blob: 36af6eeaa67eaa0532442706e668e9a268596c90 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 Torvalds1da177e2005-04-16 15:20:36 -07008#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 Dickins072441e2011-06-27 16:18:02 -070017#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080019#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080024#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070032#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070033#include <linux/oom.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060034#include <linux/frontswap.h>
35#include <linux/swapfile.h>
Mel Gormanf981c592012-07-31 16:44:47 -070036#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/pgtable.h>
39#include <asm/tlbflush.h>
40#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080041#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Hugh Dickins570a335b2009-12-14 17:58:46 -080043static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
44 unsigned char);
45static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080046static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080047
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060048DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070049static unsigned int nr_swapfiles;
Shaohua Liec8acf22013-02-22 16:34:38 -080050atomic_long_t nr_swap_pages;
51/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070052long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070053static int least_priority;
Shaohua Liec8acf22013-02-22 16:34:38 -080054static atomic_t highest_priority_index = ATOMIC_INIT(-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static const char Bad_file[] = "Bad swap file entry ";
57static const char Unused_file[] = "Unused swap file entry ";
58static const char Bad_offset[] = "Bad swap offset entry ";
59static const char Unused_offset[] = "Unused swap offset entry ";
60
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060061struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060063struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Ingo Molnarfc0abb12006-01-18 17:42:33 -080065static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Kay Sievers66d7dd52010-10-26 14:22:06 -070067static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
68/* Activity counter to indicate that a swapon or swapoff has occurred */
69static atomic_t proc_poll_event = ATOMIC_INIT(0);
70
Hugh Dickins8d69aae2009-12-14 17:58:45 -080071static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070072{
Hugh Dickins570a335b2009-12-14 17:58:46 -080073 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070074}
75
Hugh Dickinsefa90a92009-12-14 17:58:41 -080076/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070077static int
78__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
79{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080080 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070081 struct page *page;
82 int ret = 0;
83
Shaohua Li33806f02013-02-22 16:34:37 -080084 page = find_get_page(swap_address_space(entry), entry.val);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070085 if (!page)
86 return 0;
87 /*
88 * This function is called from scan_swap_map() and it's called
89 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
90 * We have to use trylock for avoiding deadlock. This is a special
91 * case and you should use try_to_free_swap() with explicit lock_page()
92 * in usual operations.
93 */
94 if (trylock_page(page)) {
95 ret = try_to_free_swap(page);
96 unlock_page(page);
97 }
98 page_cache_release(page);
99 return ret;
100}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800103 * swapon tell device that all the old swap contents can be discarded,
104 * to allow the swap device to optimize its wear-levelling.
105 */
106static int discard_swap(struct swap_info_struct *si)
107{
108 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800109 sector_t start_block;
110 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800111 int err = 0;
112
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800113 /* Do not discard the swap header page! */
114 se = &si->first_swap_extent;
115 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
116 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
117 if (nr_blocks) {
118 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200119 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800120 if (err)
121 return err;
122 cond_resched();
123 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800124
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800125 list_for_each_entry(se, &si->first_swap_extent.list, list) {
126 start_block = se->start_block << (PAGE_SHIFT - 9);
127 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800128
129 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200130 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800131 if (err)
132 break;
133
134 cond_resched();
135 }
136 return err; /* That will often be -EOPNOTSUPP */
137}
138
Hugh Dickins7992fde2009-01-06 14:39:53 -0800139/*
140 * swap allocation tell device that a cluster of swap can now be discarded,
141 * to allow the swap device to optimize its wear-levelling.
142 */
143static void discard_swap_cluster(struct swap_info_struct *si,
144 pgoff_t start_page, pgoff_t nr_pages)
145{
146 struct swap_extent *se = si->curr_swap_extent;
147 int found_extent = 0;
148
149 while (nr_pages) {
150 struct list_head *lh;
151
152 if (se->start_page <= start_page &&
153 start_page < se->start_page + se->nr_pages) {
154 pgoff_t offset = start_page - se->start_page;
155 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800156 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800157
158 if (nr_blocks > nr_pages)
159 nr_blocks = nr_pages;
160 start_page += nr_blocks;
161 nr_pages -= nr_blocks;
162
163 if (!found_extent++)
164 si->curr_swap_extent = se;
165
166 start_block <<= PAGE_SHIFT - 9;
167 nr_blocks <<= PAGE_SHIFT - 9;
168 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200169 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800170 break;
171 }
172
173 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800174 se = list_entry(lh, struct swap_extent, list);
175 }
176}
177
178static int wait_for_discard(void *word)
179{
180 schedule();
181 return 0;
182}
183
Hugh Dickins048c27f2005-09-03 15:54:40 -0700184#define SWAPFILE_CLUSTER 256
185#define LATENCY_LIMIT 256
186
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700187static unsigned long scan_swap_map(struct swap_info_struct *si,
188 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800190 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800191 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800192 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700193 int latency_ration = LATENCY_LIMIT;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800194 int found_free_cluster = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800196 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700197 * We try to cluster swap pages by allocating them sequentially
198 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
199 * way, however, we resort to first-free allocation, starting
200 * a new cluster. This prevents us from scattering swap pages
201 * all over the entire swap partition, so that we reduce
202 * overall disk seek times between swap pages. -- sct
203 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800204 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700205 */
206
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700207 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800208 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800209
210 if (unlikely(!si->cluster_nr--)) {
211 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
212 si->cluster_nr = SWAPFILE_CLUSTER - 1;
213 goto checks;
214 }
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -0700215 if (si->flags & SWP_PAGE_DISCARD) {
Hugh Dickins7992fde2009-01-06 14:39:53 -0800216 /*
217 * Start range check on racing allocations, in case
218 * they overlap the cluster we eventually decide on
219 * (we scan without swap_lock to allow preemption).
220 * It's hardly conceivable that cluster_nr could be
221 * wrapped during our scan, but don't depend on it.
222 */
223 if (si->lowest_alloc)
224 goto checks;
225 si->lowest_alloc = si->max;
226 si->highest_alloc = 0;
227 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800228 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700229
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800230 /*
231 * If seek is expensive, start searching for new cluster from
232 * start of partition, to minimize the span of allocated swap.
233 * But if seek is cheap, search from our current position, so
234 * that swap is allocated from all over the partition: if the
235 * Flash Translation Layer only remaps within limited zones,
236 * we don't want to wear out the first zone too quickly.
237 */
238 if (!(si->flags & SWP_SOLIDSTATE))
239 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700240 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
241
242 /* Locate the first empty (unaligned) cluster */
243 for (; last_in_cluster <= si->highest_bit; offset++) {
244 if (si->swap_map[offset])
245 last_in_cluster = offset + SWAPFILE_CLUSTER;
246 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800247 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800248 offset -= SWAPFILE_CLUSTER - 1;
249 si->cluster_next = offset;
250 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800251 found_free_cluster = 1;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800252 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700254 if (unlikely(--latency_ration < 0)) {
255 cond_resched();
256 latency_ration = LATENCY_LIMIT;
257 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700258 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800259
260 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800261 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
262
263 /* Locate the first empty (unaligned) cluster */
264 for (; last_in_cluster < scan_base; offset++) {
265 if (si->swap_map[offset])
266 last_in_cluster = offset + SWAPFILE_CLUSTER;
267 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800268 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800269 offset -= SWAPFILE_CLUSTER - 1;
270 si->cluster_next = offset;
271 si->cluster_nr = SWAPFILE_CLUSTER - 1;
272 found_free_cluster = 1;
273 goto checks;
274 }
275 if (unlikely(--latency_ration < 0)) {
276 cond_resched();
277 latency_ration = LATENCY_LIMIT;
278 }
279 }
280
281 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800282 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800283 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800284 si->lowest_alloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700286
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800287checks:
288 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700289 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700290 if (!si->highest_bit)
291 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800292 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800293 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700294
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700295 /* reuse swap entry of cache-only swap if not busy. */
296 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700297 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800298 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700299 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800300 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700301 /* entry was freed successfully, try to use this again */
302 if (swap_was_freed)
303 goto checks;
304 goto scan; /* check next one */
305 }
306
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800307 if (si->swap_map[offset])
308 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700309
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800310 if (offset == si->lowest_bit)
311 si->lowest_bit++;
312 if (offset == si->highest_bit)
313 si->highest_bit--;
314 si->inuse_pages++;
315 if (si->inuse_pages == si->pages) {
316 si->lowest_bit = si->max;
317 si->highest_bit = 0;
318 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800319 si->swap_map[offset] = usage;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800320 si->cluster_next = offset + 1;
321 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800322
323 if (si->lowest_alloc) {
324 /*
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -0700325 * Only set when SWP_PAGE_DISCARD, and there's a scan
Hugh Dickins7992fde2009-01-06 14:39:53 -0800326 * for a free cluster in progress or just completed.
