blob: d76b2a18f044a07b8428e463528a91041790d37e [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>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080041#include <linux/swap_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;
Chris Wilsonfb0fec52015-12-04 15:58:53 +000051/*
52 * Some modules use swappable objects and may try to swap them out under
53 * memory pressure (via the shrinker). Before doing so, they may wish to
54 * check to see if any swap space is available.
55 */
56EXPORT_SYMBOL_GPL(nr_swap_pages);
Shaohua Liec8acf22013-02-22 16:34:38 -080057/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070058long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070059static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061static const char Bad_file[] = "Bad swap file entry ";
62static const char Unused_file[] = "Unused swap file entry ";
63static const char Bad_offset[] = "Bad swap offset entry ";
64static const char Unused_offset[] = "Unused swap offset entry ";
65
Dan Streetmanadfab832014-06-04 16:09:53 -070066/*
67 * all active swap_info_structs
68 * protected with swap_lock, and ordered by priority.
69 */
Dan Streetman18ab4d42014-06-04 16:09:59 -070070PLIST_HEAD(swap_active_head);
71
72/*
73 * all available (active, not full) swap_info_structs
74 * protected with swap_avail_lock, ordered by priority.
75 * This is used by get_swap_page() instead of swap_active_head
76 * because swap_active_head includes all swap_info_structs,
77 * but get_swap_page() doesn't need to look at full ones.
78 * This uses its own lock instead of swap_lock because when a
79 * swap_info_struct changes between not-full/full, it needs to
80 * add/remove itself to/from this list, but the swap_info_struct->lock
81 * is held and the locking order requires swap_lock to be taken
82 * before any swap_info_struct->lock.
83 */
84static PLIST_HEAD(swap_avail_head);
85static DEFINE_SPINLOCK(swap_avail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060087struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Ingo Molnarfc0abb12006-01-18 17:42:33 -080089static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Kay Sievers66d7dd52010-10-26 14:22:06 -070091static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
92/* Activity counter to indicate that a swapon or swapoff has occurred */
93static atomic_t proc_poll_event = ATOMIC_INIT(0);
94
Hugh Dickins8d69aae2009-12-14 17:58:45 -080095static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070096{
Hugh Dickins570a335b2009-12-14 17:58:46 -080097 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070098}
99
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800100/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700101static int
102__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
103{
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800104 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700105 struct page *page;
106 int ret = 0;
107
Huang Yingf6ab1f72016-10-07 17:00:21 -0700108 page = find_get_page(swap_address_space(entry), swp_offset(entry));
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700109 if (!page)
110 return 0;
111 /*
112 * This function is called from scan_swap_map() and it's called
113 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
114 * We have to use trylock for avoiding deadlock. This is a special
115 * case and you should use try_to_free_swap() with explicit lock_page()
116 * in usual operations.
117 */
118 if (trylock_page(page)) {
119 ret = try_to_free_swap(page);
120 unlock_page(page);
121 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300122 put_page(page);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700123 return ret;
124}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800127 * swapon tell device that all the old swap contents can be discarded,
128 * to allow the swap device to optimize its wear-levelling.
129 */
130static int discard_swap(struct swap_info_struct *si)
131{
132 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800133 sector_t start_block;
134 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800135 int err = 0;
136
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800137 /* Do not discard the swap header page! */
138 se = &si->first_swap_extent;
139 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
140 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
141 if (nr_blocks) {
142 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200143 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800144 if (err)
145 return err;
146 cond_resched();
147 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800148
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800149 list_for_each_entry(se, &si->first_swap_extent.list, list) {
150 start_block = se->start_block << (PAGE_SHIFT - 9);
151 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800152
153 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200154 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800155 if (err)
156 break;
157
158 cond_resched();
159 }
160 return err; /* That will often be -EOPNOTSUPP */
161}
162
Hugh Dickins7992fde2009-01-06 14:39:53 -0800163/*
164 * swap allocation tell device that a cluster of swap can now be discarded,
165 * to allow the swap device to optimize its wear-levelling.
166 */
167static void discard_swap_cluster(struct swap_info_struct *si,
168 pgoff_t start_page, pgoff_t nr_pages)
169{
170 struct swap_extent *se = si->curr_swap_extent;
171 int found_extent = 0;
172
173 while (nr_pages) {
Hugh Dickins7992fde2009-01-06 14:39:53 -0800174 if (se->start_page <= start_page &&
175 start_page < se->start_page + se->nr_pages) {
176 pgoff_t offset = start_page - se->start_page;
177 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800178 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800179
180 if (nr_blocks > nr_pages)
181 nr_blocks = nr_pages;
182 start_page += nr_blocks;
183 nr_pages -= nr_blocks;
184
185 if (!found_extent++)
186 si->curr_swap_extent = se;
187
188 start_block <<= PAGE_SHIFT - 9;
189 nr_blocks <<= PAGE_SHIFT - 9;
190 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200191 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800192 break;
193 }
194
Geliang Tanga8ae4992016-01-14 15:20:45 -0800195 se = list_next_entry(se, list);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800196 }
197}
198
Hugh Dickins048c27f2005-09-03 15:54:40 -0700199#define SWAPFILE_CLUSTER 256
200#define LATENCY_LIMIT 256
201
Shaohua Li2a8f9442013-09-11 14:20:28 -0700202static inline void cluster_set_flag(struct swap_cluster_info *info,
203 unsigned int flag)
204{
205 info->flags = flag;
206}
207
208static inline unsigned int cluster_count(struct swap_cluster_info *info)
209{
210 return info->data;
211}
212
213static inline void cluster_set_count(struct swap_cluster_info *info,
214 unsigned int c)
215{
216 info->data = c;
217}
218
219static inline void cluster_set_count_flag(struct swap_cluster_info *info,
220 unsigned int c, unsigned int f)
221{
222 info->flags = f;
223 info->data = c;
224}
225
226static inline unsigned int cluster_next(struct swap_cluster_info *info)
227{
228 return info->data;
229}
230
231static inline void cluster_set_next(struct swap_cluster_info *info,
232 unsigned int n)
233{
234 info->data = n;
235}
236
237static inline void cluster_set_next_flag(struct swap_cluster_info *info,
238 unsigned int n, unsigned int f)
239{
240 info->flags = f;
241 info->data = n;
242}
243
244static inline bool cluster_is_free(struct swap_cluster_info *info)
245{
246 return info->flags & CLUSTER_FLAG_FREE;
247}
248
249static inline bool cluster_is_null(struct swap_cluster_info *info)
250{
251 return info->flags & CLUSTER_FLAG_NEXT_NULL;
252}
253
254static inline void cluster_set_null(struct swap_cluster_info *info)
255{
256 info->flags = CLUSTER_FLAG_NEXT_NULL;
257 info->data = 0;
258}
259
Huang Ying6b534912016-10-07 16:58:42 -0700260static inline bool cluster_list_empty(struct swap_cluster_list *list)
261{
262 return cluster_is_null(&list->head);
263}
264
265static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
266{
267 return cluster_next(&list->head);
268}
269
270static void cluster_list_init(struct swap_cluster_list *list)
271{
272 cluster_set_null(&list->head);
273 cluster_set_null(&list->tail);
274}
275
276static void cluster_list_add_tail(struct swap_cluster_list *list,
277 struct swap_cluster_info *ci,
278 unsigned int idx)
279{
280 if (cluster_list_empty(list)) {
281 cluster_set_next_flag(&list->head, idx, 0);
282 cluster_set_next_flag(&list->tail, idx, 0);
283 } else {
284 unsigned int tail = cluster_next(&list->tail);
285
286 cluster_set_next(&ci[tail], idx);
287 cluster_set_next_flag(&list->tail, idx, 0);
288 }
289}
290
291static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
292 struct swap_cluster_info *ci)
293{
294 unsigned int idx;
295
296 idx = cluster_next(&list->head);
297 if (cluster_next(&list->tail) == idx) {
298 cluster_set_null(&list->head);
299 cluster_set_null(&list->tail);
300 } else
301 cluster_set_next_flag(&list->head,
302 cluster_next(&ci[idx]), 0);
303
304 return idx;
305}
306
Shaohua Li815c2c52013-09-11 14:20:30 -0700307/* Add a cluster to discard list and schedule it to do discard */
308static void swap_cluster_schedule_discard(struct swap_info_struct *si,
309 unsigned int idx)
310{
311 /*
312 * If scan_swap_map() can't find a free cluster, it will check
313 * si->swap_map directly. To make sure the discarding cluster isn't
314 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
315 * will be cleared after discard
316 */
317 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
318 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
319
Huang Ying6b534912016-10-07 16:58:42 -0700320 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
Shaohua Li815c2c52013-09-11 14:20:30 -0700321
322 schedule_work(&si->discard_work);
323}
324
325/*
326 * Doing discard actually. After a cluster discard is finished, the cluster
327 * will be added to free cluster list. caller should hold si->lock.
328*/
329static void swap_do_scheduled_discard(struct swap_info_struct *si)
330{
331 struct swap_cluster_info *info;
332 unsigned int idx;
333
334 info = si->cluster_info;
335
Huang Ying6b534912016-10-07 16:58:42 -0700336 while (!cluster_list_empty(&si->discard_clusters)) {
337 idx = cluster_list_del_first(&si->discard_clusters, info);
Shaohua Li815c2c52013-09-11 14:20:30 -0700338 spin_unlock(&si->lock);
339
340 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
341 SWAPFILE_CLUSTER);
342
343 spin_lock(&si->lock);
344 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -0700345 cluster_list_add_tail(&si->free_clusters, info, idx);
Shaohua Li815c2c52013-09-11 14:20:30 -0700346 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
347 0, SWAPFILE_CLUSTER);
348 }
349}
350
351static void swap_discard_work(struct work_struct *work)
352{
353 struct swap_info_struct *si;
354
355 si = container_of(work, struct swap_info_struct, discard_work);
356
357 spin_lock(&si->lock);
358 swap_do_scheduled_discard(si);
359 spin_unlock(&si->lock);
360}
361
Shaohua Li2a8f9442013-09-11 14:20:28 -0700362/*
363 * The cluster corresponding to page_nr will be used. The cluster will be
364 * removed from free cluster list and its usage counter will be increased.
