blob: 71cd288b200179d4962222337f3947342fcebe57 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/mm.h>
9#include <linux/hugetlb.h>
10#include <linux/mman.h>
11#include <linux/slab.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/vmalloc.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
Hugh Dickins072441e2011-06-27 16:18:02 -070017#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080019#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080024#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070032#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070033#include <linux/oom.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060034#include <linux/frontswap.h>
35#include <linux/swapfile.h>
Mel Gormanf981c592012-07-31 16:44:47 -070036#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/pgtable.h>
39#include <asm/tlbflush.h>
40#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080041#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Hugh Dickins570a335b2009-12-14 17:58:46 -080043static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
44 unsigned char);
45static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080046static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080047
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060048DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070049static unsigned int nr_swapfiles;
Hugh Dickinsb9627162009-01-06 14:39:41 -080050long nr_swap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070052static int least_priority;
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
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060059struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060061struct 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 Dickins570a335b2009-12-14 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 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700226 spin_unlock(&swap_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) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700245 spin_lock(&swap_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) {
266 spin_lock(&swap_lock);
267 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;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700280 spin_lock(&swap_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;
296 spin_unlock(&swap_lock);
297 swap_was_freed = __try_to_reclaim_swap(si, offset);
298 spin_lock(&swap_lock);
299 /* 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;
338 spin_unlock(&swap_lock);
339
340 if (offset < last_in_cluster)
341 discard_swap_cluster(si, offset,
342 last_in_cluster - offset + 1);
343
344 spin_lock(&swap_lock);
345 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 */
358 spin_unlock(&swap_lock);
359 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
360 wait_for_discard, TASK_UNINTERRUPTIBLE);
361 spin_lock(&swap_lock);
362 } 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:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700377 spin_unlock(&swap_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]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700380 spin_lock(&swap_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) {
384 spin_lock(&swap_lock);
385 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]) {
395 spin_lock(&swap_lock);
396 goto checks;
397 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700398 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
399 spin_lock(&swap_lock);
400 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 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700407 spin_lock(&swap_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Hugh Dickins5d337b92005-09-03 15:54:41 -0700421 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700423 goto noswap;
424 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700426 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800427 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700428 next = si->next;
429 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800430 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700431 next = swap_list.head;
432 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700434
435 if (!si->highest_bit)
436 continue;
437 if (!(si->flags & SWP_WRITEOK))
438 continue;
439
440 swap_list.next = next;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700441 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800442 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700443 if (offset) {
444 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700445 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700446 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700447 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700449
450 nr_swap_pages++;
451noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700452 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700453 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
Hugh Dickins910321e2010-09-09 16:38:07 -0700456/* The only caller of this function is now susupend routine */
457swp_entry_t get_swap_page_of_type(int type)
458{
459 struct swap_info_struct *si;
460 pgoff_t offset;
461
462 spin_lock(&swap_lock);
463 si = swap_info[type];
464 if (si && (si->flags & SWP_WRITEOK)) {
465 nr_swap_pages--;
466 /* This is called for allocating swap entry, not cache */
467 offset = scan_swap_map(si, 1);
468 if (offset) {
469 spin_unlock(&swap_lock);
470 return swp_entry(type, offset);
471 }
472 nr_swap_pages++;
473 }
474 spin_unlock(&swap_lock);
475 return (swp_entry_t) {0};
476}
477
Hugh Dickins73c34b62009-12-14 17:58:43 -0800478static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800480 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 unsigned long offset, type;
482
483 if (!entry.val)
484 goto out;
485 type = swp_type(entry);
486 if (type >= nr_swapfiles)
487 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800488 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (!(p->flags & SWP_USED))
490 goto bad_device;
491 offset = swp_offset(entry);
492 if (offset >= p->max)
493 goto bad_offset;
494 if (!p->swap_map[offset])
495 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700496 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 return p;
498
499bad_free:
500 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
501 goto out;
502bad_offset:
503 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
504 goto out;
505bad_device:
506 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
507 goto out;
508bad_nofile:
509 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
510out:
511 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800512}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800514static unsigned char swap_entry_free(struct swap_info_struct *p,
515 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Hugh Dickins253d5532009-12-14 17:58:44 -0800517 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800518 unsigned char count;
519 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700521 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800522 has_cache = count & SWAP_HAS_CACHE;
523 count &= ~SWAP_HAS_CACHE;
524
525 if (usage == SWAP_HAS_CACHE) {
526 VM_BUG_ON(!has_cache);
527 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800528 } else if (count == SWAP_MAP_SHMEM) {
529 /*
530 * Or we could insist on shmem.c using a special
531 * swap_shmem_free() and free_shmem_swap_and_cache()...
532 */
533 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800534 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
535 if (count == COUNT_CONTINUED) {
536 if (swap_count_continued(p, offset, count))
537 count = SWAP_MAP_MAX | COUNT_CONTINUED;
538 else
539 count = SWAP_MAP_MAX;
540 } else
541 count--;
542 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800543
544 if (!count)
545 mem_cgroup_uncharge_swap(entry);
546
547 usage = count | has_cache;
548 p->swap_map[offset] = usage;
549
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700550 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800551 if (!usage) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700552 if (offset < p->lowest_bit)
553 p->lowest_bit = offset;
554 if (offset > p->highest_bit)
555 p->highest_bit = offset;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800556 if (swap_list.next >= 0 &&
557 p->prio > swap_info[swap_list.next]->prio)
558 swap_list.next = p->type;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700559 nr_swap_pages++;
560 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600561 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700562 if (p->flags & SWP_BLKDEV) {
563 struct gendisk *disk = p->bdev->bd_disk;
564 if (disk->fops->swap_slot_free_notify)
565 disk->fops->swap_slot_free_notify(p->bdev,
566 offset);
567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800569
570 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571}
572
573/*
574 * Caller has made sure that the swapdevice corresponding to entry
575 * is still around or has not been recycled.
576 */
577void swap_free(swp_entry_t entry)
578{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800579 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 p = swap_info_get(entry);
582 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800583 swap_entry_free(p, entry, 1);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700584 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 }
586}
587
588/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700589 * Called after dropping swapcache to decrease refcnt to swap entries.
