 | BUG_ON(page == head); |
| 2517 | map = kmap_atomic(page, KM_USER0) + offset; |
| 2518 | } |
| 2519 | BUG_ON(*map == 0); |
| 2520 | *map -= 1; |
| 2521 | if (*map == 0) |
| 2522 | count = 0; |
| 2523 | kunmap_atomic(map, KM_USER0); |
| 2524 | page = list_entry(page->lru.prev, struct page, lru); |
| 2525 | while (page != head) { |
| 2526 | map = kmap_atomic(page, KM_USER0) + offset; |
| 2527 | *map = SWAP_CONT_MAX | count; |
| 2528 | count = COUNT_CONTINUED; |
| 2529 | kunmap_atomic(map, KM_USER0); |
| 2530 | page = list_entry(page->lru.prev, struct page, lru); |
| 2531 | } |
| 2532 | return count == COUNT_CONTINUED; |
| 2533 | } |
| 2534 | } |
| 2535 | |
| 2536 | /* |
| 2537 | * free_swap_count_continuations - swapoff free all the continuation pages |
| 2538 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. |
| 2539 | */ |
| 2540 | static void free_swap_count_continuations(struct swap_info_struct *si) |
| 2541 | { |
| 2542 | pgoff_t offset; |
| 2543 | |
| 2544 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { |
| 2545 | struct page *head; |
| 2546 | head = vmalloc_to_page(si->swap_map + offset); |
| 2547 | if (page_private(head)) { |
| 2548 | struct list_head *this, *next; |
| 2549 | list_for_each_safe(this, next, &head->lru) { |
| 2550 | struct page *page; |
| 2551 | page = list_entry(this, struct page, lru); |
| 2552 | list_del(this); |
| 2553 | __free_page(page); |
| 2554 | } |
| 2555 | } |
| 2556 | } |
| 2557 | } |