365 */
366static void inc_cluster_info_page(struct swap_info_struct *p,
367 struct swap_cluster_info *cluster_info, unsigned long page_nr)
368{
369 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
370
371 if (!cluster_info)
372 return;
373 if (cluster_is_free(&cluster_info[idx])) {
Huang Ying6b534912016-10-07 16:58:42 -0700374 VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx);
375 cluster_list_del_first(&p->free_clusters, cluster_info);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700376 cluster_set_count_flag(&cluster_info[idx], 0, 0);
377 }
378
379 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
380 cluster_set_count(&cluster_info[idx],
381 cluster_count(&cluster_info[idx]) + 1);
382}
383
384/*
385 * The cluster corresponding to page_nr decreases one usage. If the usage
386 * counter becomes 0, which means no page in the cluster is in using, we can
387 * optionally discard the cluster and add it to free cluster list.
388 */
389static void dec_cluster_info_page(struct swap_info_struct *p,
390 struct swap_cluster_info *cluster_info, unsigned long page_nr)
391{
392 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
393
394 if (!cluster_info)
395 return;
396
397 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
398 cluster_set_count(&cluster_info[idx],
399 cluster_count(&cluster_info[idx]) - 1);
400
401 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700402 /*
403 * If the swap is discardable, prepare discard the cluster
404 * instead of free it immediately. The cluster will be freed
405 * after discard.
406 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700407 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
408 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700409 swap_cluster_schedule_discard(p, idx);
410 return;
411 }
412
Shaohua Li2a8f9442013-09-11 14:20:28 -0700413 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -0700414 cluster_list_add_tail(&p->free_clusters, cluster_info, idx);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700415 }
416}
417
418/*
419 * It's possible scan_swap_map() uses a free cluster in the middle of free
420 * cluster list. Avoiding such abuse to avoid list corruption.
421 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700422static bool
423scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700424 unsigned long offset)
425{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700426 struct percpu_cluster *percpu_cluster;
427 bool conflict;
428
Shaohua Li2a8f9442013-09-11 14:20:28 -0700429 offset /= SWAPFILE_CLUSTER;
Huang Ying6b534912016-10-07 16:58:42 -0700430 conflict = !cluster_list_empty(&si->free_clusters) &&
431 offset != cluster_list_first(&si->free_clusters) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700432 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700433
434 if (!conflict)
435 return false;
436
437 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
438 cluster_set_null(&percpu_cluster->index);
439 return true;
440}
441
442/*
443 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
444 * might involve allocating a new cluster for current CPU too.
445 */
446static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
447 unsigned long *offset, unsigned long *scan_base)
448{
449 struct percpu_cluster *cluster;
450 bool found_free;
451 unsigned long tmp;
452
453new_cluster:
454 cluster = this_cpu_ptr(si->percpu_cluster);
455 if (cluster_is_null(&cluster->index)) {
Huang Ying6b534912016-10-07 16:58:42 -0700456 if (!cluster_list_empty(&si->free_clusters)) {
457 cluster->index = si->free_clusters.head;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700458 cluster->next = cluster_next(&cluster->index) *
459 SWAPFILE_CLUSTER;
Huang Ying6b534912016-10-07 16:58:42 -0700460 } else if (!cluster_list_empty(&si->discard_clusters)) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700461 /*
462 * we don't have free cluster but have some clusters in
463 * discarding, do discard now and reclaim them
464 */
465 swap_do_scheduled_discard(si);
466 *scan_base = *offset = si->cluster_next;
467 goto new_cluster;
468 } else
469 return;
470 }
471
472 found_free = false;
473
474 /*
475 * Other CPUs can use our cluster if they can't find a free cluster,
476 * check if there is still free entry in the cluster
477 */
478 tmp = cluster->next;
479 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
480 SWAPFILE_CLUSTER) {
481 if (!si->swap_map[tmp]) {
482 found_free = true;
483 break;
484 }
485 tmp++;
486 }
487 if (!found_free) {
488 cluster_set_null(&cluster->index);
489 goto new_cluster;
490 }
491 cluster->next = tmp + 1;
492 *offset = tmp;
493 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700494}
495
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700496static unsigned long scan_swap_map(struct swap_info_struct *si,
497 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800499 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800500 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800501 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700502 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800504 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700505 * We try to cluster swap pages by allocating them sequentially
506 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
507 * way, however, we resort to first-free allocation, starting
508 * a new cluster. This prevents us from scattering swap pages
509 * all over the entire swap partition, so that we reduce
510 * overall disk seek times between swap pages. -- sct
511 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800512 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700513 */
514
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700515 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800516 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800517
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700518 /* SSD algorithm */
519 if (si->cluster_info) {
520 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
521 goto checks;
522 }
523
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800524 if (unlikely(!si->cluster_nr--)) {
525 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
526 si->cluster_nr = SWAPFILE_CLUSTER - 1;
527 goto checks;
528 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700529
Shaohua Liec8acf22013-02-22 16:34:38 -0800530 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700531
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800532 /*
533 * If seek is expensive, start searching for new cluster from
534 * start of partition, to minimize the span of allocated swap.
Chen Yucong50088c42014-06-04 16:10:57 -0700535 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
536 * case, just handled by scan_swap_map_try_ssd_cluster() above.
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800537 */
Chen Yucong50088c42014-06-04 16:10:57 -0700538 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700539 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
540
541 /* Locate the first empty (unaligned) cluster */
542 for (; last_in_cluster <= si->highest_bit; offset++) {
543 if (si->swap_map[offset])
544 last_in_cluster = offset + SWAPFILE_CLUSTER;
545 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800546 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800547 offset -= SWAPFILE_CLUSTER - 1;
548 si->cluster_next = offset;
549 si->cluster_nr = SWAPFILE_CLUSTER - 1;
550 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700552 if (unlikely(--latency_ration < 0)) {
553 cond_resched();
554 latency_ration = LATENCY_LIMIT;
555 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700556 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800557
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800558 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800559 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800560 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700562
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800563checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700564 if (si->cluster_info) {
565 while (scan_swap_map_ssd_cluster_conflict(si, offset))
566 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
567 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800568 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700569 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700570 if (!si->highest_bit)
571 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800572 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800573 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700574
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700575 /* reuse swap entry of cache-only swap if not busy. */
576 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700577 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800578 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700579 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800580 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700581 /* entry was freed successfully, try to use this again */
582 if (swap_was_freed)
583 goto checks;
584 goto scan; /* check next one */
585 }
586
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800587 if (si->swap_map[offset])
588 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700589
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800590 if (offset == si->lowest_bit)
591 si->lowest_bit++;
592 if (offset == si->highest_bit)
593 si->highest_bit--;
594 si->inuse_pages++;
595 if (si->inuse_pages == si->pages) {
596 si->lowest_bit = si->max;
597 si->highest_bit = 0;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700598 spin_lock(&swap_avail_lock);
599 plist_del(&si->avail_list, &swap_avail_head);
600 spin_unlock(&swap_avail_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800601 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800602 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700603 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800604 si->cluster_next = offset + 1;
605 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800606
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800607 return offset;
608
609scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800610 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700611 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700612 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800613 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700614 goto checks;
615 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700616 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800617 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700618 goto checks;
619 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700620 if (unlikely(--latency_ration < 0)) {
621 cond_resched();
622 latency_ration = LATENCY_LIMIT;
623 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700624 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800625 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800626 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800627 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800628 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800629 goto checks;
630 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700631 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800632 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700633 goto checks;
634 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800635 if (unlikely(--latency_ration < 0)) {
636 cond_resched();
637 latency_ration = LATENCY_LIMIT;
638 }
Jamie Liua5998062014-01-23 15:53:40 -0800639 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800640 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800641 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700642
643no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700644 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 return 0;
646}
647
648swp_entry_t get_swap_page(void)
649{
Dan Streetmanadfab832014-06-04 16:09:53 -0700650 struct swap_info_struct *si, *next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700651 pgoff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Shaohua Liec8acf22013-02-22 16:34:38 -0800653 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700654 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800655 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Dan Streetman18ab4d42014-06-04 16:09:59 -0700657 spin_lock(&swap_avail_lock);
658
659start_over:
660 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
661 /* requeue si to after same-priority siblings */
662 plist_requeue(&si->avail_list, &swap_avail_head);
663 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800664 spin_lock(&si->lock);
Dan Streetmanadfab832014-06-04 16:09:53 -0700665 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
Dan Streetman18ab4d42014-06-04 16:09:59 -0700666 spin_lock(&swap_avail_lock);
667 if (plist_node_empty(&si->avail_list)) {
668 spin_unlock(&si->lock);
669 goto nextsi;
670 }
671 WARN(!si->highest_bit,
672 "swap_info %d in list but !highest_bit\n",
673 si->type);
674 WARN(!(si->flags & SWP_WRITEOK),
675 "swap_info %d in list but !SWP_WRITEOK\n",
676 si->type);
677 plist_del(&si->avail_list, &swap_avail_head);
Shaohua Liec8acf22013-02-22 16:34:38 -0800678 spin_unlock(&si->lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700679 goto nextsi;
Shaohua Liec8acf22013-02-22 16:34:38 -0800680 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700681
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700682 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800683 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800684 spin_unlock(&si->lock);
685 if (offset)
Dan Streetmanadfab832014-06-04 16:09:53 -0700686 return swp_entry(si->type, offset);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700687 pr_debug("scan_swap_map of si %d failed to find offset\n",
688 si->type);
689 spin_lock(&swap_avail_lock);
690nextsi:
Dan Streetmanadfab832014-06-04 16:09:53 -0700691 /*
692 * if we got here, it's likely that si was almost full before,
693 * and since scan_swap_map() can drop the si->lock, multiple
694 * callers probably all tried to get a page from the same si
Dan Streetman18ab4d42014-06-04 16:09:59 -0700695 * and it filled up before we could get one; or, the si filled
696 * up between us dropping swap_avail_lock and taking si->lock.