590 */
591void swapcache_free(swp_entry_t entry, struct page *page)
592{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700593 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800594 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700595
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700596 p = swap_info_get(entry);
597 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800598 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
599 if (page)
600 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700601 spin_unlock(&swap_lock);
602 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700603}
604
605/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700606 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800607 * This does not give an exact answer when swap count is continued,
608 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700610int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700612 int count = 0;
613 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 swp_entry_t entry;
615
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700616 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 p = swap_info_get(entry);
618 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700619 count = swap_count(p->swap_map[swp_offset(entry)]);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700620 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700622 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623}
624
625/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800626 * We can write to an anon page without COW if there are no other references
627 * to it. And as a side-effect, free up its swap: because the old content
628 * on disk will never be read, and seeking back there to write new content
629 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800631int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700633 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Hugh Dickins51726b12009-01-06 14:39:25 -0800635 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800636 if (unlikely(PageKsm(page)))
637 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700638 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800639 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700640 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800641 if (count == 1 && !PageWriteback(page)) {
642 delete_from_swap_cache(page);
643 SetPageDirty(page);
644 }
645 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800646 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800650 * If swap is getting full, or if there are no more mappings of this page,
651 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800653int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
Hugh Dickins51726b12009-01-06 14:39:25 -0800655 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 if (!PageSwapCache(page))
658 return 0;
659 if (PageWriteback(page))
660 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800661 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return 0;
663
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700664 /*
665 * Once hibernation has begun to create its image of memory,
666 * there's a danger that one of the calls to try_to_free_swap()
667 * - most probably a call from __try_to_reclaim_swap() while
668 * hibernation is allocating its own swap pages for the image,
669 * but conceivably even a call from memory reclaim - will free
670 * the swap from a page which has already been recorded in the
671 * image as a clean swapcache page, and then reuse its swap for
672 * another page of the image. On waking from hibernation, the
673 * original page might be freed under memory pressure, then
674 * later read back in from swap, now with the wrong data.
675 *
Mel Gormanf90ac392012-01-10 15:07:15 -0800676 * Hibration suspends storage while it is writing the image
677 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700678 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800679 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700680 return 0;
681
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800682 delete_from_swap_cache(page);
683 SetPageDirty(page);
684 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700685}
686
687/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 * Free the swap entry like above, but also try to
689 * free the page cache entry if it is the last user.
690 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800691int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800693 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 struct page *page = NULL;
695
Andi Kleena7420aa2009-09-16 11:50:05 +0200696 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800697 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 p = swap_info_get(entry);
700 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800701 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Nick Piggin93fac702006-03-31 02:29:56 -0800702 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700703 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800704 page_cache_release(page);
705 page = NULL;
706 }
707 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700708 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
710 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800711 /*
712 * Not mapped elsewhere, or swap space full? Free it!
713 * Also recheck PageSwapCache now page is locked (above).
714 */
Nick Piggin93fac702006-03-31 02:29:56 -0800715 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800716 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 delete_from_swap_cache(page);
718 SetPageDirty(page);
719 }
720 unlock_page(page);
721 page_cache_release(page);
722 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800723 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
725
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200726#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800727/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800728 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800729 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800730 * @offset - number of the PAGE_SIZE-sized block of the device, starting
731 * from 0, in which the swap header is expected to be located.
732 *
733 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800734 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800735int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800736{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800737 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800738 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800739
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800740 if (device)
741 bdev = bdget(device);
742
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800743 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800744 for (type = 0; type < nr_swapfiles; type++) {
745 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800746
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800747 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800748 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700749
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800750 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800751 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200752 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800753
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800754 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800755 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800756 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800757 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800758 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800759
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800760 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800761 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200762 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800763
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800764 spin_unlock(&swap_lock);
765 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800766 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800767 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800768 }
769 }
770 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800771 if (bdev)
772 bdput(bdev);
773
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800774 return -ENODEV;
775}
776
777/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800778 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
779 * corresponding to given index in swap_info (swap type).
780 */
781sector_t swapdev_block(int type, pgoff_t offset)
782{
783 struct block_device *bdev;
784
785 if ((unsigned int)type >= nr_swapfiles)
786 return 0;
787 if (!(swap_info[type]->flags & SWP_WRITEOK))
788 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800789 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800790}
791
792/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800793 * Return either the total number of swap pages of given type, or the number
794 * of free pages of that type (depending on @free)
795 *
796 * This is needed for software suspend
797 */
798unsigned int count_swap_pages(int type, int free)
799{
800 unsigned int n = 0;
801
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800802 spin_lock(&swap_lock);
803 if ((unsigned int)type < nr_swapfiles) {
804 struct swap_info_struct *sis = swap_info[type];
805
806 if (sis->flags & SWP_WRITEOK) {
807 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800808 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800809 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800810 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800811 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800812 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800813 return n;
814}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800815#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700818 * No need to decide whether this PTE shares the swap entry with others,
819 * just let do_wp_page work it out if a write is requested later - to
820 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800822static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 unsigned long addr, swp_entry_t entry, struct page *page)
824{
Johannes Weiner72835c82012-01-12 17:18:32 -0800825 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800826 spinlock_t *ptl;
827 pte_t *pte;
828 int ret = 1;
829
Johannes Weiner72835c82012-01-12 17:18:32 -0800830 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
831 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800832 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800833 goto out_nolock;
834 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800835
836 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
837 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
Johannes Weiner5d84c772012-07-31 16:45:28 -0700838 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800839 ret = 0;
840 goto out;
841 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800842
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800843 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800844 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 get_page(page);
846 set_pte_at(vma->vm_mm, addr, pte,
847 pte_mkold(mk_pte(page, vma->vm_page_prot)));
848 page_add_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -0800849 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 swap_free(entry);
851 /*
852 * Move the page to the active list so it is not
853 * immediately swapped out again after swapon.
854 */
855 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800856out:
857 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800858out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800859 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
861
862static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
863 unsigned long addr, unsigned long end,
864 swp_entry_t entry, struct page *page)
865{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700867 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800868 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
Hugh Dickins044d66c2008-02-07 00:14:04 -0800870 /*
871 * We don't actually need pte lock while scanning for swp_pte: since
872 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
873 * page table while we're scanning; though it could get zapped, and on
874 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
875 * of unmatched parts which look like swp_pte, so unuse_pte must
876 * recheck under pte lock. Scanning without pte lock lets it be
877 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
878 */
879 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 do {
881 /*
882 * swapoff spends a _lot_ of time in this loop!
883 * Test inline before going to call unuse_pte.
