blob: 9ae4c8d36a3b712c925e11fda5efacc7cde0998f [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>
Mel Gormanfdd4d472012-07-31 16:44:47 -070034#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36#include <asm/pgtable.h>
37#include <asm/tlbflush.h>
38#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080039#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Hugh Dickins570a3352009-12-14 17:58:46 -080041static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
42 unsigned char);
43static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080044static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a3352009-12-14 17:58:46 -080045
Adrian Bunk7c363b82008-07-25 19:46:24 -070046static DEFINE_SPINLOCK(swap_lock);
47static unsigned int nr_swapfiles;
Shaohua Li1abbca72013-02-22 16:34:38 -080048atomic_long_t nr_swap_pages;
49/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070050long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070051static int least_priority;
Shaohua Li1abbca72013-02-22 16:34:38 -080052static atomic_t highest_priority_index = ATOMIC_INIT(-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static const char Bad_file[] = "Bad swap file entry ";
55static const char Unused_file[] = "Unused swap file entry ";
56static const char Bad_offset[] = "Bad swap offset entry ";
57static const char Unused_offset[] = "Unused swap offset entry ";
58
Adrian Bunk7c363b82008-07-25 19:46:24 -070059static struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Hugh Dickinsefa90a92009-12-14 17:58:41 -080061static struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Ingo Molnarfc0abb12006-01-18 17:42:33 -080063static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Kay Sievers66d7dd52010-10-26 14:22:06 -070065static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
66/* Activity counter to indicate that a swapon or swapoff has occurred */
67static atomic_t proc_poll_event = ATOMIC_INIT(0);
68
Hugh Dickins8d69aae2009-12-14 17:58:45 -080069static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070070{
Hugh Dickins570a3352009-12-14 17:58:46 -080071 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070072}
73
Hugh Dickinsefa90a92009-12-14 17:58:41 -080074/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070075static int
76__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
77{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080078 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070079 struct page *page;
80 int ret = 0;
81
82 page = find_get_page(&swapper_space, entry.val);
83 if (!page)
84 return 0;
85 /*
86 * This function is called from scan_swap_map() and it's called
87 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
88 * We have to use trylock for avoiding deadlock. This is a special
89 * case and you should use try_to_free_swap() with explicit lock_page()
90 * in usual operations.
91 */
92 if (trylock_page(page)) {
93 ret = try_to_free_swap(page);
94 unlock_page(page);
95 }
96 page_cache_release(page);
97 return ret;
98}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070099
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800101 * swapon tell device that all the old swap contents can be discarded,
102 * to allow the swap device to optimize its wear-levelling.
103 */
104static int discard_swap(struct swap_info_struct *si)
105{
106 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800107 sector_t start_block;
108 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800109 int err = 0;
110
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800111 /* Do not discard the swap header page! */
112 se = &si->first_swap_extent;
113 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
114 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
115 if (nr_blocks) {
116 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200117 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800118 if (err)
119 return err;
120 cond_resched();
121 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800122
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800123 list_for_each_entry(se, &si->first_swap_extent.list, list) {
124 start_block = se->start_block << (PAGE_SHIFT - 9);
125 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800126
127 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200128 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800129 if (err)
130 break;
131
132 cond_resched();
133 }
134 return err; /* That will often be -EOPNOTSUPP */
135}
136
Hugh Dickins7992fde2009-01-06 14:39:53 -0800137/*
138 * swap allocation tell device that a cluster of swap can now be discarded,
139 * to allow the swap device to optimize its wear-levelling.
140 */
141static void discard_swap_cluster(struct swap_info_struct *si,
142 pgoff_t start_page, pgoff_t nr_pages)
143{
144 struct swap_extent *se = si->curr_swap_extent;
145 int found_extent = 0;
146
147 while (nr_pages) {
148 struct list_head *lh;
149
150 if (se->start_page <= start_page &&
151 start_page < se->start_page + se->nr_pages) {
152 pgoff_t offset = start_page - se->start_page;
153 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800154 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800155
156 if (nr_blocks > nr_pages)
157 nr_blocks = nr_pages;
158 start_page += nr_blocks;
159 nr_pages -= nr_blocks;
160
161 if (!found_extent++)
162 si->curr_swap_extent = se;
163
164 start_block <<= PAGE_SHIFT - 9;
165 nr_blocks <<= PAGE_SHIFT - 9;
166 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200167 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800168 break;
169 }
170
171 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800172 se = list_entry(lh, struct swap_extent, list);
173 }
174}
175
176static int wait_for_discard(void *word)
177{
178 schedule();
179 return 0;
180}
181
Hugh Dickins048c27f2005-09-03 15:54:40 -0700182#define SWAPFILE_CLUSTER 256
183#define LATENCY_LIMIT 256
184
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700185static unsigned long scan_swap_map(struct swap_info_struct *si,
186 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800188 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800189 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800190 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700191 int latency_ration = LATENCY_LIMIT;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800192 int found_free_cluster = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800194 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700195 * We try to cluster swap pages by allocating them sequentially
196 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
197 * way, however, we resort to first-free allocation, starting
198 * a new cluster. This prevents us from scattering swap pages
199 * all over the entire swap partition, so that we reduce
200 * overall disk seek times between swap pages. -- sct
201 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800202 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700203 */
204
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700205 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800206 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800207
208 if (unlikely(!si->cluster_nr--)) {
209 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
210 si->cluster_nr = SWAPFILE_CLUSTER - 1;
211 goto checks;
212 }
Hugh Dickins7992fde2009-01-06 14:39:53 -0800213 if (si->flags & SWP_DISCARDABLE) {
214 /*
215 * Start range check on racing allocations, in case
216 * they overlap the cluster we eventually decide on
217 * (we scan without swap_lock to allow preemption).
218 * It's hardly conceivable that cluster_nr could be
219 * wrapped during our scan, but don't depend on it.
220 */
221 if (si->lowest_alloc)
222 goto checks;
223 si->lowest_alloc = si->max;
224 si->highest_alloc = 0;
225 }
Shaohua Li1abbca72013-02-22 16:34:38 -0800226 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700227
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800228 /*
229 * If seek is expensive, start searching for new cluster from
230 * start of partition, to minimize the span of allocated swap.
231 * But if seek is cheap, search from our current position, so
232 * that swap is allocated from all over the partition: if the
233 * Flash Translation Layer only remaps within limited zones,
234 * we don't want to wear out the first zone too quickly.
235 */
236 if (!(si->flags & SWP_SOLIDSTATE))
237 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700238 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
239
240 /* Locate the first empty (unaligned) cluster */
241 for (; last_in_cluster <= si->highest_bit; offset++) {
242 if (si->swap_map[offset])
243 last_in_cluster = offset + SWAPFILE_CLUSTER;
244 else if (offset == last_in_cluster) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800245 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800246 offset -= SWAPFILE_CLUSTER - 1;
247 si->cluster_next = offset;
248 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800249 found_free_cluster = 1;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800250 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700252 if (unlikely(--latency_ration < 0)) {
253 cond_resched();
254 latency_ration = LATENCY_LIMIT;
255 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700256 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800257
258 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800259 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
260
261 /* Locate the first empty (unaligned) cluster */
262 for (; last_in_cluster < scan_base; offset++) {
263 if (si->swap_map[offset])
264 last_in_cluster = offset + SWAPFILE_CLUSTER;
265 else if (offset == last_in_cluster) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800266 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800267 offset -= SWAPFILE_CLUSTER - 1;
268 si->cluster_next = offset;
269 si->cluster_nr = SWAPFILE_CLUSTER - 1;
270 found_free_cluster = 1;
271 goto checks;
272 }
273 if (unlikely(--latency_ration < 0)) {
274 cond_resched();
275 latency_ration = LATENCY_LIMIT;
276 }
277 }
278
279 offset = scan_base;
Shaohua Li1abbca72013-02-22 16:34:38 -0800280 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800281 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800282 si->lowest_alloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700284
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800285checks:
286 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700287 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700288 if (!si->highest_bit)
289 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800290 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800291 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700292
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700293 /* reuse swap entry of cache-only swap if not busy. */
294 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700295 int swap_was_freed;
Shaohua Li1abbca72013-02-22 16:34:38 -0800296 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700297 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Li1abbca72013-02-22 16:34:38 -0800298 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700299 /* entry was freed successfully, try to use this again */
300 if (swap_was_freed)
301 goto checks;
302 goto scan; /* check next one */
303 }
304
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800305 if (si->swap_map[offset])
306 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700307
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800308 if (offset == si->lowest_bit)
309 si->lowest_bit++;
310 if (offset == si->highest_bit)
311 si->highest_bit--;
312 si->inuse_pages++;
313 if (si->inuse_pages == si->pages) {
314 si->lowest_bit = si->max;
315 si->highest_bit = 0;
316 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800317 si->swap_map[offset] = usage;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800318 si->cluster_next = offset + 1;
319 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800320
321 if (si->lowest_alloc) {
322 /*
323 * Only set when SWP_DISCARDABLE, and there's a scan
324 * for a free cluster in progress or just completed.