697 * Since we dropped the swap_avail_lock, the swap_avail_head
698 * list may have been modified; so if next is still in the
699 * swap_avail_head list then try it, otherwise start over.
Dan Streetmanadfab832014-06-04 16:09:53 -0700700 */
Dan Streetman18ab4d42014-06-04 16:09:59 -0700701 if (plist_node_empty(&next->avail_list))
702 goto start_over;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700704
Dan Streetman18ab4d42014-06-04 16:09:59 -0700705 spin_unlock(&swap_avail_lock);
706
Shaohua Liec8acf22013-02-22 16:34:38 -0800707 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700708noswap:
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700709 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800712/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700713swp_entry_t get_swap_page_of_type(int type)
714{
715 struct swap_info_struct *si;
716 pgoff_t offset;
717
Hugh Dickins910321e2010-09-09 16:38:07 -0700718 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800719 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700720 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800721 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700722 /* This is called for allocating swap entry, not cache */
723 offset = scan_swap_map(si, 1);
724 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800725 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700726 return swp_entry(type, offset);
727 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800728 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700729 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800730 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700731 return (swp_entry_t) {0};
732}
733
Hugh Dickins73c34b62009-12-14 17:58:43 -0800734static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800736 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 unsigned long offset, type;
738
739 if (!entry.val)
740 goto out;
741 type = swp_type(entry);
742 if (type >= nr_swapfiles)
743 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800744 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (!(p->flags & SWP_USED))
746 goto bad_device;
747 offset = swp_offset(entry);
748 if (offset >= p->max)
749 goto bad_offset;
750 if (!p->swap_map[offset])
751 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800752 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return p;
754
755bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700756 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 goto out;
758bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700759 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 goto out;
761bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700762 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 goto out;
764bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700765 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766out:
767 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800768}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800770static unsigned char swap_entry_free(struct swap_info_struct *p,
771 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Hugh Dickins253d5532009-12-14 17:58:44 -0800773 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800774 unsigned char count;
775 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700777 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800778 has_cache = count & SWAP_HAS_CACHE;
779 count &= ~SWAP_HAS_CACHE;
780
781 if (usage == SWAP_HAS_CACHE) {
782 VM_BUG_ON(!has_cache);
783 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800784 } else if (count == SWAP_MAP_SHMEM) {
785 /*
786 * Or we could insist on shmem.c using a special
787 * swap_shmem_free() and free_shmem_swap_and_cache()...
788 */
789 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800790 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
791 if (count == COUNT_CONTINUED) {
792 if (swap_count_continued(p, offset, count))
793 count = SWAP_MAP_MAX | COUNT_CONTINUED;
794 else
795 count = SWAP_MAP_MAX;
796 } else
797 count--;
798 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800799
Hugh Dickins253d5532009-12-14 17:58:44 -0800800 usage = count | has_cache;
801 p->swap_map[offset] = usage;
802
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700803 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800804 if (!usage) {
Vladimir Davydov37e84352016-01-20 15:02:56 -0800805 mem_cgroup_uncharge_swap(entry);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700806 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700807 if (offset < p->lowest_bit)
808 p->lowest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700809 if (offset > p->highest_bit) {
810 bool was_full = !p->highest_bit;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700811 p->highest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700812 if (was_full && (p->flags & SWP_WRITEOK)) {
813 spin_lock(&swap_avail_lock);
814 WARN_ON(!plist_node_empty(&p->avail_list));
815 if (plist_node_empty(&p->avail_list))
816 plist_add(&p->avail_list,
817 &swap_avail_head);
818 spin_unlock(&swap_avail_lock);
819 }
820 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800821 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700822 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600823 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700824 if (p->flags & SWP_BLKDEV) {
825 struct gendisk *disk = p->bdev->bd_disk;
826 if (disk->fops->swap_slot_free_notify)
827 disk->fops->swap_slot_free_notify(p->bdev,
828 offset);
829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800831
832 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
835/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800836 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 * is still around or has not been recycled.
838 */
839void swap_free(swp_entry_t entry)
840{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800841 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 p = swap_info_get(entry);
844 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800845 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800846 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 }
848}
849
850/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700851 * Called after dropping swapcache to decrease refcnt to swap entries.
852 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700853void swapcache_free(swp_entry_t entry)
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700854{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700855 struct swap_info_struct *p;
856
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700857 p = swap_info_get(entry);
858 if (p) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700859 swap_entry_free(p, entry, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800860 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700861 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700862}
863
864/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700865 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800866 * This does not give an exact answer when swap count is continued,
867 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700869int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700871 int count = 0;
872 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 swp_entry_t entry;
874
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700875 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 p = swap_info_get(entry);
877 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700878 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800879 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700881 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
883
884/*
Minchan Kim8334b962015-09-08 15:00:24 -0700885 * How many references to @entry are currently swapped out?
886 * This considers COUNT_CONTINUED so it returns exact answer.
887 */
888int swp_swapcount(swp_entry_t entry)
889{
890 int count, tmp_count, n;
891 struct swap_info_struct *p;
892 struct page *page;
893 pgoff_t offset;
894 unsigned char *map;
895
896 p = swap_info_get(entry);
897 if (!p)
898 return 0;
899
900 count = swap_count(p->swap_map[swp_offset(entry)]);
901 if (!(count & COUNT_CONTINUED))
902 goto out;
903
904 count &= ~COUNT_CONTINUED;
905 n = SWAP_MAP_MAX + 1;
906
907 offset = swp_offset(entry);
908 page = vmalloc_to_page(p->swap_map + offset);
909 offset &= ~PAGE_MASK;
910 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
911
912 do {
Geliang Tanga8ae4992016-01-14 15:20:45 -0800913 page = list_next_entry(page, lru);
Minchan Kim8334b962015-09-08 15:00:24 -0700914 map = kmap_atomic(page);
915 tmp_count = map[offset];
916 kunmap_atomic(map);
917
918 count += (tmp_count & ~COUNT_CONTINUED) * n;
919 n *= (SWAP_CONT_MAX + 1);
920 } while (tmp_count & COUNT_CONTINUED);
921out:
922 spin_unlock(&p->lock);
923 return count;
924}
925
926/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800927 * We can write to an anon page without COW if there are no other references
928 * to it. And as a side-effect, free up its swap: because the old content
929 * on disk will never be read, and seeking back there to write new content
930 * later would only waste time away from clustering.
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700931 *
932 * NOTE: total_mapcount should not be relied upon by the caller if
933 * reuse_swap_page() returns false, but it may be always overwritten
934 * (see the other implementation for CONFIG_SWAP=n).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 */
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700936bool reuse_swap_page(struct page *page, int *total_mapcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700938 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Sasha Levin309381fea2014-01-23 15:52:54 -0800940 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800941 if (unlikely(PageKsm(page)))
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700942 return false;
943 count = page_trans_huge_mapcount(page, total_mapcount);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800944 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700945 count += page_swapcount(page);
Minchan Kim6ca29ee2017-01-10 16:58:15 -0800946 if (count != 1)
947 goto out;
948 if (!PageWriteback(page)) {
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800949 delete_from_swap_cache(page);
950 SetPageDirty(page);
Minchan Kim6ca29ee2017-01-10 16:58:15 -0800951 } else {
952 swp_entry_t entry;
953 struct swap_info_struct *p;
954
955 entry.val = page_private(page);
956 p = swap_info_get(entry);
957 if (p->flags & SWP_STABLE_WRITES) {
958 spin_unlock(&p->lock);
959 return false;
960 }
961 spin_unlock(&p->lock);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800962 }
963 }
Minchan Kim6ca29ee2017-01-10 16:58:15 -0800964out:
Hugh Dickins5ad64682009-12-14 17:59:24 -0800965 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966}
967
968/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800969 * If swap is getting full, or if there are no more mappings of this page,
970 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800972int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973{
Sasha Levin309381fea2014-01-23 15:52:54 -0800974 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 if (!PageSwapCache(page))
977 return 0;
978 if (PageWriteback(page))
979 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800980 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 return 0;
982
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700983 /*
984 * Once hibernation has begun to create its image of memory,
985 * there's a danger that one of the calls to try_to_free_swap()
986 * - most probably a call from __try_to_reclaim_swap() while
987 * hibernation is allocating its own swap pages for the image,
988 * but conceivably even a call from memory reclaim - will free
989 * the swap from a page which has already been recorded in the
990 * image as a clean swapcache page, and then reuse its swap for
991 * another page of the image. On waking from hibernation, the
992 * original page might be freed under memory pressure, then
993 * later read back in from swap, now with the wrong data.
994 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800995 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -0800996 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700997 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800998 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700999 return 0;
1000
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001001 delete_from_swap_cache(page);
1002 SetPageDirty(page);
1003 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -07001004}
1005
1006/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 * Free the swap entry like above, but also try to
1008 * free the page cache entry if it is the last user.
1009 */
Hugh Dickins2509ef22009-01-06 14:40:10 -08001010int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011{
Hugh Dickins2509ef22009-01-06 14:40:10 -08001012 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 struct page *page = NULL;
1014
Andi Kleena7420aa2009-09-16 11:50:05 +02001015 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -08001016 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -07001017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 p = swap_info_get(entry);
1019 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -08001020 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -08001021 page = find_get_page(swap_address_space(entry),
Huang Yingf6ab1f72016-10-07 17:00:21 -07001022 swp_offset(entry));
Nick Piggin8413ac92008-10-18 20:26:59 -07001023 if (page && !trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001024 put_page(page);
Nick Piggin93fac702006-03-31 02:29:56 -08001025 page = NULL;
1026 }
1027 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001028 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
1030 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001031 /*
1032 * Not mapped elsewhere, or swap space full? Free it!