327 */
328 if (found_free_cluster) {
329 /*
330 * To optimize wear-levelling, discard the
331 * old data of the cluster, taking care not to
332 * discard any of its pages that have already
333 * been allocated by racing tasks (offset has
334 * already stepped over any at the beginning).
335 */
336 if (offset < si->highest_alloc &&
337 si->lowest_alloc <= last_in_cluster)
338 last_in_cluster = si->lowest_alloc - 1;
339 si->flags |= SWP_DISCARDING;
Shaohua Liec8acf22013-02-22 16:34:38 -0800340 spin_unlock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800341
342 if (offset < last_in_cluster)
343 discard_swap_cluster(si, offset,
344 last_in_cluster - offset + 1);
345
Shaohua Liec8acf22013-02-22 16:34:38 -0800346 spin_lock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800347 si->lowest_alloc = 0;
348 si->flags &= ~SWP_DISCARDING;
349
350 smp_mb(); /* wake_up_bit advises this */
351 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
352
353 } else if (si->flags & SWP_DISCARDING) {
354 /*
355 * Delay using pages allocated by racing tasks
356 * until the whole discard has been issued. We
357 * could defer that delay until swap_writepage,
358 * but it's easier to keep this self-contained.
359 */
Shaohua Liec8acf22013-02-22 16:34:38 -0800360 spin_unlock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800361 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
362 wait_for_discard, TASK_UNINTERRUPTIBLE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800363 spin_lock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800364 } else {
365 /*
366 * Note pages allocated by racing tasks while
367 * scan for a free cluster is in progress, so
368 * that its final discard can exclude them.
369 */
370 if (offset < si->lowest_alloc)
371 si->lowest_alloc = offset;
372 if (offset > si->highest_alloc)
373 si->highest_alloc = offset;
374 }
375 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800376 return offset;
377
378scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800379 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700380 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700381 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800382 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700383 goto checks;
384 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700385 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800386 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700387 goto checks;
388 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700389 if (unlikely(--latency_ration < 0)) {
390 cond_resched();
391 latency_ration = LATENCY_LIMIT;
392 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700393 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800394 offset = si->lowest_bit;
395 while (++offset < scan_base) {
396 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800397 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800398 goto checks;
399 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700400 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800401 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700402 goto checks;
403 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800404 if (unlikely(--latency_ration < 0)) {
405 cond_resched();
406 latency_ration = LATENCY_LIMIT;
407 }
408 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800409 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700410
411no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700412 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 return 0;
414}
415
416swp_entry_t get_swap_page(void)
417{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700418 struct swap_info_struct *si;
419 pgoff_t offset;
420 int type, next;
421 int wrapped = 0;
Shaohua Liec8acf22013-02-22 16:34:38 -0800422 int hp_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
Hugh Dickins5d337b92005-09-03 15:54:41 -0700424 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800425 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700426 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800427 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700429 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800430 hp_index = atomic_xchg(&highest_priority_index, -1);
431 /*
432 * highest_priority_index records current highest priority swap
433 * type which just frees swap entries. If its priority is
434 * higher than that of swap_list.next swap type, we use it. It
435 * isn't protected by swap_lock, so it can be an invalid value
436 * if the corresponding swap type is swapoff. We double check
437 * the flags here. It's even possible the swap type is swapoff
438 * and swapon again and its priority is changed. In such rare
439 * case, low prority swap type might be used, but eventually
440 * high priority swap will be used after several rounds of
441 * swap.
442 */
443 if (hp_index != -1 && hp_index != type &&
444 swap_info[type]->prio < swap_info[hp_index]->prio &&
445 (swap_info[hp_index]->flags & SWP_WRITEOK)) {
446 type = hp_index;
447 swap_list.next = type;
448 }
449
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800450 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700451 next = si->next;
452 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800453 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700454 next = swap_list.head;
455 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700457
Shaohua Liec8acf22013-02-22 16:34:38 -0800458 spin_lock(&si->lock);
459 if (!si->highest_bit) {
460 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700461 continue;
Shaohua Liec8acf22013-02-22 16:34:38 -0800462 }
463 if (!(si->flags & SWP_WRITEOK)) {
464 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700465 continue;
Shaohua Liec8acf22013-02-22 16:34:38 -0800466 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700467
468 swap_list.next = next;
Shaohua Liec8acf22013-02-22 16:34:38 -0800469
470 spin_unlock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700471 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800472 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800473 spin_unlock(&si->lock);
474 if (offset)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700475 return swp_entry(type, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800476 spin_lock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700477 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700479
Shaohua Liec8acf22013-02-22 16:34:38 -0800480 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700481noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700482 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700483 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
485
Hugh Dickins910321e2010-09-09 16:38:07 -0700486/* The only caller of this function is now susupend routine */
487swp_entry_t get_swap_page_of_type(int type)
488{
489 struct swap_info_struct *si;
490 pgoff_t offset;
491
Hugh Dickins910321e2010-09-09 16:38:07 -0700492 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800493 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700494 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800495 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700496 /* This is called for allocating swap entry, not cache */
497 offset = scan_swap_map(si, 1);
498 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800499 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700500 return swp_entry(type, offset);
501 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800502 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700503 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800504 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700505 return (swp_entry_t) {0};
506}
507
Hugh Dickins73c34b62009-12-14 17:58:43 -0800508static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800510 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 unsigned long offset, type;
512
513 if (!entry.val)
514 goto out;
515 type = swp_type(entry);
516 if (type >= nr_swapfiles)
517 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800518 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 if (!(p->flags & SWP_USED))
520 goto bad_device;
521 offset = swp_offset(entry);
522 if (offset >= p->max)
523 goto bad_offset;
524 if (!p->swap_map[offset])
525 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800526 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 return p;
528
529bad_free:
530 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
531 goto out;
532bad_offset:
533 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
534 goto out;
535bad_device:
536 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
537 goto out;
538bad_nofile:
539 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
540out:
541 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800542}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
Shaohua Liec8acf22013-02-22 16:34:38 -0800544/*
545 * This swap type frees swap entry, check if it is the highest priority swap
546 * type which just frees swap entry. get_swap_page() uses
547 * highest_priority_index to search highest priority swap type. The
548 * swap_info_struct.lock can't protect us if there are multiple swap types
549 * active, so we use atomic_cmpxchg.
550 */
551static void set_highest_priority_index(int type)
552{
553 int old_hp_index, new_hp_index;
554
555 do {
556 old_hp_index = atomic_read(&highest_priority_index);
557 if (old_hp_index != -1 &&
558 swap_info[old_hp_index]->prio >= swap_info[type]->prio)
559 break;
560 new_hp_index = type;
561 } while (atomic_cmpxchg(&highest_priority_index,
562 old_hp_index, new_hp_index) != old_hp_index);
563}
564
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800565static unsigned char swap_entry_free(struct swap_info_struct *p,
566 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
Hugh Dickins253d5532009-12-14 17:58:44 -0800568 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800569 unsigned char count;
570 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700572 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800573 has_cache = count & SWAP_HAS_CACHE;
574 count &= ~SWAP_HAS_CACHE;
575
576 if (usage == SWAP_HAS_CACHE) {
577 VM_BUG_ON(!has_cache);
578 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800579 } else if (count == SWAP_MAP_SHMEM) {
580 /*
581 * Or we could insist on shmem.c using a special
582 * swap_shmem_free() and free_shmem_swap_and_cache()...