884 */
885 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800886 pte_unmap(pte);
887 ret = unuse_pte(vma, pmd, addr, entry, page);
888 if (ret)
889 goto out;
890 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 }
892 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800893 pte_unmap(pte - 1);
894out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800895 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
897
898static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
899 unsigned long addr, unsigned long end,
900 swp_entry_t entry, struct page *page)
901{
902 pmd_t *pmd;
903 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800904 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 pmd = pmd_offset(pud, addr);
907 do {
908 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700909 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800911 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
912 if (ret)
913 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 } while (pmd++, addr = next, addr != end);
915 return 0;
916}
917
918static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
919 unsigned long addr, unsigned long end,
920 swp_entry_t entry, struct page *page)
921{
922 pud_t *pud;
923 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800924 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926 pud = pud_offset(pgd, addr);
927 do {
928 next = pud_addr_end(addr, end);
929 if (pud_none_or_clear_bad(pud))
930 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800931 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
932 if (ret)
933 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 } while (pud++, addr = next, addr != end);
935 return 0;
936}
937
938static int unuse_vma(struct vm_area_struct *vma,
939 swp_entry_t entry, struct page *page)
940{
941 pgd_t *pgd;
942 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800943 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800945 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 addr = page_address_in_vma(page, vma);
947 if (addr == -EFAULT)
948 return 0;
949 else
950 end = addr + PAGE_SIZE;
951 } else {
952 addr = vma->vm_start;
953 end = vma->vm_end;
954 }
955
956 pgd = pgd_offset(vma->vm_mm, addr);
957 do {
958 next = pgd_addr_end(addr, end);
959 if (pgd_none_or_clear_bad(pgd))
960 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800961 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
962 if (ret)
963 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 } while (pgd++, addr = next, addr != end);
965 return 0;
966}
967
968static int unuse_mm(struct mm_struct *mm,
969 swp_entry_t entry, struct page *page)
970{
971 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800972 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
974 if (!down_read_trylock(&mm->mmap_sem)) {
975 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -0700976 * Activate page so shrink_inactive_list is unlikely to unmap
977 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700979 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 unlock_page(page);
981 down_read(&mm->mmap_sem);
982 lock_page(page);
983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800985 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 break;
987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800989 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990}
991
992/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600993 * Scan swap_map (or frontswap_map if frontswap parameter is true)
994 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 * Recycle to start on reaching the end, returning 0 when empty.
996 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700997static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600998 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001000 unsigned int max = si->max;
1001 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001002 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001005 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 * for whether an entry is in use, not modifying it; false
1007 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001008 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 */
1010 for (;;) {
1011 if (++i >= max) {
1012 if (!prev) {
1013 i = 0;
1014 break;
1015 }
1016 /*
1017 * No entries in use at top of swap_map,
1018 * loop back to start and recheck there.
1019 */
1020 max = prev + 1;
1021 prev = 0;
1022 i = 1;
1023 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001024 if (frontswap) {
1025 if (frontswap_test(si, i))
1026 break;
1027 else
1028 continue;
1029 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001031 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 break;
1033 }
1034 return i;
1035}
1036
1037/*
1038 * We completely avoid races by reading each swap page in advance,
1039 * and then search for the process using it. All the necessary
1040 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001041 *
1042 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1043 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001045int try_to_unuse(unsigned int type, bool frontswap,
1046 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001048 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001050 unsigned char *swap_map;
1051 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 struct page *page;
1053 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001054 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057 /*
1058 * When searching mms for an entry, a good strategy is to
1059 * start at the first mm we freed the previous entry from
1060 * (though actually we don't notice whether we or coincidence
1061 * freed the entry). Initialize this start_mm with a hold.
1062 *
1063 * A simpler strategy would be to start at the last mm we
1064 * freed the previous entry from; but that would take less
1065 * advantage of mmlist ordering, which clusters forked mms
1066 * together, child after parent. If we race with dup_mmap(), we
1067 * prefer to resolve parent before child, lest we miss entries
1068 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001069 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 */
1071 start_mm = &init_mm;
1072 atomic_inc(&init_mm.mm_users);
1073
1074 /*
1075 * Keep on scanning until all entries have gone. Usually,
1076 * one pass through swap_map is enough, but not necessarily:
1077 * there are races when an instance of an entry might be missed.
1078 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001079 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 if (signal_pending(current)) {
1081 retval = -EINTR;
1082 break;
1083 }
1084
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001085 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 * Get a page for the entry, using the existing swap
1087 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001088 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 */
1090 swap_map = &si->swap_map[i];
1091 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001092 page = read_swap_cache_async(entry,
1093 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (!page) {
1095 /*
1096 * Either swap_duplicate() failed because entry
1097 * has been freed independently, and will not be
1098 * reused since sys_swapoff() already disabled
1099 * allocation from here, or alloc_page() failed.
1100 */
1101 if (!*swap_map)
1102 continue;
1103 retval = -ENOMEM;
1104 break;
1105 }
1106
1107 /*
1108 * Don't hold on to start_mm if it looks like exiting.
1109 */
1110 if (atomic_read(&start_mm->mm_users) == 1) {
1111 mmput(start_mm);
1112 start_mm = &init_mm;
1113 atomic_inc(&init_mm.mm_users);
1114 }
1115
1116 /*
1117 * Wait for and lock page. When do_swap_page races with
1118 * try_to_unuse, do_swap_page can handle the fault much
1119 * faster than try_to_unuse can locate the entry. This
1120 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1121 * defer to do_swap_page in such a case - in some tests,
1122 * do_swap_page and try_to_unuse repeatedly compete.
1123 */
1124 wait_on_page_locked(page);
1125 wait_on_page_writeback(page);
1126 lock_page(page);
1127 wait_on_page_writeback(page);
1128
1129 /*
1130 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001133 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1134 retval = shmem_unuse(entry, page);
1135 /* page has already been unlocked and released */
1136 if (retval < 0)
1137 break;
1138 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001140 if (swap_count(swcount) && start_mm != &init_mm)
1141 retval = unuse_mm(start_mm, entry, page);
1142
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001143 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 int set_start_mm = (*swap_map >= swcount);
1145 struct list_head *p = &start_mm->mmlist;
1146 struct mm_struct *new_start_mm = start_mm;
1147 struct mm_struct *prev_mm = start_mm;
1148 struct mm_struct *mm;
1149
1150 atomic_inc(&new_start_mm->mm_users);
1151 atomic_inc(&prev_mm->mm_users);
1152 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001153 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 (p = p->next) != &start_mm->mmlist) {
1155 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001156 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 spin_unlock(&mmlist_lock);
1159 mmput(prev_mm);
1160 prev_mm = mm;
1161
1162 cond_resched();
1163
1164 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001165 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001167 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001169 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001171
Bo Liu32c5fc12009-11-02 16:50:33 +00001172 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 mmput(new_start_mm);
1174 atomic_inc(&mm->mm_users);
1175 new_start_mm = mm;
1176 set_start_mm = 0;
1177 }
1178 spin_lock(&mmlist_lock);
1179 }
1180 spin_unlock(&mmlist_lock);
1181 mmput(prev_mm);
1182 mmput(start_mm);
1183 start_mm = new_start_mm;
1184 }
1185 if (retval) {
1186 unlock_page(page);
1187 page_cache_release(page);
1188 break;
1189 }
1190
1191 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 * If a reference remains (rare), we would like to leave
1193 * the page in the swap cache; but try_to_unmap could
1194 * then re-duplicate the entry once we drop page lock,
1195 * so we might loop indefinitely; also, that page could
1196 * not be swapped out to other storage meanwhile. So:
1197 * delete from cache even if there's another reference,
1198 * after ensuring that the data has been saved to disk -
1199 * since if the reference remains (rarer), it will be
1200 * read from disk into another page. Splitting into two
1201 * pages would be incorrect if swap supported "shared
1202 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001203 *
1204 * Given how unuse_vma() targets one particular offset
1205 * in an anon_vma, once the anon_vma has been determined,
1206 * this splitting happens to be just what is needed to
1207 * handle where KSM pages have been swapped out: re-reading
1208 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001210 if (swap_count(*swap_map) &&
1211 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 struct writeback_control wbc = {
1213 .sync_mode = WB_SYNC_NONE,
1214 };
1215
1216 swap_writepage(page, &wbc);
1217 lock_page(page);
1218 wait_on_page_writeback(page);
1219 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001220
1221 /*
1222 * It is conceivable that a racing task removed this page from
1223 * swap cache just before we acquired the page lock at the top,
1224 * or while we dropped it in unuse_mm(). The page might even
1225 * be back in swap cache on another swap area: that we must not
1226 * delete, since it may not have been written out to swap yet.