1033 * Also recheck PageSwapCache now page is locked (above).
1034 */
Nick Piggin93fac702006-03-31 02:29:56 -08001035 if (PageSwapCache(page) && !PageWriteback(page) &&
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08001036 (!page_mapped(page) || mem_cgroup_swap_full(page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 delete_from_swap_cache(page);
1038 SetPageDirty(page);
1039 }
1040 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001041 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 }
Hugh Dickins2509ef22009-01-06 14:40:10 -08001043 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044}
1045
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001046#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001047/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001048 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001049 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001050 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1051 * from 0, in which the swap header is expected to be located.
1052 *
1053 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001054 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001055int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001056{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001057 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001058 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001059
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001060 if (device)
1061 bdev = bdget(device);
1062
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001063 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001064 for (type = 0; type < nr_swapfiles; type++) {
1065 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001066
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001067 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001068 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001069
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001070 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001071 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001072 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001073
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001074 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001075 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001076 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001077 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001078 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001079
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001080 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001081 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001082 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001083
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001084 spin_unlock(&swap_lock);
1085 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001086 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001087 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001088 }
1089 }
1090 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001091 if (bdev)
1092 bdput(bdev);
1093
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001094 return -ENODEV;
1095}
1096
1097/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001098 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1099 * corresponding to given index in swap_info (swap type).
1100 */
1101sector_t swapdev_block(int type, pgoff_t offset)
1102{
1103 struct block_device *bdev;
1104
1105 if ((unsigned int)type >= nr_swapfiles)
1106 return 0;
1107 if (!(swap_info[type]->flags & SWP_WRITEOK))
1108 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001109 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001110}
1111
1112/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001113 * Return either the total number of swap pages of given type, or the number
1114 * of free pages of that type (depending on @free)
1115 *
1116 * This is needed for software suspend
1117 */
1118unsigned int count_swap_pages(int type, int free)
1119{
1120 unsigned int n = 0;
1121
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001122 spin_lock(&swap_lock);
1123 if ((unsigned int)type < nr_swapfiles) {
1124 struct swap_info_struct *sis = swap_info[type];
1125
Shaohua Liec8acf22013-02-22 16:34:38 -08001126 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001127 if (sis->flags & SWP_WRITEOK) {
1128 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001129 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001130 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001131 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001132 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001133 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001134 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001135 return n;
1136}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001137#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001138
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001139static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001140{
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001141 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001142}
1143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001145 * No need to decide whether this PTE shares the swap entry with others,
1146 * just let do_wp_page work it out if a write is requested later - to
1147 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001149static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 unsigned long addr, swp_entry_t entry, struct page *page)
1151{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001152 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001153 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001154 spinlock_t *ptl;
1155 pte_t *pte;
1156 int ret = 1;
1157
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001158 swapcache = page;
1159 page = ksm_might_need_to_copy(page, vma, addr);
1160 if (unlikely(!page))
1161 return -ENOMEM;
1162
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001163 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1164 &memcg, false)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001165 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001166 goto out_nolock;
1167 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001168
1169 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001170 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001171 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001172 ret = 0;
1173 goto out;
1174 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001175
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001176 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001177 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 get_page(page);
1179 set_pte_at(vma->vm_mm, addr, pte,
1180 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001181 if (page == swapcache) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001182 page_add_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001183 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001184 } else { /* ksm created a completely new copy */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001185 page_add_new_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001186 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001187 lru_cache_add_active_or_unevictable(page, vma);
1188 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 swap_free(entry);
1190 /*
1191 * Move the page to the active list so it is not
1192 * immediately swapped out again after swapon.
1193 */
1194 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001195out:
1196 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001197out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001198 if (page != swapcache) {
1199 unlock_page(page);
1200 put_page(page);
1201 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001202 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203}
1204
1205static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1206 unsigned long addr, unsigned long end,
1207 swp_entry_t entry, struct page *page)
1208{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001210 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001211 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Hugh Dickins044d66c2008-02-07 00:14:04 -08001213 /*
1214 * We don't actually need pte lock while scanning for swp_pte: since
1215 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1216 * page table while we're scanning; though it could get zapped, and on
1217 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1218 * of unmatched parts which look like swp_pte, so unuse_pte must
1219 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001220 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001221 */
1222 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 do {
1224 /*
1225 * swapoff spends a _lot_ of time in this loop!
1226 * Test inline before going to call unuse_pte.
1227 */
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001228 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001229 pte_unmap(pte);
1230 ret = unuse_pte(vma, pmd, addr, entry, page);
1231 if (ret)
1232 goto out;
1233 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
1235 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001236 pte_unmap(pte - 1);
1237out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001238 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239}
1240
1241static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1242 unsigned long addr, unsigned long end,
1243 swp_entry_t entry, struct page *page)
1244{
1245 pmd_t *pmd;
1246 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001247 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 pmd = pmd_offset(pud, addr);
1250 do {
1251 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001252 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001254 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1255 if (ret)
1256 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 } while (pmd++, addr = next, addr != end);
1258 return 0;
1259}
1260
1261static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1262 unsigned long addr, unsigned long end,
1263 swp_entry_t entry, struct page *page)
1264{
1265 pud_t *pud;
1266 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001267 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
1269 pud = pud_offset(pgd, addr);
1270 do {
1271 next = pud_addr_end(addr, end);
1272 if (pud_none_or_clear_bad(pud))
1273 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001274 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1275 if (ret)
1276 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 } while (pud++, addr = next, addr != end);
1278 return 0;
1279}
1280
1281static int unuse_vma(struct vm_area_struct *vma,
1282 swp_entry_t entry, struct page *page)
1283{
1284 pgd_t *pgd;
1285 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001286 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001288 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 addr = page_address_in_vma(page, vma);
1290 if (addr == -EFAULT)
1291 return 0;
1292 else
1293 end = addr + PAGE_SIZE;
1294 } else {
1295 addr = vma->vm_start;
1296 end = vma->vm_end;
1297 }
1298
1299 pgd = pgd_offset(vma->vm_mm, addr);
1300 do {
1301 next = pgd_addr_end(addr, end);
1302 if (pgd_none_or_clear_bad(pgd))
1303 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001304 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1305 if (ret)
1306 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 } while (pgd++, addr = next, addr != end);
1308 return 0;
1309}
1310
1311static int unuse_mm(struct mm_struct *mm,
1312 swp_entry_t entry, struct page *page)
1313{
1314 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001315 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317 if (!down_read_trylock(&mm->mmap_sem)) {
1318 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001319 * Activate page so shrink_inactive_list is unlikely to unmap
1320 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001322 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 unlock_page(page);
1324 down_read(&mm->mmap_sem);
1325 lock_page(page);
1326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001328 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 break;
1330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001332 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333}
1334
1335/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001336 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1337 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 * Recycle to start on reaching the end, returning 0 when empty.
1339 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001340static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001341 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001343 unsigned int max = si->max;
1344 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001345 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001348 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * for whether an entry is in use, not modifying it; false
1350 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001351 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 */
1353 for (;;) {
1354 if (++i >= max) {
1355 if (!prev) {
1356 i = 0;
1357 break;
1358 }
1359 /*
1360 * No entries in use at top of swap_map,
1361 * loop back to start and recheck there.
1362 */
1363 max = prev + 1;
1364 prev = 0;
1365 i = 1;
1366 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001367 if (frontswap) {
1368 if (frontswap_test(si, i))
1369 break;
1370 else
1371 continue;
1372 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001373 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001374 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 break;
1376 }
1377 return i;
1378}
1379
1380/*
1381 * We completely avoid races by reading each swap page in advance,
1382 * and then search for the process using it. All the necessary
1383 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001384 *
1385 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1386 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001388int try_to_unuse(unsigned int type, bool frontswap,
1389 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001391 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001393 volatile unsigned char *swap_map; /* swap_map is accessed without
1394 * locking. Mark it as volatile
1395 * to prevent compiler doing
1396 * something odd.
1397 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001398 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 struct page *page;
1400 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001401 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
1404 /*
1405 * When searching mms for an entry, a good strategy is to
1406 * start at the first mm we freed the previous entry from
1407 * (though actually we don't notice whether we or coincidence
1408 * freed the entry). Initialize this start_mm with a hold.
1409 *
1410 * A simpler strategy would be to start at the last mm we
1411 * freed the previous entry from; but that would take less
1412 * advantage of mmlist ordering, which clusters forked mms
1413 * together, child after parent. If we race with dup_mmap(), we
1414 * prefer to resolve parent before child, lest we miss entries
1415 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001416 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 */
1418 start_mm = &init_mm;
1419 atomic_inc(&init_mm.mm_users);
1420
1421 /*
1422 * Keep on scanning until all entries have gone. Usually,
1423 * one pass through swap_map is enough, but not necessarily:
1424 * there are races when an instance of an entry might be missed.
1425 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001426 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (signal_pending(current)) {
1428 retval = -EINTR;
1429 break;
1430 }
1431
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001432 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 * Get a page for the entry, using the existing swap
1434 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001435 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 */
1437 swap_map = &si->swap_map[i];
1438 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001439 page = read_swap_cache_async(entry,
1440 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 if (!page) {
1442 /*
1443 * Either swap_duplicate() failed because entry
1444 * has been freed independently, and will not be
1445 * reused since sys_swapoff() already disabled
1446 * allocation from here, or alloc_page() failed.