583 */
584 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800585 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
586 if (count == COUNT_CONTINUED) {
587 if (swap_count_continued(p, offset, count))
588 count = SWAP_MAP_MAX | COUNT_CONTINUED;
589 else
590 count = SWAP_MAP_MAX;
591 } else
592 count--;
593 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800594
595 if (!count)
596 mem_cgroup_uncharge_swap(entry);
597
598 usage = count | has_cache;
599 p->swap_map[offset] = usage;
600
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700601 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800602 if (!usage) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700603 if (offset < p->lowest_bit)
604 p->lowest_bit = offset;
605 if (offset > p->highest_bit)
606 p->highest_bit = offset;
Shaohua Liec8acf22013-02-22 16:34:38 -0800607 set_highest_priority_index(p->type);
608 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700609 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600610 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700611 if (p->flags & SWP_BLKDEV) {
612 struct gendisk *disk = p->bdev->bd_disk;
613 if (disk->fops->swap_slot_free_notify)
614 disk->fops->swap_slot_free_notify(p->bdev,
615 offset);
616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800618
619 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
622/*
623 * Caller has made sure that the swapdevice corresponding to entry
624 * is still around or has not been recycled.
625 */
626void swap_free(swp_entry_t entry)
627{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800628 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 p = swap_info_get(entry);
631 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800632 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800633 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635}
636
637/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700638 * Called after dropping swapcache to decrease refcnt to swap entries.
639 */
640void swapcache_free(swp_entry_t entry, struct page *page)
641{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700642 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800643 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700644
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700645 p = swap_info_get(entry);
646 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800647 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
648 if (page)
649 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
Shaohua Liec8acf22013-02-22 16:34:38 -0800650 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700651 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700652}
653
654/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700655 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800656 * This does not give an exact answer when swap count is continued,
657 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700659int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700661 int count = 0;
662 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 swp_entry_t entry;
664
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700665 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 p = swap_info_get(entry);
667 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700668 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800669 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700671 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
674/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800675 * We can write to an anon page without COW if there are no other references
676 * to it. And as a side-effect, free up its swap: because the old content
677 * on disk will never be read, and seeking back there to write new content
678 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800680int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700682 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Hugh Dickins51726b12009-01-06 14:39:25 -0800684 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800685 if (unlikely(PageKsm(page)))
686 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700687 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800688 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700689 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800690 if (count == 1 && !PageWriteback(page)) {
691 delete_from_swap_cache(page);
692 SetPageDirty(page);
693 }
694 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800695 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
698/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800699 * If swap is getting full, or if there are no more mappings of this page,
700 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800702int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
Hugh Dickins51726b12009-01-06 14:39:25 -0800704 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
706 if (!PageSwapCache(page))
707 return 0;
708 if (PageWriteback(page))
709 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800710 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return 0;
712
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700713 /*
714 * Once hibernation has begun to create its image of memory,
715 * there's a danger that one of the calls to try_to_free_swap()
716 * - most probably a call from __try_to_reclaim_swap() while
717 * hibernation is allocating its own swap pages for the image,
718 * but conceivably even a call from memory reclaim - will free
719 * the swap from a page which has already been recorded in the
720 * image as a clean swapcache page, and then reuse its swap for
721 * another page of the image. On waking from hibernation, the
722 * original page might be freed under memory pressure, then
723 * later read back in from swap, now with the wrong data.
724 *
Mel Gormanf90ac392012-01-10 15:07:15 -0800725 * Hibration suspends storage while it is writing the image
726 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700727 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800728 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700729 return 0;
730
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800731 delete_from_swap_cache(page);
732 SetPageDirty(page);
733 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700734}
735
736/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 * Free the swap entry like above, but also try to
738 * free the page cache entry if it is the last user.
739 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800740int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800742 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 struct page *page = NULL;
744
Andi Kleena7420aa2009-09-16 11:50:05 +0200745 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800746 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 p = swap_info_get(entry);
749 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800750 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -0800751 page = find_get_page(swap_address_space(entry),
752 entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700753 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800754 page_cache_release(page);
755 page = NULL;
756 }
757 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800758 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
760 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800761 /*
762 * Not mapped elsewhere, or swap space full? Free it!
763 * Also recheck PageSwapCache now page is locked (above).
764 */
Nick Piggin93fac702006-03-31 02:29:56 -0800765 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800766 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 delete_from_swap_cache(page);
768 SetPageDirty(page);
769 }
770 unlock_page(page);
771 page_cache_release(page);
772 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800773 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774}
775
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200776#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800777/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800778 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800779 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800780 * @offset - number of the PAGE_SIZE-sized block of the device, starting
781 * from 0, in which the swap header is expected to be located.
782 *
783 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800784 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800785int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800786{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800787 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800788 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800789
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800790 if (device)
791 bdev = bdget(device);
792
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800793 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800794 for (type = 0; type < nr_swapfiles; type++) {
795 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800796
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800797 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800798 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700799
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800800 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800801 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200802 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800803
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800804 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800805 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800806 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800807 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800808 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800809
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800810 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800811 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200812 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800813
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800814 spin_unlock(&swap_lock);
815 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800816 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800817 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800818 }
819 }
820 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800821 if (bdev)
822 bdput(bdev);
823
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800824 return -ENODEV;
825}
826
827/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800828 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
829 * corresponding to given index in swap_info (swap type).
830 */
831sector_t swapdev_block(int type, pgoff_t offset)
832{
833 struct block_device *bdev;
834
835 if ((unsigned int)type >= nr_swapfiles)
836 return 0;
837 if (!(swap_info[type]->flags & SWP_WRITEOK))
838 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800839 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800840}
841
842/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800843 * Return either the total number of swap pages of given type, or the number
844 * of free pages of that type (depending on @free)
845 *
846 * This is needed for software suspend
847 */
848unsigned int count_swap_pages(int type, int free)
849{
850 unsigned int n = 0;
851
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800852 spin_lock(&swap_lock);
853 if ((unsigned int)type < nr_swapfiles) {
854 struct swap_info_struct *sis = swap_info[type];
855
Shaohua Liec8acf22013-02-22 16:34:38 -0800856 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800857 if (sis->flags & SWP_WRITEOK) {
858 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800859 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800860 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800861 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800862 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800863 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800864 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800865 return n;
866}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800867#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700870 * No need to decide whether this PTE shares the swap entry with others,
871 * just let do_wp_page work it out if a write is requested later - to
872 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800874static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 unsigned long addr, swp_entry_t entry, struct page *page)
876{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -0800877 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -0800878 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800879 spinlock_t *ptl;
880 pte_t *pte;
881 int ret = 1;
882
Hugh Dickins9e16b7f2013-02-22 16:36:09 -0800883 swapcache = page;
884 page = ksm_might_need_to_copy(page, vma, addr);
885 if (unlikely(!page))
886 return -ENOMEM;
887
Johannes Weiner72835c82012-01-12 17:18:32 -0800888 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
889 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800890 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800891 goto out_nolock;
892 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800893
894 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
895 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
Johannes Weiner5d84c772012-07-31 16:45:28 -0700896 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800897 ret = 0;
898 goto out;
899 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800900
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800901 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800902 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 get_page(page);
904 set_pte_at(vma->vm_mm, addr, pte,
905 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Hugh Dickins9e16b7f2013-02-22 16:36:09 -0800906 if (page == swapcache)
907 page_add_anon_rmap(page, vma, addr);
908 else /* ksm created a completely new copy */
909 page_add_new_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -0800910 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 swap_free(entry);
912 /*
913 * Move the page to the active list so it is not
914 * immediately swapped out again after swapon.
915 */
916 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800917out:
918 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800919out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -0800920 if (page != swapcache) {
921 unlock_page(page);
922 put_page(page);
923 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800924 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925}
926
927static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
928 unsigned long addr, unsigned long end,
929 swp_entry_t entry, struct page *page)
930{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700932 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800933 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Hugh Dickins044d66c2008-02-07 00:14:04 -0800935 /*
936 * We don't actually need pte lock while scanning for swp_pte: since
937 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
938 * page table while we're scanning; though it could get zapped, and on
939 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
940 * of unmatched parts which look like swp_pte, so unuse_pte must
941 * recheck under pte lock. Scanning without pte lock lets it be
942 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
943 */
944 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 do {
946 /*
947 * swapoff spends a _lot_ of time in this loop!
948 * Test inline before going to call unuse_pte.
