blob: 4a7f7e6992b665b64052ea77bbb1cd9be0aa00b4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/mm.h>
9#include <linux/hugetlb.h>
10#include <linux/mman.h>
11#include <linux/slab.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/vmalloc.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
Hugh Dickins072441e2011-06-27 16:18:02 -070017#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080019#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080024#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070032#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070033#include <linux/oom.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060034#include <linux/frontswap.h>
35#include <linux/swapfile.h>
Mel Gormanf981c592012-07-31 16:44:47 -070036#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/pgtable.h>
39#include <asm/tlbflush.h>
40#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080041#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Hugh Dickins570a335b2009-12-14 17:58:46 -080043static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
44 unsigned char);
45static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080046static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080047
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060048DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070049static unsigned int nr_swapfiles;
Shaohua Liec8acf22013-02-22 16:34:38 -080050atomic_long_t nr_swap_pages;
51/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070052long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070053static int least_priority;
Shaohua Liec8acf22013-02-22 16:34:38 -080054static atomic_t highest_priority_index = ATOMIC_INIT(-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static const char Bad_file[] = "Bad swap file entry ";
57static const char Unused_file[] = "Unused swap file entry ";
58static const char Bad_offset[] = "Bad swap offset entry ";
59static const char Unused_offset[] = "Unused swap offset entry ";
60
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060061struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060063struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Ingo Molnarfc0abb12006-01-18 17:42:33 -080065static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Kay Sievers66d7dd52010-10-26 14:22:06 -070067static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
68/* Activity counter to indicate that a swapon or swapoff has occurred */
69static atomic_t proc_poll_event = ATOMIC_INIT(0);
70
Hugh Dickins8d69aae2009-12-14 17:58:45 -080071static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070072{
Hugh Dickins570a335b2009-12-14 17:58:46 -080073 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070074}
75
Hugh Dickinsefa90a92009-12-14 17:58:41 -080076/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070077static int
78__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
79{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080080 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070081 struct page *page;
82 int ret = 0;
83
Shaohua Li33806f02013-02-22 16:34:37 -080084 page = find_get_page(swap_address_space(entry), entry.val);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070085 if (!page)
86 return 0;
87 /*
88 * This function is called from scan_swap_map() and it's called
89 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
90 * We have to use trylock for avoiding deadlock. This is a special
91 * case and you should use try_to_free_swap() with explicit lock_page()
92 * in usual operations.
93 */
94 if (trylock_page(page)) {
95 ret = try_to_free_swap(page);
96 unlock_page(page);
97 }
98 page_cache_release(page);
99 return ret;
100}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800103 * swapon tell device that all the old swap contents can be discarded,
104 * to allow the swap device to optimize its wear-levelling.
105 */
106static int discard_swap(struct swap_info_struct *si)
107{
108 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800109 sector_t start_block;
110 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800111 int err = 0;
112
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800113 /* Do not discard the swap header page! */
114 se = &si->first_swap_extent;
115 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
116 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
117 if (nr_blocks) {
118 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200119 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800120 if (err)
121 return err;
122 cond_resched();
123 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800124
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800125 list_for_each_entry(se, &si->first_swap_extent.list, list) {
126 start_block = se->start_block << (PAGE_SHIFT - 9);
127 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800128
129 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200130 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800131 if (err)
132 break;
133
134 cond_resched();
135 }
136 return err; /* That will often be -EOPNOTSUPP */
137}
138
Hugh Dickins7992fde2009-01-06 14:39:53 -0800139/*
140 * swap allocation tell device that a cluster of swap can now be discarded,
141 * to allow the swap device to optimize its wear-levelling.
142 */
143static void discard_swap_cluster(struct swap_info_struct *si,
144 pgoff_t start_page, pgoff_t nr_pages)
145{
146 struct swap_extent *se = si->curr_swap_extent;
147 int found_extent = 0;
148
149 while (nr_pages) {
150 struct list_head *lh;
151
152 if (se->start_page <= start_page &&
153 start_page < se->start_page + se->nr_pages) {
154 pgoff_t offset = start_page - se->start_page;
155 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800156 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800157
158 if (nr_blocks > nr_pages)
159 nr_blocks = nr_pages;
160 start_page += nr_blocks;
161 nr_pages -= nr_blocks;
162
163 if (!found_extent++)
164 si->curr_swap_extent = se;
165
166 start_block <<= PAGE_SHIFT - 9;
167 nr_blocks <<= PAGE_SHIFT - 9;
168 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200169 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800170 break;
171 }
172
173 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800174 se = list_entry(lh, struct swap_extent, list);
175 }
176}
177
Hugh Dickins048c27f2005-09-03 15:54:40 -0700178#define SWAPFILE_CLUSTER 256
179#define LATENCY_LIMIT 256
180
Shaohua Li2a8f9442013-09-11 14:20:28 -0700181static inline void cluster_set_flag(struct swap_cluster_info *info,
182 unsigned int flag)
183{
184 info->flags = flag;
185}
186
187static inline unsigned int cluster_count(struct swap_cluster_info *info)
188{
189 return info->data;
190}
191
192static inline void cluster_set_count(struct swap_cluster_info *info,
193 unsigned int c)
194{
195 info->data = c;
196}
197
198static inline void cluster_set_count_flag(struct swap_cluster_info *info,
199 unsigned int c, unsigned int f)
200{
201 info->flags = f;
202 info->data = c;
203}
204
205static inline unsigned int cluster_next(struct swap_cluster_info *info)
206{
207 return info->data;
208}
209
210static inline void cluster_set_next(struct swap_cluster_info *info,
211 unsigned int n)
212{
213 info->data = n;
214}
215
216static inline void cluster_set_next_flag(struct swap_cluster_info *info,
217 unsigned int n, unsigned int f)
218{
219 info->flags = f;
220 info->data = n;
221}
222
223static inline bool cluster_is_free(struct swap_cluster_info *info)
224{
225 return info->flags & CLUSTER_FLAG_FREE;
226}
227
228static inline bool cluster_is_null(struct swap_cluster_info *info)
229{
230 return info->flags & CLUSTER_FLAG_NEXT_NULL;
231}
232
233static inline void cluster_set_null(struct swap_cluster_info *info)
234{
235 info->flags = CLUSTER_FLAG_NEXT_NULL;
236 info->data = 0;
237}
238
Shaohua Li815c2c52013-09-11 14:20:30 -0700239/* Add a cluster to discard list and schedule it to do discard */
240static void swap_cluster_schedule_discard(struct swap_info_struct *si,
241 unsigned int idx)
242{
243 /*
244 * If scan_swap_map() can't find a free cluster, it will check
245 * si->swap_map directly. To make sure the discarding cluster isn't
246 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
247 * will be cleared after discard
248 */
249 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
250 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
251
252 if (cluster_is_null(&si->discard_cluster_head)) {
253 cluster_set_next_flag(&si->discard_cluster_head,
254 idx, 0);
255 cluster_set_next_flag(&si->discard_cluster_tail,
256 idx, 0);
257 } else {
258 unsigned int tail = cluster_next(&si->discard_cluster_tail);
259 cluster_set_next(&si->cluster_info[tail], idx);
260 cluster_set_next_flag(&si->discard_cluster_tail,
261 idx, 0);
262 }
263
264 schedule_work(&si->discard_work);
265}
266
267/*
268 * Doing discard actually. After a cluster discard is finished, the cluster
269 * will be added to free cluster list. caller should hold si->lock.
270*/
271static void swap_do_scheduled_discard(struct swap_info_struct *si)
272{
273 struct swap_cluster_info *info;
274 unsigned int idx;
275
276 info = si->cluster_info;
277
278 while (!cluster_is_null(&si->discard_cluster_head)) {
279 idx = cluster_next(&si->discard_cluster_head);
280
281 cluster_set_next_flag(&si->discard_cluster_head,
282 cluster_next(&info[idx]), 0);
283 if (cluster_next(&si->discard_cluster_tail) == idx) {
284 cluster_set_null(&si->discard_cluster_head);
285 cluster_set_null(&si->discard_cluster_tail);
286 }
287 spin_unlock(&si->lock);
288
289 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
290 SWAPFILE_CLUSTER);
291
292 spin_lock(&si->lock);
293 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
294 if (cluster_is_null(&si->free_cluster_head)) {
295 cluster_set_next_flag(&si->free_cluster_head,
296 idx, 0);
297 cluster_set_next_flag(&si->free_cluster_tail,
298 idx, 0);
299 } else {
300 unsigned int tail;
301
302 tail = cluster_next(&si->free_cluster_tail);
303 cluster_set_next(&info[tail], idx);
304 cluster_set_next_flag(&si->free_cluster_tail,
305 idx, 0);
306 }
307 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
308 0, SWAPFILE_CLUSTER);
309 }
310}
311
312static void swap_discard_work(struct work_struct *work)
313{
314 struct swap_info_struct *si;
315
316 si = container_of(work, struct swap_info_struct, discard_work);
317
318 spin_lock(&si->lock);
319 swap_do_scheduled_discard(si);
320 spin_unlock(&si->lock);
321}
322
Shaohua Li2a8f9442013-09-11 14:20:28 -0700323/*
324 * The cluster corresponding to page_nr will be used. The cluster will be
325 * removed from free cluster list and its usage counter will be increased.