1447 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001448 swcount = *swap_map;
1449 /*
1450 * We don't hold lock here, so the swap entry could be
1451 * SWAP_MAP_BAD (when the cluster is discarding).
1452 * Instead of fail out, We can just skip the swap
1453 * entry because swapoff will wait for discarding
1454 * finish anyway.
1455 */
1456 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 continue;
1458 retval = -ENOMEM;
1459 break;
1460 }
1461
1462 /*
1463 * Don't hold on to start_mm if it looks like exiting.
1464 */
1465 if (atomic_read(&start_mm->mm_users) == 1) {
1466 mmput(start_mm);
1467 start_mm = &init_mm;
1468 atomic_inc(&init_mm.mm_users);
1469 }
1470
1471 /*
1472 * Wait for and lock page. When do_swap_page races with
1473 * try_to_unuse, do_swap_page can handle the fault much
1474 * faster than try_to_unuse can locate the entry. This
1475 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1476 * defer to do_swap_page in such a case - in some tests,
1477 * do_swap_page and try_to_unuse repeatedly compete.
1478 */
1479 wait_on_page_locked(page);
1480 wait_on_page_writeback(page);
1481 lock_page(page);
1482 wait_on_page_writeback(page);
1483
1484 /*
1485 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001488 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1489 retval = shmem_unuse(entry, page);
1490 /* page has already been unlocked and released */
1491 if (retval < 0)
1492 break;
1493 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001495 if (swap_count(swcount) && start_mm != &init_mm)
1496 retval = unuse_mm(start_mm, entry, page);
1497
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001498 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 int set_start_mm = (*swap_map >= swcount);
1500 struct list_head *p = &start_mm->mmlist;
1501 struct mm_struct *new_start_mm = start_mm;
1502 struct mm_struct *prev_mm = start_mm;
1503 struct mm_struct *mm;
1504
1505 atomic_inc(&new_start_mm->mm_users);
1506 atomic_inc(&prev_mm->mm_users);
1507 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001508 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 (p = p->next) != &start_mm->mmlist) {
1510 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001511 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 spin_unlock(&mmlist_lock);
1514 mmput(prev_mm);
1515 prev_mm = mm;
1516
1517 cond_resched();
1518
1519 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001520 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001522 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001524 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001526
Bo Liu32c5fc12009-11-02 16:50:33 +00001527 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 mmput(new_start_mm);
1529 atomic_inc(&mm->mm_users);
1530 new_start_mm = mm;
1531 set_start_mm = 0;
1532 }
1533 spin_lock(&mmlist_lock);
1534 }
1535 spin_unlock(&mmlist_lock);
1536 mmput(prev_mm);
1537 mmput(start_mm);
1538 start_mm = new_start_mm;
1539 }
1540 if (retval) {
1541 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001542 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 break;
1544 }
1545
1546 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 * If a reference remains (rare), we would like to leave
1548 * the page in the swap cache; but try_to_unmap could
1549 * then re-duplicate the entry once we drop page lock,
1550 * so we might loop indefinitely; also, that page could
1551 * not be swapped out to other storage meanwhile. So:
1552 * delete from cache even if there's another reference,
1553 * after ensuring that the data has been saved to disk -
1554 * since if the reference remains (rarer), it will be
1555 * read from disk into another page. Splitting into two
1556 * pages would be incorrect if swap supported "shared
1557 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001558 *
1559 * Given how unuse_vma() targets one particular offset
1560 * in an anon_vma, once the anon_vma has been determined,
1561 * this splitting happens to be just what is needed to
1562 * handle where KSM pages have been swapped out: re-reading
1563 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001565 if (swap_count(*swap_map) &&
1566 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 struct writeback_control wbc = {
1568 .sync_mode = WB_SYNC_NONE,
1569 };
1570
1571 swap_writepage(page, &wbc);
1572 lock_page(page);
1573 wait_on_page_writeback(page);
1574 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001575
1576 /*
1577 * It is conceivable that a racing task removed this page from
1578 * swap cache just before we acquired the page lock at the top,
1579 * or while we dropped it in unuse_mm(). The page might even
1580 * be back in swap cache on another swap area: that we must not
1581 * delete, since it may not have been written out to swap yet.
1582 */
1583 if (PageSwapCache(page) &&
1584 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001585 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 /*
1588 * So we could skip searching mms once swap count went
1589 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001590 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 */
1592 SetPageDirty(page);
1593 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001594 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
1596 /*
1597 * Make sure that we aren't completely killing
1598 * interactive performance.
1599 */
1600 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001601 if (frontswap && pages_to_unuse > 0) {
1602 if (!--pages_to_unuse)
1603 break;
1604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 }
1606
1607 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 return retval;
1609}
1610
1611/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001612 * After a successful try_to_unuse, if no swap is now in use, we know
1613 * we can empty the mmlist. swap_lock must be held on entry and exit.
1614 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 * added to the mmlist just after page_duplicate - before would be racy.
1616 */
1617static void drain_mmlist(void)
1618{
1619 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001620 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001622 for (type = 0; type < nr_swapfiles; type++)
1623 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 return;
1625 spin_lock(&mmlist_lock);
1626 list_for_each_safe(p, next, &init_mm.mmlist)
1627 list_del_init(p);
1628 spin_unlock(&mmlist_lock);
1629}
1630
1631/*
1632 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001633 * corresponds to page offset for the specified swap entry.
1634 * Note that the type of this function is sector_t, but it returns page offset
1635 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001637static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001639 struct swap_info_struct *sis;
1640 struct swap_extent *start_se;
1641 struct swap_extent *se;
1642 pgoff_t offset;
1643
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001644 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001645 *bdev = sis->bdev;
1646
1647 offset = swp_offset(entry);
1648 start_se = sis->curr_swap_extent;
1649 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
1651 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 if (se->start_page <= offset &&
1653 offset < (se->start_page + se->nr_pages)) {
1654 return se->start_block + (offset - se->start_page);
1655 }
Geliang Tanga8ae4992016-01-14 15:20:45 -08001656 se = list_next_entry(se, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 sis->curr_swap_extent = se;
1658 BUG_ON(se == start_se); /* It *must* be present */
1659 }
1660}
1661
1662/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001663 * Returns the page offset into bdev for the specified page's swap entry.
1664 */
1665sector_t map_swap_page(struct page *page, struct block_device **bdev)
1666{
1667 swp_entry_t entry;
1668 entry.val = page_private(page);
1669 return map_swap_entry(entry, bdev);
1670}
1671
1672/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 * Free all of a swapdev's extent information
1674 */
1675static void destroy_swap_extents(struct swap_info_struct *sis)
1676{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001677 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 struct swap_extent *se;
1679
Geliang Tanga8ae4992016-01-14 15:20:45 -08001680 se = list_first_entry(&sis->first_swap_extent.list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 struct swap_extent, list);
1682 list_del(&se->list);
1683 kfree(se);
1684 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001685
1686 if (sis->flags & SWP_FILE) {
1687 struct file *swap_file = sis->swap_file;
1688 struct address_space *mapping = swap_file->f_mapping;
1689
1690 sis->flags &= ~SWP_FILE;
1691 mapping->a_ops->swap_deactivate(swap_file);
1692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693}
1694
1695/*
1696 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001697 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001699 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 */
Mel Gormana509bc12012-07-31 16:44:57 -07001701int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1703 unsigned long nr_pages, sector_t start_block)
1704{
1705 struct swap_extent *se;
1706 struct swap_extent *new_se;
1707 struct list_head *lh;
1708
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001709 if (start_page == 0) {
1710 se = &sis->first_swap_extent;
1711 sis->curr_swap_extent = se;
1712 se->start_page = 0;
1713 se->nr_pages = nr_pages;
1714 se->start_block = start_block;
1715 return 1;
1716 } else {
1717 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001719 BUG_ON(se->start_page + se->nr_pages != start_page);
1720 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 /* Merge it */
1722 se->nr_pages += nr_pages;
1723 return 0;
1724 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 }
1726
1727 /*
1728 * No merge. Insert a new extent, preserving ordering.
1729 */
1730 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1731 if (new_se == NULL)
1732 return -ENOMEM;
1733 new_se->start_page = start_page;
1734 new_se->nr_pages = nr_pages;
1735 new_se->start_block = start_block;
1736
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001737 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001738 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739}
1740
1741/*
1742 * A `swap extent' is a simple thing which maps a contiguous range of pages
1743 * onto a contiguous range of disk blocks. An ordered list of swap extents
1744 * is built at swapon time and is then used at swap_writepage/swap_readpage
1745 * time for locating where on disk a page belongs.
1746 *
1747 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1748 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1749 * swap files identically.
1750 *
1751 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1752 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1753 * swapfiles are handled *identically* after swapon time.
1754 *
1755 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1756 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1757 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1758 * requirements, they are simply tossed out - we will never use those blocks
1759 * for swapping.
1760 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001761 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1763 * which will scribble on the fs.
1764 *
1765 * The amount of disk space which a single swap extent represents varies.
1766 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1767 * extents in the list. To avoid much list walking, we cache the previous
1768 * search location in `curr_swap_extent', and start new searches from there.
1769 * This is extremely effective. The average number of iterations in
1770 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1771 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001772static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773{
Mel Gorman62c230b2012-07-31 16:44:55 -07001774 struct file *swap_file = sis->swap_file;
1775 struct address_space *mapping = swap_file->f_mapping;
1776 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 int ret;
1778
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 if (S_ISBLK(inode->i_mode)) {
1780 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001781 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001782 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 }
1784
Mel Gorman62c230b2012-07-31 16:44:55 -07001785 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001786 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001787 if (!ret) {
1788 sis->flags |= SWP_FILE;
1789 ret = add_swap_extent(sis, 0, sis->max, 0);
1790 *span = sis->pages;
1791 }
Mel Gormana509bc12012-07-31 16:44:57 -07001792 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001793 }
1794
Mel Gormana509bc12012-07-31 16:44:57 -07001795 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796}
1797
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001798static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001799 unsigned char *swap_map,
1800 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001801{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001802 if (prio >= 0)
1803 p->prio = prio;
1804 else
1805 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001806 /*
1807 * the plist prio is negated because plist ordering is
1808 * low-to-high, while swap ordering is high-to-low
1809 */
1810 p->list.prio = -p->prio;
1811 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001812 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001813 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001814 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001815 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001816 total_swap_pages += p->pages;
1817
Dan Streetmanadfab832014-06-04 16:09:53 -07001818 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001819 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07001820 * both lists are plists, and thus priority ordered.