949 */
950 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800951 pte_unmap(pte);
952 ret = unuse_pte(vma, pmd, addr, entry, page);
953 if (ret)
954 goto out;
955 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 }
957 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800958 pte_unmap(pte - 1);
959out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800960 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961}
962
963static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
964 unsigned long addr, unsigned long end,
965 swp_entry_t entry, struct page *page)
966{
967 pmd_t *pmd;
968 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800969 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 pmd = pmd_offset(pud, addr);
972 do {
973 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700974 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800976 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
977 if (ret)
978 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 } while (pmd++, addr = next, addr != end);
980 return 0;
981}
982
983static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
984 unsigned long addr, unsigned long end,
985 swp_entry_t entry, struct page *page)
986{
987 pud_t *pud;
988 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800989 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
991 pud = pud_offset(pgd, addr);
992 do {
993 next = pud_addr_end(addr, end);
994 if (pud_none_or_clear_bad(pud))
995 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800996 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
997 if (ret)
998 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 } while (pud++, addr = next, addr != end);
1000 return 0;
1001}
1002
1003static int unuse_vma(struct vm_area_struct *vma,
1004 swp_entry_t entry, struct page *page)
1005{
1006 pgd_t *pgd;
1007 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001008 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001010 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 addr = page_address_in_vma(page, vma);
1012 if (addr == -EFAULT)
1013 return 0;
1014 else
1015 end = addr + PAGE_SIZE;
1016 } else {
1017 addr = vma->vm_start;
1018 end = vma->vm_end;
1019 }
1020
1021 pgd = pgd_offset(vma->vm_mm, addr);
1022 do {
1023 next = pgd_addr_end(addr, end);
1024 if (pgd_none_or_clear_bad(pgd))
1025 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001026 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1027 if (ret)
1028 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 } while (pgd++, addr = next, addr != end);
1030 return 0;
1031}
1032
1033static int unuse_mm(struct mm_struct *mm,
1034 swp_entry_t entry, struct page *page)
1035{
1036 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001037 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039 if (!down_read_trylock(&mm->mmap_sem)) {
1040 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001041 * Activate page so shrink_inactive_list is unlikely to unmap
1042 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001044 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 unlock_page(page);
1046 down_read(&mm->mmap_sem);
1047 lock_page(page);
1048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001050 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 break;
1052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001054 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055}
1056
1057/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001058 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1059 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 * Recycle to start on reaching the end, returning 0 when empty.
1061 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001062static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001063 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001065 unsigned int max = si->max;
1066 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001067 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001070 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 * for whether an entry is in use, not modifying it; false
1072 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001073 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 */
1075 for (;;) {
1076 if (++i >= max) {
1077 if (!prev) {
1078 i = 0;
1079 break;
1080 }
1081 /*
1082 * No entries in use at top of swap_map,
1083 * loop back to start and recheck there.
1084 */
1085 max = prev + 1;
1086 prev = 0;
1087 i = 1;
1088 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001089 if (frontswap) {
1090 if (frontswap_test(si, i))
1091 break;
1092 else
1093 continue;
1094 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001096 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 break;
1098 }
1099 return i;
1100}
1101
1102/*
1103 * We completely avoid races by reading each swap page in advance,
1104 * and then search for the process using it. All the necessary
1105 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001106 *
1107 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1108 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001110int try_to_unuse(unsigned int type, bool frontswap,
1111 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001113 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001115 unsigned char *swap_map;
1116 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 struct page *page;
1118 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001119 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
1122 /*
1123 * When searching mms for an entry, a good strategy is to
1124 * start at the first mm we freed the previous entry from
1125 * (though actually we don't notice whether we or coincidence
1126 * freed the entry). Initialize this start_mm with a hold.
1127 *
1128 * A simpler strategy would be to start at the last mm we
1129 * freed the previous entry from; but that would take less
1130 * advantage of mmlist ordering, which clusters forked mms
1131 * together, child after parent. If we race with dup_mmap(), we
1132 * prefer to resolve parent before child, lest we miss entries
1133 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001134 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 */
1136 start_mm = &init_mm;
1137 atomic_inc(&init_mm.mm_users);
1138
1139 /*
1140 * Keep on scanning until all entries have gone. Usually,
1141 * one pass through swap_map is enough, but not necessarily:
1142 * there are races when an instance of an entry might be missed.
1143 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001144 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 if (signal_pending(current)) {
1146 retval = -EINTR;
1147 break;
1148 }
1149
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001150 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 * Get a page for the entry, using the existing swap
1152 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001153 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 */
1155 swap_map = &si->swap_map[i];
1156 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001157 page = read_swap_cache_async(entry,
1158 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 if (!page) {
1160 /*
1161 * Either swap_duplicate() failed because entry
1162 * has been freed independently, and will not be
1163 * reused since sys_swapoff() already disabled
1164 * allocation from here, or alloc_page() failed.
1165 */
1166 if (!*swap_map)
1167 continue;
1168 retval = -ENOMEM;
1169 break;
1170 }
1171
1172 /*
1173 * Don't hold on to start_mm if it looks like exiting.
1174 */
1175 if (atomic_read(&start_mm->mm_users) == 1) {
1176 mmput(start_mm);
1177 start_mm = &init_mm;
1178 atomic_inc(&init_mm.mm_users);
1179 }
1180
1181 /*
1182 * Wait for and lock page. When do_swap_page races with
1183 * try_to_unuse, do_swap_page can handle the fault much
1184 * faster than try_to_unuse can locate the entry. This
1185 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1186 * defer to do_swap_page in such a case - in some tests,
1187 * do_swap_page and try_to_unuse repeatedly compete.
1188 */
1189 wait_on_page_locked(page);
1190 wait_on_page_writeback(page);
1191 lock_page(page);
1192 wait_on_page_writeback(page);
1193
1194 /*
1195 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001198 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1199 retval = shmem_unuse(entry, page);
1200 /* page has already been unlocked and released */
1201 if (retval < 0)
1202 break;
1203 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001205 if (swap_count(swcount) && start_mm != &init_mm)
1206 retval = unuse_mm(start_mm, entry, page);
1207
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001208 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 int set_start_mm = (*swap_map >= swcount);
1210 struct list_head *p = &start_mm->mmlist;
1211 struct mm_struct *new_start_mm = start_mm;
1212 struct mm_struct *prev_mm = start_mm;
1213 struct mm_struct *mm;
1214
1215 atomic_inc(&new_start_mm->mm_users);
1216 atomic_inc(&prev_mm->mm_users);
1217 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001218 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 (p = p->next) != &start_mm->mmlist) {
1220 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001221 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 spin_unlock(&mmlist_lock);
1224 mmput(prev_mm);
1225 prev_mm = mm;
1226
1227 cond_resched();
1228
1229 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001230 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001232 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001234 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001236
Bo Liu32c5fc12009-11-02 16:50:33 +00001237 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 mmput(new_start_mm);
1239 atomic_inc(&mm->mm_users);
1240 new_start_mm = mm;
1241 set_start_mm = 0;
1242 }
1243 spin_lock(&mmlist_lock);
1244 }
1245 spin_unlock(&mmlist_lock);
1246 mmput(prev_mm);
1247 mmput(start_mm);
1248 start_mm = new_start_mm;
1249 }
1250 if (retval) {
1251 unlock_page(page);
1252 page_cache_release(page);
1253 break;
1254 }
1255
1256 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 * If a reference remains (rare), we would like to leave
1258 * the page in the swap cache; but try_to_unmap could
1259 * then re-duplicate the entry once we drop page lock,
1260 * so we might loop indefinitely; also, that page could
1261 * not be swapped out to other storage meanwhile. So:
1262 * delete from cache even if there's another reference,
1263 * after ensuring that the data has been saved to disk -
1264 * since if the reference remains (rarer), it will be
1265 * read from disk into another page. Splitting into two
1266 * pages would be incorrect if swap supported "shared
1267 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001268 *
1269 * Given how unuse_vma() targets one particular offset
1270 * in an anon_vma, once the anon_vma has been determined,
1271 * this splitting happens to be just what is needed to
1272 * handle where KSM pages have been swapped out: re-reading
1273 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001275 if (swap_count(*swap_map) &&
1276 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 struct writeback_control wbc = {
1278 .sync_mode = WB_SYNC_NONE,
1279 };
1280
1281 swap_writepage(page, &wbc);
1282 lock_page(page);
1283 wait_on_page_writeback(page);
1284 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001285
1286 /*
1287 * It is conceivable that a racing task removed this page from
1288 * swap cache just before we acquired the page lock at the top,
1289 * or while we dropped it in unuse_mm(). The page might even
1290 * be back in swap cache on another swap area: that we must not
1291 * delete, since it may not have been written out to swap yet.