325 */
326 if (found_free_cluster) {
327 /*
328 * To optimize wear-levelling, discard the
329 * old data of the cluster, taking care not to
330 * discard any of its pages that have already
331 * been allocated by racing tasks (offset has
332 * already stepped over any at the beginning).
333 */
334 if (offset < si->highest_alloc &&
335 si->lowest_alloc <= last_in_cluster)
336 last_in_cluster = si->lowest_alloc - 1;
337 si->flags |= SWP_DISCARDING;
Shaohua Li1abbca72013-02-22 16:34:38 -0800338 spin_unlock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800339
340 if (offset < last_in_cluster)
341 discard_swap_cluster(si, offset,
342 last_in_cluster - offset + 1);
343
Shaohua Li1abbca72013-02-22 16:34:38 -0800344 spin_lock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800345 si->lowest_alloc = 0;
346 si->flags &= ~SWP_DISCARDING;
347
348 smp_mb(); /* wake_up_bit advises this */
349 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
350
351 } else if (si->flags & SWP_DISCARDING) {
352 /*
353 * Delay using pages allocated by racing tasks
354 * until the whole discard has been issued. We
355 * could defer that delay until swap_writepage,
356 * but it's easier to keep this self-contained.
357 */
Shaohua Li1abbca72013-02-22 16:34:38 -0800358 spin_unlock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800359 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
360 wait_for_discard, TASK_UNINTERRUPTIBLE);
Shaohua Li1abbca72013-02-22 16:34:38 -0800361 spin_lock(&si->lock);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800362 } else {
363 /*
364 * Note pages allocated by racing tasks while
365 * scan for a free cluster is in progress, so
366 * that its final discard can exclude them.
367 */
368 if (offset < si->lowest_alloc)
369 si->lowest_alloc = offset;
370 if (offset > si->highest_alloc)
371 si->highest_alloc = offset;
372 }
373 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800374 return offset;
375
376scan:
Shaohua Li1abbca72013-02-22 16:34:38 -0800377 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700378 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700379 if (!si->swap_map[offset]) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800380 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700381 goto checks;
382 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700383 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800384 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700385 goto checks;
386 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700387 if (unlikely(--latency_ration < 0)) {
388 cond_resched();
389 latency_ration = LATENCY_LIMIT;
390 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700391 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800392 offset = si->lowest_bit;
393 while (++offset < scan_base) {
394 if (!si->swap_map[offset]) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800395 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800396 goto checks;
397 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700398 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800399 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700400 goto checks;
401 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800402 if (unlikely(--latency_ration < 0)) {
403 cond_resched();
404 latency_ration = LATENCY_LIMIT;
405 }
406 }
Shaohua Li1abbca72013-02-22 16:34:38 -0800407 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700408
409no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700410 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 return 0;
412}
413
414swp_entry_t get_swap_page(void)
415{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700416 struct swap_info_struct *si;
417 pgoff_t offset;
418 int type, next;
419 int wrapped = 0;
Shaohua Li1abbca72013-02-22 16:34:38 -0800420 int hp_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
Hugh Dickins5d337b92005-09-03 15:54:41 -0700422 spin_lock(&swap_lock);
Shaohua Li1abbca72013-02-22 16:34:38 -0800423 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700424 goto noswap;
Shaohua Li1abbca72013-02-22 16:34:38 -0800425 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700427 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800428 hp_index = atomic_xchg(&highest_priority_index, -1);
429 /*
430 * highest_priority_index records current highest priority swap
431 * type which just frees swap entries. If its priority is
432 * higher than that of swap_list.next swap type, we use it. It
433 * isn't protected by swap_lock, so it can be an invalid value
434 * if the corresponding swap type is swapoff. We double check
435 * the flags here. It's even possible the swap type is swapoff
436 * and swapon again and its priority is changed. In such rare
437 * case, low prority swap type might be used, but eventually
438 * high priority swap will be used after several rounds of
439 * swap.
440 */
441 if (hp_index != -1 && hp_index != type &&
442 swap_info[type]->prio < swap_info[hp_index]->prio &&
443 (swap_info[hp_index]->flags & SWP_WRITEOK)) {
444 type = hp_index;
445 swap_list.next = type;
446 }
447
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800448 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700449 next = si->next;
450 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800451 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700452 next = swap_list.head;
453 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700455
Shaohua Li1abbca72013-02-22 16:34:38 -0800456 spin_lock(&si->lock);
457 if (!si->highest_bit) {
458 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700459 continue;
Shaohua Li1abbca72013-02-22 16:34:38 -0800460 }
461 if (!(si->flags & SWP_WRITEOK)) {
462 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700463 continue;
Shaohua Li1abbca72013-02-22 16:34:38 -0800464 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700465
466 swap_list.next = next;
Shaohua Li1abbca72013-02-22 16:34:38 -0800467
468 spin_unlock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700469 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800470 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Li1abbca72013-02-22 16:34:38 -0800471 spin_unlock(&si->lock);
472 if (offset)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700473 return swp_entry(type, offset);
Shaohua Li1abbca72013-02-22 16:34:38 -0800474 spin_lock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700475 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700477
Shaohua Li1abbca72013-02-22 16:34:38 -0800478 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700479noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700480 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700481 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
Hugh Dickins910321e2010-09-09 16:38:07 -0700484/* The only caller of this function is now susupend routine */
485swp_entry_t get_swap_page_of_type(int type)
486{
487 struct swap_info_struct *si;
488 pgoff_t offset;
489
Hugh Dickins910321e2010-09-09 16:38:07 -0700490 si = swap_info[type];
Shaohua Li1abbca72013-02-22 16:34:38 -0800491 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700492 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800493 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700494 /* This is called for allocating swap entry, not cache */
495 offset = scan_swap_map(si, 1);
496 if (offset) {
Shaohua Li1abbca72013-02-22 16:34:38 -0800497 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700498 return swp_entry(type, offset);
499 }
Shaohua Li1abbca72013-02-22 16:34:38 -0800500 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700501 }
Shaohua Li1abbca72013-02-22 16:34:38 -0800502 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700503 return (swp_entry_t) {0};
504}
505
Hugh Dickins73c34b62009-12-14 17:58:43 -0800506static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800508 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 unsigned long offset, type;
510
511 if (!entry.val)
512 goto out;
513 type = swp_type(entry);
514 if (type >= nr_swapfiles)
515 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800516 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (!(p->flags & SWP_USED))
518 goto bad_device;
519 offset = swp_offset(entry);
520 if (offset >= p->max)
521 goto bad_offset;
522 if (!p->swap_map[offset])
523 goto bad_free;
Shaohua Li1abbca72013-02-22 16:34:38 -0800524 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return p;
526
527bad_free:
528 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
529 goto out;
530bad_offset:
531 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
532 goto out;
533bad_device:
534 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
535 goto out;
536bad_nofile:
537 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
538out:
539 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800540}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Shaohua Li1abbca72013-02-22 16:34:38 -0800542/*
543 * This swap type frees swap entry, check if it is the highest priority swap
544 * type which just frees swap entry. get_swap_page() uses
545 * highest_priority_index to search highest priority swap type. The
546 * swap_info_struct.lock can't protect us if there are multiple swap types
547 * active, so we use atomic_cmpxchg.
548 */
549static void set_highest_priority_index(int type)
550{
551 int old_hp_index, new_hp_index;
552
553 do {
554 old_hp_index = atomic_read(&highest_priority_index);
555 if (old_hp_index != -1 &&
556 swap_info[old_hp_index]->prio >= swap_info[type]->prio)
557 break;
558 new_hp_index = type;
559 } while (atomic_cmpxchg(&highest_priority_index,
560 old_hp_index, new_hp_index) != old_hp_index);
561}
562
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800563static unsigned char swap_entry_free(struct swap_info_struct *p,
564 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Hugh Dickins253d5532009-12-14 17:58:44 -0800566 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800567 unsigned char count;
568 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700570 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800571 has_cache = count & SWAP_HAS_CACHE;
572 count &= ~SWAP_HAS_CACHE;
573
574 if (usage == SWAP_HAS_CACHE) {
575 VM_BUG_ON(!has_cache);
576 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800577 } else if (count == SWAP_MAP_SHMEM) {
578 /*
579 * Or we could insist on shmem.c using a special
580 * swap_shmem_free() and free_shmem_swap_and_cache()...