1821 * swap_active_head needs to be priority ordered for swapoff(),
1822 * which on removal of any swap_info_struct with an auto-assigned
1823 * (i.e. negative) priority increments the auto-assigned priority
1824 * of any lower-priority swap_info_structs.
1825 * swap_avail_head needs to be priority ordered for get_swap_page(),
1826 * which allocates swap pages from the highest available priority
1827 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07001828 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07001829 plist_add(&p->list, &swap_active_head);
1830 spin_lock(&swap_avail_lock);
1831 plist_add(&p->avail_list, &swap_avail_head);
1832 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001833}
1834
1835static void enable_swap_info(struct swap_info_struct *p, int prio,
1836 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001837 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001838 unsigned long *frontswap_map)
1839{
Minchan Kim4f898492013-04-30 15:26:54 -07001840 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001841 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001842 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001843 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001844 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001845 spin_unlock(&swap_lock);
1846}
1847
1848static void reinsert_swap_info(struct swap_info_struct *p)
1849{
1850 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001851 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001852 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001853 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001854 spin_unlock(&swap_lock);
1855}
1856
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001857SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001859 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001860 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001861 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001862 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 struct file *swap_file, *victim;
1864 struct address_space *mapping;
1865 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001866 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07001867 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001868 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 if (!capable(CAP_SYS_ADMIN))
1871 return -EPERM;
1872
Al Viro191c5422012-02-13 03:58:52 +00001873 BUG_ON(!current->mm);
1874
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001877 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
Jeff Layton669abf42012-10-10 16:43:10 -04001879 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 err = PTR_ERR(victim);
1881 if (IS_ERR(victim))
1882 goto out;
1883
1884 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001885 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07001886 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001887 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001888 if (p->swap_file->f_mapping == mapping) {
1889 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07001891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 }
Dan Streetmanadfab832014-06-04 16:09:53 -07001894 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001896 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 goto out_dput;
1898 }
Al Viro191c5422012-02-13 03:58:52 +00001899 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 vm_unacct_memory(p->pages);
1901 else {
1902 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001903 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 goto out_dput;
1905 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001906 spin_lock(&swap_avail_lock);
1907 plist_del(&p->avail_list, &swap_avail_head);
1908 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001909 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001910 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001911 struct swap_info_struct *si = p;
1912
Dan Streetman18ab4d42014-06-04 16:09:59 -07001913 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001914 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001915 si->list.prio--;
1916 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07001917 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001918 least_priority++;
1919 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001920 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08001921 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 total_swap_pages -= p->pages;
1923 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001924 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001925 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001926
David Rientjese1e12d22012-12-11 16:02:56 -08001927 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07001928 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08001929 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 if (err) {
1932 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001933 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 goto out_dput;
1935 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001936
Shaohua Li815c2c52013-09-11 14:20:30 -07001937 flush_work(&p->discard_work);
1938
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001939 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001940 if (p->flags & SWP_CONTINUED)
1941 free_swap_count_continuations(p);
1942
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001943 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001944 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001945 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001947
1948 /* wait for anyone still in scan_swap_map */
1949 p->highest_bit = 0; /* cuts scans short */
1950 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001951 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001952 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001953 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001954 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001955 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001956 }
1957
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001959 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 p->swap_file = NULL;
1961 p->max = 0;
1962 swap_map = p->swap_map;
1963 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001964 cluster_info = p->cluster_info;
1965 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07001966 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08001967 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001968 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001969 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08001970 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001971 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001972 free_percpu(p->percpu_cluster);
1973 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001975 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001976 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001977 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07001978 swap_cgroup_swapoff(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 inode = mapping->host;
1981 if (S_ISBLK(inode->i_mode)) {
1982 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001983 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001984 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 } else {
Al Viro59551022016-01-22 15:40:57 -05001986 inode_lock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 inode->i_flags &= ~S_SWAPFILE;
Al Viro59551022016-01-22 15:40:57 -05001988 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 }
1990 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08001991
1992 /*
1993 * Clear the SWP_USED flag after all resources are freed so that swapon
1994 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1995 * not hold p->lock after we cleared its SWP_WRITEOK.
1996 */
1997 spin_lock(&swap_lock);
1998 p->flags = 0;
1999 spin_unlock(&swap_lock);
2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002002 atomic_inc(&proc_poll_event);
2003 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
2005out_dput:
2006 filp_close(victim, NULL);
2007out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08002008 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 return err;
2010}
2011
2012#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07002013static unsigned swaps_poll(struct file *file, poll_table *wait)
2014{
Kay Sieversf1514632011-07-12 20:48:39 +02002015 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002016
2017 poll_wait(file, &proc_poll_wait, wait);
2018
Kay Sieversf1514632011-07-12 20:48:39 +02002019 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2020 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002021 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2022 }
2023
2024 return POLLIN | POLLRDNORM;
2025}
2026
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027/* iterator */
2028static void *swap_start(struct seq_file *swap, loff_t *pos)
2029{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002030 struct swap_info_struct *si;
2031 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 loff_t l = *pos;
2033
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002034 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002036 if (!l)
2037 return SEQ_START_TOKEN;
2038
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002039 for (type = 0; type < nr_swapfiles; type++) {
2040 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2041 si = swap_info[type];
2042 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002044 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002045 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 }
2047
2048 return NULL;
2049}
2050
2051static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2052{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002053 struct swap_info_struct *si = v;
2054 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002056 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002057 type = 0;
2058 else
2059 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002060
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002061 for (; type < nr_swapfiles; type++) {
2062 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2063 si = swap_info[type];
2064 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 continue;
2066 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002067 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 }
2069
2070 return NULL;
2071}
2072
2073static void swap_stop(struct seq_file *swap, void *v)
2074{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002075 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076}
2077
2078static int swap_show(struct seq_file *swap, void *v)
2079{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002080 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 struct file *file;
2082 int len;
2083
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002084 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002085 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2086 return 0;
2087 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002089 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002090 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002091 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002092 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002093 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002095 si->pages << (PAGE_SHIFT - 10),
2096 si->inuse_pages << (PAGE_SHIFT - 10),
2097 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return 0;
2099}
2100
Helge Deller15ad7cd2006-12-06 20:40:36 -08002101static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 .start = swap_start,
2103 .next = swap_next,
2104 .stop = swap_stop,
2105 .show = swap_show
2106};
2107
2108static int swaps_open(struct inode *inode, struct file *file)
2109{
Kay Sieversf1514632011-07-12 20:48:39 +02002110 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002111 int ret;
2112
Kay Sievers66d7dd52010-10-26 14:22:06 -07002113 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002114 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002115 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002116
Kay Sieversf1514632011-07-12 20:48:39 +02002117 seq = file->private_data;
2118 seq->poll_event = atomic_read(&proc_poll_event);
2119 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}
2121
Helge Deller15ad7cd2006-12-06 20:40:36 -08002122static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 .open = swaps_open,
2124 .read = seq_read,
2125 .llseek = seq_lseek,
2126 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002127 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128};
2129
2130static int __init procswaps_init(void)
2131{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002132 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 return 0;
2134}
2135__initcall(procswaps_init);
2136#endif /* CONFIG_PROC_FS */
2137
Jan Beulich17963162008-12-16 11:35:24 +00002138#ifdef MAX_SWAPFILES_CHECK
2139static int __init max_swapfiles_check(void)
2140{
2141 MAX_SWAPFILES_CHECK();
2142 return 0;
2143}
2144late_initcall(max_swapfiles_check);
2145#endif
2146
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002147static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002149 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002151
2152 p = kzalloc(sizeof(*p), GFP_KERNEL);
2153 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002154 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002155
Hugh Dickins5d337b92005-09-03 15:54:41 -07002156 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002157 for (type = 0; type < nr_swapfiles; type++) {
2158 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002160 }
Christoph Lameter06972122006-06-23 02:03:35 -07002161 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002162 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002163 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002164 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002166 if (type >= nr_swapfiles) {
2167 p->type = type;
2168 swap_info[type] = p;
2169 /*
2170 * Write swap_info[type] before nr_swapfiles, in case a
2171 * racing procfs swap_start() or swap_next() is reading them.
2172 * (We never shrink nr_swapfiles, we never free this entry.)
2173 */
2174 smp_wmb();
2175 nr_swapfiles++;
2176 } else {
2177 kfree(p);
2178 p = swap_info[type];
2179 /*
2180 * Do not memset this entry: a racing procfs swap_next()
2181 * would be relying on p->type to remain valid.