1292 */
1293 if (PageSwapCache(page) &&
1294 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001295 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297 /*
1298 * So we could skip searching mms once swap count went
1299 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001300 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 */
1302 SetPageDirty(page);
1303 unlock_page(page);
1304 page_cache_release(page);
1305
1306 /*
1307 * Make sure that we aren't completely killing
1308 * interactive performance.
1309 */
1310 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001311 if (frontswap && pages_to_unuse > 0) {
1312 if (!--pages_to_unuse)
1313 break;
1314 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 }
1316
1317 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 return retval;
1319}
1320
1321/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001322 * After a successful try_to_unuse, if no swap is now in use, we know
1323 * we can empty the mmlist. swap_lock must be held on entry and exit.
1324 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 * added to the mmlist just after page_duplicate - before would be racy.
1326 */
1327static void drain_mmlist(void)
1328{
1329 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001330 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001332 for (type = 0; type < nr_swapfiles; type++)
1333 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 return;
1335 spin_lock(&mmlist_lock);
1336 list_for_each_safe(p, next, &init_mm.mmlist)
1337 list_del_init(p);
1338 spin_unlock(&mmlist_lock);
1339}
1340
1341/*
1342 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001343 * corresponds to page offset for the specified swap entry.
1344 * Note that the type of this function is sector_t, but it returns page offset
1345 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001347static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001349 struct swap_info_struct *sis;
1350 struct swap_extent *start_se;
1351 struct swap_extent *se;
1352 pgoff_t offset;
1353
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001354 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001355 *bdev = sis->bdev;
1356
1357 offset = swp_offset(entry);
1358 start_se = sis->curr_swap_extent;
1359 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
1361 for ( ; ; ) {
1362 struct list_head *lh;
1363
1364 if (se->start_page <= offset &&
1365 offset < (se->start_page + se->nr_pages)) {
1366 return se->start_block + (offset - se->start_page);
1367 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001368 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 se = list_entry(lh, struct swap_extent, list);
1370 sis->curr_swap_extent = se;
1371 BUG_ON(se == start_se); /* It *must* be present */
1372 }
1373}
1374
1375/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001376 * Returns the page offset into bdev for the specified page's swap entry.
1377 */
1378sector_t map_swap_page(struct page *page, struct block_device **bdev)
1379{
1380 swp_entry_t entry;
1381 entry.val = page_private(page);
1382 return map_swap_entry(entry, bdev);
1383}
1384
1385/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 * Free all of a swapdev's extent information
1387 */
1388static void destroy_swap_extents(struct swap_info_struct *sis)
1389{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001390 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 struct swap_extent *se;
1392
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001393 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 struct swap_extent, list);
1395 list_del(&se->list);
1396 kfree(se);
1397 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001398
1399 if (sis->flags & SWP_FILE) {
1400 struct file *swap_file = sis->swap_file;
1401 struct address_space *mapping = swap_file->f_mapping;
1402
1403 sis->flags &= ~SWP_FILE;
1404 mapping->a_ops->swap_deactivate(swap_file);
1405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406}
1407
1408/*
1409 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001410 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001412 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 */
Mel Gormana509bc12012-07-31 16:44:57 -07001414int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1416 unsigned long nr_pages, sector_t start_block)
1417{
1418 struct swap_extent *se;
1419 struct swap_extent *new_se;
1420 struct list_head *lh;
1421
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001422 if (start_page == 0) {
1423 se = &sis->first_swap_extent;
1424 sis->curr_swap_extent = se;
1425 se->start_page = 0;
1426 se->nr_pages = nr_pages;
1427 se->start_block = start_block;
1428 return 1;
1429 } else {
1430 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001432 BUG_ON(se->start_page + se->nr_pages != start_page);
1433 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 /* Merge it */
1435 se->nr_pages += nr_pages;
1436 return 0;
1437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 }
1439
1440 /*
1441 * No merge. Insert a new extent, preserving ordering.
1442 */
1443 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1444 if (new_se == NULL)
1445 return -ENOMEM;
1446 new_se->start_page = start_page;
1447 new_se->nr_pages = nr_pages;
1448 new_se->start_block = start_block;
1449
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001450 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001451 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
1454/*
1455 * A `swap extent' is a simple thing which maps a contiguous range of pages
1456 * onto a contiguous range of disk blocks. An ordered list of swap extents
1457 * is built at swapon time and is then used at swap_writepage/swap_readpage
1458 * time for locating where on disk a page belongs.
1459 *
1460 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1461 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1462 * swap files identically.
1463 *
1464 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1465 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1466 * swapfiles are handled *identically* after swapon time.
1467 *
1468 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1469 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1470 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1471 * requirements, they are simply tossed out - we will never use those blocks
1472 * for swapping.
1473 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001474 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1476 * which will scribble on the fs.
1477 *
1478 * The amount of disk space which a single swap extent represents varies.
1479 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1480 * extents in the list. To avoid much list walking, we cache the previous
1481 * search location in `curr_swap_extent', and start new searches from there.
1482 * This is extremely effective. The average number of iterations in
1483 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1484 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001485static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486{
Mel Gorman62c230b2012-07-31 16:44:55 -07001487 struct file *swap_file = sis->swap_file;
1488 struct address_space *mapping = swap_file->f_mapping;
1489 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 int ret;
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 if (S_ISBLK(inode->i_mode)) {
1493 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001494 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001495 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 }
1497
Mel Gorman62c230b2012-07-31 16:44:55 -07001498 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001499 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001500 if (!ret) {
1501 sis->flags |= SWP_FILE;
1502 ret = add_swap_extent(sis, 0, sis->max, 0);
1503 *span = sis->pages;
1504 }
Mel Gormana509bc12012-07-31 16:44:57 -07001505 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001506 }
1507
Mel Gormana509bc12012-07-31 16:44:57 -07001508 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509}
1510
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001511static void _enable_swap_info(struct swap_info_struct *p, int prio,
Minchan Kim4f898492013-04-30 15:26:54 -07001512 unsigned char *swap_map)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001513{
1514 int i, prev;
1515
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001516 if (prio >= 0)
1517 p->prio = prio;
1518 else
1519 p->prio = --least_priority;
1520 p->swap_map = swap_map;
1521 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001522 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001523 total_swap_pages += p->pages;
1524
1525 /* insert swap space into swap_list: */
1526 prev = -1;
1527 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1528 if (p->prio >= swap_info[i]->prio)
1529 break;
1530 prev = i;
1531 }
1532 p->next = i;
1533 if (prev < 0)
1534 swap_list.head = swap_list.next = p->type;
1535 else
1536 swap_info[prev]->next = p->type;
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001537}
1538
1539static void enable_swap_info(struct swap_info_struct *p, int prio,
1540 unsigned char *swap_map,
1541 unsigned long *frontswap_map)
1542{
Minchan Kim4f898492013-04-30 15:26:54 -07001543 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001544 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001545 spin_lock(&p->lock);
Minchan Kim4f898492013-04-30 15:26:54 -07001546 _enable_swap_info(p, prio, swap_map);
Shaohua Liec8acf22013-02-22 16:34:38 -08001547 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001548 spin_unlock(&swap_lock);
1549}
1550
1551static void reinsert_swap_info(struct swap_info_struct *p)
1552{
1553 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001554 spin_lock(&p->lock);
Minchan Kim4f898492013-04-30 15:26:54 -07001555 _enable_swap_info(p, p->prio, p->swap_map);
Shaohua Liec8acf22013-02-22 16:34:38 -08001556 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001557 spin_unlock(&swap_lock);
1558}
1559
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001560SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001562 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001563 unsigned char *swap_map;
Minchan Kim4f898492013-04-30 15:26:54 -07001564 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 struct file *swap_file, *victim;
1566 struct address_space *mapping;
1567 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001568 struct filename *pathname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 int i, type, prev;
1570 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (!capable(CAP_SYS_ADMIN))
1573 return -EPERM;
1574
Al Viro191c5422012-02-13 03:58:52 +00001575 BUG_ON(!current->mm);
1576
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001579 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
Jeff Layton669abf42012-10-10 16:43:10 -04001581 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 err = PTR_ERR(victim);
1583 if (IS_ERR(victim))
1584 goto out;
1585
1586 mapping = victim->f_mapping;