326 */
327static void inc_cluster_info_page(struct swap_info_struct *p,
328 struct swap_cluster_info *cluster_info, unsigned long page_nr)
329{
330 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
331
332 if (!cluster_info)
333 return;
334 if (cluster_is_free(&cluster_info[idx])) {
335 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
336 cluster_set_next_flag(&p->free_cluster_head,
337 cluster_next(&cluster_info[idx]), 0);
338 if (cluster_next(&p->free_cluster_tail) == idx) {
339 cluster_set_null(&p->free_cluster_tail);
340 cluster_set_null(&p->free_cluster_head);
341 }
342 cluster_set_count_flag(&cluster_info[idx], 0, 0);
343 }
344
345 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
346 cluster_set_count(&cluster_info[idx],
347 cluster_count(&cluster_info[idx]) + 1);
348}
349
350/*
351 * The cluster corresponding to page_nr decreases one usage. If the usage
352 * counter becomes 0, which means no page in the cluster is in using, we can
353 * optionally discard the cluster and add it to free cluster list.
354 */
355static void dec_cluster_info_page(struct swap_info_struct *p,
356 struct swap_cluster_info *cluster_info, unsigned long page_nr)
357{
358 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
359
360 if (!cluster_info)
361 return;
362
363 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
364 cluster_set_count(&cluster_info[idx],
365 cluster_count(&cluster_info[idx]) - 1);
366
367 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700368 /*
369 * If the swap is discardable, prepare discard the cluster
370 * instead of free it immediately. The cluster will be freed
371 * after discard.
372 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700373 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
374 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700375 swap_cluster_schedule_discard(p, idx);
376 return;
377 }
378
Shaohua Li2a8f9442013-09-11 14:20:28 -0700379 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
380 if (cluster_is_null(&p->free_cluster_head)) {
381 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
382 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
383 } else {
384 unsigned int tail = cluster_next(&p->free_cluster_tail);
385 cluster_set_next(&cluster_info[tail], idx);
386 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
387 }
388 }
389}
390
391/*
392 * It's possible scan_swap_map() uses a free cluster in the middle of free
393 * cluster list. Avoiding such abuse to avoid list corruption.
394 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700395static bool
396scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700397 unsigned long offset)
398{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700399 struct percpu_cluster *percpu_cluster;
400 bool conflict;
401
Shaohua Li2a8f9442013-09-11 14:20:28 -0700402 offset /= SWAPFILE_CLUSTER;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700403 conflict = !cluster_is_null(&si->free_cluster_head) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700404 offset != cluster_next(&si->free_cluster_head) &&
405 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700406
407 if (!conflict)
408 return false;
409
410 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
411 cluster_set_null(&percpu_cluster->index);
412 return true;
413}
414
415/*
416 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
417 * might involve allocating a new cluster for current CPU too.
418 */
419static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
420 unsigned long *offset, unsigned long *scan_base)
421{
422 struct percpu_cluster *cluster;
423 bool found_free;
424 unsigned long tmp;
425
426new_cluster:
427 cluster = this_cpu_ptr(si->percpu_cluster);
428 if (cluster_is_null(&cluster->index)) {
429 if (!cluster_is_null(&si->free_cluster_head)) {
430 cluster->index = si->free_cluster_head;
431 cluster->next = cluster_next(&cluster->index) *
432 SWAPFILE_CLUSTER;
433 } else if (!cluster_is_null(&si->discard_cluster_head)) {
434 /*
435 * we don't have free cluster but have some clusters in
436 * discarding, do discard now and reclaim them
437 */
438 swap_do_scheduled_discard(si);
439 *scan_base = *offset = si->cluster_next;
440 goto new_cluster;
441 } else
442 return;
443 }
444
445 found_free = false;
446
447 /*
448 * Other CPUs can use our cluster if they can't find a free cluster,
449 * check if there is still free entry in the cluster
450 */
451 tmp = cluster->next;
452 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
453 SWAPFILE_CLUSTER) {
454 if (!si->swap_map[tmp]) {
455 found_free = true;
456 break;
457 }
458 tmp++;
459 }
460 if (!found_free) {
461 cluster_set_null(&cluster->index);
462 goto new_cluster;
463 }
464 cluster->next = tmp + 1;
465 *offset = tmp;
466 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700467}
468
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700469static unsigned long scan_swap_map(struct swap_info_struct *si,
470 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800472 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800473 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800474 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700475 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800477 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700478 * We try to cluster swap pages by allocating them sequentially
479 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
480 * way, however, we resort to first-free allocation, starting
481 * a new cluster. This prevents us from scattering swap pages
482 * all over the entire swap partition, so that we reduce
483 * overall disk seek times between swap pages. -- sct
484 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800485 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700486 */
487
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700488 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800489 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800490
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700491 /* SSD algorithm */
492 if (si->cluster_info) {
493 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
494 goto checks;
495 }
496
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800497 if (unlikely(!si->cluster_nr--)) {
498 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
499 si->cluster_nr = SWAPFILE_CLUSTER - 1;
500 goto checks;
501 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700502
Shaohua Liec8acf22013-02-22 16:34:38 -0800503 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700504
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800505 /*
506 * If seek is expensive, start searching for new cluster from
507 * start of partition, to minimize the span of allocated swap.
508 * But if seek is cheap, search from our current position, so
509 * that swap is allocated from all over the partition: if the
510 * Flash Translation Layer only remaps within limited zones,
511 * we don't want to wear out the first zone too quickly.
512 */
513 if (!(si->flags & SWP_SOLIDSTATE))
514 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700515 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
516
517 /* Locate the first empty (unaligned) cluster */
518 for (; last_in_cluster <= si->highest_bit; offset++) {
519 if (si->swap_map[offset])
520 last_in_cluster = offset + SWAPFILE_CLUSTER;
521 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800522 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800523 offset -= SWAPFILE_CLUSTER - 1;
524 si->cluster_next = offset;
525 si->cluster_nr = SWAPFILE_CLUSTER - 1;
526 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700528 if (unlikely(--latency_ration < 0)) {
529 cond_resched();
530 latency_ration = LATENCY_LIMIT;
531 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700532 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800533
534 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800535 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
536
537 /* Locate the first empty (unaligned) cluster */
538 for (; last_in_cluster < scan_base; offset++) {
539 if (si->swap_map[offset])
540 last_in_cluster = offset + SWAPFILE_CLUSTER;
541 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800542 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800543 offset -= SWAPFILE_CLUSTER - 1;
544 si->cluster_next = offset;
545 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800546 goto checks;
547 }
548 if (unlikely(--latency_ration < 0)) {
549 cond_resched();
550 latency_ration = LATENCY_LIMIT;
551 }
552 }
553
554 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800555 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800556 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700558
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800559checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700560 if (si->cluster_info) {
561 while (scan_swap_map_ssd_cluster_conflict(si, offset))
562 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
563 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800564 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700565 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700566 if (!si->highest_bit)
567 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800568 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800569 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700570
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700571 /* reuse swap entry of cache-only swap if not busy. */
572 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700573 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800574 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700575 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800576 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700577 /* entry was freed successfully, try to use this again */
578 if (swap_was_freed)
579 goto checks;
580 goto scan; /* check next one */
581 }
582
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800583 if (si->swap_map[offset])
584 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700585
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800586 if (offset == si->lowest_bit)
587 si->lowest_bit++;
588 if (offset == si->highest_bit)
589 si->highest_bit--;
590 si->inuse_pages++;
591 if (si->inuse_pages == si->pages) {
592 si->lowest_bit = si->max;
593 si->highest_bit = 0;
594 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800595 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700596 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800597 si->cluster_next = offset + 1;
598 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800599
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800600 return offset;
601
602scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800603 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700604 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700605 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800606 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700607 goto checks;
608 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700609 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800610 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700611 goto checks;
612 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700613 if (unlikely(--latency_ration < 0)) {
614 cond_resched();
615 latency_ration = LATENCY_LIMIT;
616 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700617 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800618 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800619 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800620 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800621 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800622 goto checks;
623 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700624 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800625 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700626 goto checks;
627 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800628 if (unlikely(--latency_ration < 0)) {
629 cond_resched();
630 latency_ration = LATENCY_LIMIT;
631 }
Jamie Liua5998062014-01-23 15:53:40 -0800632 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800633 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800634 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700635
636no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700637 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 return 0;
639}
640
641swp_entry_t get_swap_page(void)
642{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700643 struct swap_info_struct *si;
644 pgoff_t offset;
645 int type, next;
646 int wrapped = 0;
Shaohua Liec8acf22013-02-22 16:34:38 -0800647 int hp_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Hugh Dickins5d337b92005-09-03 15:54:41 -0700649 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800650 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700651 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800652 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700654 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800655 hp_index = atomic_xchg(&highest_priority_index, -1);
656 /*
657 * highest_priority_index records current highest priority swap
658 * type which just frees swap entries. If its priority is
659 * higher than that of swap_list.next swap type, we use it. It
660 * isn't protected by swap_lock, so it can be an invalid value
661 * if the corresponding swap type is swapoff. We double check
662 * the flags here. It's even possible the swap type is swapoff
663 * and swapon again and its priority is changed. In such rare
664 * case, low prority swap type might be used, but eventually
665 * high priority swap will be used after several rounds of
666 * swap.