2182 */
2183 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002184 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002185 plist_node_init(&p->list, 0);
2186 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002188 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002189 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002190
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002191 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002192}
2193
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002194static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2195{
2196 int error;
2197
2198 if (S_ISBLK(inode->i_mode)) {
2199 p->bdev = bdgrab(I_BDEV(inode));
2200 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002201 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002202 if (error < 0) {
2203 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002204 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002205 }
2206 p->old_block_size = block_size(p->bdev);
2207 error = set_blocksize(p->bdev, PAGE_SIZE);
2208 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002209 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002210 p->flags |= SWP_BLKDEV;
2211 } else if (S_ISREG(inode->i_mode)) {
2212 p->bdev = inode->i_sb->s_bdev;
Al Viro59551022016-01-22 15:40:57 -05002213 inode_lock(inode);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002214 if (IS_SWAPFILE(inode))
2215 return -EBUSY;
2216 } else
2217 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002218
2219 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002220}
2221
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002222static unsigned long read_swap_header(struct swap_info_struct *p,
2223 union swap_header *swap_header,
2224 struct inode *inode)
2225{
2226 int i;
2227 unsigned long maxpages;
2228 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002229 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002230
2231 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002232 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002233 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002234 }
2235
2236 /* swap partition endianess hack... */
2237 if (swab32(swap_header->info.version) == 1) {
2238 swab32s(&swap_header->info.version);
2239 swab32s(&swap_header->info.last_page);
2240 swab32s(&swap_header->info.nr_badpages);
Jann Horndd111be2016-11-10 10:46:19 -08002241 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2242 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002243 for (i = 0; i < swap_header->info.nr_badpages; i++)
2244 swab32s(&swap_header->info.badpages[i]);
2245 }
2246 /* Check the swap header's sub-version */
2247 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002248 pr_warn("Unable to handle swap header version %d\n",
2249 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002250 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002251 }
2252
2253 p->lowest_bit = 1;
2254 p->cluster_next = 1;
2255 p->cluster_nr = 0;
2256
2257 /*
2258 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002259 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002260 * of bits for the swap offset in the swp_entry_t type, and
2261 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002262 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002263 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002264 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002265 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002266 * offset is extracted. This will mask all the bits from
2267 * the initial ~0UL mask that can't be encoded in either
2268 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002269 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002270 */
2271 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002272 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002273 last_page = swap_header->info.last_page;
2274 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002275 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002276 maxpages << (PAGE_SHIFT - 10),
2277 last_page << (PAGE_SHIFT - 10));
2278 }
2279 if (maxpages > last_page) {
2280 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002281 /* p->max is an unsigned int: don't overflow it */
2282 if ((unsigned int)maxpages == 0)
2283 maxpages = UINT_MAX;
2284 }
2285 p->highest_bit = maxpages - 1;
2286
2287 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002288 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002289 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2290 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002291 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002292 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002293 }
2294 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002295 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002296 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002297 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002298
2299 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002300}
2301
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002302static int setup_swap_map_and_extents(struct swap_info_struct *p,
2303 union swap_header *swap_header,
2304 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002305 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002306 unsigned long maxpages,
2307 sector_t *span)
2308{
2309 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002310 unsigned int nr_good_pages;
2311 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002312 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2313 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002314
2315 nr_good_pages = maxpages - 1; /* omit header page */
2316
Huang Ying6b534912016-10-07 16:58:42 -07002317 cluster_list_init(&p->free_clusters);
2318 cluster_list_init(&p->discard_clusters);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002319
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002320 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2321 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002322 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2323 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002324 if (page_nr < maxpages) {
2325 swap_map[page_nr] = SWAP_MAP_BAD;
2326 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002327 /*
2328 * Haven't marked the cluster free yet, no list
2329 * operation involved
2330 */
2331 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002332 }
2333 }
2334
Shaohua Li2a8f9442013-09-11 14:20:28 -07002335 /* Haven't marked the cluster free yet, no list operation involved */
2336 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2337 inc_cluster_info_page(p, cluster_info, i);
2338
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002339 if (nr_good_pages) {
2340 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002341 /*
2342 * Not mark the cluster free yet, no list
2343 * operation involved
2344 */
2345 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002346 p->max = maxpages;
2347 p->pages = nr_good_pages;
2348 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002349 if (nr_extents < 0)
2350 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002351 nr_good_pages = p->pages;
2352 }
2353 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002354 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002355 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002356 }
2357
Shaohua Li2a8f9442013-09-11 14:20:28 -07002358 if (!cluster_info)
2359 return nr_extents;
2360
2361 for (i = 0; i < nr_clusters; i++) {
2362 if (!cluster_count(&cluster_info[idx])) {
2363 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -07002364 cluster_list_add_tail(&p->free_clusters, cluster_info,
2365 idx);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002366 }
2367 idx++;
2368 if (idx == nr_clusters)
2369 idx = 0;
2370 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002371 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002372}
2373
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002374/*
2375 * Helper to sys_swapon determining if a given swap
2376 * backing device queue supports DISCARD operations.
2377 */
2378static bool swap_discardable(struct swap_info_struct *si)
2379{
2380 struct request_queue *q = bdev_get_queue(si->bdev);
2381
2382 if (!q || !blk_queue_discard(q))
2383 return false;
2384
2385 return true;
2386}
2387
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002388SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2389{
2390 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002391 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002392 struct file *swap_file = NULL;
2393 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002394 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002395 int error;
2396 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002397 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002398 sector_t span;
2399 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002400 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002401 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002402 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002403 struct page *page = NULL;
2404 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002405
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002406 if (swap_flags & ~SWAP_FLAGS_VALID)
2407 return -EINVAL;
2408
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002409 if (!capable(CAP_SYS_ADMIN))
2410 return -EPERM;
2411
2412 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002413 if (IS_ERR(p))
2414 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002415
Shaohua Li815c2c52013-09-11 14:20:30 -07002416 INIT_WORK(&p->discard_work, swap_discard_work);
2417
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002420 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002422 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 }
Jeff Layton669abf42012-10-10 16:43:10 -04002424 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002426 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002428 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 }
2430
2431 p->swap_file = swap_file;
2432 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002433 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002434
Al Viro59551022016-01-22 15:40:57 -05002435 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002436 error = claim_swapfile(p, inode);
2437 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 /*
2441 * Read the swap header.
2442 */
2443 if (!mapping->a_ops->readpage) {
2444 error = -EINVAL;
2445 goto bad_swap;
2446 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002447 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 if (IS_ERR(page)) {
2449 error = PTR_ERR(page);
2450 goto bad_swap;
2451 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002452 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002454 maxpages = read_swap_header(p, swap_header, inode);
2455 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 error = -EINVAL;
2457 goto bad_swap;
2458 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002459
Hugh Dickins81e33972009-01-06 14:39:49 -08002460 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002461 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002462 if (!swap_map) {
2463 error = -ENOMEM;
2464 goto bad_swap;
2465 }
Minchan Kim6ca29ee2017-01-10 16:58:15 -08002466
2467 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
2468 p->flags |= SWP_STABLE_WRITES;
2469
Shaohua Li2a8f9442013-09-11 14:20:28 -07002470 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002471 int cpu;
2472
Shaohua Li2a8f9442013-09-11 14:20:28 -07002473 p->flags |= SWP_SOLIDSTATE;
2474 /*
2475 * select a random position to start with to help wear leveling
2476 * SSD
2477 */
2478 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2479
2480 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2481 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2482 if (!cluster_info) {
2483 error = -ENOMEM;
2484 goto bad_swap;
2485 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002486 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2487 if (!p->percpu_cluster) {
2488 error = -ENOMEM;
2489 goto bad_swap;
2490 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002491 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002492 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002493 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002494 cluster_set_null(&cluster->index);
2495 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002496 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002497
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002498 error = swap_cgroup_swapon(p->type, maxpages);
2499 if (error)
2500 goto bad_swap;
2501
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002502 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002503 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002504 if (unlikely(nr_extents < 0)) {
2505 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 goto bad_swap;
2507 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002508 /* frontswap enabled? set up bit-per-page map for frontswap */
Vlastimil Babka8ea1d2a2016-07-26 15:24:42 -07002509 if (IS_ENABLED(CONFIG_FRONTSWAP))
Akinobu Mita7b579762013-06-12 14:05:08 -07002510 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Shaohua Li2a8f9442013-09-11 14:20:28 -07002512 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2513 /*
2514 * When discard is enabled for swap with no particular
2515 * policy flagged, we set all swap discard flags here in
2516 * order to sustain backward compatibility with older
2517 * swapon(8) releases.
2518 */
2519 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2520 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002521
Shaohua Li2a8f9442013-09-11 14:20:28 -07002522 /*
2523 * By flagging sys_swapon, a sysadmin can tell us to
2524 * either do single-time area discards only, or to just
2525 * perform discards for released swap page-clusters.
2526 * Now it's time to adjust the p->flags accordingly.
2527 */
2528 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2529 p->flags &= ~SWP_PAGE_DISCARD;
2530 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2531 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002532
Shaohua Li2a8f9442013-09-11 14:20:28 -07002533 /* issue a swapon-time discard if it's still required */
2534 if (p->flags & SWP_AREA_DISCARD) {
2535 int err = discard_swap(p);
2536 if (unlikely(err))
2537 pr_err("swapon: discard_swap(%p): %d\n",
2538 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002539 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002540 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002541
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002542 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002543 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002544 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002545 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002546 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002547 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002548
Joe Perches756a025f02016-03-17 14:19:47 -07002549 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002550 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002551 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2552 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002553 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002554 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2555 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002556 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002557
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002558 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002559 atomic_inc(&proc_poll_event);
2560 wake_up_interruptible(&proc_poll_wait);
2561
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002562 if (S_ISREG(inode->i_mode))
2563 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 error = 0;
2565 goto out;
2566bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002567 free_percpu(p->percpu_cluster);
2568 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002569 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002570 set_blocksize(p->bdev, p->old_block_size);
2571 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002573 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002574 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002575 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002578 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002580 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002581 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002582 if (inode && S_ISREG(inode->i_mode)) {
Al Viro59551022016-01-22 15:40:57 -05002583 inode_unlock(inode);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002584 inode = NULL;
2585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588out:
2589 if (page && !IS_ERR(page)) {
2590 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002591 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 }
2593 if (name)
2594 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002595 if (inode && S_ISREG(inode->i_mode))
Al Viro59551022016-01-22 15:40:57 -05002596 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 return error;
2598}
2599
2600void si_swapinfo(struct sysinfo *val)
2601{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002602 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 unsigned long nr_to_be_unused = 0;
2604
Hugh Dickins5d337b92005-09-03 15:54:41 -07002605 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002606 for (type = 0; type < nr_swapfiles; type++) {
2607 struct swap_info_struct *si = swap_info[type];
2608
2609 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2610 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002612 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002614 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615}
2616
2617/*
2618 * Verify that a swap entry is valid and increment its swap map count.