1227 */
1228 if (PageSwapCache(page) &&
1229 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001230 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
1232 /*
1233 * So we could skip searching mms once swap count went
1234 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001235 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 */
1237 SetPageDirty(page);
1238 unlock_page(page);
1239 page_cache_release(page);
1240
1241 /*
1242 * Make sure that we aren't completely killing
1243 * interactive performance.
1244 */
1245 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001246 if (frontswap && pages_to_unuse > 0) {
1247 if (!--pages_to_unuse)
1248 break;
1249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
1251
1252 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 return retval;
1254}
1255
1256/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001257 * After a successful try_to_unuse, if no swap is now in use, we know
1258 * we can empty the mmlist. swap_lock must be held on entry and exit.
1259 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 * added to the mmlist just after page_duplicate - before would be racy.
1261 */
1262static void drain_mmlist(void)
1263{
1264 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001265 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001267 for (type = 0; type < nr_swapfiles; type++)
1268 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 return;
1270 spin_lock(&mmlist_lock);
1271 list_for_each_safe(p, next, &init_mm.mmlist)
1272 list_del_init(p);
1273 spin_unlock(&mmlist_lock);
1274}
1275
1276/*
1277 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001278 * corresponds to page offset for the specified swap entry.
1279 * Note that the type of this function is sector_t, but it returns page offset
1280 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001282static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001284 struct swap_info_struct *sis;
1285 struct swap_extent *start_se;
1286 struct swap_extent *se;
1287 pgoff_t offset;
1288
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001289 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001290 *bdev = sis->bdev;
1291
1292 offset = swp_offset(entry);
1293 start_se = sis->curr_swap_extent;
1294 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 for ( ; ; ) {
1297 struct list_head *lh;
1298
1299 if (se->start_page <= offset &&
1300 offset < (se->start_page + se->nr_pages)) {
1301 return se->start_block + (offset - se->start_page);
1302 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001303 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 se = list_entry(lh, struct swap_extent, list);
1305 sis->curr_swap_extent = se;
1306 BUG_ON(se == start_se); /* It *must* be present */
1307 }
1308}
1309
1310/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001311 * Returns the page offset into bdev for the specified page's swap entry.
1312 */
1313sector_t map_swap_page(struct page *page, struct block_device **bdev)
1314{
1315 swp_entry_t entry;
1316 entry.val = page_private(page);
1317 return map_swap_entry(entry, bdev);
1318}
1319
1320/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 * Free all of a swapdev's extent information
1322 */
1323static void destroy_swap_extents(struct swap_info_struct *sis)
1324{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001325 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 struct swap_extent *se;
1327
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001328 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 struct swap_extent, list);
1330 list_del(&se->list);
1331 kfree(se);
1332 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001333
1334 if (sis->flags & SWP_FILE) {
1335 struct file *swap_file = sis->swap_file;
1336 struct address_space *mapping = swap_file->f_mapping;
1337
1338 sis->flags &= ~SWP_FILE;
1339 mapping->a_ops->swap_deactivate(swap_file);
1340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341}
1342
1343/*
1344 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001345 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001347 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 */
Mel Gormana509bc12012-07-31 16:44:57 -07001349int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1351 unsigned long nr_pages, sector_t start_block)
1352{
1353 struct swap_extent *se;
1354 struct swap_extent *new_se;
1355 struct list_head *lh;
1356
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001357 if (start_page == 0) {
1358 se = &sis->first_swap_extent;
1359 sis->curr_swap_extent = se;
1360 se->start_page = 0;
1361 se->nr_pages = nr_pages;
1362 se->start_block = start_block;
1363 return 1;
1364 } else {
1365 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001367 BUG_ON(se->start_page + se->nr_pages != start_page);
1368 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 /* Merge it */
1370 se->nr_pages += nr_pages;
1371 return 0;
1372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 }
1374
1375 /*
1376 * No merge. Insert a new extent, preserving ordering.
1377 */
1378 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1379 if (new_se == NULL)
1380 return -ENOMEM;
1381 new_se->start_page = start_page;
1382 new_se->nr_pages = nr_pages;
1383 new_se->start_block = start_block;
1384
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001385 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001386 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
1389/*
1390 * A `swap extent' is a simple thing which maps a contiguous range of pages
1391 * onto a contiguous range of disk blocks. An ordered list of swap extents
1392 * is built at swapon time and is then used at swap_writepage/swap_readpage
1393 * time for locating where on disk a page belongs.
1394 *
1395 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1396 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1397 * swap files identically.
1398 *
1399 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1400 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1401 * swapfiles are handled *identically* after swapon time.
1402 *
1403 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1404 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1405 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1406 * requirements, they are simply tossed out - we will never use those blocks
1407 * for swapping.
1408 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001409 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1411 * which will scribble on the fs.
1412 *
1413 * The amount of disk space which a single swap extent represents varies.
1414 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1415 * extents in the list. To avoid much list walking, we cache the previous
1416 * search location in `curr_swap_extent', and start new searches from there.