581 */
582 count = 0;
Hugh Dickins570a3352009-12-14 17:58:46 -0800583 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
584 if (count == COUNT_CONTINUED) {
585 if (swap_count_continued(p, offset, count))
586 count = SWAP_MAP_MAX | COUNT_CONTINUED;
587 else
588 count = SWAP_MAP_MAX;
589 } else
590 count--;
591 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800592
593 if (!count)
594 mem_cgroup_uncharge_swap(entry);
595
596 usage = count | has_cache;
597 p->swap_map[offset] = usage;
598
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700599 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800600 if (!usage) {
Nitin Guptab3a27d02010-05-17 11:02:43 +0530601 struct gendisk *disk = p->bdev->bd_disk;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700602 if (offset < p->lowest_bit)
603 p->lowest_bit = offset;
604 if (offset > p->highest_bit)
605 p->highest_bit = offset;
Shaohua Li1abbca72013-02-22 16:34:38 -0800606 set_highest_priority_index(p->type);
607 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700608 p->inuse_pages--;
Nitin Guptab3a27d02010-05-17 11:02:43 +0530609 if ((p->flags & SWP_BLKDEV) &&
610 disk->fops->swap_slot_free_notify)
611 disk->fops->swap_slot_free_notify(p->bdev, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800613
614 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
616
617/*
618 * Caller has made sure that the swapdevice corresponding to entry
619 * is still around or has not been recycled.
620 */
621void swap_free(swp_entry_t entry)
622{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800623 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
625 p = swap_info_get(entry);
626 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800627 swap_entry_free(p, entry, 1);
Shaohua Li1abbca72013-02-22 16:34:38 -0800628 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 }
630}
631
632/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700633 * Called after dropping swapcache to decrease refcnt to swap entries.
634 */
635void swapcache_free(swp_entry_t entry, struct page *page)
636{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700637 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800638 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700639
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700640 p = swap_info_get(entry);
641 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800642 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
643 if (page)
644 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
Shaohua Li1abbca72013-02-22 16:34:38 -0800645 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700646 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700647}
648
649/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700650 * How many references to page are currently swapped out?
Hugh Dickins570a3352009-12-14 17:58:46 -0800651 * This does not give an exact answer when swap count is continued,
652 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 */
Mel Gormanfdd4d472012-07-31 16:44:47 -0700654int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700656 int count = 0;
657 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 swp_entry_t entry;
659
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700660 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 p = swap_info_get(entry);
662 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700663 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Li1abbca72013-02-22 16:34:38 -0800664 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700666 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
669/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800670 * We can write to an anon page without COW if there are no other references
671 * to it. And as a side-effect, free up its swap: because the old content
672 * on disk will never be read, and seeking back there to write new content
673 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800675int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700677 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Hugh Dickins51726b12009-01-06 14:39:25 -0800679 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800680 if (unlikely(PageKsm(page)))
681 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700682 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800683 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700684 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800685 if (count == 1 && !PageWriteback(page)) {
686 delete_from_swap_cache(page);
687 SetPageDirty(page);
688 }
689 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800690 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
693/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800694 * If swap is getting full, or if there are no more mappings of this page,
695 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800697int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Hugh Dickins51726b12009-01-06 14:39:25 -0800699 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 if (!PageSwapCache(page))
702 return 0;
703 if (PageWriteback(page))
704 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800705 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 return 0;
707
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700708 /*
709 * Once hibernation has begun to create its image of memory,
710 * there's a danger that one of the calls to try_to_free_swap()
711 * - most probably a call from __try_to_reclaim_swap() while
712 * hibernation is allocating its own swap pages for the image,
713 * but conceivably even a call from memory reclaim - will free
714 * the swap from a page which has already been recorded in the
715 * image as a clean swapcache page, and then reuse its swap for
716 * another page of the image. On waking from hibernation, the
717 * original page might be freed under memory pressure, then
718 * later read back in from swap, now with the wrong data.
719 *
Mel Gormanf90ac392012-01-10 15:07:15 -0800720 * Hibration suspends storage while it is writing the image
721 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700722 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800723 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700724 return 0;
725
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800726 delete_from_swap_cache(page);
727 SetPageDirty(page);
728 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700729}
730
731/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 * Free the swap entry like above, but also try to
733 * free the page cache entry if it is the last user.
734 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800735int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800737 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 struct page *page = NULL;
739
Andi Kleena7420aa2009-09-16 11:50:05 +0200740 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800741 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700742
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 p = swap_info_get(entry);
744 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800745 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Nick Piggin93fac702006-03-31 02:29:56 -0800746 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700747 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800748 page_cache_release(page);
749 page = NULL;
750 }
751 }
Shaohua Li1abbca72013-02-22 16:34:38 -0800752 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 }
754 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800755 /*
756 * Not mapped elsewhere, or swap space full? Free it!
757 * Also recheck PageSwapCache now page is locked (above).
758 */
Nick Piggin93fac702006-03-31 02:29:56 -0800759 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800760 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 delete_from_swap_cache(page);
762 SetPageDirty(page);
763 }
764 unlock_page(page);
765 page_cache_release(page);
766 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800767 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768}
769
Daisuke Nishimura02491442010-03-10 15:22:17 -0800770#ifdef CONFIG_CGROUP_MEM_RES_CTLR
771/**
772 * mem_cgroup_count_swap_user - count the user of a swap entry
773 * @ent: the swap entry to be checked
774 * @pagep: the pointer for the swap cache page of the entry to be stored
775 *
776 * Returns the number of the user of the swap entry. The number is valid only
777 * for swaps of anonymous pages.
778 * If the entry is found on swap cache, the page is stored to pagep with
779 * refcount of it being incremented.
780 */
781int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep)
782{
783 struct page *page;
784 struct swap_info_struct *p;
785 int count = 0;
786
787 page = find_get_page(&swapper_space, ent.val);
788 if (page)
789 count += page_mapcount(page);
790 p = swap_info_get(ent);
791 if (p) {
792 count += swap_count(p->swap_map[swp_offset(ent)]);
793 spin_unlock(&swap_lock);
794 }
795
796 *pagep = page;
797 return count;
798}
799#endif
800
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200801#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800802/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800803 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800804 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800805 * @offset - number of the PAGE_SIZE-sized block of the device, starting
806 * from 0, in which the swap header is expected to be located.
807 *
808 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800809 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800810int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800811{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800812 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800813 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800814
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800815 if (device)
816 bdev = bdget(device);
817
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800818 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800819 for (type = 0; type < nr_swapfiles; type++) {
820 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800821
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800822 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800823 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700824
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800825 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800826 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200827 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800828
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800829 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800830 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800831 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800832 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800833 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800834
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800835 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800836 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200837 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800838
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800839 spin_unlock(&swap_lock);
840 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800841 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800842 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800843 }
844 }
845 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800846 if (bdev)
847 bdput(bdev);
848
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800849 return -ENODEV;
850}
851
852/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800853 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
854 * corresponding to given index in swap_info (swap type).
855 */
856sector_t swapdev_block(int type, pgoff_t offset)
857{
858 struct block_device *bdev;
859
860 if ((unsigned int)type >= nr_swapfiles)
861 return 0;
862 if (!(swap_info[type]->flags & SWP_WRITEOK))
863 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800864 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800865}
866
867/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800868 * Return either the total number of swap pages of given type, or the number
869 * of free pages of that type (depending on @free)
870 *
871 * This is needed for software suspend
872 */
873unsigned int count_swap_pages(int type, int free)
874{
875 unsigned int n = 0;
876
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800877 spin_lock(&swap_lock);
878 if ((unsigned int)type < nr_swapfiles) {
879 struct swap_info_struct *sis = swap_info[type];
880
Shaohua Li1abbca72013-02-22 16:34:38 -0800881 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800882 if (sis->flags & SWP_WRITEOK) {
883 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800884 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800885 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800886 }
Shaohua Li1abbca72013-02-22 16:34:38 -0800887 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800888 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800889 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800890 return n;
891}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800892#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700895 * No need to decide whether this PTE shares the swap entry with others,
896 * just let do_wp_page work it out if a write is requested later - to
897 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800899static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 unsigned long addr, swp_entry_t entry, struct page *page)
901{
Johannes Weiner72835c82012-01-12 17:18:32 -0800902 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800903 spinlock_t *ptl;
904 pte_t *pte;
905 int ret = 1;
906
Johannes Weiner72835c82012-01-12 17:18:32 -0800907 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
908 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800909 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800910 goto out_nolock;
911 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800912
913 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
914 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
915 if (ret > 0)
Johannes Weiner72835c82012-01-12 17:18:32 -0800916 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800917 ret = 0;
918 goto out;
919 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800920
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800921 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800922 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 get_page(page);
924 set_pte_at(vma->vm_mm, addr, pte,
925 pte_mkold(mk_pte(page, vma->vm_page_prot)));
926 page_add_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -0800927 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 swap_free(entry);
929 /*
930 * Move the page to the active list so it is not
931 * immediately swapped out again after swapon.
932 */
933 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800934out:
935 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800936out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800937 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938}
939
940static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
941 unsigned long addr, unsigned long end,
942 swp_entry_t entry, struct page *page)
943{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700945 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800946 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
Hugh Dickins044d66c2008-02-07 00:14:04 -0800948 /*
949 * We don't actually need pte lock while scanning for swp_pte: since
950 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
951 * page table while we're scanning; though it could get zapped, and on
952 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
953 * of unmatched parts which look like swp_pte, so unuse_pte must
954 * recheck under pte lock. Scanning without pte lock lets it be
955 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
956 */
957 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 do {
959 /*
960 * swapoff spends a _lot_ of time in this loop!