2619 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002620 * Returns error code in following case.
2621 * - success -> 0
2622 * - swp_entry is invalid -> EINVAL
2623 * - swp_entry is migration entry -> EINVAL
2624 * - swap-cache reference is requested but there is already one. -> EEXIST
2625 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002626 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002628static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002630 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002632 unsigned char count;
2633 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002634 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635
Andi Kleena7420aa2009-09-16 11:50:05 +02002636 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002637 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002638
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 type = swp_type(entry);
2640 if (type >= nr_swapfiles)
2641 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002642 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 offset = swp_offset(entry);
2644
Shaohua Liec8acf22013-02-22 16:34:38 -08002645 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002646 if (unlikely(offset >= p->max))
2647 goto unlock_out;
2648
Hugh Dickins253d5532009-12-14 17:58:44 -08002649 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002650
2651 /*
2652 * swapin_readahead() doesn't check if a swap entry is valid, so the
2653 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2654 */
2655 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2656 err = -ENOENT;
2657 goto unlock_out;
2658 }
2659
Hugh Dickins253d5532009-12-14 17:58:44 -08002660 has_cache = count & SWAP_HAS_CACHE;
2661 count &= ~SWAP_HAS_CACHE;
2662 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002663
Hugh Dickins253d5532009-12-14 17:58:44 -08002664 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002665
2666 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002667 if (!has_cache && count)
2668 has_cache = SWAP_HAS_CACHE;
2669 else if (has_cache) /* someone else added cache */
2670 err = -EEXIST;
2671 else /* no users remaining */
2672 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002673
2674 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002675
Hugh Dickins570a335b2009-12-14 17:58:46 -08002676 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2677 count += usage;
2678 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002679 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002680 else if (swap_count_continued(p, offset, count))
2681 count = COUNT_CONTINUED;
2682 else
2683 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002684 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002685 err = -ENOENT; /* unused swap entry */
2686
2687 p->swap_map[offset] = count | has_cache;
2688
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002689unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002690 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002692 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693
2694bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002695 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 goto out;
2697}
Hugh Dickins253d5532009-12-14 17:58:44 -08002698
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002699/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002700 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2701 * (in which case its reference count is never incremented).
2702 */
2703void swap_shmem_alloc(swp_entry_t entry)
2704{
2705 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2706}
2707
2708/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002709 * Increase reference count of swap entry by 1.
2710 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2711 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2712 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2713 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002714 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002715int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002716{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002717 int err = 0;
2718
2719 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2720 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2721 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002722}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002724/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002725 * @entry: swap entry for which we allocate swap cache.
2726 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002727 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002728 * This can return error codes. Returns 0 at success.
2729 * -EBUSY means there is a swap cache.
2730 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002731 */
2732int swapcache_prepare(swp_entry_t entry)
2733{
Hugh Dickins253d5532009-12-14 17:58:44 -08002734 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002735}
2736
Mel Gormanf981c592012-07-31 16:44:47 -07002737struct swap_info_struct *page_swap_info(struct page *page)
2738{
2739 swp_entry_t swap = { .val = page_private(page) };
Mel Gormanf981c592012-07-31 16:44:47 -07002740 return swap_info[swp_type(swap)];
2741}
2742
2743/*
2744 * out-of-line __page_file_ methods to avoid include hell.
2745 */
2746struct address_space *__page_file_mapping(struct page *page)
2747{
Sasha Levin309381fea2014-01-23 15:52:54 -08002748 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002749 return page_swap_info(page)->swap_file->f_mapping;
2750}
2751EXPORT_SYMBOL_GPL(__page_file_mapping);
2752
2753pgoff_t __page_file_index(struct page *page)
2754{
2755 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002756 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002757 return swp_offset(swap);
2758}
2759EXPORT_SYMBOL_GPL(__page_file_index);
2760
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002762 * add_swap_count_continuation - called when a swap count is duplicated
2763 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2764 * page of the original vmalloc'ed swap_map, to hold the continuation count
2765 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2766 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2767 *
2768 * These continuation pages are seldom referenced: the common paths all work
2769 * on the original swap_map, only referring to a continuation page when the
2770 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2771 *
2772 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2773 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2774 * can be called after dropping locks.
2775 */
2776int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2777{
2778 struct swap_info_struct *si;
2779 struct page *head;
2780 struct page *page;
2781 struct page *list_page;
2782 pgoff_t offset;
2783 unsigned char count;
2784
2785 /*
2786 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2787 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2788 */
2789 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2790
2791 si = swap_info_get(entry);
2792 if (!si) {
2793 /*
2794 * An acceptable race has occurred since the failing
2795 * __swap_duplicate(): the swap entry has been freed,
2796 * perhaps even the whole swap_map cleared for swapoff.
2797 */
2798 goto outer;
2799 }
2800
2801 offset = swp_offset(entry);
2802 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2803
2804 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2805 /*
2806 * The higher the swap count, the more likely it is that tasks
2807 * will race to add swap count continuation: we need to avoid
2808 * over-provisioning.
2809 */
2810 goto out;
2811 }
2812
2813 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002814 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002815 return -ENOMEM;
2816 }
2817
2818 /*
2819 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002820 * no architecture is using highmem pages for kernel page tables: so it
2821 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002822 */
2823 head = vmalloc_to_page(si->swap_map + offset);
2824 offset &= ~PAGE_MASK;
2825
2826 /*
2827 * Page allocation does not initialize the page's lru field,
2828 * but it does always reset its private field.
2829 */
2830 if (!page_private(head)) {
2831 BUG_ON(count & COUNT_CONTINUED);
2832 INIT_LIST_HEAD(&head->lru);
2833 set_page_private(head, SWP_CONTINUED);
2834 si->flags |= SWP_CONTINUED;
2835 }
2836
2837 list_for_each_entry(list_page, &head->lru, lru) {
2838 unsigned char *map;
2839
2840 /*
2841 * If the previous map said no continuation, but we've found
2842 * a continuation page, free our allocation and use this one.
2843 */
2844 if (!(count & COUNT_CONTINUED))
2845 goto out;
2846
Cong Wang9b04c5f2011-11-25 23:14:39 +08002847 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002848 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002849 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002850
2851 /*
2852 * If this continuation count now has some space in it,
2853 * free our allocation and use this one.
2854 */
2855 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2856 goto out;
2857 }
2858
2859 list_add_tail(&page->lru, &head->lru);
2860 page = NULL; /* now it's attached, don't free it */
2861out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002862 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002863outer:
2864 if (page)
2865 __free_page(page);
2866 return 0;
2867}
2868
2869/*
2870 * swap_count_continued - when the original swap_map count is incremented
2871 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2872 * into, carry if so, or else fail until a new continuation page is allocated;
2873 * when the original swap_map count is decremented from 0 with continuation,
2874 * borrow from the continuation and report whether it still holds more.
2875 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2876 */
2877static bool swap_count_continued(struct swap_info_struct *si,
2878 pgoff_t offset, unsigned char count)
2879{
2880 struct page *head;
2881 struct page *page;
2882 unsigned char *map;
2883
2884 head = vmalloc_to_page(si->swap_map + offset);
2885 if (page_private(head) != SWP_CONTINUED) {
2886 BUG_ON(count & COUNT_CONTINUED);
2887 return false; /* need to add count continuation */
2888 }
2889
2890 offset &= ~PAGE_MASK;
2891 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002892 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002893
2894 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2895 goto init_map; /* jump over SWAP_CONT_MAX checks */
2896
2897 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2898 /*
2899 * Think of how you add 1 to 999
2900 */
2901 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002902 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002903 page = list_entry(page->lru.next, struct page, lru);
2904 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002905 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002906 }
2907 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002908 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002909 page = list_entry(page->lru.next, struct page, lru);
2910 if (page == head)
2911 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002912 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002913init_map: *map = 0; /* we didn't zero the page */
2914 }
2915 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002916 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002917 page = list_entry(page->lru.prev, struct page, lru);
2918 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002919 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002920 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002921 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002922 page = list_entry(page->lru.prev, struct page, lru);
2923 }
2924 return true; /* incremented */
2925
2926 } else { /* decrementing */
2927 /*
2928 * Think of how you subtract 1 from 1000
2929 */
2930 BUG_ON(count != COUNT_CONTINUED);
2931 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002932 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002933 page = list_entry(page->lru.next, struct page, lru);
2934 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002935 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002936 }
2937 BUG_ON(*map == 0);
2938 *map -= 1;
2939 if (*map == 0)
2940 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002941 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002942 page = list_entry(page->lru.prev, struct page, lru);
2943 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002944 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002945 *map = SWAP_CONT_MAX | count;
2946 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002947 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002948 page = list_entry(page->lru.prev, struct page, lru);
2949 }
2950 return count == COUNT_CONTINUED;
2951 }
2952}
2953
2954/*
2955 * free_swap_count_continuations - swapoff free all the continuation pages
2956 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2957 */
2958static void free_swap_count_continuations(struct swap_info_struct *si)
2959{
2960 pgoff_t offset;
2961
2962 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2963 struct page *head;
2964 head = vmalloc_to_page(si->swap_map + offset);
2965 if (page_private(head)) {
Geliang Tang0d576d22016-01-14 15:21:49 -08002966 struct page *page, *next;
2967
2968 list_for_each_entry_safe(page, next, &head->lru, lru) {
2969 list_del(&page->lru);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002970 __free_page(page);
2971 }
2972 }
2973 }
2974}