667 */
668 if (hp_index != -1 && hp_index != type &&
669 swap_info[type]->prio < swap_info[hp_index]->prio &&
670 (swap_info[hp_index]->flags & SWP_WRITEOK)) {
671 type = hp_index;
672 swap_list.next = type;
673 }
674
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800675 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700676 next = si->next;
677 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800678 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700679 next = swap_list.head;
680 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700682
Shaohua Liec8acf22013-02-22 16:34:38 -0800683 spin_lock(&si->lock);
684 if (!si->highest_bit) {
685 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700686 continue;
Shaohua Liec8acf22013-02-22 16:34:38 -0800687 }
688 if (!(si->flags & SWP_WRITEOK)) {
689 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700690 continue;
Shaohua Liec8acf22013-02-22 16:34:38 -0800691 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700692
693 swap_list.next = next;
Shaohua Liec8acf22013-02-22 16:34:38 -0800694
695 spin_unlock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700696 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800697 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800698 spin_unlock(&si->lock);
699 if (offset)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700700 return swp_entry(type, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800701 spin_lock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700702 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700704
Shaohua Liec8acf22013-02-22 16:34:38 -0800705 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700706noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700707 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700708 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709}
710
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800711/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700712swp_entry_t get_swap_page_of_type(int type)
713{
714 struct swap_info_struct *si;
715 pgoff_t offset;
716
Hugh Dickins910321e2010-09-09 16:38:07 -0700717 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800718 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700719 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800720 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700721 /* This is called for allocating swap entry, not cache */
722 offset = scan_swap_map(si, 1);
723 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800724 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700725 return swp_entry(type, offset);
726 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800727 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700728 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800729 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700730 return (swp_entry_t) {0};
731}
732
Hugh Dickins73c34b62009-12-14 17:58:43 -0800733static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800735 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 unsigned long offset, type;
737
738 if (!entry.val)
739 goto out;
740 type = swp_type(entry);
741 if (type >= nr_swapfiles)
742 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800743 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (!(p->flags & SWP_USED))
745 goto bad_device;
746 offset = swp_offset(entry);
747 if (offset >= p->max)
748 goto bad_offset;
749 if (!p->swap_map[offset])
750 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800751 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 return p;
753
754bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700755 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 goto out;
757bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700758 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 goto out;
760bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700761 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 goto out;
763bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700764 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765out:
766 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800767}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Shaohua Liec8acf22013-02-22 16:34:38 -0800769/*
770 * This swap type frees swap entry, check if it is the highest priority swap
771 * type which just frees swap entry. get_swap_page() uses
772 * highest_priority_index to search highest priority swap type. The
773 * swap_info_struct.lock can't protect us if there are multiple swap types
774 * active, so we use atomic_cmpxchg.
775 */
776static void set_highest_priority_index(int type)
777{
778 int old_hp_index, new_hp_index;
779
780 do {
781 old_hp_index = atomic_read(&highest_priority_index);
782 if (old_hp_index != -1 &&
783 swap_info[old_hp_index]->prio >= swap_info[type]->prio)
784 break;
785 new_hp_index = type;
786 } while (atomic_cmpxchg(&highest_priority_index,
787 old_hp_index, new_hp_index) != old_hp_index);
788}
789
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800790static unsigned char swap_entry_free(struct swap_info_struct *p,
791 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
Hugh Dickins253d5532009-12-14 17:58:44 -0800793 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800794 unsigned char count;
795 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700797 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800798 has_cache = count & SWAP_HAS_CACHE;
799 count &= ~SWAP_HAS_CACHE;
800
801 if (usage == SWAP_HAS_CACHE) {
802 VM_BUG_ON(!has_cache);
803 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800804 } else if (count == SWAP_MAP_SHMEM) {
805 /*
806 * Or we could insist on shmem.c using a special
807 * swap_shmem_free() and free_shmem_swap_and_cache()...
808 */
809 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800810 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
811 if (count == COUNT_CONTINUED) {
812 if (swap_count_continued(p, offset, count))
813 count = SWAP_MAP_MAX | COUNT_CONTINUED;
814 else
815 count = SWAP_MAP_MAX;
816 } else
817 count--;
818 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800819
820 if (!count)
821 mem_cgroup_uncharge_swap(entry);
822
823 usage = count | has_cache;
824 p->swap_map[offset] = usage;
825
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700826 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800827 if (!usage) {
Shaohua Li2a8f9442013-09-11 14:20:28 -0700828 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700829 if (offset < p->lowest_bit)
830 p->lowest_bit = offset;
831 if (offset > p->highest_bit)
832 p->highest_bit = offset;
Shaohua Liec8acf22013-02-22 16:34:38 -0800833 set_highest_priority_index(p->type);
834 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700835 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600836 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700837 if (p->flags & SWP_BLKDEV) {
838 struct gendisk *disk = p->bdev->bd_disk;
839 if (disk->fops->swap_slot_free_notify)
840 disk->fops->swap_slot_free_notify(p->bdev,
841 offset);
842 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800844
845 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846}
847
848/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800849 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 * is still around or has not been recycled.
851 */
852void swap_free(swp_entry_t entry)
853{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800854 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856 p = swap_info_get(entry);
857 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800858 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800859 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
861}
862
863/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700864 * Called after dropping swapcache to decrease refcnt to swap entries.
865 */
866void swapcache_free(swp_entry_t entry, struct page *page)
867{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700868 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800869 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700870
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700871 p = swap_info_get(entry);
872 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800873 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
874 if (page)
875 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
Shaohua Liec8acf22013-02-22 16:34:38 -0800876 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700877 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700878}
879
880/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700881 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800882 * This does not give an exact answer when swap count is continued,
883 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700885int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700887 int count = 0;
888 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 swp_entry_t entry;
890
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700891 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 p = swap_info_get(entry);
893 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700894 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800895 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700897 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
900/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800901 * We can write to an anon page without COW if there are no other references
902 * to it. And as a side-effect, free up its swap: because the old content
903 * on disk will never be read, and seeking back there to write new content
904 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800906int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700908 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Sasha Levin309381fea2014-01-23 15:52:54 -0800910 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800911 if (unlikely(PageKsm(page)))
912 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700913 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800914 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700915 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800916 if (count == 1 && !PageWriteback(page)) {
917 delete_from_swap_cache(page);
918 SetPageDirty(page);
919 }
920 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800921 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922}
923
924/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800925 * If swap is getting full, or if there are no more mappings of this page,
926 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800928int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929{
Sasha Levin309381fea2014-01-23 15:52:54 -0800930 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 if (!PageSwapCache(page))
933 return 0;
934 if (PageWriteback(page))
935 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800936 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return 0;
938
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700939 /*
940 * Once hibernation has begun to create its image of memory,
941 * there's a danger that one of the calls to try_to_free_swap()
942 * - most probably a call from __try_to_reclaim_swap() while
943 * hibernation is allocating its own swap pages for the image,
944 * but conceivably even a call from memory reclaim - will free
945 * the swap from a page which has already been recorded in the
946 * image as a clean swapcache page, and then reuse its swap for
947 * another page of the image. On waking from hibernation, the
948 * original page might be freed under memory pressure, then
949 * later read back in from swap, now with the wrong data.
950 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800951 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -0800952 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700953 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800954 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700955 return 0;
956
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800957 delete_from_swap_cache(page);
958 SetPageDirty(page);
959 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700960}
961
962/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 * Free the swap entry like above, but also try to
964 * free the page cache entry if it is the last user.
965 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800966int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800968 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 struct page *page = NULL;
970
Andi Kleena7420aa2009-09-16 11:50:05 +0200971 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800972 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 p = swap_info_get(entry);
975 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800976 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -0800977 page = find_get_page(swap_address_space(entry),
978 entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700979 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800980 page_cache_release(page);
981 page = NULL;
982 }
983 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800984 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 }
986 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800987 /*
988 * Not mapped elsewhere, or swap space full? Free it!
989 * Also recheck PageSwapCache now page is locked (above).
990 */
Nick Piggin93fac702006-03-31 02:29:56 -0800991 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800992 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 delete_from_swap_cache(page);
994 SetPageDirty(page);
995 }
996 unlock_page(page);
997 page_cache_release(page);
998 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800999 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001002#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001003/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001004 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001005 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001006 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1007 * from 0, in which the swap header is expected to be located.
1008 *
1009 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001010 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001011int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001012{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001013 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001014 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001015
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001016 if (device)
1017 bdev = bdget(device);
1018
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001019 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001020 for (type = 0; type < nr_swapfiles; type++) {
1021 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001022
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001023 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001024 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001025
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001026 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001027 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001028 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001029
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001030 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001031 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001032 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001033 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001034 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001035
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001036 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001037 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001038 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001039
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001040 spin_unlock(&swap_lock);
1041 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001042 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001043 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001044 }
1045 }
1046 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001047 if (bdev)
1048 bdput(bdev);
1049
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001050 return -ENODEV;
1051}
1052
1053/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001054 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1055 * corresponding to given index in swap_info (swap type).
1056 */
1057sector_t swapdev_block(int type, pgoff_t offset)
1058{
1059 struct block_device *bdev;
1060
1061 if ((unsigned int)type >= nr_swapfiles)
1062 return 0;
1063 if (!(swap_info[type]->flags & SWP_WRITEOK))
1064 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001065 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001066}
1067
1068/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001069 * Return either the total number of swap pages of given type, or the number
1070 * of free pages of that type (depending on @free)
1071 *
1072 * This is needed for software suspend
1073 */
1074unsigned int count_swap_pages(int type, int free)
1075{
1076 unsigned int n = 0;
1077
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001078 spin_lock(&swap_lock);
1079 if ((unsigned int)type < nr_swapfiles) {
1080 struct swap_info_struct *sis = swap_info[type];
1081
Shaohua Liec8acf22013-02-22 16:34:38 -08001082 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001083 if (sis->flags & SWP_WRITEOK) {
1084 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001085 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001086 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001087 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001088 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001089 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001090 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001091 return n;
1092}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001093#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001094
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001095static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1096{
1097#ifdef CONFIG_MEM_SOFT_DIRTY
1098 /*
1099 * When pte keeps soft dirty bit the pte generated
1100 * from swap entry does not has it, still it's same
1101 * pte from logical point of view.