961 * Test inline before going to call unuse_pte.
962 */
963 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800964 pte_unmap(pte);
965 ret = unuse_pte(vma, pmd, addr, entry, page);
966 if (ret)
967 goto out;
968 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 }
970 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800971 pte_unmap(pte - 1);
972out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800973 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974}
975
976static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
977 unsigned long addr, unsigned long end,
978 swp_entry_t entry, struct page *page)
979{
980 pmd_t *pmd;
981 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800982 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 pmd = pmd_offset(pud, addr);
985 do {
986 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700987 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800989 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
990 if (ret)
991 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 } while (pmd++, addr = next, addr != end);
993 return 0;
994}
995
996static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
997 unsigned long addr, unsigned long end,
998 swp_entry_t entry, struct page *page)
999{
1000 pud_t *pud;
1001 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001002 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004 pud = pud_offset(pgd, addr);
1005 do {
1006 next = pud_addr_end(addr, end);
1007 if (pud_none_or_clear_bad(pud))
1008 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001009 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1010 if (ret)
1011 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 } while (pud++, addr = next, addr != end);
1013 return 0;
1014}
1015
1016static int unuse_vma(struct vm_area_struct *vma,
1017 swp_entry_t entry, struct page *page)
1018{
1019 pgd_t *pgd;
1020 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001021 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001023 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 addr = page_address_in_vma(page, vma);
1025 if (addr == -EFAULT)
1026 return 0;
1027 else
1028 end = addr + PAGE_SIZE;
1029 } else {
1030 addr = vma->vm_start;
1031 end = vma->vm_end;
1032 }
1033
1034 pgd = pgd_offset(vma->vm_mm, addr);
1035 do {
1036 next = pgd_addr_end(addr, end);
1037 if (pgd_none_or_clear_bad(pgd))
1038 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001039 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1040 if (ret)
1041 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 } while (pgd++, addr = next, addr != end);
1043 return 0;
1044}
1045
1046static int unuse_mm(struct mm_struct *mm,
1047 swp_entry_t entry, struct page *page)
1048{
1049 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001050 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 if (!down_read_trylock(&mm->mmap_sem)) {
1053 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001054 * Activate page so shrink_inactive_list is unlikely to unmap
1055 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001057 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 unlock_page(page);
1059 down_read(&mm->mmap_sem);
1060 lock_page(page);
1061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001063 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 break;
1065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001067 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068}
1069
1070/*
1071 * Scan swap_map from current position to next entry still in use.
1072 * Recycle to start on reaching the end, returning 0 when empty.
1073 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001074static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1075 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001077 unsigned int max = si->max;
1078 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001079 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001082 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 * for whether an entry is in use, not modifying it; false
1084 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001085 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 */
1087 for (;;) {
1088 if (++i >= max) {
1089 if (!prev) {
1090 i = 0;
1091 break;
1092 }
1093 /*
1094 * No entries in use at top of swap_map,
1095 * loop back to start and recheck there.
1096 */
1097 max = prev + 1;
1098 prev = 0;
1099 i = 1;
1100 }
1101 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001102 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 break;
1104 }
1105 return i;
1106}
1107
1108/*
1109 * We completely avoid races by reading each swap page in advance,
1110 * and then search for the process using it. All the necessary
1111 * page table adjustments can then be made atomically.
1112 */
1113static int try_to_unuse(unsigned int type)
1114{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001115 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001117 unsigned char *swap_map;
1118 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 struct page *page;
1120 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001121 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
1124 /*
1125 * When searching mms for an entry, a good strategy is to
1126 * start at the first mm we freed the previous entry from
1127 * (though actually we don't notice whether we or coincidence
1128 * freed the entry). Initialize this start_mm with a hold.
1129 *
1130 * A simpler strategy would be to start at the last mm we
1131 * freed the previous entry from; but that would take less
1132 * advantage of mmlist ordering, which clusters forked mms
1133 * together, child after parent. If we race with dup_mmap(), we
1134 * prefer to resolve parent before child, lest we miss entries
1135 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a3352009-12-14 17:58:46 -08001136 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 */
1138 start_mm = &init_mm;
1139 atomic_inc(&init_mm.mm_users);
1140
1141 /*
1142 * Keep on scanning until all entries have gone. Usually,
1143 * one pass through swap_map is enough, but not necessarily:
1144 * there are races when an instance of an entry might be missed.
1145 */
1146 while ((i = find_next_to_unuse(si, i)) != 0) {
1147 if (signal_pending(current)) {
1148 retval = -EINTR;
1149 break;
1150 }
1151
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001152 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 * Get a page for the entry, using the existing swap
1154 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001155 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 */
1157 swap_map = &si->swap_map[i];
1158 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001159 page = read_swap_cache_async(entry,
1160 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (!page) {
1162 /*
1163 * Either swap_duplicate() failed because entry
1164 * has been freed independently, and will not be
1165 * reused since sys_swapoff() already disabled
1166 * allocation from here, or alloc_page() failed.
1167 */
1168 if (!*swap_map)
1169 continue;
1170 retval = -ENOMEM;
1171 break;
1172 }
1173
1174 /*
1175 * Don't hold on to start_mm if it looks like exiting.
1176 */
1177 if (atomic_read(&start_mm->mm_users) == 1) {
1178 mmput(start_mm);
1179 start_mm = &init_mm;
1180 atomic_inc(&init_mm.mm_users);
1181 }
1182
1183 /*
1184 * Wait for and lock page. When do_swap_page races with
1185 * try_to_unuse, do_swap_page can handle the fault much
1186 * faster than try_to_unuse can locate the entry. This
1187 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1188 * defer to do_swap_page in such a case - in some tests,
1189 * do_swap_page and try_to_unuse repeatedly compete.
1190 */
1191 wait_on_page_locked(page);
1192 wait_on_page_writeback(page);
1193 lock_page(page);
1194 wait_on_page_writeback(page);
1195
1196 /*
1197 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001200 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1201 retval = shmem_unuse(entry, page);
1202 /* page has already been unlocked and released */
1203 if (retval < 0)
1204 break;
1205 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001207 if (swap_count(swcount) && start_mm != &init_mm)
1208 retval = unuse_mm(start_mm, entry, page);
1209
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001210 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 int set_start_mm = (*swap_map >= swcount);
1212 struct list_head *p = &start_mm->mmlist;
1213 struct mm_struct *new_start_mm = start_mm;
1214 struct mm_struct *prev_mm = start_mm;
1215 struct mm_struct *mm;
1216
1217 atomic_inc(&new_start_mm->mm_users);
1218 atomic_inc(&prev_mm->mm_users);
1219 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001220 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 (p = p->next) != &start_mm->mmlist) {
1222 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001223 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 spin_unlock(&mmlist_lock);
1226 mmput(prev_mm);
1227 prev_mm = mm;
1228
1229 cond_resched();
1230
1231 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001232 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001234 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001236 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001238
Bo Liu32c5fc12009-11-02 16:50:33 +00001239 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 mmput(new_start_mm);
1241 atomic_inc(&mm->mm_users);
1242 new_start_mm = mm;
1243 set_start_mm = 0;
1244 }
1245 spin_lock(&mmlist_lock);
1246 }
1247 spin_unlock(&mmlist_lock);
1248 mmput(prev_mm);
1249 mmput(start_mm);
1250 start_mm = new_start_mm;
1251 }
1252 if (retval) {
1253 unlock_page(page);
1254 page_cache_release(page);
1255 break;
1256 }
1257
1258 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 * If a reference remains (rare), we would like to leave
1260 * the page in the swap cache; but try_to_unmap could
1261 * then re-duplicate the entry once we drop page lock,
1262 * so we might loop indefinitely; also, that page could
1263 * not be swapped out to other storage meanwhile. So:
1264 * delete from cache even if there's another reference,
1265 * after ensuring that the data has been saved to disk -
1266 * since if the reference remains (rarer), it will be
1267 * read from disk into another page. Splitting into two
1268 * pages would be incorrect if swap supported "shared
1269 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001270 *
1271 * Given how unuse_vma() targets one particular offset
1272 * in an anon_vma, once the anon_vma has been determined,
1273 * this splitting happens to be just what is needed to
1274 * handle where KSM pages have been swapped out: re-reading
1275 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001277 if (swap_count(*swap_map) &&
1278 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 struct writeback_control wbc = {
1280 .sync_mode = WB_SYNC_NONE,
1281 };
1282
1283 swap_writepage(page, &wbc);
1284 lock_page(page);
1285 wait_on_page_writeback(page);
1286 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001287
1288 /*
1289 * It is conceivable that a racing task removed this page from
1290 * swap cache just before we acquired the page lock at the top,
1291 * or while we dropped it in unuse_mm(). The page might even
1292 * be back in swap cache on another swap area: that we must not
1293 * delete, since it may not have been written out to swap yet.