1417 * This is extremely effective. The average number of iterations in
1418 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1419 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001420static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421{
Mel Gorman62c230b2012-07-31 16:44:55 -07001422 struct file *swap_file = sis->swap_file;
1423 struct address_space *mapping = swap_file->f_mapping;
1424 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 int ret;
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (S_ISBLK(inode->i_mode)) {
1428 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001429 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001430 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 }
1432
Mel Gorman62c230b2012-07-31 16:44:55 -07001433 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001434 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001435 if (!ret) {
1436 sis->flags |= SWP_FILE;
1437 ret = add_swap_extent(sis, 0, sis->max, 0);
1438 *span = sis->pages;
1439 }
Mel Gormana509bc12012-07-31 16:44:57 -07001440 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001441 }
1442
Mel Gormana509bc12012-07-31 16:44:57 -07001443 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444}
1445
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001446static void enable_swap_info(struct swap_info_struct *p, int prio,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001447 unsigned char *swap_map,
1448 unsigned long *frontswap_map)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001449{
1450 int i, prev;
1451
1452 spin_lock(&swap_lock);
1453 if (prio >= 0)
1454 p->prio = prio;
1455 else
1456 p->prio = --least_priority;
1457 p->swap_map = swap_map;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001458 frontswap_map_set(p, frontswap_map);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001459 p->flags |= SWP_WRITEOK;
1460 nr_swap_pages += p->pages;
1461 total_swap_pages += p->pages;
1462
1463 /* insert swap space into swap_list: */
1464 prev = -1;
1465 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1466 if (p->prio >= swap_info[i]->prio)
1467 break;
1468 prev = i;
1469 }
1470 p->next = i;
1471 if (prev < 0)
1472 swap_list.head = swap_list.next = p->type;
1473 else
1474 swap_info[prev]->next = p->type;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001475 frontswap_init(p->type);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001476 spin_unlock(&swap_lock);
1477}
1478
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001479SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001481 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001482 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 struct file *swap_file, *victim;
1484 struct address_space *mapping;
1485 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001486 struct filename *pathname;
David Rientjes72788c32011-05-24 17:11:40 -07001487 int oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 int i, type, prev;
1489 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 if (!capable(CAP_SYS_ADMIN))
1492 return -EPERM;
1493
Al Viro191c5422012-02-13 03:58:52 +00001494 BUG_ON(!current->mm);
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 pathname = getname(specialfile);
1497 err = PTR_ERR(pathname);
1498 if (IS_ERR(pathname))
1499 goto out;
1500
Jeff Layton669abf42012-10-10 16:43:10 -04001501 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 err = PTR_ERR(victim);
1503 if (IS_ERR(victim))
1504 goto out;
1505
1506 mapping = victim->f_mapping;
1507 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001508 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001509 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1510 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001511 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 if (p->swap_file->f_mapping == mapping)
1513 break;
1514 }
1515 prev = type;
1516 }
1517 if (type < 0) {
1518 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001519 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 goto out_dput;
1521 }
Al Viro191c5422012-02-13 03:58:52 +00001522 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 vm_unacct_memory(p->pages);
1524 else {
1525 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001526 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 goto out_dput;
1528 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001529 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001531 else
1532 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 if (type == swap_list.next) {
1534 /* just pick something that's safe... */
1535 swap_list.next = swap_list.head;
1536 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001537 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001538 for (i = p->next; i >= 0; i = swap_info[i]->next)
1539 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001540 least_priority++;
1541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 nr_swap_pages -= p->pages;
1543 total_swap_pages -= p->pages;
1544 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001545 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001546
David Rientjes72788c32011-05-24 17:11:40 -07001547 oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001548 err = try_to_unuse(type, false, 0); /* force all pages to be unused */
David Rientjes43362a42011-10-31 17:07:18 -07001549 compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX, oom_score_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 if (err) {
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001552 /*
1553 * reading p->prio and p->swap_map outside the lock is
1554 * safe here because only sys_swapon and sys_swapoff
1555 * change them, and there can be no other sys_swapon or
1556 * sys_swapoff for this swap_info_struct at this point.
1557 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 /* re-insert swap space back into swap_list */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001559 enable_swap_info(p, p->prio, p->swap_map, frontswap_map_get(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 goto out_dput;
1561 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001562
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001563 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001564 if (p->flags & SWP_CONTINUED)
1565 free_swap_count_continuations(p);
1566
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001567 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001568 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001570
1571 /* wait for anyone still in scan_swap_map */
1572 p->highest_bit = 0; /* cuts scans short */
1573 while (p->flags >= SWP_SCANNING) {
1574 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001575 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001576 spin_lock(&swap_lock);
1577 }
1578
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 swap_file = p->swap_file;
1580 p->swap_file = NULL;
1581 p->max = 0;
1582 swap_map = p->swap_map;
1583 p->swap_map = NULL;
1584 p->flags = 0;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001585 frontswap_invalidate_area(type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001586 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001587 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 vfree(swap_map);
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001589 vfree(frontswap_map_get(p));
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001590 /* Destroy swap account informatin */
1591 swap_cgroup_swapoff(type);
1592
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 inode = mapping->host;
1594 if (S_ISBLK(inode->i_mode)) {
1595 struct block_device *bdev = I_BDEV(inode);
1596 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001597 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001599 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001601 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 }
1603 filp_close(swap_file, NULL);
1604 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001605 atomic_inc(&proc_poll_event);
1606 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608out_dput:
1609 filp_close(victim, NULL);
1610out:
1611 return err;
1612}
1613
1614#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001615static unsigned swaps_poll(struct file *file, poll_table *wait)
1616{
Kay Sieversf1514632011-07-12 20:48:39 +02001617 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001618
1619 poll_wait(file, &proc_poll_wait, wait);
1620
Kay Sieversf1514632011-07-12 20:48:39 +02001621 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1622 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001623 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1624 }
1625
1626 return POLLIN | POLLRDNORM;
1627}
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629/* iterator */
1630static void *swap_start(struct seq_file *swap, loff_t *pos)
1631{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001632 struct swap_info_struct *si;
1633 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 loff_t l = *pos;
1635
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001636 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001638 if (!l)
1639 return SEQ_START_TOKEN;
1640
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001641 for (type = 0; type < nr_swapfiles; type++) {
1642 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1643 si = swap_info[type];
1644 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001646 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001647 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
1649
1650 return NULL;
1651}
1652
1653static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1654{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001655 struct swap_info_struct *si = v;
1656 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001658 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001659 type = 0;
1660 else
1661 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001662
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001663 for (; type < nr_swapfiles; type++) {
1664 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1665 si = swap_info[type];
1666 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 continue;
1668 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001669 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
1671
1672 return NULL;
1673}
1674
1675static void swap_stop(struct seq_file *swap, void *v)
1676{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001677 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678}
1679
1680static int swap_show(struct seq_file *swap, void *v)
1681{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001682 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 struct file *file;
1684 int len;
1685
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001686 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001687 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1688 return 0;
1689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001691 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001692 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001693 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001694 len < 40 ? 40 - len : 1, " ",