1102 */
1103 pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1104 return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1105#else
1106 return pte_same(pte, swp_pte);
1107#endif
1108}
1109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001111 * No need to decide whether this PTE shares the swap entry with others,
1112 * just let do_wp_page work it out if a write is requested later - to
1113 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001115static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 unsigned long addr, swp_entry_t entry, struct page *page)
1117{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001118 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001119 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001120 spinlock_t *ptl;
1121 pte_t *pte;
1122 int ret = 1;
1123
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001124 swapcache = page;
1125 page = ksm_might_need_to_copy(page, vma, addr);
1126 if (unlikely(!page))
1127 return -ENOMEM;
1128
Johannes Weiner72835c82012-01-12 17:18:32 -08001129 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
1130 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001131 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001132 goto out_nolock;
1133 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001134
1135 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001136 if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
Johannes Weiner5d84c772012-07-31 16:45:28 -07001137 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001138 ret = 0;
1139 goto out;
1140 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001141
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001142 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001143 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 get_page(page);
1145 set_pte_at(vma->vm_mm, addr, pte,
1146 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001147 if (page == swapcache)
1148 page_add_anon_rmap(page, vma, addr);
1149 else /* ksm created a completely new copy */
1150 page_add_new_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -08001151 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 swap_free(entry);
1153 /*
1154 * Move the page to the active list so it is not
1155 * immediately swapped out again after swapon.
1156 */
1157 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001158out:
1159 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001160out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001161 if (page != swapcache) {
1162 unlock_page(page);
1163 put_page(page);
1164 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001165 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166}
1167
1168static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1169 unsigned long addr, unsigned long end,
1170 swp_entry_t entry, struct page *page)
1171{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001173 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001174 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
Hugh Dickins044d66c2008-02-07 00:14:04 -08001176 /*
1177 * We don't actually need pte lock while scanning for swp_pte: since
1178 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1179 * page table while we're scanning; though it could get zapped, and on
1180 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1181 * of unmatched parts which look like swp_pte, so unuse_pte must
1182 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001183 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001184 */
1185 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 do {
1187 /*
1188 * swapoff spends a _lot_ of time in this loop!
1189 * Test inline before going to call unuse_pte.
1190 */
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001191 if (unlikely(maybe_same_pte(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001192 pte_unmap(pte);
1193 ret = unuse_pte(vma, pmd, addr, entry, page);
1194 if (ret)
1195 goto out;
1196 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 }
1198 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001199 pte_unmap(pte - 1);
1200out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001201 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202}
1203
1204static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1205 unsigned long addr, unsigned long end,
1206 swp_entry_t entry, struct page *page)
1207{
1208 pmd_t *pmd;
1209 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001210 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
1212 pmd = pmd_offset(pud, addr);
1213 do {
1214 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001215 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001217 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1218 if (ret)
1219 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 } while (pmd++, addr = next, addr != end);
1221 return 0;
1222}
1223
1224static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1225 unsigned long addr, unsigned long end,
1226 swp_entry_t entry, struct page *page)
1227{
1228 pud_t *pud;
1229 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001230 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
1232 pud = pud_offset(pgd, addr);
1233 do {
1234 next = pud_addr_end(addr, end);
1235 if (pud_none_or_clear_bad(pud))
1236 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001237 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1238 if (ret)
1239 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 } while (pud++, addr = next, addr != end);
1241 return 0;
1242}
1243
1244static int unuse_vma(struct vm_area_struct *vma,
1245 swp_entry_t entry, struct page *page)
1246{
1247 pgd_t *pgd;
1248 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001249 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001251 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 addr = page_address_in_vma(page, vma);
1253 if (addr == -EFAULT)
1254 return 0;
1255 else
1256 end = addr + PAGE_SIZE;
1257 } else {
1258 addr = vma->vm_start;
1259 end = vma->vm_end;
1260 }
1261
1262 pgd = pgd_offset(vma->vm_mm, addr);
1263 do {
1264 next = pgd_addr_end(addr, end);
1265 if (pgd_none_or_clear_bad(pgd))
1266 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001267 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1268 if (ret)
1269 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 } while (pgd++, addr = next, addr != end);
1271 return 0;
1272}
1273
1274static int unuse_mm(struct mm_struct *mm,
1275 swp_entry_t entry, struct page *page)
1276{
1277 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001278 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
1280 if (!down_read_trylock(&mm->mmap_sem)) {
1281 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001282 * Activate page so shrink_inactive_list is unlikely to unmap
1283 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001285 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 unlock_page(page);
1287 down_read(&mm->mmap_sem);
1288 lock_page(page);
1289 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001291 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 break;
1293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001295 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296}
1297
1298/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001299 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1300 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 * Recycle to start on reaching the end, returning 0 when empty.
1302 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001303static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001304 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001306 unsigned int max = si->max;
1307 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001308 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001311 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 * for whether an entry is in use, not modifying it; false
1313 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001314 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 */
1316 for (;;) {
1317 if (++i >= max) {
1318 if (!prev) {
1319 i = 0;
1320 break;
1321 }
1322 /*
1323 * No entries in use at top of swap_map,
1324 * loop back to start and recheck there.
1325 */
1326 max = prev + 1;
1327 prev = 0;
1328 i = 1;
1329 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001330 if (frontswap) {
1331 if (frontswap_test(si, i))
1332 break;
1333 else
1334 continue;
1335 }
Shaohua Liedfe23d2013-09-11 14:20:31 -07001336 count = ACCESS_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001337 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 break;
1339 }
1340 return i;
1341}
1342
1343/*
1344 * We completely avoid races by reading each swap page in advance,
1345 * and then search for the process using it. All the necessary
1346 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001347 *
1348 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1349 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001351int try_to_unuse(unsigned int type, bool frontswap,
1352 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001354 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001356 volatile unsigned char *swap_map; /* swap_map is accessed without
1357 * locking. Mark it as volatile
1358 * to prevent compiler doing
1359 * something odd.
1360 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001361 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 struct page *page;
1363 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001364 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 /*
1368 * When searching mms for an entry, a good strategy is to
1369 * start at the first mm we freed the previous entry from
1370 * (though actually we don't notice whether we or coincidence
1371 * freed the entry). Initialize this start_mm with a hold.
1372 *
1373 * A simpler strategy would be to start at the last mm we
1374 * freed the previous entry from; but that would take less
1375 * advantage of mmlist ordering, which clusters forked mms
1376 * together, child after parent. If we race with dup_mmap(), we
1377 * prefer to resolve parent before child, lest we miss entries
1378 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001379 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 */
1381 start_mm = &init_mm;
1382 atomic_inc(&init_mm.mm_users);
1383
1384 /*
1385 * Keep on scanning until all entries have gone. Usually,
1386 * one pass through swap_map is enough, but not necessarily:
1387 * there are races when an instance of an entry might be missed.
1388 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001389 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 if (signal_pending(current)) {
1391 retval = -EINTR;
1392 break;
1393 }
1394
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001395 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 * Get a page for the entry, using the existing swap
1397 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001398 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 */
1400 swap_map = &si->swap_map[i];
1401 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001402 page = read_swap_cache_async(entry,
1403 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 if (!page) {
1405 /*
1406 * Either swap_duplicate() failed because entry
1407 * has been freed independently, and will not be
1408 * reused since sys_swapoff() already disabled
1409 * allocation from here, or alloc_page() failed.
1410 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001411 swcount = *swap_map;
1412 /*
1413 * We don't hold lock here, so the swap entry could be
1414 * SWAP_MAP_BAD (when the cluster is discarding).
1415 * Instead of fail out, We can just skip the swap
1416 * entry because swapoff will wait for discarding
1417 * finish anyway.
1418 */
1419 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 continue;
1421 retval = -ENOMEM;
1422 break;
1423 }
1424
1425 /*
1426 * Don't hold on to start_mm if it looks like exiting.
1427 */
1428 if (atomic_read(&start_mm->mm_users) == 1) {
1429 mmput(start_mm);
1430 start_mm = &init_mm;
1431 atomic_inc(&init_mm.mm_users);
1432 }
1433
1434 /*
1435 * Wait for and lock page. When do_swap_page races with
1436 * try_to_unuse, do_swap_page can handle the fault much
1437 * faster than try_to_unuse can locate the entry. This
1438 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1439 * defer to do_swap_page in such a case - in some tests,
1440 * do_swap_page and try_to_unuse repeatedly compete.