1294 */
1295 if (PageSwapCache(page) &&
1296 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001297 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
1299 /*
1300 * So we could skip searching mms once swap count went
1301 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001302 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 */
1304 SetPageDirty(page);
1305 unlock_page(page);
1306 page_cache_release(page);
1307
1308 /*
1309 * Make sure that we aren't completely killing
1310 * interactive performance.
1311 */
1312 cond_resched();
1313 }
1314
1315 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 return retval;
1317}
1318
1319/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001320 * After a successful try_to_unuse, if no swap is now in use, we know
1321 * we can empty the mmlist. swap_lock must be held on entry and exit.
1322 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 * added to the mmlist just after page_duplicate - before would be racy.
1324 */
1325static void drain_mmlist(void)
1326{
1327 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001328 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001330 for (type = 0; type < nr_swapfiles; type++)
1331 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 return;
1333 spin_lock(&mmlist_lock);
1334 list_for_each_safe(p, next, &init_mm.mmlist)
1335 list_del_init(p);
1336 spin_unlock(&mmlist_lock);
1337}
1338
1339/*
1340 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001341 * corresponds to page offset for the specified swap entry.
1342 * Note that the type of this function is sector_t, but it returns page offset
1343 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001345static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001347 struct swap_info_struct *sis;
1348 struct swap_extent *start_se;
1349 struct swap_extent *se;
1350 pgoff_t offset;
1351
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001352 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001353 *bdev = sis->bdev;
1354
1355 offset = swp_offset(entry);
1356 start_se = sis->curr_swap_extent;
1357 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 for ( ; ; ) {
1360 struct list_head *lh;
1361
1362 if (se->start_page <= offset &&
1363 offset < (se->start_page + se->nr_pages)) {
1364 return se->start_block + (offset - se->start_page);
1365 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001366 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 se = list_entry(lh, struct swap_extent, list);
1368 sis->curr_swap_extent = se;
1369 BUG_ON(se == start_se); /* It *must* be present */
1370 }
1371}
1372
1373/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001374 * Returns the page offset into bdev for the specified page's swap entry.
1375 */
1376sector_t map_swap_page(struct page *page, struct block_device **bdev)
1377{
1378 swp_entry_t entry;
1379 entry.val = page_private(page);
1380 return map_swap_entry(entry, bdev);
1381}
1382
1383/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 * Free all of a swapdev's extent information
1385 */
1386static void destroy_swap_extents(struct swap_info_struct *sis)
1387{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001388 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 struct swap_extent *se;
1390
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001391 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 struct swap_extent, list);
1393 list_del(&se->list);
1394 kfree(se);
1395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396}
1397
1398/*
1399 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001400 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001402 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 */
1404static int
1405add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1406 unsigned long nr_pages, sector_t start_block)
1407{
1408 struct swap_extent *se;
1409 struct swap_extent *new_se;
1410 struct list_head *lh;
1411
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001412 if (start_page == 0) {
1413 se = &sis->first_swap_extent;
1414 sis->curr_swap_extent = se;
1415 se->start_page = 0;
1416 se->nr_pages = nr_pages;
1417 se->start_block = start_block;
1418 return 1;
1419 } else {
1420 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001422 BUG_ON(se->start_page + se->nr_pages != start_page);
1423 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 /* Merge it */
1425 se->nr_pages += nr_pages;
1426 return 0;
1427 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 }
1429
1430 /*
1431 * No merge. Insert a new extent, preserving ordering.
1432 */
1433 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1434 if (new_se == NULL)
1435 return -ENOMEM;
1436 new_se->start_page = start_page;
1437 new_se->nr_pages = nr_pages;
1438 new_se->start_block = start_block;
1439
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001440 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001441 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
1443
1444/*
1445 * A `swap extent' is a simple thing which maps a contiguous range of pages
1446 * onto a contiguous range of disk blocks. An ordered list of swap extents
1447 * is built at swapon time and is then used at swap_writepage/swap_readpage
1448 * time for locating where on disk a page belongs.
1449 *
1450 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1451 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1452 * swap files identically.
1453 *
1454 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1455 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1456 * swapfiles are handled *identically* after swapon time.
1457 *
1458 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1459 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1460 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1461 * requirements, they are simply tossed out - we will never use those blocks
1462 * for swapping.
1463 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001464 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1466 * which will scribble on the fs.
1467 *
1468 * The amount of disk space which a single swap extent represents varies.
1469 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1470 * extents in the list. To avoid much list walking, we cache the previous
1471 * search location in `curr_swap_extent', and start new searches from there.
1472 * This is extremely effective. The average number of iterations in
1473 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1474 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001475static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476{
1477 struct inode *inode;
1478 unsigned blocks_per_page;
1479 unsigned long page_no;
1480 unsigned blkbits;
1481 sector_t probe_block;
1482 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001483 sector_t lowest_block = -1;
1484 sector_t highest_block = 0;
1485 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 int ret;
1487
1488 inode = sis->swap_file->f_mapping->host;
1489 if (S_ISBLK(inode->i_mode)) {
1490 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001491 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001492 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 }
1494
1495 blkbits = inode->i_blkbits;
1496 blocks_per_page = PAGE_SIZE >> blkbits;
1497
1498 /*
1499 * Map all the blocks into the extent list. This code doesn't try
1500 * to be very smart.
1501 */
1502 probe_block = 0;
1503 page_no = 0;
1504 last_block = i_size_read(inode) >> blkbits;
1505 while ((probe_block + blocks_per_page) <= last_block &&
1506 page_no < sis->max) {
1507 unsigned block_in_page;
1508 sector_t first_block;
1509
1510 first_block = bmap(inode, probe_block);
1511 if (first_block == 0)
1512 goto bad_bmap;
1513
1514 /*
1515 * It must be PAGE_SIZE aligned on-disk
1516 */
1517 if (first_block & (blocks_per_page - 1)) {
1518 probe_block++;
1519 goto reprobe;
1520 }
1521
1522 for (block_in_page = 1; block_in_page < blocks_per_page;
1523 block_in_page++) {
1524 sector_t block;
1525
1526 block = bmap(inode, probe_block + block_in_page);
1527 if (block == 0)
1528 goto bad_bmap;
1529 if (block != first_block + block_in_page) {
1530 /* Discontiguity */
1531 probe_block++;
1532 goto reprobe;
1533 }
1534 }
1535
Hugh Dickins53092a72005-09-03 15:54:34 -07001536 first_block >>= (PAGE_SHIFT - blkbits);
1537 if (page_no) { /* exclude the header page */
1538 if (first_block < lowest_block)
1539 lowest_block = first_block;
1540 if (first_block > highest_block)
1541 highest_block = first_block;
1542 }
1543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 /*
1545 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1546 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001547 ret = add_swap_extent(sis, page_no, 1, first_block);
1548 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001550 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 page_no++;
1552 probe_block += blocks_per_page;
1553reprobe:
1554 continue;
1555 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001556 ret = nr_extents;
1557 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001559 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001561 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001563out:
1564 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565bad_bmap:
1566 printk(KERN_ERR "swapon: swapfile has holes\n");
1567 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001568 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001571static void enable_swap_info(struct swap_info_struct *p, int prio,
1572 unsigned char *swap_map)
1573{
1574 int i, prev;
1575
1576 spin_lock(&swap_lock);
1577 if (prio >= 0)
1578 p->prio = prio;
1579 else
1580 p->prio = --least_priority;
1581 p->swap_map = swap_map;
1582 p->flags |= SWP_WRITEOK;
Shaohua Li1abbca72013-02-22 16:34:38 -08001583 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001584 total_swap_pages += p->pages;
1585
1586 /* insert swap space into swap_list: */
1587 prev = -1;
1588 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1589 if (p->prio >= swap_info[i]->prio)
1590 break;
1591 prev = i;
1592 }
1593 p->next = i;
1594 if (prev < 0)
1595 swap_list.head = swap_list.next = p->type;
1596 else
1597 swap_info[prev]->next = p->type;
1598 spin_unlock(&swap_lock);
1599}
1600
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001601SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001603 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001604 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 struct file *swap_file, *victim;
1606 struct address_space *mapping;
1607 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001608 char *pathname;
David Rientjes72788c32011-05-24 17:11:40 -07001609 int oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 int i, type, prev;
1611 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001612
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 if (!capable(CAP_SYS_ADMIN))
1614 return -EPERM;
1615
Al Viro191c5422012-02-13 03:58:52 +00001616 BUG_ON(!current->mm);
1617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 pathname = getname(specialfile);
1619 err = PTR_ERR(pathname);
1620 if (IS_ERR(pathname))
1621 goto out;
1622
1623 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1624 putname(pathname);
1625 err = PTR_ERR(victim);
1626 if (IS_ERR(victim))
1627 goto out;
1628
1629 mapping = victim->f_mapping;
1630 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001631 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001632 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1633 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001634 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 if (p->swap_file->f_mapping == mapping)
1636 break;
1637 }
1638 prev = type;
1639 }
1640 if (type < 0) {
1641 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001642 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 goto out_dput;
1644 }
Al Viro191c5422012-02-13 03:58:52 +00001645 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 vm_unacct_memory(p->pages);
1647 else {
1648 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001649 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 goto out_dput;
1651 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001652 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001654 else
1655 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 if (type == swap_list.next) {
1657 /* just pick something that's safe... */
1658 swap_list.next = swap_list.head;
1659 }
Shaohua Li1abbca72013-02-22 16:34:38 -08001660 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001661 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001662 for (i = p->next; i >= 0; i = swap_info[i]->next)
1663 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001664 least_priority++;
1665 }
Shaohua Li1abbca72013-02-22 16:34:38 -08001666 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 total_swap_pages -= p->pages;
1668 p->flags &= ~SWP_WRITEOK;
Shaohua Li1abbca72013-02-22 16:34:38 -08001669 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001670 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001671
David Rientjes72788c32011-05-24 17:11:40 -07001672 oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 err = try_to_unuse(type);
David Rientjes43362a42011-10-31 17:07:18 -07001674 compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX, oom_score_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 if (err) {
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001677 /*
1678 * reading p->prio and p->swap_map outside the lock is
1679 * safe here because only sys_swapon and sys_swapoff
1680 * change them, and there can be no other sys_swapon or
1681 * sys_swapoff for this swap_info_struct at this point.