1695 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001697 si->pages << (PAGE_SHIFT - 10),
1698 si->inuse_pages << (PAGE_SHIFT - 10),
1699 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 return 0;
1701}
1702
Helge Deller15ad7cd2006-12-06 20:40:36 -08001703static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 .start = swap_start,
1705 .next = swap_next,
1706 .stop = swap_stop,
1707 .show = swap_show
1708};
1709
1710static int swaps_open(struct inode *inode, struct file *file)
1711{
Kay Sieversf1514632011-07-12 20:48:39 +02001712 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001713 int ret;
1714
Kay Sievers66d7dd52010-10-26 14:22:06 -07001715 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02001716 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07001717 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001718
Kay Sieversf1514632011-07-12 20:48:39 +02001719 seq = file->private_data;
1720 seq->poll_event = atomic_read(&proc_poll_event);
1721 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722}
1723
Helge Deller15ad7cd2006-12-06 20:40:36 -08001724static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 .open = swaps_open,
1726 .read = seq_read,
1727 .llseek = seq_lseek,
1728 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001729 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730};
1731
1732static int __init procswaps_init(void)
1733{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001734 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 return 0;
1736}
1737__initcall(procswaps_init);
1738#endif /* CONFIG_PROC_FS */
1739
Jan Beulich17963162008-12-16 11:35:24 +00001740#ifdef MAX_SWAPFILES_CHECK
1741static int __init max_swapfiles_check(void)
1742{
1743 MAX_SWAPFILES_CHECK();
1744 return 0;
1745}
1746late_initcall(max_swapfiles_check);
1747#endif
1748
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001749static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001751 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001753
1754 p = kzalloc(sizeof(*p), GFP_KERNEL);
1755 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001756 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001757
Hugh Dickins5d337b92005-09-03 15:54:41 -07001758 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001759 for (type = 0; type < nr_swapfiles; type++) {
1760 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001762 }
Christoph Lameter06972122006-06-23 02:03:35 -07001763 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001764 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001765 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07001766 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001768 if (type >= nr_swapfiles) {
1769 p->type = type;
1770 swap_info[type] = p;
1771 /*
1772 * Write swap_info[type] before nr_swapfiles, in case a
1773 * racing procfs swap_start() or swap_next() is reading them.
1774 * (We never shrink nr_swapfiles, we never free this entry.)
1775 */
1776 smp_wmb();
1777 nr_swapfiles++;
1778 } else {
1779 kfree(p);
1780 p = swap_info[type];
1781 /*
1782 * Do not memset this entry: a racing procfs swap_next()
1783 * would be relying on p->type to remain valid.
1784 */
1785 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001786 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001789 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001790
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001791 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001792}
1793
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001794static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1795{
1796 int error;
1797
1798 if (S_ISBLK(inode->i_mode)) {
1799 p->bdev = bdgrab(I_BDEV(inode));
1800 error = blkdev_get(p->bdev,
1801 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1802 sys_swapon);
1803 if (error < 0) {
1804 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001805 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001806 }
1807 p->old_block_size = block_size(p->bdev);
1808 error = set_blocksize(p->bdev, PAGE_SIZE);
1809 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001810 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001811 p->flags |= SWP_BLKDEV;
1812 } else if (S_ISREG(inode->i_mode)) {
1813 p->bdev = inode->i_sb->s_bdev;
1814 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001815 if (IS_SWAPFILE(inode))
1816 return -EBUSY;
1817 } else
1818 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001819
1820 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001821}
1822
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001823static unsigned long read_swap_header(struct swap_info_struct *p,
1824 union swap_header *swap_header,
1825 struct inode *inode)
1826{
1827 int i;
1828 unsigned long maxpages;
1829 unsigned long swapfilepages;
1830
1831 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1832 printk(KERN_ERR "Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001833 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001834 }
1835
1836 /* swap partition endianess hack... */
1837 if (swab32(swap_header->info.version) == 1) {
1838 swab32s(&swap_header->info.version);
1839 swab32s(&swap_header->info.last_page);
1840 swab32s(&swap_header->info.nr_badpages);
1841 for (i = 0; i < swap_header->info.nr_badpages; i++)
1842 swab32s(&swap_header->info.badpages[i]);
1843 }
1844 /* Check the swap header's sub-version */
1845 if (swap_header->info.version != 1) {
1846 printk(KERN_WARNING
1847 "Unable to handle swap header version %d\n",
1848 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001849 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001850 }
1851
1852 p->lowest_bit = 1;
1853 p->cluster_next = 1;
1854 p->cluster_nr = 0;
1855
1856 /*
1857 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07001858 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001859 * of bits for the swap offset in the swp_entry_t type, and
1860 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07001861 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001862 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001863 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001864 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001865 * offset is extracted. This will mask all the bits from
1866 * the initial ~0UL mask that can't be encoded in either
1867 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07001868 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001869 */
1870 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07001871 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001872 if (maxpages > swap_header->info.last_page) {
1873 maxpages = swap_header->info.last_page + 1;
1874 /* p->max is an unsigned int: don't overflow it */
1875 if ((unsigned int)maxpages == 0)
1876 maxpages = UINT_MAX;
1877 }
1878 p->highest_bit = maxpages - 1;
1879
1880 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001881 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001882 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1883 if (swapfilepages && maxpages > swapfilepages) {
1884 printk(KERN_WARNING
1885 "Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001886 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001887 }
1888 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001889 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001890 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001891 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001892
1893 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001894}
1895
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001896static int setup_swap_map_and_extents(struct swap_info_struct *p,
1897 union swap_header *swap_header,
1898 unsigned char *swap_map,
1899 unsigned long maxpages,
1900 sector_t *span)
1901{
1902 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001903 unsigned int nr_good_pages;
1904 int nr_extents;
1905
1906 nr_good_pages = maxpages - 1; /* omit header page */
1907
1908 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1909 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001910 if (page_nr == 0 || page_nr > swap_header->info.last_page)
1911 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001912 if (page_nr < maxpages) {
1913 swap_map[page_nr] = SWAP_MAP_BAD;
1914 nr_good_pages--;
1915 }
1916 }
1917
1918 if (nr_good_pages) {
1919 swap_map[0] = SWAP_MAP_BAD;
1920 p->max = maxpages;
1921 p->pages = nr_good_pages;
1922 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001923 if (nr_extents < 0)
1924 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001925 nr_good_pages = p->pages;
1926 }
1927 if (!nr_good_pages) {
1928 printk(KERN_WARNING "Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001929 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001930 }
1931
1932 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001933}
1934
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001935SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
1936{
1937 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04001938 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001939 struct file *swap_file = NULL;
1940 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001941 int i;
1942 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001943 int error;
1944 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001945 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001946 sector_t span;
1947 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001948 unsigned char *swap_map = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001949 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001950 struct page *page = NULL;
1951 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001952
Hugh Dickinsd15cab92012-03-28 14:42:42 -07001953 if (swap_flags & ~SWAP_FLAGS_VALID)
1954 return -EINVAL;
1955
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001956 if (!capable(CAP_SYS_ADMIN))
1957 return -EPERM;
1958
1959 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07001960 if (IS_ERR(p))
1961 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001962
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07001965 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07001967 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 }
Jeff Layton669abf42012-10-10 16:43:10 -04001969 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07001971 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07001973 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 }
1975
1976 p->swap_file = swap_file;
1977 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001980 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07001982 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07001984 if (mapping == q->swap_file->f_mapping) {
1985 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07001987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 }
1989
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03001990 inode = mapping->host;
1991 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001992 error = claim_swapfile(p, inode);
1993 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 /*
1997 * Read the swap header.
