blob: 2210de290b54d160d31afc937471225077bf083a [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);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800946 if (count == 1 && !PageWriteback(page)) {
947 delete_from_swap_cache(page);
948 SetPageDirty(page);
949 }
950 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800951 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952}
953
954/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800955 * If swap is getting full, or if there are no more mappings of this page,
956 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800958int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959{
Sasha Levin309381fea2014-01-23 15:52:54 -0800960 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
962 if (!PageSwapCache(page))
963 return 0;
964 if (PageWriteback(page))
965 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800966 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 return 0;
968
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700969 /*
970 * Once hibernation has begun to create its image of memory,
971 * there's a danger that one of the calls to try_to_free_swap()
972 * - most probably a call from __try_to_reclaim_swap() while
973 * hibernation is allocating its own swap pages for the image,
974 * but conceivably even a call from memory reclaim - will free
975 * the swap from a page which has already been recorded in the
976 * image as a clean swapcache page, and then reuse its swap for
977 * another page of the image. On waking from hibernation, the
978 * original page might be freed under memory pressure, then
979 * later read back in from swap, now with the wrong data.
980 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800981 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -0800982 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700983 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800984 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700985 return 0;
986
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800987 delete_from_swap_cache(page);
988 SetPageDirty(page);
989 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700990}
991
992/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 * Free the swap entry like above, but also try to
994 * free the page cache entry if it is the last user.
995 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800996int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800998 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 struct page *page = NULL;
1000
Andi Kleena7420aa2009-09-16 11:50:05 +02001001 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -08001002 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -07001003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 p = swap_info_get(entry);
1005 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -08001006 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -08001007 page = find_get_page(swap_address_space(entry),
Huang Yingf6ab1f72016-10-07 17:00:21 -07001008 swp_offset(entry));
Nick Piggin8413ac92008-10-18 20:26:59 -07001009 if (page && !trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001010 put_page(page);
Nick Piggin93fac702006-03-31 02:29:56 -08001011 page = NULL;
1012 }
1013 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001014 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 }
1016 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001017 /*
1018 * Not mapped elsewhere, or swap space full? Free it!
1019 * Also recheck PageSwapCache now page is locked (above).
1020 */
Nick Piggin93fac702006-03-31 02:29:56 -08001021 if (PageSwapCache(page) && !PageWriteback(page) &&
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08001022 (!page_mapped(page) || mem_cgroup_swap_full(page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 delete_from_swap_cache(page);
1024 SetPageDirty(page);
1025 }
1026 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001027 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 }
Hugh Dickins2509ef22009-01-06 14:40:10 -08001029 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030}
1031
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001032#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001033/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001034 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001035 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001036 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1037 * from 0, in which the swap header is expected to be located.
1038 *
1039 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001040 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001041int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001042{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001043 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001044 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001045
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001046 if (device)
1047 bdev = bdget(device);
1048
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001049 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001050 for (type = 0; type < nr_swapfiles; type++) {
1051 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001052
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001053 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001054 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001055
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001056 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001057 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001058 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001059
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001060 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001061 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001062 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001063 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001064 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001065
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001066 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001067 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001068 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001069
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001070 spin_unlock(&swap_lock);
1071 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001072 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001073 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001074 }
1075 }
1076 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001077 if (bdev)
1078 bdput(bdev);
1079
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001080 return -ENODEV;
1081}
1082
1083/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001084 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1085 * corresponding to given index in swap_info (swap type).
1086 */
1087sector_t swapdev_block(int type, pgoff_t offset)
1088{
1089 struct block_device *bdev;
1090
1091 if ((unsigned int)type >= nr_swapfiles)
1092 return 0;
1093 if (!(swap_info[type]->flags & SWP_WRITEOK))
1094 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001095 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001096}
1097
1098/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001099 * Return either the total number of swap pages of given type, or the number
1100 * of free pages of that type (depending on @free)
1101 *
1102 * This is needed for software suspend
1103 */
1104unsigned int count_swap_pages(int type, int free)
1105{
1106 unsigned int n = 0;
1107
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001108 spin_lock(&swap_lock);
1109 if ((unsigned int)type < nr_swapfiles) {
1110 struct swap_info_struct *sis = swap_info[type];
1111
Shaohua Liec8acf22013-02-22 16:34:38 -08001112 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001113 if (sis->flags & SWP_WRITEOK) {
1114 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001115 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001116 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001117 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001118 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001119 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001120 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001121 return n;
1122}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001123#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001124
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001125static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001126{
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001127 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001128}
1129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001131 * No need to decide whether this PTE shares the swap entry with others,
1132 * just let do_wp_page work it out if a write is requested later - to
1133 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001135static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 unsigned long addr, swp_entry_t entry, struct page *page)
1137{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001138 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001139 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001140 spinlock_t *ptl;
1141 pte_t *pte;
1142 int ret = 1;
1143
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001144 swapcache = page;
1145 page = ksm_might_need_to_copy(page, vma, addr);
1146 if (unlikely(!page))
1147 return -ENOMEM;
1148
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001149 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1150 &memcg, false)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001151 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001152 goto out_nolock;
1153 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001154
1155 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001156 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001157 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001158 ret = 0;
1159 goto out;
1160 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001161
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001162 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001163 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 get_page(page);
1165 set_pte_at(vma->vm_mm, addr, pte,
1166 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001167 if (page == swapcache) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001168 page_add_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001169 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001170 } else { /* ksm created a completely new copy */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001171 page_add_new_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001172 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001173 lru_cache_add_active_or_unevictable(page, vma);
1174 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 swap_free(entry);
1176 /*
1177 * Move the page to the active list so it is not
1178 * immediately swapped out again after swapon.