1441 */
1442 wait_on_page_locked(page);
1443 wait_on_page_writeback(page);
1444 lock_page(page);
1445 wait_on_page_writeback(page);
1446
1447 /*
1448 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001451 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1452 retval = shmem_unuse(entry, page);
1453 /* page has already been unlocked and released */
1454 if (retval < 0)
1455 break;
1456 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001458 if (swap_count(swcount) && start_mm != &init_mm)
1459 retval = unuse_mm(start_mm, entry, page);
1460
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001461 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 int set_start_mm = (*swap_map >= swcount);
1463 struct list_head *p = &start_mm->mmlist;
1464 struct mm_struct *new_start_mm = start_mm;
1465 struct mm_struct *prev_mm = start_mm;
1466 struct mm_struct *mm;
1467
1468 atomic_inc(&new_start_mm->mm_users);
1469 atomic_inc(&prev_mm->mm_users);
1470 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001471 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 (p = p->next) != &start_mm->mmlist) {
1473 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001474 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 spin_unlock(&mmlist_lock);
1477 mmput(prev_mm);
1478 prev_mm = mm;
1479
1480 cond_resched();
1481
1482 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001483 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001485 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001487 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001489
Bo Liu32c5fc12009-11-02 16:50:33 +00001490 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 mmput(new_start_mm);
1492 atomic_inc(&mm->mm_users);
1493 new_start_mm = mm;
1494 set_start_mm = 0;
1495 }
1496 spin_lock(&mmlist_lock);
1497 }
1498 spin_unlock(&mmlist_lock);
1499 mmput(prev_mm);
1500 mmput(start_mm);
1501 start_mm = new_start_mm;
1502 }
1503 if (retval) {
1504 unlock_page(page);
1505 page_cache_release(page);
1506 break;
1507 }
1508
1509 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 * If a reference remains (rare), we would like to leave
1511 * the page in the swap cache; but try_to_unmap could
1512 * then re-duplicate the entry once we drop page lock,
1513 * so we might loop indefinitely; also, that page could
1514 * not be swapped out to other storage meanwhile. So:
1515 * delete from cache even if there's another reference,
1516 * after ensuring that the data has been saved to disk -
1517 * since if the reference remains (rarer), it will be
1518 * read from disk into another page. Splitting into two
1519 * pages would be incorrect if swap supported "shared
1520 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001521 *
1522 * Given how unuse_vma() targets one particular offset
1523 * in an anon_vma, once the anon_vma has been determined,
1524 * this splitting happens to be just what is needed to
1525 * handle where KSM pages have been swapped out: re-reading
1526 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001528 if (swap_count(*swap_map) &&
1529 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 struct writeback_control wbc = {
1531 .sync_mode = WB_SYNC_NONE,
1532 };
1533
1534 swap_writepage(page, &wbc);
1535 lock_page(page);
1536 wait_on_page_writeback(page);
1537 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001538
1539 /*
1540 * It is conceivable that a racing task removed this page from
1541 * swap cache just before we acquired the page lock at the top,
1542 * or while we dropped it in unuse_mm(). The page might even
1543 * be back in swap cache on another swap area: that we must not
1544 * delete, since it may not have been written out to swap yet.
1545 */
1546 if (PageSwapCache(page) &&
1547 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001548 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 /*
1551 * So we could skip searching mms once swap count went
1552 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001553 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 */
1555 SetPageDirty(page);
1556 unlock_page(page);
1557 page_cache_release(page);
1558
1559 /*
1560 * Make sure that we aren't completely killing
1561 * interactive performance.
1562 */
1563 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001564 if (frontswap && pages_to_unuse > 0) {
1565 if (!--pages_to_unuse)
1566 break;
1567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 }
1569
1570 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 return retval;
1572}
1573
1574/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001575 * After a successful try_to_unuse, if no swap is now in use, we know
1576 * we can empty the mmlist. swap_lock must be held on entry and exit.
1577 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 * added to the mmlist just after page_duplicate - before would be racy.
1579 */
1580static void drain_mmlist(void)
1581{
1582 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001583 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001585 for (type = 0; type < nr_swapfiles; type++)
1586 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 return;
1588 spin_lock(&mmlist_lock);
1589 list_for_each_safe(p, next, &init_mm.mmlist)
1590 list_del_init(p);
1591 spin_unlock(&mmlist_lock);
1592}
1593
1594/*
1595 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001596 * corresponds to page offset for the specified swap entry.
1597 * Note that the type of this function is sector_t, but it returns page offset
1598 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001600static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001602 struct swap_info_struct *sis;
1603 struct swap_extent *start_se;
1604 struct swap_extent *se;
1605 pgoff_t offset;
1606
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001607 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001608 *bdev = sis->bdev;
1609
1610 offset = swp_offset(entry);
1611 start_se = sis->curr_swap_extent;
1612 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
1614 for ( ; ; ) {
1615 struct list_head *lh;
1616
1617 if (se->start_page <= offset &&
1618 offset < (se->start_page + se->nr_pages)) {
1619 return se->start_block + (offset - se->start_page);
1620 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001621 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 se = list_entry(lh, struct swap_extent, list);
1623 sis->curr_swap_extent = se;
1624 BUG_ON(se == start_se); /* It *must* be present */
1625 }
1626}
1627
1628/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001629 * Returns the page offset into bdev for the specified page's swap entry.
1630 */
1631sector_t map_swap_page(struct page *page, struct block_device **bdev)
1632{
1633 swp_entry_t entry;
1634 entry.val = page_private(page);
1635 return map_swap_entry(entry, bdev);
1636}
1637
1638/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 * Free all of a swapdev's extent information
1640 */
1641static void destroy_swap_extents(struct swap_info_struct *sis)
1642{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001643 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 struct swap_extent *se;
1645
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001646 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 struct swap_extent, list);
1648 list_del(&se->list);
1649 kfree(se);
1650 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001651
1652 if (sis->flags & SWP_FILE) {
1653 struct file *swap_file = sis->swap_file;
1654 struct address_space *mapping = swap_file->f_mapping;
1655
1656 sis->flags &= ~SWP_FILE;
1657 mapping->a_ops->swap_deactivate(swap_file);
1658 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
1661/*
1662 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001663 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001665 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 */
Mel Gormana509bc12012-07-31 16:44:57 -07001667int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1669 unsigned long nr_pages, sector_t start_block)
1670{
1671 struct swap_extent *se;
1672 struct swap_extent *new_se;
1673 struct list_head *lh;
1674
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001675 if (start_page == 0) {
1676 se = &sis->first_swap_extent;
1677 sis->curr_swap_extent = se;
1678 se->start_page = 0;
1679 se->nr_pages = nr_pages;
1680 se->start_block = start_block;
1681 return 1;
1682 } else {
1683 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001685 BUG_ON(se->start_page + se->nr_pages != start_page);
1686 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 /* Merge it */
1688 se->nr_pages += nr_pages;
1689 return 0;
1690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 }
1692
1693 /*
1694 * No merge. Insert a new extent, preserving ordering.
1695 */
1696 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1697 if (new_se == NULL)
1698 return -ENOMEM;
1699 new_se->start_page = start_page;
1700 new_se->nr_pages = nr_pages;
1701 new_se->start_block = start_block;
1702
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001703 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001704 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705}
1706
1707/*
1708 * A `swap extent' is a simple thing which maps a contiguous range of pages
1709 * onto a contiguous range of disk blocks. An ordered list of swap extents
1710 * is built at swapon time and is then used at swap_writepage/swap_readpage
1711 * time for locating where on disk a page belongs.
1712 *
1713 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1714 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1715 * swap files identically.
1716 *
1717 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1718 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1719 * swapfiles are handled *identically* after swapon time.
1720 *
1721 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1722 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1723 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1724 * requirements, they are simply tossed out - we will never use those blocks
1725 * for swapping.
1726 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001727 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1729 * which will scribble on the fs.
1730 *
1731 * The amount of disk space which a single swap extent represents varies.
1732 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1733 * extents in the list. To avoid much list walking, we cache the previous
1734 * search location in `curr_swap_extent', and start new searches from there.