1587 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001588 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001589 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1590 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001591 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 if (p->swap_file->f_mapping == mapping)
1593 break;
1594 }
1595 prev = type;
1596 }
1597 if (type < 0) {
1598 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001599 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 goto out_dput;
1601 }
Al Viro191c5422012-02-13 03:58:52 +00001602 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 vm_unacct_memory(p->pages);
1604 else {
1605 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001606 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 goto out_dput;
1608 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001609 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001611 else
1612 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 if (type == swap_list.next) {
1614 /* just pick something that's safe... */
1615 swap_list.next = swap_list.head;
1616 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001617 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001618 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001619 for (i = p->next; i >= 0; i = swap_info[i]->next)
1620 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001621 least_priority++;
1622 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001623 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 total_swap_pages -= p->pages;
1625 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001626 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001627 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001628
David Rientjese1e12d22012-12-11 16:02:56 -08001629 set_current_oom_origin();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001630 err = try_to_unuse(type, false, 0); /* force all pages to be unused */
David Rientjese1e12d22012-12-11 16:02:56 -08001631 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 if (err) {
1634 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001635 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 goto out_dput;
1637 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001638
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001639 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001640 if (p->flags & SWP_CONTINUED)
1641 free_swap_count_continuations(p);
1642
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001643 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001644 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001645 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001647
1648 /* wait for anyone still in scan_swap_map */
1649 p->highest_bit = 0; /* cuts scans short */
1650 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001651 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001652 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001653 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001654 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001655 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001656 }
1657
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 swap_file = p->swap_file;
1659 p->swap_file = NULL;
1660 p->max = 0;
1661 swap_map = p->swap_map;
1662 p->swap_map = NULL;
1663 p->flags = 0;
Minchan Kim4f898492013-04-30 15:26:54 -07001664 frontswap_map = frontswap_map_get(p);
1665 frontswap_map_set(p, NULL);
Shaohua Liec8acf22013-02-22 16:34:38 -08001666 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001667 spin_unlock(&swap_lock);
Minchan Kim4f898492013-04-30 15:26:54 -07001668 frontswap_invalidate_area(type);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001669 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 vfree(swap_map);
Minchan Kim4f898492013-04-30 15:26:54 -07001671 vfree(frontswap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001672 /* Destroy swap account informatin */
1673 swap_cgroup_swapoff(type);
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 inode = mapping->host;
1676 if (S_ISBLK(inode->i_mode)) {
1677 struct block_device *bdev = I_BDEV(inode);
1678 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001679 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001681 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001683 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
1685 filp_close(swap_file, NULL);
1686 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001687 atomic_inc(&proc_poll_event);
1688 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
1690out_dput:
1691 filp_close(victim, NULL);
1692out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001693 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 return err;
1695}
1696
1697#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001698static unsigned swaps_poll(struct file *file, poll_table *wait)
1699{
Kay Sieversf1514632011-07-12 20:48:39 +02001700 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001701
1702 poll_wait(file, &proc_poll_wait, wait);
1703
Kay Sieversf1514632011-07-12 20:48:39 +02001704 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1705 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001706 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1707 }
1708
1709 return POLLIN | POLLRDNORM;
1710}
1711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712/* iterator */
1713static void *swap_start(struct seq_file *swap, loff_t *pos)
1714{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001715 struct swap_info_struct *si;
1716 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 loff_t l = *pos;
1718
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001719 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001721 if (!l)
1722 return SEQ_START_TOKEN;
1723
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001724 for (type = 0; type < nr_swapfiles; type++) {
1725 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1726 si = swap_info[type];
1727 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001729 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001730 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 }
1732
1733 return NULL;
1734}
1735
1736static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1737{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001738 struct swap_info_struct *si = v;
1739 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001741 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001742 type = 0;
1743 else
1744 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001745
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001746 for (; type < nr_swapfiles; type++) {
1747 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1748 si = swap_info[type];
1749 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 continue;
1751 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001752 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 }
1754
1755 return NULL;
1756}
1757
1758static void swap_stop(struct seq_file *swap, void *v)
1759{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001760 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761}
1762
1763static int swap_show(struct seq_file *swap, void *v)
1764{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001765 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 struct file *file;
1767 int len;
1768
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001769 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001770 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1771 return 0;
1772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001774 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001775 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001776 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001777 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05001778 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001780 si->pages << (PAGE_SHIFT - 10),
1781 si->inuse_pages << (PAGE_SHIFT - 10),
1782 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 return 0;
1784}
1785
Helge Deller15ad7cd2006-12-06 20:40:36 -08001786static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 .start = swap_start,
1788 .next = swap_next,
1789 .stop = swap_stop,
1790 .show = swap_show
1791};
1792
1793static int swaps_open(struct inode *inode, struct file *file)
1794{
Kay Sieversf1514632011-07-12 20:48:39 +02001795 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001796 int ret;
1797
Kay Sievers66d7dd52010-10-26 14:22:06 -07001798 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02001799 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07001800 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001801
Kay Sieversf1514632011-07-12 20:48:39 +02001802 seq = file->private_data;
1803 seq->poll_event = atomic_read(&proc_poll_event);
1804 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805}
1806
Helge Deller15ad7cd2006-12-06 20:40:36 -08001807static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 .open = swaps_open,
1809 .read = seq_read,
1810 .llseek = seq_lseek,
1811 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001812 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813};
1814
1815static int __init procswaps_init(void)
1816{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001817 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 return 0;
1819}
1820__initcall(procswaps_init);
1821#endif /* CONFIG_PROC_FS */
1822
Jan Beulich17963162008-12-16 11:35:24 +00001823#ifdef MAX_SWAPFILES_CHECK
1824static int __init max_swapfiles_check(void)
1825{
1826 MAX_SWAPFILES_CHECK();
1827 return 0;
1828}
1829late_initcall(max_swapfiles_check);
1830#endif
1831
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001832static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001834 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001836
1837 p = kzalloc(sizeof(*p), GFP_KERNEL);
1838 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001839 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001840
Hugh Dickins5d337b92005-09-03 15:54:41 -07001841 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001842 for (type = 0; type < nr_swapfiles; type++) {
1843 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001845 }
Christoph Lameter06972122006-06-23 02:03:35 -07001846 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001847 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001848 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07001849 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001851 if (type >= nr_swapfiles) {
1852 p->type = type;
1853 swap_info[type] = p;
1854 /*
1855 * Write swap_info[type] before nr_swapfiles, in case a
1856 * racing procfs swap_start() or swap_next() is reading them.
1857 * (We never shrink nr_swapfiles, we never free this entry.)
1858 */
1859 smp_wmb();
1860 nr_swapfiles++;
1861 } else {
1862 kfree(p);
1863 p = swap_info[type];
1864 /*
1865 * Do not memset this entry: a racing procfs swap_next()
1866 * would be relying on p->type to remain valid.