1998 */
1999 if (!mapping->a_ops->readpage) {
2000 error = -EINVAL;
2001 goto bad_swap;
2002 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002003 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 if (IS_ERR(page)) {
2005 error = PTR_ERR(page);
2006 goto bad_swap;
2007 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002008 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002010 maxpages = read_swap_header(p, swap_header, inode);
2011 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 error = -EINVAL;
2013 goto bad_swap;
2014 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002015
Hugh Dickins81e33972009-01-06 14:39:49 -08002016 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002017 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002018 if (!swap_map) {
2019 error = -ENOMEM;
2020 goto bad_swap;
2021 }
2022
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002023 error = swap_cgroup_swapon(p->type, maxpages);
2024 if (error)
2025 goto bad_swap;
2026
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002027 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2028 maxpages, &span);
2029 if (unlikely(nr_extents < 0)) {
2030 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 goto bad_swap;
2032 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002033 /* frontswap enabled? set up bit-per-page map for frontswap */
2034 if (frontswap_enabled)
2035 frontswap_map = vzalloc(maxpages / sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002037 if (p->bdev) {
2038 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2039 p->flags |= SWP_SOLIDSTATE;
2040 p->cluster_next = 1 + (random32() % p->highest_bit);
2041 }
Shaohua Li052b1982012-03-21 16:34:17 -07002042 if ((swap_flags & SWAP_FLAG_DISCARD) && discard_swap(p) == 0)
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002043 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002044 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002045
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002046 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002047 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002048 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002049 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002050 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002051 enable_swap_info(p, prio, swap_map, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002052
2053 printk(KERN_INFO "Adding %uk swap on %s. "
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002054 "Priority:%d extents:%d across:%lluk %s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002055 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002056 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2057 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002058 (p->flags & SWP_DISCARDABLE) ? "D" : "",
2059 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002060
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002061 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002062 atomic_inc(&proc_poll_event);
2063 wake_up_interruptible(&proc_poll_wait);
2064
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002065 if (S_ISREG(inode->i_mode))
2066 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 error = 0;
2068 goto out;
2069bad_swap:
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002070 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002071 set_blocksize(p->bdev, p->old_block_size);
2072 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002074 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002075 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002076 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002079 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 vfree(swap_map);
Mel Gorman52c50562011-03-22 16:30:08 -07002081 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002082 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002083 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002084 inode = NULL;
2085 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002087 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088out:
2089 if (page && !IS_ERR(page)) {
2090 kunmap(page);
2091 page_cache_release(page);
2092 }
2093 if (name)
2094 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002095 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002096 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 return error;
2098}
2099
2100void si_swapinfo(struct sysinfo *val)
2101{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002102 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 unsigned long nr_to_be_unused = 0;
2104
Hugh Dickins5d337b92005-09-03 15:54:41 -07002105 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002106 for (type = 0; type < nr_swapfiles; type++) {
2107 struct swap_info_struct *si = swap_info[type];
2108
2109 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2110 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
2112 val->freeswap = nr_swap_pages + nr_to_be_unused;
2113 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002114 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115}
2116
2117/*
2118 * Verify that a swap entry is valid and increment its swap map count.
2119 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002120 * Returns error code in following case.
2121 * - success -> 0
2122 * - swp_entry is invalid -> EINVAL
2123 * - swp_entry is migration entry -> EINVAL
2124 * - swap-cache reference is requested but there is already one. -> EEXIST
2125 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002126 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002128static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002130 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002132 unsigned char count;
2133 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002134 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
Andi Kleena7420aa2009-09-16 11:50:05 +02002136 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002137 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 type = swp_type(entry);
2140 if (type >= nr_swapfiles)
2141 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002142 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 offset = swp_offset(entry);
2144
Hugh Dickins5d337b92005-09-03 15:54:41 -07002145 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002146 if (unlikely(offset >= p->max))
2147 goto unlock_out;
2148
Hugh Dickins253d5532009-12-14 17:58:44 -08002149 count = p->swap_map[offset];
2150 has_cache = count & SWAP_HAS_CACHE;
2151 count &= ~SWAP_HAS_CACHE;
2152 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002153
Hugh Dickins253d5532009-12-14 17:58:44 -08002154 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002155
2156 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002157 if (!has_cache && count)
2158 has_cache = SWAP_HAS_CACHE;
2159 else if (has_cache) /* someone else added cache */
2160 err = -EEXIST;
2161 else /* no users remaining */
2162 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002163
2164 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002165
Hugh Dickins570a335b2009-12-14 17:58:46 -08002166 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2167 count += usage;
2168 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002169 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002170 else if (swap_count_continued(p, offset, count))
2171 count = COUNT_CONTINUED;
2172 else
2173 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002174 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002175 err = -ENOENT; /* unused swap entry */
2176
2177 p->swap_map[offset] = count | has_cache;
2178
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002179unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002180 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002182 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
2184bad_file:
2185 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2186 goto out;
2187}
Hugh Dickins253d5532009-12-14 17:58:44 -08002188
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002189/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002190 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2191 * (in which case its reference count is never incremented).
2192 */
2193void swap_shmem_alloc(swp_entry_t entry)
2194{
2195 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2196}
2197
2198/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002199 * Increase reference count of swap entry by 1.
2200 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2201 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2202 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2203 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002204 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002205int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002206{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002207 int err = 0;
2208
2209 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2210 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2211 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002212}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002214/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002215 * @entry: swap entry for which we allocate swap cache.
2216 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002217 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002218 * This can return error codes. Returns 0 at success.
2219 * -EBUSY means there is a swap cache.