1735 * This is extremely effective. The average number of iterations in
1736 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1737 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001738static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
Mel Gorman62c230b2012-07-31 16:44:55 -07001740 struct file *swap_file = sis->swap_file;
1741 struct address_space *mapping = swap_file->f_mapping;
1742 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 int ret;
1744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 if (S_ISBLK(inode->i_mode)) {
1746 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001747 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001748 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 }
1750
Mel Gorman62c230b2012-07-31 16:44:55 -07001751 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001752 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001753 if (!ret) {
1754 sis->flags |= SWP_FILE;
1755 ret = add_swap_extent(sis, 0, sis->max, 0);
1756 *span = sis->pages;
1757 }
Mel Gormana509bc12012-07-31 16:44:57 -07001758 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001759 }
1760
Mel Gormana509bc12012-07-31 16:44:57 -07001761 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762}
1763
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001764static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001765 unsigned char *swap_map,
1766 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001767{
1768 int i, prev;
1769
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001770 if (prio >= 0)
1771 p->prio = prio;
1772 else
1773 p->prio = --least_priority;
1774 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001775 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001776 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001777 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001778 total_swap_pages += p->pages;
1779
1780 /* insert swap space into swap_list: */
1781 prev = -1;
1782 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1783 if (p->prio >= swap_info[i]->prio)
1784 break;
1785 prev = i;
1786 }
1787 p->next = i;
1788 if (prev < 0)
1789 swap_list.head = swap_list.next = p->type;
1790 else
1791 swap_info[prev]->next = p->type;
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001792}
1793
1794static void enable_swap_info(struct swap_info_struct *p, int prio,
1795 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001796 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001797 unsigned long *frontswap_map)
1798{
Minchan Kim4f898492013-04-30 15:26:54 -07001799 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001800 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001801 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001802 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001803 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001804 spin_unlock(&swap_lock);
1805}
1806
1807static void reinsert_swap_info(struct swap_info_struct *p)
1808{
1809 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001810 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001811 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001812 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001813 spin_unlock(&swap_lock);
1814}
1815
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001816SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001818 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001819 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001820 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001821 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 struct file *swap_file, *victim;
1823 struct address_space *mapping;
1824 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001825 struct filename *pathname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 int i, type, prev;
1827 int err;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001828 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001829
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 if (!capable(CAP_SYS_ADMIN))
1831 return -EPERM;
1832
Al Viro191c5422012-02-13 03:58:52 +00001833 BUG_ON(!current->mm);
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001837 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
Jeff Layton669abf42012-10-10 16:43:10 -04001839 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 err = PTR_ERR(victim);
1841 if (IS_ERR(victim))
1842 goto out;
1843
1844 mapping = victim->f_mapping;
1845 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001846 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001847 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1848 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001849 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 if (p->swap_file->f_mapping == mapping)
1851 break;
1852 }
1853 prev = type;
1854 }
1855 if (type < 0) {
1856 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001857 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 goto out_dput;
1859 }
Al Viro191c5422012-02-13 03:58:52 +00001860 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 vm_unacct_memory(p->pages);
1862 else {
1863 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001864 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 goto out_dput;
1866 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001867 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001869 else
1870 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 if (type == swap_list.next) {
1872 /* just pick something that's safe... */
1873 swap_list.next = swap_list.head;
1874 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001875 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001876 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001877 for (i = p->next; i >= 0; i = swap_info[i]->next)
1878 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001879 least_priority++;
1880 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001881 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 total_swap_pages -= p->pages;
1883 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001884 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001885 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001886
David Rientjese1e12d22012-12-11 16:02:56 -08001887 set_current_oom_origin();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001888 err = try_to_unuse(type, false, 0); /* force all pages to be unused */
David Rientjese1e12d22012-12-11 16:02:56 -08001889 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 if (err) {
1892 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001893 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 goto out_dput;
1895 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001896
Shaohua Li815c2c52013-09-11 14:20:30 -07001897 flush_work(&p->discard_work);
1898
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001899 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001900 if (p->flags & SWP_CONTINUED)
1901 free_swap_count_continuations(p);
1902
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001903 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001904 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001905 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001907
1908 /* wait for anyone still in scan_swap_map */
1909 p->highest_bit = 0; /* cuts scans short */
1910 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001911 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001912 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001913 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001914 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001915 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001916 }
1917
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001919 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 p->swap_file = NULL;
1921 p->max = 0;
1922 swap_map = p->swap_map;
1923 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001924 cluster_info = p->cluster_info;
1925 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07001926 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08001927 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001928 spin_unlock(&swap_lock);
Minchan Kim4f898492013-04-30 15:26:54 -07001929 frontswap_invalidate_area(type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08001930 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001931 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001932 free_percpu(p->percpu_cluster);
1933 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001935 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001936 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001937 /* Destroy swap account information */
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001938 swap_cgroup_swapoff(type);
1939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 inode = mapping->host;
1941 if (S_ISBLK(inode->i_mode)) {
1942 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001943 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001944 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001946 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001948 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 }
1950 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08001951
1952 /*
1953 * Clear the SWP_USED flag after all resources are freed so that swapon
1954 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1955 * not hold p->lock after we cleared its SWP_WRITEOK.
1956 */
1957 spin_lock(&swap_lock);
1958 p->flags = 0;
1959 spin_unlock(&swap_lock);
1960
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001962 atomic_inc(&proc_poll_event);
1963 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
1965out_dput:
1966 filp_close(victim, NULL);
1967out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001968 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 return err;
1970}
1971
1972#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001973static unsigned swaps_poll(struct file *file, poll_table *wait)
1974{
Kay Sieversf1514632011-07-12 20:48:39 +02001975 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001976
1977 poll_wait(file, &proc_poll_wait, wait);
1978
Kay Sieversf1514632011-07-12 20:48:39 +02001979 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1980 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001981 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1982 }
1983
1984 return POLLIN | POLLRDNORM;
1985}
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987/* iterator */
1988static void *swap_start(struct seq_file *swap, loff_t *pos)
1989{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001990 struct swap_info_struct *si;
1991 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 loff_t l = *pos;
1993
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001994 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001996 if (!l)
1997 return SEQ_START_TOKEN;
1998
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001999 for (type = 0; type < nr_swapfiles; type++) {
2000 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2001 si = swap_info[type];
2002 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002004 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002005 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 }
2007
2008 return NULL;
2009}
2010
2011static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2012{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002013 struct swap_info_struct *si = v;
2014 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002016 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002017 type = 0;
2018 else
2019 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002020
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002021 for (; type < nr_swapfiles; type++) {
2022 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2023 si = swap_info[type];
2024 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 continue;
2026 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002027 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 }
2029
2030 return NULL;
2031}
2032
2033static void swap_stop(struct seq_file *swap, void *v)
2034{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002035 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036}
2037
2038static int swap_show(struct seq_file *swap, void *v)
2039{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002040 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 struct file *file;
2042 int len;
2043
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002044 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002045 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2046 return 0;
2047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002049 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08002050 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002051 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002052 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002053 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002055 si->pages << (PAGE_SHIFT - 10),
2056 si->inuse_pages << (PAGE_SHIFT - 10),
2057 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 return 0;
2059}
2060
Helge Deller15ad7cd2006-12-06 20:40:36 -08002061static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 .start = swap_start,
2063 .next = swap_next,
2064 .stop = swap_stop,
2065 .show = swap_show
2066};
2067
2068static int swaps_open(struct inode *inode, struct file *file)
2069{
Kay Sieversf1514632011-07-12 20:48:39 +02002070 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002071 int ret;
2072
Kay Sievers66d7dd52010-10-26 14:22:06 -07002073 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002074 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002075 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002076
Kay Sieversf1514632011-07-12 20:48:39 +02002077 seq = file->private_data;
2078 seq->poll_event = atomic_read(&proc_poll_event);
2079 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080}
2081
Helge Deller15ad7cd2006-12-06 20:40:36 -08002082static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 .open = swaps_open,
2084 .read = seq_read,
2085 .llseek = seq_lseek,
2086 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002087 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088};
2089
2090static int __init procswaps_init(void)
2091{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002092 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 return 0;
2094}
2095__initcall(procswaps_init);
2096#endif /* CONFIG_PROC_FS */
2097
Jan Beulich17963162008-12-16 11:35:24 +00002098#ifdef MAX_SWAPFILES_CHECK
2099static int __init max_swapfiles_check(void)
2100{
2101 MAX_SWAPFILES_CHECK();
2102 return 0;
2103}
2104late_initcall(max_swapfiles_check);
2105#endif
2106
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002107static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002109 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002111
2112 p = kzalloc(sizeof(*p), GFP_KERNEL);
2113 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002114 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002115
Hugh Dickins5d337b92005-09-03 15:54:41 -07002116 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002117 for (type = 0; type < nr_swapfiles; type++) {
2118 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002120 }
Christoph Lameter06972122006-06-23 02:03:35 -07002121 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002122 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002123 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002124 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002126 if (type >= nr_swapfiles) {
2127 p->type = type;
2128 swap_info[type] = p;
2129 /*
2130 * Write swap_info[type] before nr_swapfiles, in case a
2131 * racing procfs swap_start() or swap_next() is reading them.
2132 * (We never shrink nr_swapfiles, we never free this entry.)
2133 */
2134 smp_wmb();
2135 nr_swapfiles++;
2136 } else {
2137 kfree(p);
2138 p = swap_info[type];
2139 /*
2140 * Do not memset this entry: a racing procfs swap_next()
2141 * would be relying on p->type to remain valid.