1682 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 /* re-insert swap space back into swap_list */
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001684 enable_swap_info(p, p->prio, p->swap_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 goto out_dput;
1686 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001687
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001688 destroy_swap_extents(p);
Hugh Dickins570a3352009-12-14 17:58:46 -08001689 if (p->flags & SWP_CONTINUED)
1690 free_swap_count_continuations(p);
1691
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001692 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001693 spin_lock(&swap_lock);
Shaohua Li1abbca72013-02-22 16:34:38 -08001694 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001696
1697 /* wait for anyone still in scan_swap_map */
1698 p->highest_bit = 0; /* cuts scans short */
1699 while (p->flags >= SWP_SCANNING) {
Shaohua Li1abbca72013-02-22 16:34:38 -08001700 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001701 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001702 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001703 spin_lock(&swap_lock);
Shaohua Li1abbca72013-02-22 16:34:38 -08001704 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001705 }
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 swap_file = p->swap_file;
1708 p->swap_file = NULL;
1709 p->max = 0;
1710 swap_map = p->swap_map;
1711 p->swap_map = NULL;
1712 p->flags = 0;
Shaohua Li1abbca72013-02-22 16:34:38 -08001713 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001714 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001715 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 vfree(swap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001717 /* Destroy swap account informatin */
1718 swap_cgroup_swapoff(type);
1719
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 inode = mapping->host;
1721 if (S_ISBLK(inode->i_mode)) {
1722 struct block_device *bdev = I_BDEV(inode);
1723 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001724 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001726 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001728 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 }
1730 filp_close(swap_file, NULL);
1731 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001732 atomic_inc(&proc_poll_event);
1733 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
1735out_dput:
1736 filp_close(victim, NULL);
1737out:
1738 return err;
1739}
1740
1741#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001742static unsigned swaps_poll(struct file *file, poll_table *wait)
1743{
Kay Sieversf1514632011-07-12 20:48:39 +02001744 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001745
1746 poll_wait(file, &proc_poll_wait, wait);
1747
Kay Sieversf1514632011-07-12 20:48:39 +02001748 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1749 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001750 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1751 }
1752
1753 return POLLIN | POLLRDNORM;
1754}
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756/* iterator */
1757static void *swap_start(struct seq_file *swap, loff_t *pos)
1758{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001759 struct swap_info_struct *si;
1760 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 loff_t l = *pos;
1762
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001763 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001765 if (!l)
1766 return SEQ_START_TOKEN;
1767
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001768 for (type = 0; type < nr_swapfiles; type++) {
1769 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1770 si = swap_info[type];
1771 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001773 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001774 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 }
1776
1777 return NULL;
1778}
1779
1780static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1781{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001782 struct swap_info_struct *si = v;
1783 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001785 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001786 type = 0;
1787 else
1788 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001789
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001790 for (; type < nr_swapfiles; type++) {
1791 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1792 si = swap_info[type];
1793 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 continue;
1795 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001796 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 }
1798
1799 return NULL;
1800}
1801
1802static void swap_stop(struct seq_file *swap, void *v)
1803{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001804 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805}
1806
1807static int swap_show(struct seq_file *swap, void *v)
1808{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001809 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 struct file *file;
1811 int len;
1812
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001813 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001814 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1815 return 0;
1816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001818 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001819 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001820 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001821 len < 40 ? 40 - len : 1, " ",
1822 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001824 si->pages << (PAGE_SHIFT - 10),
1825 si->inuse_pages << (PAGE_SHIFT - 10),
1826 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 return 0;
1828}
1829
Helge Deller15ad7cd2006-12-06 20:40:36 -08001830static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 .start = swap_start,
1832 .next = swap_next,
1833 .stop = swap_stop,
1834 .show = swap_show
1835};
1836
1837static int swaps_open(struct inode *inode, struct file *file)
1838{
Kay Sieversf1514632011-07-12 20:48:39 +02001839 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001840 int ret;
1841
Kay Sievers66d7dd52010-10-26 14:22:06 -07001842 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02001843 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07001844 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001845
Kay Sieversf1514632011-07-12 20:48:39 +02001846 seq = file->private_data;
1847 seq->poll_event = atomic_read(&proc_poll_event);
1848 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849}
1850
Helge Deller15ad7cd2006-12-06 20:40:36 -08001851static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 .open = swaps_open,
1853 .read = seq_read,
1854 .llseek = seq_lseek,
1855 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001856 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857};
1858
1859static int __init procswaps_init(void)
1860{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001861 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 return 0;
1863}
1864__initcall(procswaps_init);
1865#endif /* CONFIG_PROC_FS */
1866
Jan Beulich17963162008-12-16 11:35:24 +00001867#ifdef MAX_SWAPFILES_CHECK
1868static int __init max_swapfiles_check(void)
1869{
1870 MAX_SWAPFILES_CHECK();
1871 return 0;
1872}
1873late_initcall(max_swapfiles_check);
1874#endif
1875
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001876static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001878 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001880
1881 p = kzalloc(sizeof(*p), GFP_KERNEL);
1882 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001883 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001884
Hugh Dickins5d337b92005-09-03 15:54:41 -07001885 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001886 for (type = 0; type < nr_swapfiles; type++) {
1887 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001889 }
Christoph Lameter06972122006-06-23 02:03:35 -07001890 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001891 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001892 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07001893 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001895 if (type >= nr_swapfiles) {
1896 p->type = type;
1897 swap_info[type] = p;
1898 /*
1899 * Write swap_info[type] before nr_swapfiles, in case a
1900 * racing procfs swap_start() or swap_next() is reading them.
1901 * (We never shrink nr_swapfiles, we never free this entry.)
1902 */
1903 smp_wmb();
1904 nr_swapfiles++;
1905 } else {
1906 kfree(p);
1907 p = swap_info[type];
1908 /*
1909 * Do not memset this entry: a racing procfs swap_next()
1910 * would be relying on p->type to remain valid.