1867 */
1868 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001869 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001872 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001873 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001874
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001875 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001876}
1877
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001878static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1879{
1880 int error;
1881
1882 if (S_ISBLK(inode->i_mode)) {
1883 p->bdev = bdgrab(I_BDEV(inode));
1884 error = blkdev_get(p->bdev,
1885 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1886 sys_swapon);
1887 if (error < 0) {
1888 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001889 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001890 }
1891 p->old_block_size = block_size(p->bdev);
1892 error = set_blocksize(p->bdev, PAGE_SIZE);
1893 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001894 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001895 p->flags |= SWP_BLKDEV;
1896 } else if (S_ISREG(inode->i_mode)) {
1897 p->bdev = inode->i_sb->s_bdev;
1898 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001899 if (IS_SWAPFILE(inode))
1900 return -EBUSY;
1901 } else
1902 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001903
1904 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001905}
1906
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001907static unsigned long read_swap_header(struct swap_info_struct *p,
1908 union swap_header *swap_header,
1909 struct inode *inode)
1910{
1911 int i;
1912 unsigned long maxpages;
1913 unsigned long swapfilepages;
1914
1915 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1916 printk(KERN_ERR "Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001917 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001918 }
1919
1920 /* swap partition endianess hack... */
1921 if (swab32(swap_header->info.version) == 1) {
1922 swab32s(&swap_header->info.version);
1923 swab32s(&swap_header->info.last_page);
1924 swab32s(&swap_header->info.nr_badpages);
1925 for (i = 0; i < swap_header->info.nr_badpages; i++)
1926 swab32s(&swap_header->info.badpages[i]);
1927 }
1928 /* Check the swap header's sub-version */
1929 if (swap_header->info.version != 1) {
1930 printk(KERN_WARNING
1931 "Unable to handle swap header version %d\n",
1932 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001933 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001934 }
1935
1936 p->lowest_bit = 1;
1937 p->cluster_next = 1;
1938 p->cluster_nr = 0;
1939
1940 /*
1941 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07001942 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001943 * of bits for the swap offset in the swp_entry_t type, and
1944 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07001945 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001946 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001947 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001948 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001949 * offset is extracted. This will mask all the bits from
1950 * the initial ~0UL mask that can't be encoded in either
1951 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07001952 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001953 */
1954 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07001955 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001956 if (maxpages > swap_header->info.last_page) {
1957 maxpages = swap_header->info.last_page + 1;
1958 /* p->max is an unsigned int: don't overflow it */
1959 if ((unsigned int)maxpages == 0)
1960 maxpages = UINT_MAX;
1961 }
1962 p->highest_bit = maxpages - 1;
1963
1964 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001965 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001966 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1967 if (swapfilepages && maxpages > swapfilepages) {
1968 printk(KERN_WARNING
1969 "Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001970 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001971 }
1972 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001973 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001974 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001975 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001976
1977 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001978}
1979
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001980static int setup_swap_map_and_extents(struct swap_info_struct *p,
1981 union swap_header *swap_header,
1982 unsigned char *swap_map,
1983 unsigned long maxpages,
1984 sector_t *span)
1985{
1986 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001987 unsigned int nr_good_pages;
1988 int nr_extents;
1989
1990 nr_good_pages = maxpages - 1; /* omit header page */
1991
1992 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1993 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001994 if (page_nr == 0 || page_nr > swap_header->info.last_page)
1995 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001996 if (page_nr < maxpages) {
1997 swap_map[page_nr] = SWAP_MAP_BAD;
1998 nr_good_pages--;
1999 }
2000 }
2001
2002 if (nr_good_pages) {
2003 swap_map[0] = SWAP_MAP_BAD;
2004 p->max = maxpages;
2005 p->pages = nr_good_pages;
2006 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002007 if (nr_extents < 0)
2008 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002009 nr_good_pages = p->pages;
2010 }
2011 if (!nr_good_pages) {
2012 printk(KERN_WARNING "Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002013 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002014 }
2015
2016 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002017}
2018
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002019/*
2020 * Helper to sys_swapon determining if a given swap
2021 * backing device queue supports DISCARD operations.
2022 */
2023static bool swap_discardable(struct swap_info_struct *si)
2024{
2025 struct request_queue *q = bdev_get_queue(si->bdev);
2026
2027 if (!q || !blk_queue_discard(q))
2028 return false;
2029
2030 return true;
2031}
2032
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002033SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2034{
2035 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002036 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002037 struct file *swap_file = NULL;
2038 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002039 int i;
2040 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002041 int error;
2042 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002043 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002044 sector_t span;
2045 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002046 unsigned char *swap_map = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002047 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002048 struct page *page = NULL;
2049 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002050
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002051 if (swap_flags & ~SWAP_FLAGS_VALID)
2052 return -EINVAL;
2053
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002054 if (!capable(CAP_SYS_ADMIN))
2055 return -EPERM;
2056
2057 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002058 if (IS_ERR(p))
2059 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002063 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002065 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 }
Jeff Layton669abf42012-10-10 16:43:10 -04002067 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002069 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002071 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073
2074 p->swap_file = swap_file;
2075 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002078 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002080 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002082 if (mapping == q->swap_file->f_mapping) {
2083 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002085 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 }
2087
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002088 inode = mapping->host;
2089 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002090 error = claim_swapfile(p, inode);
2091 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 /*
2095 * Read the swap header.
2096 */
2097 if (!mapping->a_ops->readpage) {
2098 error = -EINVAL;
2099 goto bad_swap;
2100 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002101 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 if (IS_ERR(page)) {
2103 error = PTR_ERR(page);
2104 goto bad_swap;
2105 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002106 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002108 maxpages = read_swap_header(p, swap_header, inode);
2109 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 error = -EINVAL;
2111 goto bad_swap;
2112 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002113
Hugh Dickins81e33972009-01-06 14:39:49 -08002114 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002115 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002116 if (!swap_map) {
2117 error = -ENOMEM;
2118 goto bad_swap;
2119 }
2120
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002121 error = swap_cgroup_swapon(p->type, maxpages);
2122 if (error)
2123 goto bad_swap;
2124
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002125 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2126 maxpages, &span);
2127 if (unlikely(nr_extents < 0)) {
2128 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 goto bad_swap;
2130 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002131 /* frontswap enabled? set up bit-per-page map for frontswap */
2132 if (frontswap_enabled)
Akinobu Mita7b579762013-06-12 14:05:08 -07002133 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002135 if (p->bdev) {
2136 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2137 p->flags |= SWP_SOLIDSTATE;
Akinobu Mitad3d30412013-04-29 16:21:29 -07002138 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002139 }
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002140
2141 if ((swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2142 /*
2143 * When discard is enabled for swap with no particular
2144 * policy flagged, we set all swap discard flags here in
2145 * order to sustain backward compatibility with older
2146 * swapon(8) releases.
2147 */
2148 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2149 SWP_PAGE_DISCARD);
2150
2151 /*
2152 * By flagging sys_swapon, a sysadmin can tell us to
2153 * either do single-time area discards only, or to just
2154 * perform discards for released swap page-clusters.
2155 * Now it's time to adjust the p->flags accordingly.
2156 */
2157 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2158 p->flags &= ~SWP_PAGE_DISCARD;
2159 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2160 p->flags &= ~SWP_AREA_DISCARD;
2161
2162 /* issue a swapon-time discard if it's still required */
2163 if (p->flags & SWP_AREA_DISCARD) {
2164 int err = discard_swap(p);
2165 if (unlikely(err))
2166 printk(KERN_ERR
2167 "swapon: discard_swap(%p): %d\n",
2168 p, err);
2169 }
2170 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002171 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002172
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002173 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002174 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002175 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002176 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002177 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002178 enable_swap_info(p, prio, swap_map, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002179
2180 printk(KERN_INFO "Adding %uk swap on %s. "
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002181 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002182 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002183 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2184 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002185 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002186 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2187 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002188 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002189
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002190 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002191 atomic_inc(&proc_poll_event);
2192 wake_up_interruptible(&proc_poll_wait);
2193
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002194 if (S_ISREG(inode->i_mode))
2195 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 error = 0;
2197 goto out;
2198bad_swap:
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002199 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002200 set_blocksize(p->bdev, p->old_block_size);
2201 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002203 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002204 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002205 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002208 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 vfree(swap_map);
Mel Gorman52c50562011-03-22 16:30:08 -07002210 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002211 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002212 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002213 inode = NULL;
2214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217out:
2218 if (page && !IS_ERR(page)) {
2219 kunmap(page);
2220 page_cache_release(page);
2221 }
2222 if (name)
2223 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002224 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002225 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 return error;
2227}
2228
2229void si_swapinfo(struct sysinfo *val)
2230{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002231 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 unsigned long nr_to_be_unused = 0;
2233
Hugh Dickins5d337b92005-09-03 15:54:41 -07002234 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002235 for (type = 0; type < nr_swapfiles; type++) {
2236 struct swap_info_struct *si = swap_info[type];
2237
2238 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2239 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002241 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002243 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244}
2245
2246/*
2247 * Verify that a swap entry is valid and increment its swap map count.
2248 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002249 * Returns error code in following case.