2220 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002221 */
2222int swapcache_prepare(swp_entry_t entry)
2223{
Hugh Dickins253d5532009-12-14 17:58:44 -08002224 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002225}
2226
Mel Gormanf981c592012-07-31 16:44:47 -07002227struct swap_info_struct *page_swap_info(struct page *page)
2228{
2229 swp_entry_t swap = { .val = page_private(page) };
2230 BUG_ON(!PageSwapCache(page));
2231 return swap_info[swp_type(swap)];
2232}
2233
2234/*
2235 * out-of-line __page_file_ methods to avoid include hell.
2236 */
2237struct address_space *__page_file_mapping(struct page *page)
2238{
2239 VM_BUG_ON(!PageSwapCache(page));
2240 return page_swap_info(page)->swap_file->f_mapping;
2241}
2242EXPORT_SYMBOL_GPL(__page_file_mapping);
2243
2244pgoff_t __page_file_index(struct page *page)
2245{
2246 swp_entry_t swap = { .val = page_private(page) };
2247 VM_BUG_ON(!PageSwapCache(page));
2248 return swp_offset(swap);
2249}
2250EXPORT_SYMBOL_GPL(__page_file_index);
2251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002253 * add_swap_count_continuation - called when a swap count is duplicated
2254 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2255 * page of the original vmalloc'ed swap_map, to hold the continuation count
2256 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2257 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2258 *
2259 * These continuation pages are seldom referenced: the common paths all work
2260 * on the original swap_map, only referring to a continuation page when the
2261 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2262 *
2263 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2264 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2265 * can be called after dropping locks.
2266 */
2267int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2268{
2269 struct swap_info_struct *si;
2270 struct page *head;
2271 struct page *page;
2272 struct page *list_page;
2273 pgoff_t offset;
2274 unsigned char count;
2275
2276 /*
2277 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2278 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2279 */
2280 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2281
2282 si = swap_info_get(entry);
2283 if (!si) {
2284 /*
2285 * An acceptable race has occurred since the failing
2286 * __swap_duplicate(): the swap entry has been freed,
2287 * perhaps even the whole swap_map cleared for swapoff.
2288 */
2289 goto outer;
2290 }
2291
2292 offset = swp_offset(entry);
2293 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2294
2295 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2296 /*
2297 * The higher the swap count, the more likely it is that tasks
2298 * will race to add swap count continuation: we need to avoid
2299 * over-provisioning.
2300 */
2301 goto out;
2302 }
2303
2304 if (!page) {
2305 spin_unlock(&swap_lock);
2306 return -ENOMEM;
2307 }
2308
2309 /*
2310 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2311 * no architecture is using highmem pages for kernel pagetables: so it
2312 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2313 */
2314 head = vmalloc_to_page(si->swap_map + offset);
2315 offset &= ~PAGE_MASK;
2316
2317 /*
2318 * Page allocation does not initialize the page's lru field,
2319 * but it does always reset its private field.
2320 */
2321 if (!page_private(head)) {
2322 BUG_ON(count & COUNT_CONTINUED);
2323 INIT_LIST_HEAD(&head->lru);
2324 set_page_private(head, SWP_CONTINUED);
2325 si->flags |= SWP_CONTINUED;
2326 }
2327
2328 list_for_each_entry(list_page, &head->lru, lru) {
2329 unsigned char *map;
2330
2331 /*
2332 * If the previous map said no continuation, but we've found
2333 * a continuation page, free our allocation and use this one.
2334 */
2335 if (!(count & COUNT_CONTINUED))
2336 goto out;
2337
Cong Wang9b04c5f2011-11-25 23:14:39 +08002338 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002339 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002340 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002341
2342 /*
2343 * If this continuation count now has some space in it,
2344 * free our allocation and use this one.
2345 */
2346 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2347 goto out;
2348 }
2349
2350 list_add_tail(&page->lru, &head->lru);
2351 page = NULL; /* now it's attached, don't free it */
2352out:
2353 spin_unlock(&swap_lock);
2354outer:
2355 if (page)
2356 __free_page(page);
2357 return 0;
2358}
2359
2360/*
2361 * swap_count_continued - when the original swap_map count is incremented
2362 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2363 * into, carry if so, or else fail until a new continuation page is allocated;
2364 * when the original swap_map count is decremented from 0 with continuation,
2365 * borrow from the continuation and report whether it still holds more.
2366 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2367 */
2368static bool swap_count_continued(struct swap_info_struct *si,
2369 pgoff_t offset, unsigned char count)
2370{
2371 struct page *head;
2372 struct page *page;
2373 unsigned char *map;
2374
2375 head = vmalloc_to_page(si->swap_map + offset);
2376 if (page_private(head) != SWP_CONTINUED) {
2377 BUG_ON(count & COUNT_CONTINUED);
2378 return false; /* need to add count continuation */
2379 }
2380
2381 offset &= ~PAGE_MASK;
2382 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002383 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002384
2385 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2386 goto init_map; /* jump over SWAP_CONT_MAX checks */
2387
2388 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2389 /*
2390 * Think of how you add 1 to 999
2391 */
2392 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002393 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002394 page = list_entry(page->lru.next, struct page, lru);
2395 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002396 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002397 }
2398 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002399 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002400 page = list_entry(page->lru.next, struct page, lru);
2401 if (page == head)
2402 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002403 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002404init_map: *map = 0; /* we didn't zero the page */
2405 }
2406 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002407 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002408 page = list_entry(page->lru.prev, struct page, lru);
2409 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002410 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002411 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002412 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002413 page = list_entry(page->lru.prev, struct page, lru);
2414 }
2415 return true; /* incremented */
2416
2417 } else { /* decrementing */
2418 /*
2419 * Think of how you subtract 1 from 1000
2420 */
2421 BUG_ON(count != COUNT_CONTINUED);
2422 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002423 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002424 page = list_entry(page->lru.next, struct page, lru);
2425 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002426 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002427 }
2428 BUG_ON(*map == 0);
2429 *map -= 1;
2430 if (*map == 0)
2431 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002432 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002433 page = list_entry(page->lru.prev, struct page, lru);
2434 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002435 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002436 *map = SWAP_CONT_MAX | count;
2437 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002438 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002439 page = list_entry(page->lru.prev, struct page, lru);
2440 }
2441 return count == COUNT_CONTINUED;
2442 }
2443}
2444
2445/*
2446 * free_swap_count_continuations - swapoff free all the continuation pages
2447 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2448 */
2449static void free_swap_count_continuations(struct swap_info_struct *si)
2450{
2451 pgoff_t offset;
2452
2453 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2454 struct page *head;
2455 head = vmalloc_to_page(si->swap_map + offset);
2456 if (page_private(head)) {
2457 struct list_head *this, *next;
2458 list_for_each_safe(this, next, &head->lru) {
2459 struct page *page;
2460 page = list_entry(this, struct page, lru);
2461 list_del(this);
2462 __free_page(page);
2463 }
2464 }
2465 }
2466}