1911 */
1912 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001913 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001916 spin_unlock(&swap_lock);
Shaohua Li1abbca72013-02-22 16:34:38 -08001917 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001918
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001919 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001920}
1921
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001922static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1923{
1924 int error;
1925
1926 if (S_ISBLK(inode->i_mode)) {
1927 p->bdev = bdgrab(I_BDEV(inode));
1928 error = blkdev_get(p->bdev,
1929 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1930 sys_swapon);
1931 if (error < 0) {
1932 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001933 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001934 }
1935 p->old_block_size = block_size(p->bdev);
1936 error = set_blocksize(p->bdev, PAGE_SIZE);
1937 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001938 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001939 p->flags |= SWP_BLKDEV;
1940 } else if (S_ISREG(inode->i_mode)) {
1941 p->bdev = inode->i_sb->s_bdev;
1942 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001943 if (IS_SWAPFILE(inode))
1944 return -EBUSY;
1945 } else
1946 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001947
1948 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001949}
1950
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001951static unsigned long read_swap_header(struct swap_info_struct *p,
1952 union swap_header *swap_header,
1953 struct inode *inode)
1954{
1955 int i;
1956 unsigned long maxpages;
1957 unsigned long swapfilepages;
1958
1959 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1960 printk(KERN_ERR "Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001961 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001962 }
1963
1964 /* swap partition endianess hack... */
1965 if (swab32(swap_header->info.version) == 1) {
1966 swab32s(&swap_header->info.version);
1967 swab32s(&swap_header->info.last_page);
1968 swab32s(&swap_header->info.nr_badpages);
1969 for (i = 0; i < swap_header->info.nr_badpages; i++)
1970 swab32s(&swap_header->info.badpages[i]);
1971 }
1972 /* Check the swap header's sub-version */
1973 if (swap_header->info.version != 1) {
1974 printk(KERN_WARNING
1975 "Unable to handle swap header version %d\n",
1976 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001977 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001978 }
1979
1980 p->lowest_bit = 1;
1981 p->cluster_next = 1;
1982 p->cluster_nr = 0;
1983
1984 /*
1985 * Find out how many pages are allowed for a single swap
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001986 * device. There are three limiting factors: 1) the number
1987 * of bits for the swap offset in the swp_entry_t type, and
1988 * 2) the number of bits in the swap pte as defined by the
1989 * the different architectures, and 3) the number of free bits
1990 * in an exceptional radix_tree entry. In order to find the
1991 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001992 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001993 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001994 * offset is extracted. This will mask all the bits from
1995 * the initial ~0UL mask that can't be encoded in either
1996 * the swp_entry_t or the architecture definition of a
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001997 * swap pte. Then the same is done for a radix_tree entry.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001998 */
1999 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002000 swp_entry_to_pte(swp_entry(0, ~0UL))));
2001 maxpages = swp_offset(radix_to_swp_entry(
2002 swp_to_radix_entry(swp_entry(0, maxpages)))) + 1;
2003
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002004 if (maxpages > swap_header->info.last_page) {
2005 maxpages = swap_header->info.last_page + 1;
2006 /* p->max is an unsigned int: don't overflow it */
2007 if ((unsigned int)maxpages == 0)
2008 maxpages = UINT_MAX;
2009 }
2010 p->highest_bit = maxpages - 1;
2011
2012 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002013 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002014 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2015 if (swapfilepages && maxpages > swapfilepages) {
2016 printk(KERN_WARNING
2017 "Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002018 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002019 }
2020 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002021 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002022 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002023 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002024
2025 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002026}
2027
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002028static int setup_swap_map_and_extents(struct swap_info_struct *p,
2029 union swap_header *swap_header,
2030 unsigned char *swap_map,
2031 unsigned long maxpages,
2032 sector_t *span)
2033{
2034 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002035 unsigned int nr_good_pages;
2036 int nr_extents;
2037
2038 nr_good_pages = maxpages - 1; /* omit header page */
2039
2040 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2041 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002042 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2043 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002044 if (page_nr < maxpages) {
2045 swap_map[page_nr] = SWAP_MAP_BAD;
2046 nr_good_pages--;
2047 }
2048 }
2049
2050 if (nr_good_pages) {
2051 swap_map[0] = SWAP_MAP_BAD;
2052 p->max = maxpages;
2053 p->pages = nr_good_pages;
2054 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002055 if (nr_extents < 0)
2056 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002057 nr_good_pages = p->pages;
2058 }
2059 if (!nr_good_pages) {
2060 printk(KERN_WARNING "Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002061 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002062 }
2063
2064 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002065}
2066
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002067SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2068{
2069 struct swap_info_struct *p;
Cesar Eduardo Barros28b36bd2011-03-22 16:33:20 -07002070 char *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002071 struct file *swap_file = NULL;
2072 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002073 int i;
2074 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002075 int error;
2076 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002077 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002078 sector_t span;
2079 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002080 unsigned char *swap_map = NULL;
2081 struct page *page = NULL;
2082 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002083
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002084 if (swap_flags & ~SWAP_FLAGS_VALID)
2085 return -EINVAL;
2086
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002087 if (!capable(CAP_SYS_ADMIN))
2088 return -EPERM;
2089
2090 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002091 if (IS_ERR(p))
2092 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002093
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002096 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002098 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 }
2100 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002102 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002104 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 }
2106
2107 p->swap_file = swap_file;
2108 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002111 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002113 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002115 if (mapping == q->swap_file->f_mapping) {
2116 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 }
2120
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002121 inode = mapping->host;
2122 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002123 error = claim_swapfile(p, inode);
2124 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 /*
2128 * Read the swap header.
2129 */
2130 if (!mapping->a_ops->readpage) {
2131 error = -EINVAL;
2132 goto bad_swap;
2133 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002134 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 if (IS_ERR(page)) {
2136 error = PTR_ERR(page);
2137 goto bad_swap;
2138 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002139 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002141 maxpages = read_swap_header(p, swap_header, inode);
2142 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 error = -EINVAL;
2144 goto bad_swap;
2145 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002146
Hugh Dickins81e33972009-01-06 14:39:49 -08002147 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002148 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002149 if (!swap_map) {
2150 error = -ENOMEM;
2151 goto bad_swap;
2152 }
2153
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002154 error = swap_cgroup_swapon(p->type, maxpages);
2155 if (error)
2156 goto bad_swap;
2157
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002158 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2159 maxpages, &span);
2160 if (unlikely(nr_extents < 0)) {
2161 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 goto bad_swap;
2163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002165 if (p->bdev) {
2166 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2167 p->flags |= SWP_SOLIDSTATE;
2168 p->cluster_next = 1 + (random32() % p->highest_bit);
2169 }
Shaohua Li052b1982012-03-21 16:34:17 -07002170 if ((swap_flags & SWAP_FLAG_DISCARD) && discard_swap(p) == 0)
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002171 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002172 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002173
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002174 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002175 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002176 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002177 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002178 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002179 enable_swap_info(p, prio, swap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002180
2181 printk(KERN_INFO "Adding %uk swap on %s. "
2182 "Priority:%d extents:%d across:%lluk %s%s\n",
2183 p->pages<<(PAGE_SHIFT-10), name, p->prio,
2184 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2185 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2186 (p->flags & SWP_DISCARDABLE) ? "D" : "");
2187
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002188 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002189 atomic_inc(&proc_poll_event);
2190 wake_up_interruptible(&proc_poll_wait);
2191
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002192 if (S_ISREG(inode->i_mode))
2193 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 error = 0;
2195 goto out;
2196bad_swap:
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002197 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002198 set_blocksize(p->bdev, p->old_block_size);
2199 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002201 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002202 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002203 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002206 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 vfree(swap_map);
Mel Gorman52c50562011-03-22 16:30:08 -07002208 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002209 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002210 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002211 inode = NULL;
2212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215out:
2216 if (page && !IS_ERR(page)) {
2217 kunmap(page);
2218 page_cache_release(page);
2219 }
2220 if (name)
2221 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002222 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002223 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 return error;
2225}
2226
2227void si_swapinfo(struct sysinfo *val)
2228{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002229 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 unsigned long nr_to_be_unused = 0;
2231
Hugh Dickins5d337b92005-09-03 15:54:41 -07002232 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002233 for (type = 0; type < nr_swapfiles; type++) {
2234 struct swap_info_struct *si = swap_info[type];
2235
2236 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2237 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 }
Shaohua Li1abbca72013-02-22 16:34:38 -08002239 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002241 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242}
2243
2244/*
2245 * Verify that a swap entry is valid and increment its swap map count.
2246 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002247 * Returns error code in following case.
2248 * - success -> 0
2249 * - swp_entry is invalid -> EINVAL
2250 * - swp_entry is migration entry -> EINVAL
2251 * - swap-cache reference is requested but there is already one. -> EEXIST
2252 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a3352009-12-14 17:58:46 -08002253 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002255static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002257 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002259 unsigned char count;
2260 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002261 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262
Andi Kleena7420aa2009-09-16 11:50:05 +02002263 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002264 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 type = swp_type(entry);
2267 if (type >= nr_swapfiles)
2268 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002269 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 offset = swp_offset(entry);
2271
Shaohua Li1abbca72013-02-22 16:34:38 -08002272 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002273 if (unlikely(offset >= p->max))
2274 goto unlock_out;
2275
Hugh Dickins253d5532009-12-14 17:58:44 -08002276 count = p->swap_map[offset];
2277 has_cache = count & SWAP_HAS_CACHE;
2278 count &= ~SWAP_HAS_CACHE;
2279 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002280
Hugh Dickins253d5532009-12-14 17:58:44 -08002281 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002282
2283 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002284 if (!has_cache && count)
2285 has_cache = SWAP_HAS_CACHE;
2286 else if (has_cache) /* someone else added cache */
2287 err = -EEXIST;
2288 else /* no users remaining */
2289 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002290
2291 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002292
Hugh Dickins570a3352009-12-14 17:58:46 -08002293 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2294 count += usage;
2295 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002296 err = -EINVAL;
Hugh Dickins570a3352009-12-14 17:58:46 -08002297 else if (swap_count_continued(p, offset, count))
2298 count = COUNT_CONTINUED;
2299 else
2300 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002301 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002302 err = -ENOENT; /* unused swap entry */
2303
2304 p->swap_map[offset] = count | has_cache;
2305
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002306unlock_out:
Shaohua Li1abbca72013-02-22 16:34:38 -08002307 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002309 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
2311bad_file:
2312 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2313 goto out;
2314}
Hugh Dickins253d5532009-12-14 17:58:44 -08002315
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002316/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002317 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2318 * (in which case its reference count is never incremented).