2142 */
2143 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002144 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002147 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002148 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002149
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002150 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002151}
2152
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002153static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2154{
2155 int error;
2156
2157 if (S_ISBLK(inode->i_mode)) {
2158 p->bdev = bdgrab(I_BDEV(inode));
2159 error = blkdev_get(p->bdev,
2160 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
2161 sys_swapon);
2162 if (error < 0) {
2163 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002164 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002165 }
2166 p->old_block_size = block_size(p->bdev);
2167 error = set_blocksize(p->bdev, PAGE_SIZE);
2168 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002169 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002170 p->flags |= SWP_BLKDEV;
2171 } else if (S_ISREG(inode->i_mode)) {
2172 p->bdev = inode->i_sb->s_bdev;
2173 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002174 if (IS_SWAPFILE(inode))
2175 return -EBUSY;
2176 } else
2177 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002178
2179 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002180}
2181
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002182static unsigned long read_swap_header(struct swap_info_struct *p,
2183 union swap_header *swap_header,
2184 struct inode *inode)
2185{
2186 int i;
2187 unsigned long maxpages;
2188 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002189 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002190
2191 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002192 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002193 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002194 }
2195
2196 /* swap partition endianess hack... */
2197 if (swab32(swap_header->info.version) == 1) {
2198 swab32s(&swap_header->info.version);
2199 swab32s(&swap_header->info.last_page);
2200 swab32s(&swap_header->info.nr_badpages);
2201 for (i = 0; i < swap_header->info.nr_badpages; i++)
2202 swab32s(&swap_header->info.badpages[i]);
2203 }
2204 /* Check the swap header's sub-version */
2205 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002206 pr_warn("Unable to handle swap header version %d\n",
2207 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002208 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002209 }
2210
2211 p->lowest_bit = 1;
2212 p->cluster_next = 1;
2213 p->cluster_nr = 0;
2214
2215 /*
2216 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002217 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002218 * of bits for the swap offset in the swp_entry_t type, and
2219 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002220 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002221 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002222 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002223 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002224 * offset is extracted. This will mask all the bits from
2225 * the initial ~0UL mask that can't be encoded in either
2226 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002227 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002228 */
2229 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002230 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002231 last_page = swap_header->info.last_page;
2232 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002233 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002234 maxpages << (PAGE_SHIFT - 10),
2235 last_page << (PAGE_SHIFT - 10));
2236 }
2237 if (maxpages > last_page) {
2238 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002239 /* p->max is an unsigned int: don't overflow it */
2240 if ((unsigned int)maxpages == 0)
2241 maxpages = UINT_MAX;
2242 }
2243 p->highest_bit = maxpages - 1;
2244
2245 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002246 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002247 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2248 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002249 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002250 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002251 }
2252 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002253 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002254 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002255 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002256
2257 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002258}
2259
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002260static int setup_swap_map_and_extents(struct swap_info_struct *p,
2261 union swap_header *swap_header,
2262 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002263 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002264 unsigned long maxpages,
2265 sector_t *span)
2266{
2267 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002268 unsigned int nr_good_pages;
2269 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002270 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2271 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002272
2273 nr_good_pages = maxpages - 1; /* omit header page */
2274
Shaohua Li2a8f9442013-09-11 14:20:28 -07002275 cluster_set_null(&p->free_cluster_head);
2276 cluster_set_null(&p->free_cluster_tail);
Shaohua Li815c2c52013-09-11 14:20:30 -07002277 cluster_set_null(&p->discard_cluster_head);
2278 cluster_set_null(&p->discard_cluster_tail);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002279
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002280 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2281 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002282 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2283 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002284 if (page_nr < maxpages) {
2285 swap_map[page_nr] = SWAP_MAP_BAD;
2286 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002287 /*
2288 * Haven't marked the cluster free yet, no list
2289 * operation involved
2290 */
2291 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002292 }
2293 }
2294
Shaohua Li2a8f9442013-09-11 14:20:28 -07002295 /* Haven't marked the cluster free yet, no list operation involved */
2296 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2297 inc_cluster_info_page(p, cluster_info, i);
2298
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002299 if (nr_good_pages) {
2300 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002301 /*
2302 * Not mark the cluster free yet, no list
2303 * operation involved
2304 */
2305 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002306 p->max = maxpages;
2307 p->pages = nr_good_pages;
2308 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002309 if (nr_extents < 0)
2310 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002311 nr_good_pages = p->pages;
2312 }
2313 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002314 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002315 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002316 }
2317
Shaohua Li2a8f9442013-09-11 14:20:28 -07002318 if (!cluster_info)
2319 return nr_extents;
2320
2321 for (i = 0; i < nr_clusters; i++) {
2322 if (!cluster_count(&cluster_info[idx])) {
2323 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2324 if (cluster_is_null(&p->free_cluster_head)) {
2325 cluster_set_next_flag(&p->free_cluster_head,
2326 idx, 0);
2327 cluster_set_next_flag(&p->free_cluster_tail,
2328 idx, 0);
2329 } else {
2330 unsigned int tail;
2331
2332 tail = cluster_next(&p->free_cluster_tail);
2333 cluster_set_next(&cluster_info[tail], idx);
2334 cluster_set_next_flag(&p->free_cluster_tail,
2335 idx, 0);
2336 }
2337 }
2338 idx++;
2339 if (idx == nr_clusters)
2340 idx = 0;
2341 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002342 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002343}
2344
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002345/*
2346 * Helper to sys_swapon determining if a given swap
2347 * backing device queue supports DISCARD operations.
2348 */
2349static bool swap_discardable(struct swap_info_struct *si)
2350{
2351 struct request_queue *q = bdev_get_queue(si->bdev);
2352
2353 if (!q || !blk_queue_discard(q))
2354 return false;
2355
2356 return true;
2357}
2358
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002359SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2360{
2361 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002362 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002363 struct file *swap_file = NULL;
2364 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002365 int i;
2366 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002367 int error;
2368 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002369 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002370 sector_t span;
2371 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002372 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002373 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002374 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002375 struct page *page = NULL;
2376 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002377
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002378 if (swap_flags & ~SWAP_FLAGS_VALID)
2379 return -EINVAL;
2380
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002381 if (!capable(CAP_SYS_ADMIN))
2382 return -EPERM;
2383
2384 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002385 if (IS_ERR(p))
2386 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002387
Shaohua Li815c2c52013-09-11 14:20:30 -07002388 INIT_WORK(&p->discard_work, swap_discard_work);
2389
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002392 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002394 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 }
Jeff Layton669abf42012-10-10 16:43:10 -04002396 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002398 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002400 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 }
2402
2403 p->swap_file = swap_file;
2404 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002407 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002409 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002411 if (mapping == q->swap_file->f_mapping) {
2412 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 }
2416
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002417 inode = mapping->host;
2418 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002419 error = claim_swapfile(p, inode);
2420 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 /*
2424 * Read the swap header.
2425 */
2426 if (!mapping->a_ops->readpage) {
2427 error = -EINVAL;
2428 goto bad_swap;
2429 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002430 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 if (IS_ERR(page)) {
2432 error = PTR_ERR(page);
2433 goto bad_swap;
2434 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002435 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002437 maxpages = read_swap_header(p, swap_header, inode);
2438 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 error = -EINVAL;
2440 goto bad_swap;
2441 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002442
Hugh Dickins81e33972009-01-06 14:39:49 -08002443 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002444 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002445 if (!swap_map) {
2446 error = -ENOMEM;
2447 goto bad_swap;
2448 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002449 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2450 p->flags |= SWP_SOLIDSTATE;
2451 /*
2452 * select a random position to start with to help wear leveling
2453 * SSD
2454 */
2455 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2456
2457 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2458 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2459 if (!cluster_info) {
2460 error = -ENOMEM;
2461 goto bad_swap;
2462 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002463 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2464 if (!p->percpu_cluster) {
2465 error = -ENOMEM;
2466 goto bad_swap;
2467 }
2468 for_each_possible_cpu(i) {
2469 struct percpu_cluster *cluster;
2470 cluster = per_cpu_ptr(p->percpu_cluster, i);
2471 cluster_set_null(&cluster->index);
2472 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002473 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002474
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002475 error = swap_cgroup_swapon(p->type, maxpages);
2476 if (error)
2477 goto bad_swap;
2478
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002479 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002480 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002481 if (unlikely(nr_extents < 0)) {
2482 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 goto bad_swap;
2484 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002485 /* frontswap enabled? set up bit-per-page map for frontswap */
2486 if (frontswap_enabled)
Akinobu Mita7b579762013-06-12 14:05:08 -07002487 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Shaohua Li2a8f9442013-09-11 14:20:28 -07002489 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2490 /*
2491 * When discard is enabled for swap with no particular
2492 * policy flagged, we set all swap discard flags here in
2493 * order to sustain backward compatibility with older
2494 * swapon(8) releases.
2495 */
2496 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2497 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002498
Shaohua Li2a8f9442013-09-11 14:20:28 -07002499 /*
2500 * By flagging sys_swapon, a sysadmin can tell us to
2501 * either do single-time area discards only, or to just
2502 * perform discards for released swap page-clusters.
2503 * Now it's time to adjust the p->flags accordingly.
2504 */
2505 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2506 p->flags &= ~SWP_PAGE_DISCARD;
2507 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2508 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002509
Shaohua Li2a8f9442013-09-11 14:20:28 -07002510 /* issue a swapon-time discard if it's still required */
2511 if (p->flags & SWP_AREA_DISCARD) {
2512 int err = discard_swap(p);
2513 if (unlikely(err))
2514 pr_err("swapon: discard_swap(%p): %d\n",
2515 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002516 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002517 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002518
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002519 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002520 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002521 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002522 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002523 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002524 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002525
Andrew Morton465c47f2013-09-11 14:20:17 -07002526 pr_info("Adding %uk swap on %s. "
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002527 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002528 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002529 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2530 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002531 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002532 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2533 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002534 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002535
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002536 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002537 atomic_inc(&proc_poll_event);
2538 wake_up_interruptible(&proc_poll_wait);
2539
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002540 if (S_ISREG(inode->i_mode))
2541 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 error = 0;
2543 goto out;
2544bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002545 free_percpu(p->percpu_cluster);
2546 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002547 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002548 set_blocksize(p->bdev, p->old_block_size);
2549 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002551 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002552 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002553 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002556 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002558 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002559 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002560 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002561 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002562 inode = NULL;
2563 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566out:
2567 if (page && !IS_ERR(page)) {
2568 kunmap(page);
2569 page_cache_release(page);
2570 }
2571 if (name)
2572 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002573 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002574 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 return error;
2576}
2577
2578void si_swapinfo(struct sysinfo *val)
2579{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002580 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 unsigned long nr_to_be_unused = 0;
2582
Hugh Dickins5d337b92005-09-03 15:54:41 -07002583 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002584 for (type = 0; type < nr_swapfiles; type++) {
2585 struct swap_info_struct *si = swap_info[type];
2586
2587 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2588 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002590 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002592 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593}
2594
2595/*
2596 * Verify that a swap entry is valid and increment its swap map count.