2250 * - success -> 0
2251 * - swp_entry is invalid -> EINVAL
2252 * - swp_entry is migration entry -> EINVAL
2253 * - swap-cache reference is requested but there is already one. -> EEXIST
2254 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002255 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002257static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002259 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002261 unsigned char count;
2262 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002263 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
Andi Kleena7420aa2009-09-16 11:50:05 +02002265 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002266 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002267
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 type = swp_type(entry);
2269 if (type >= nr_swapfiles)
2270 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002271 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 offset = swp_offset(entry);
2273
Shaohua Liec8acf22013-02-22 16:34:38 -08002274 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002275 if (unlikely(offset >= p->max))
2276 goto unlock_out;
2277
Hugh Dickins253d5532009-12-14 17:58:44 -08002278 count = p->swap_map[offset];
2279 has_cache = count & SWAP_HAS_CACHE;
2280 count &= ~SWAP_HAS_CACHE;
2281 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002282
Hugh Dickins253d5532009-12-14 17:58:44 -08002283 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002284
2285 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002286 if (!has_cache && count)
2287 has_cache = SWAP_HAS_CACHE;
2288 else if (has_cache) /* someone else added cache */
2289 err = -EEXIST;
2290 else /* no users remaining */
2291 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002292
2293 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002294
Hugh Dickins570a335b2009-12-14 17:58:46 -08002295 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2296 count += usage;
2297 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002298 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002299 else if (swap_count_continued(p, offset, count))
2300 count = COUNT_CONTINUED;
2301 else
2302 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002303 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002304 err = -ENOENT; /* unused swap entry */
2305
2306 p->swap_map[offset] = count | has_cache;
2307
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002308unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002309 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002311 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
2313bad_file:
2314 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2315 goto out;
2316}
Hugh Dickins253d5532009-12-14 17:58:44 -08002317
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002318/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002319 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2320 * (in which case its reference count is never incremented).
2321 */
2322void swap_shmem_alloc(swp_entry_t entry)
2323{
2324 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2325}
2326
2327/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002328 * Increase reference count of swap entry by 1.
2329 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2330 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2331 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2332 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002333 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002334int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002335{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002336 int err = 0;
2337
2338 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2339 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2340 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002341}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002343/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002344 * @entry: swap entry for which we allocate swap cache.
2345 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002346 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002347 * This can return error codes. Returns 0 at success.
2348 * -EBUSY means there is a swap cache.
2349 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002350 */
2351int swapcache_prepare(swp_entry_t entry)
2352{
Hugh Dickins253d5532009-12-14 17:58:44 -08002353 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002354}
2355
Mel Gormanf981c592012-07-31 16:44:47 -07002356struct swap_info_struct *page_swap_info(struct page *page)
2357{
2358 swp_entry_t swap = { .val = page_private(page) };
2359 BUG_ON(!PageSwapCache(page));
2360 return swap_info[swp_type(swap)];
2361}
2362
2363/*
2364 * out-of-line __page_file_ methods to avoid include hell.
2365 */
2366struct address_space *__page_file_mapping(struct page *page)
2367{
2368 VM_BUG_ON(!PageSwapCache(page));
2369 return page_swap_info(page)->swap_file->f_mapping;
2370}
2371EXPORT_SYMBOL_GPL(__page_file_mapping);
2372
2373pgoff_t __page_file_index(struct page *page)
2374{
2375 swp_entry_t swap = { .val = page_private(page) };
2376 VM_BUG_ON(!PageSwapCache(page));
2377 return swp_offset(swap);
2378}
2379EXPORT_SYMBOL_GPL(__page_file_index);
2380
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002382 * add_swap_count_continuation - called when a swap count is duplicated
2383 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2384 * page of the original vmalloc'ed swap_map, to hold the continuation count
2385 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2386 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2387 *
2388 * These continuation pages are seldom referenced: the common paths all work
2389 * on the original swap_map, only referring to a continuation page when the
2390 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2391 *
2392 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2393 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2394 * can be called after dropping locks.
2395 */
2396int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2397{
2398 struct swap_info_struct *si;
2399 struct page *head;
2400 struct page *page;
2401 struct page *list_page;
2402 pgoff_t offset;
2403 unsigned char count;
2404
2405 /*
2406 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2407 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2408 */
2409 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2410
2411 si = swap_info_get(entry);
2412 if (!si) {
2413 /*
2414 * An acceptable race has occurred since the failing
2415 * __swap_duplicate(): the swap entry has been freed,
2416 * perhaps even the whole swap_map cleared for swapoff.
2417 */
2418 goto outer;
2419 }
2420
2421 offset = swp_offset(entry);
2422 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2423
2424 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2425 /*
2426 * The higher the swap count, the more likely it is that tasks
2427 * will race to add swap count continuation: we need to avoid
2428 * over-provisioning.
2429 */
2430 goto out;
2431 }
2432
2433 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002434 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002435 return -ENOMEM;
2436 }
2437
2438 /*
2439 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2440 * no architecture is using highmem pages for kernel pagetables: so it
2441 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2442 */
2443 head = vmalloc_to_page(si->swap_map + offset);
2444 offset &= ~PAGE_MASK;
2445
2446 /*
2447 * Page allocation does not initialize the page's lru field,
2448 * but it does always reset its private field.
2449 */
2450 if (!page_private(head)) {
2451 BUG_ON(count & COUNT_CONTINUED);
2452 INIT_LIST_HEAD(&head->lru);
2453 set_page_private(head, SWP_CONTINUED);
2454 si->flags |= SWP_CONTINUED;
2455 }
2456
2457 list_for_each_entry(list_page, &head->lru, lru) {
2458 unsigned char *map;
2459
2460 /*
2461 * If the previous map said no continuation, but we've found
2462 * a continuation page, free our allocation and use this one.
2463 */
2464 if (!(count & COUNT_CONTINUED))
2465 goto out;
2466
Cong Wang9b04c5f2011-11-25 23:14:39 +08002467 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002468 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002469 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002470
2471 /*
2472 * If this continuation count now has some space in it,
2473 * free our allocation and use this one.
2474 */
2475 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2476 goto out;
2477 }
2478
2479 list_add_tail(&page->lru, &head->lru);
2480 page = NULL; /* now it's attached, don't free it */
2481out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002482 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002483outer:
2484 if (page)
2485 __free_page(page);
2486 return 0;
2487}
2488
2489/*
2490 * swap_count_continued - when the original swap_map count is incremented
2491 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2492 * into, carry if so, or else fail until a new continuation page is allocated;
2493 * when the original swap_map count is decremented from 0 with continuation,
2494 * borrow from the continuation and report whether it still holds more.
2495 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2496 */
2497static bool swap_count_continued(struct swap_info_struct *si,
2498 pgoff_t offset, unsigned char count)
2499{
2500 struct page *head;
2501 struct page *page;
2502 unsigned char *map;
2503
2504 head = vmalloc_to_page(si->swap_map + offset);
2505 if (page_private(head) != SWP_CONTINUED) {
2506 BUG_ON(count & COUNT_CONTINUED);
2507 return false; /* need to add count continuation */
2508 }
2509
2510 offset &= ~PAGE_MASK;
2511 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002512 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002513
2514 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2515 goto init_map; /* jump over SWAP_CONT_MAX checks */
2516
2517 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2518 /*
2519 * Think of how you add 1 to 999
2520 */
2521 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002522 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002523 page = list_entry(page->lru.next, struct page, lru);
2524 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002525 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002526 }
2527 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002528 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002529 page = list_entry(page->lru.next, struct page, lru);
2530 if (page == head)
2531 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002532 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002533init_map: *map = 0; /* we didn't zero the page */
2534 }
2535 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002536 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002537 page = list_entry(page->lru.prev, struct page, lru);
2538 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002539 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002540 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002541 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002542 page = list_entry(page->lru.prev, struct page, lru);
2543 }
2544 return true; /* incremented */
2545
2546 } else { /* decrementing */
2547 /*
2548 * Think of how you subtract 1 from 1000
2549 */
2550 BUG_ON(count != COUNT_CONTINUED);
2551 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002552 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002553 page = list_entry(page->lru.next, struct page, lru);
2554 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002555 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002556 }
2557 BUG_ON(*map == 0);
2558 *map -= 1;
2559 if (*map == 0)
2560 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002561 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002562 page = list_entry(page->lru.prev, struct page, lru);
2563 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002564 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002565 *map = SWAP_CONT_MAX | count;
2566 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002567 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002568 page = list_entry(page->lru.prev, struct page, lru);
2569 }
2570 return count == COUNT_CONTINUED;
2571 }
2572}
2573
2574/*
2575 * free_swap_count_continuations - swapoff free all the continuation pages
2576 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2577 */
2578static void free_swap_count_continuations(struct swap_info_struct *si)
2579{
2580 pgoff_t offset;
2581
2582 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2583 struct page *head;
2584 head = vmalloc_to_page(si->swap_map + offset);
2585 if (page_private(head)) {
2586 struct list_head *this, *next;
2587 list_for_each_safe(this, next, &head->lru) {
2588 struct page *page;
2589 page = list_entry(this, struct page, lru);
2590 list_del(this);
2591 __free_page(page);
2592 }
2593 }
2594 }
2595}