2319 */
2320void swap_shmem_alloc(swp_entry_t entry)
2321{
2322 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2323}
2324
2325/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002326 * Increase reference count of swap entry by 1.
2327 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2328 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2329 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2330 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002331 */
Hugh Dickins570a3352009-12-14 17:58:46 -08002332int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002333{
Hugh Dickins570a3352009-12-14 17:58:46 -08002334 int err = 0;
2335
2336 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2337 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2338 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002339}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002341/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002342 * @entry: swap entry for which we allocate swap cache.
2343 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002344 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002345 * This can return error codes. Returns 0 at success.
2346 * -EBUSY means there is a swap cache.
2347 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002348 */
2349int swapcache_prepare(swp_entry_t entry)
2350{
Hugh Dickins253d5532009-12-14 17:58:44 -08002351 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002352}
2353
Mel Gormanfdd4d472012-07-31 16:44:47 -07002354struct swap_info_struct *page_swap_info(struct page *page)
2355{
2356 swp_entry_t swap = { .val = page_private(page) };
2357 BUG_ON(!PageSwapCache(page));
2358 return swap_info[swp_type(swap)];
2359}
2360
2361/*
2362 * out-of-line __page_file_ methods to avoid include hell.
2363 */
2364struct address_space *__page_file_mapping(struct page *page)
2365{
2366 VM_BUG_ON(!PageSwapCache(page));
2367 return page_swap_info(page)->swap_file->f_mapping;
2368}
2369EXPORT_SYMBOL_GPL(__page_file_mapping);
2370
2371pgoff_t __page_file_index(struct page *page)
2372{
2373 swp_entry_t swap = { .val = page_private(page) };
2374 VM_BUG_ON(!PageSwapCache(page));
2375 return swp_offset(swap);
2376}
2377EXPORT_SYMBOL_GPL(__page_file_index);
2378
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379/*
Hugh Dickins570a3352009-12-14 17:58:46 -08002380 * add_swap_count_continuation - called when a swap count is duplicated
2381 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2382 * page of the original vmalloc'ed swap_map, to hold the continuation count
2383 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2384 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2385 *
2386 * These continuation pages are seldom referenced: the common paths all work
2387 * on the original swap_map, only referring to a continuation page when the
2388 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2389 *
2390 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2391 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2392 * can be called after dropping locks.
2393 */
2394int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2395{
2396 struct swap_info_struct *si;
2397 struct page *head;
2398 struct page *page;
2399 struct page *list_page;
2400 pgoff_t offset;
2401 unsigned char count;
2402
2403 /*
2404 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2405 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2406 */
2407 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2408
2409 si = swap_info_get(entry);
2410 if (!si) {
2411 /*
2412 * An acceptable race has occurred since the failing
2413 * __swap_duplicate(): the swap entry has been freed,
2414 * perhaps even the whole swap_map cleared for swapoff.
2415 */
2416 goto outer;
2417 }
2418
2419 offset = swp_offset(entry);
2420 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2421
2422 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2423 /*
2424 * The higher the swap count, the more likely it is that tasks
2425 * will race to add swap count continuation: we need to avoid
2426 * over-provisioning.
2427 */
2428 goto out;
2429 }
2430
2431 if (!page) {
Shaohua Li1abbca72013-02-22 16:34:38 -08002432 spin_unlock(&si->lock);
Hugh Dickins570a3352009-12-14 17:58:46 -08002433 return -ENOMEM;
2434 }
2435
2436 /*
2437 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2438 * no architecture is using highmem pages for kernel pagetables: so it
2439 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2440 */
2441 head = vmalloc_to_page(si->swap_map + offset);
2442 offset &= ~PAGE_MASK;
2443
2444 /*
2445 * Page allocation does not initialize the page's lru field,
2446 * but it does always reset its private field.
2447 */
2448 if (!page_private(head)) {
2449 BUG_ON(count & COUNT_CONTINUED);
2450 INIT_LIST_HEAD(&head->lru);
2451 set_page_private(head, SWP_CONTINUED);
2452 si->flags |= SWP_CONTINUED;
2453 }
2454
2455 list_for_each_entry(list_page, &head->lru, lru) {
2456 unsigned char *map;
2457
2458 /*
2459 * If the previous map said no continuation, but we've found
2460 * a continuation page, free our allocation and use this one.
2461 */
2462 if (!(count & COUNT_CONTINUED))
2463 goto out;
2464
Cong Wang9b04c5f2011-11-25 23:14:39 +08002465 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a3352009-12-14 17:58:46 -08002466 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002467 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002468
2469 /*
2470 * If this continuation count now has some space in it,
2471 * free our allocation and use this one.
2472 */
2473 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2474 goto out;
2475 }
2476
2477 list_add_tail(&page->lru, &head->lru);
2478 page = NULL; /* now it's attached, don't free it */
2479out:
Shaohua Li1abbca72013-02-22 16:34:38 -08002480 spin_unlock(&si->lock);
Hugh Dickins570a3352009-12-14 17:58:46 -08002481outer:
2482 if (page)
2483 __free_page(page);
2484 return 0;
2485}
2486
2487/*
2488 * swap_count_continued - when the original swap_map count is incremented
2489 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2490 * into, carry if so, or else fail until a new continuation page is allocated;
2491 * when the original swap_map count is decremented from 0 with continuation,
2492 * borrow from the continuation and report whether it still holds more.
2493 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2494 */
2495static bool swap_count_continued(struct swap_info_struct *si,
2496 pgoff_t offset, unsigned char count)
2497{
2498 struct page *head;
2499 struct page *page;
2500 unsigned char *map;
2501
2502 head = vmalloc_to_page(si->swap_map + offset);
2503 if (page_private(head) != SWP_CONTINUED) {
2504 BUG_ON(count & COUNT_CONTINUED);
2505 return false; /* need to add count continuation */
2506 }
2507
2508 offset &= ~PAGE_MASK;
2509 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002510 map = kmap_atomic(page) + offset;
Hugh Dickins570a3352009-12-14 17:58:46 -08002511
2512 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2513 goto init_map; /* jump over SWAP_CONT_MAX checks */
2514
2515 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2516 /*
2517 * Think of how you add 1 to 999
2518 */
2519 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002520 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002521 page = list_entry(page->lru.next, struct page, lru);
2522 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002523 map = kmap_atomic(page) + offset;
Hugh Dickins570a3352009-12-14 17:58:46 -08002524 }
2525 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002526 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002527 page = list_entry(page->lru.next, struct page, lru);
2528 if (page == head)
2529 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002530 map = kmap_atomic(page) + offset;
Hugh Dickins570a3352009-12-14 17:58:46 -08002531init_map: *map = 0; /* we didn't zero the page */
2532 }
2533 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002534 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002535 page = list_entry(page->lru.prev, struct page, lru);
2536 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002537 map = kmap_atomic(page) + offset;
Hugh Dickins570a3352009-12-14 17:58:46 -08002538 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002539 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002540 page = list_entry(page->lru.prev, struct page, lru);
2541 }
2542 return true; /* incremented */
2543
2544 } else { /* decrementing */
2545 /*
2546 * Think of how you subtract 1 from 1000
2547 */
2548 BUG_ON(count != COUNT_CONTINUED);
2549 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002550 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002551 page = list_entry(page->lru.next, struct page, lru);
2552 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002553 map = kmap_atomic(page) + offset;
Hugh Dickins570a3352009-12-14 17:58:46 -08002554 }
2555 BUG_ON(*map == 0);
2556 *map -= 1;
2557 if (*map == 0)
2558 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002559 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002560 page = list_entry(page->lru.prev, struct page, lru);
2561 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002562 map = kmap_atomic(page) + offset;
Hugh Dickins570a3352009-12-14 17:58:46 -08002563 *map = SWAP_CONT_MAX | count;
2564 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002565 kunmap_atomic(map);
Hugh Dickins570a3352009-12-14 17:58:46 -08002566 page = list_entry(page->lru.prev, struct page, lru);
2567 }
2568 return count == COUNT_CONTINUED;
2569 }
2570}
2571
2572/*
2573 * free_swap_count_continuations - swapoff free all the continuation pages
2574 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2575 */
2576static void free_swap_count_continuations(struct swap_info_struct *si)
2577{
2578 pgoff_t offset;
2579
2580 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2581 struct page *head;
2582 head = vmalloc_to_page(si->swap_map + offset);
2583 if (page_private(head)) {
2584 struct list_head *this, *next;
2585 list_for_each_safe(this, next, &head->lru) {
2586 struct page *page;
2587 page = list_entry(this, struct page, lru);
2588 list_del(this);
2589 __free_page(page);
2590 }
2591 }
2592 }
2593}