2597 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002598 * Returns error code in following case.
2599 * - success -> 0
2600 * - swp_entry is invalid -> EINVAL
2601 * - swp_entry is migration entry -> EINVAL
2602 * - swap-cache reference is requested but there is already one. -> EEXIST
2603 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002604 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002606static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002608 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002610 unsigned char count;
2611 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002612 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Andi Kleena7420aa2009-09-16 11:50:05 +02002614 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002615 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002616
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 type = swp_type(entry);
2618 if (type >= nr_swapfiles)
2619 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002620 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 offset = swp_offset(entry);
2622
Shaohua Liec8acf22013-02-22 16:34:38 -08002623 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002624 if (unlikely(offset >= p->max))
2625 goto unlock_out;
2626
Hugh Dickins253d5532009-12-14 17:58:44 -08002627 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002628
2629 /*
2630 * swapin_readahead() doesn't check if a swap entry is valid, so the
2631 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2632 */
2633 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2634 err = -ENOENT;
2635 goto unlock_out;
2636 }
2637
Hugh Dickins253d5532009-12-14 17:58:44 -08002638 has_cache = count & SWAP_HAS_CACHE;
2639 count &= ~SWAP_HAS_CACHE;
2640 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002641
Hugh Dickins253d5532009-12-14 17:58:44 -08002642 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002643
2644 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002645 if (!has_cache && count)
2646 has_cache = SWAP_HAS_CACHE;
2647 else if (has_cache) /* someone else added cache */
2648 err = -EEXIST;
2649 else /* no users remaining */
2650 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002651
2652 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002653
Hugh Dickins570a335b2009-12-14 17:58:46 -08002654 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2655 count += usage;
2656 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002657 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002658 else if (swap_count_continued(p, offset, count))
2659 count = COUNT_CONTINUED;
2660 else
2661 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002662 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002663 err = -ENOENT; /* unused swap entry */
2664
2665 p->swap_map[offset] = count | has_cache;
2666
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002667unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002668 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002670 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
2672bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002673 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 goto out;
2675}
Hugh Dickins253d5532009-12-14 17:58:44 -08002676
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002677/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002678 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2679 * (in which case its reference count is never incremented).
2680 */
2681void swap_shmem_alloc(swp_entry_t entry)
2682{
2683 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2684}
2685
2686/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002687 * Increase reference count of swap entry by 1.
2688 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2689 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2690 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2691 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002692 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002693int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002694{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002695 int err = 0;
2696
2697 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2698 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2699 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002702/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002703 * @entry: swap entry for which we allocate swap cache.
2704 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002705 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002706 * This can return error codes. Returns 0 at success.
2707 * -EBUSY means there is a swap cache.
2708 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002709 */
2710int swapcache_prepare(swp_entry_t entry)
2711{
Hugh Dickins253d5532009-12-14 17:58:44 -08002712 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002713}
2714
Mel Gormanf981c592012-07-31 16:44:47 -07002715struct swap_info_struct *page_swap_info(struct page *page)
2716{
2717 swp_entry_t swap = { .val = page_private(page) };
2718 BUG_ON(!PageSwapCache(page));
2719 return swap_info[swp_type(swap)];
2720}
2721
2722/*
2723 * out-of-line __page_file_ methods to avoid include hell.
2724 */
2725struct address_space *__page_file_mapping(struct page *page)
2726{
Sasha Levin309381fea2014-01-23 15:52:54 -08002727 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002728 return page_swap_info(page)->swap_file->f_mapping;
2729}
2730EXPORT_SYMBOL_GPL(__page_file_mapping);
2731
2732pgoff_t __page_file_index(struct page *page)
2733{
2734 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002735 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002736 return swp_offset(swap);
2737}
2738EXPORT_SYMBOL_GPL(__page_file_index);
2739
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002741 * add_swap_count_continuation - called when a swap count is duplicated
2742 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2743 * page of the original vmalloc'ed swap_map, to hold the continuation count
2744 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2745 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2746 *
2747 * These continuation pages are seldom referenced: the common paths all work
2748 * on the original swap_map, only referring to a continuation page when the
2749 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2750 *
2751 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2752 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2753 * can be called after dropping locks.
2754 */
2755int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2756{
2757 struct swap_info_struct *si;
2758 struct page *head;
2759 struct page *page;
2760 struct page *list_page;
2761 pgoff_t offset;
2762 unsigned char count;
2763
2764 /*
2765 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2766 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2767 */
2768 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2769
2770 si = swap_info_get(entry);
2771 if (!si) {
2772 /*
2773 * An acceptable race has occurred since the failing
2774 * __swap_duplicate(): the swap entry has been freed,
2775 * perhaps even the whole swap_map cleared for swapoff.
2776 */
2777 goto outer;
2778 }
2779
2780 offset = swp_offset(entry);
2781 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2782
2783 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2784 /*
2785 * The higher the swap count, the more likely it is that tasks
2786 * will race to add swap count continuation: we need to avoid
2787 * over-provisioning.
2788 */
2789 goto out;
2790 }
2791
2792 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002793 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002794 return -ENOMEM;
2795 }
2796
2797 /*
2798 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002799 * no architecture is using highmem pages for kernel page tables: so it
2800 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002801 */
2802 head = vmalloc_to_page(si->swap_map + offset);
2803 offset &= ~PAGE_MASK;
2804
2805 /*
2806 * Page allocation does not initialize the page's lru field,
2807 * but it does always reset its private field.
2808 */
2809 if (!page_private(head)) {
2810 BUG_ON(count & COUNT_CONTINUED);
2811 INIT_LIST_HEAD(&head->lru);
2812 set_page_private(head, SWP_CONTINUED);
2813 si->flags |= SWP_CONTINUED;
2814 }
2815
2816 list_for_each_entry(list_page, &head->lru, lru) {
2817 unsigned char *map;
2818
2819 /*
2820 * If the previous map said no continuation, but we've found
2821 * a continuation page, free our allocation and use this one.
2822 */
2823 if (!(count & COUNT_CONTINUED))
2824 goto out;
2825
Cong Wang9b04c5f2011-11-25 23:14:39 +08002826 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002827 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002828 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002829
2830 /*
2831 * If this continuation count now has some space in it,
2832 * free our allocation and use this one.
2833 */
2834 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2835 goto out;
2836 }
2837
2838 list_add_tail(&page->lru, &head->lru);
2839 page = NULL; /* now it's attached, don't free it */
2840out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002841 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002842outer:
2843 if (page)
2844 __free_page(page);
2845 return 0;
2846}
2847
2848/*
2849 * swap_count_continued - when the original swap_map count is incremented
2850 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2851 * into, carry if so, or else fail until a new continuation page is allocated;
2852 * when the original swap_map count is decremented from 0 with continuation,
2853 * borrow from the continuation and report whether it still holds more.
2854 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2855 */
2856static bool swap_count_continued(struct swap_info_struct *si,
2857 pgoff_t offset, unsigned char count)
2858{
2859 struct page *head;
2860 struct page *page;
2861 unsigned char *map;
2862
2863 head = vmalloc_to_page(si->swap_map + offset);
2864 if (page_private(head) != SWP_CONTINUED) {
2865 BUG_ON(count & COUNT_CONTINUED);
2866 return false; /* need to add count continuation */
2867 }
2868
2869 offset &= ~PAGE_MASK;
2870 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002871 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002872
2873 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2874 goto init_map; /* jump over SWAP_CONT_MAX checks */
2875
2876 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2877 /*
2878 * Think of how you add 1 to 999
2879 */
2880 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002881 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002882 page = list_entry(page->lru.next, struct page, lru);
2883 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002884 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002885 }
2886 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002887 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002888 page = list_entry(page->lru.next, struct page, lru);
2889 if (page == head)
2890 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002891 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002892init_map: *map = 0; /* we didn't zero the page */
2893 }
2894 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002895 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002896 page = list_entry(page->lru.prev, struct page, lru);
2897 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002898 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002899 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002900 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002901 page = list_entry(page->lru.prev, struct page, lru);
2902 }
2903 return true; /* incremented */
2904
2905 } else { /* decrementing */
2906 /*
2907 * Think of how you subtract 1 from 1000
2908 */
2909 BUG_ON(count != COUNT_CONTINUED);
2910 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002911 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002912 page = list_entry(page->lru.next, struct page, lru);
2913 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002914 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002915 }
2916 BUG_ON(*map == 0);
2917 *map -= 1;
2918 if (*map == 0)
2919 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002920 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002921 page = list_entry(page->lru.prev, struct page, lru);
2922 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002923 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002924 *map = SWAP_CONT_MAX | count;
2925 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002926 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002927 page = list_entry(page->lru.prev, struct page, lru);
2928 }
2929 return count == COUNT_CONTINUED;
2930 }
2931}
2932
2933/*
2934 * free_swap_count_continuations - swapoff free all the continuation pages
2935 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2936 */
2937static void free_swap_count_continuations(struct swap_info_struct *si)
2938{
2939 pgoff_t offset;
2940
2941 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2942 struct page *head;
2943 head = vmalloc_to_page(si->swap_map + offset);
2944 if (page_private(head)) {
2945 struct list_head *this, *next;
2946 list_for_each_safe(this, next, &head->lru) {
2947 struct page *page;
2948 page = list_entry(this, struct page, lru);
2949 list_del(this);
2950 __free_page(page);
2951 }
2952 }
2953 }
2954}