1179 */
1180 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001181out:
1182 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001183out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001184 if (page != swapcache) {
1185 unlock_page(page);
1186 put_page(page);
1187 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001188 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
1190
1191static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1192 unsigned long addr, unsigned long end,
1193 swp_entry_t entry, struct page *page)
1194{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001196 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001197 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
Hugh Dickins044d66c2008-02-07 00:14:04 -08001199 /*
1200 * We don't actually need pte lock while scanning for swp_pte: since
1201 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1202 * page table while we're scanning; though it could get zapped, and on
1203 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1204 * of unmatched parts which look like swp_pte, so unuse_pte must
1205 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001206 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001207 */
1208 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 do {
1210 /*
1211 * swapoff spends a _lot_ of time in this loop!
1212 * Test inline before going to call unuse_pte.
1213 */
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001214 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001215 pte_unmap(pte);
1216 ret = unuse_pte(vma, pmd, addr, entry, page);
1217 if (ret)
1218 goto out;
1219 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
1221 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001222 pte_unmap(pte - 1);
1223out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001224 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225}
1226
1227static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1228 unsigned long addr, unsigned long end,
1229 swp_entry_t entry, struct page *page)
1230{
1231 pmd_t *pmd;
1232 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001233 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
1235 pmd = pmd_offset(pud, addr);
1236 do {
1237 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001238 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001240 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1241 if (ret)
1242 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 } while (pmd++, addr = next, addr != end);
1244 return 0;
1245}
1246
1247static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1248 unsigned long addr, unsigned long end,
1249 swp_entry_t entry, struct page *page)
1250{
1251 pud_t *pud;
1252 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001253 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
1255 pud = pud_offset(pgd, addr);
1256 do {
1257 next = pud_addr_end(addr, end);
1258 if (pud_none_or_clear_bad(pud))
1259 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001260 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1261 if (ret)
1262 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 } while (pud++, addr = next, addr != end);
1264 return 0;
1265}
1266
1267static int unuse_vma(struct vm_area_struct *vma,
1268 swp_entry_t entry, struct page *page)
1269{
1270 pgd_t *pgd;
1271 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001272 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001274 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 addr = page_address_in_vma(page, vma);
1276 if (addr == -EFAULT)
1277 return 0;
1278 else
1279 end = addr + PAGE_SIZE;
1280 } else {
1281 addr = vma->vm_start;
1282 end = vma->vm_end;
1283 }
1284
1285 pgd = pgd_offset(vma->vm_mm, addr);
1286 do {
1287 next = pgd_addr_end(addr, end);
1288 if (pgd_none_or_clear_bad(pgd))
1289 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001290 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1291 if (ret)
1292 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 } while (pgd++, addr = next, addr != end);
1294 return 0;
1295}
1296
1297static int unuse_mm(struct mm_struct *mm,
1298 swp_entry_t entry, struct page *page)
1299{
1300 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001301 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
1303 if (!down_read_trylock(&mm->mmap_sem)) {
1304 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001305 * Activate page so shrink_inactive_list is unlikely to unmap
1306 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001308 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 unlock_page(page);
1310 down_read(&mm->mmap_sem);
1311 lock_page(page);
1312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001314 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 break;
1316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001318 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
1320
1321/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001322 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1323 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 * Recycle to start on reaching the end, returning 0 when empty.
1325 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001326static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001327 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001329 unsigned int max = si->max;
1330 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001331 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001334 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * for whether an entry is in use, not modifying it; false
1336 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001337 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 */
1339 for (;;) {
1340 if (++i >= max) {
1341 if (!prev) {
1342 i = 0;
1343 break;
1344 }
1345 /*
1346 * No entries in use at top of swap_map,
1347 * loop back to start and recheck there.
1348 */
1349 max = prev + 1;
1350 prev = 0;
1351 i = 1;
1352 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001353 if (frontswap) {
1354 if (frontswap_test(si, i))
1355 break;
1356 else
1357 continue;
1358 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001359 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001360 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 break;
1362 }
1363 return i;
1364}
1365
1366/*
1367 * We completely avoid races by reading each swap page in advance,
1368 * and then search for the process using it. All the necessary
1369 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001370 *
1371 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1372 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001374int try_to_unuse(unsigned int type, bool frontswap,
1375 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001377 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001379 volatile unsigned char *swap_map; /* swap_map is accessed without
1380 * locking. Mark it as volatile
1381 * to prevent compiler doing
1382 * something odd.
1383 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001384 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 struct page *page;
1386 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001387 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
1390 /*
1391 * When searching mms for an entry, a good strategy is to
1392 * start at the first mm we freed the previous entry from
1393 * (though actually we don't notice whether we or coincidence
1394 * freed the entry). Initialize this start_mm with a hold.
1395 *
1396 * A simpler strategy would be to start at the last mm we
1397 * freed the previous entry from; but that would take less
1398 * advantage of mmlist ordering, which clusters forked mms
1399 * together, child after parent. If we race with dup_mmap(), we
1400 * prefer to resolve parent before child, lest we miss entries
1401 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001402 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 */
1404 start_mm = &init_mm;
1405 atomic_inc(&init_mm.mm_users);
1406
1407 /*
1408 * Keep on scanning until all entries have gone. Usually,
1409 * one pass through swap_map is enough, but not necessarily:
1410 * there are races when an instance of an entry might be missed.
1411 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001412 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (signal_pending(current)) {
1414 retval = -EINTR;
1415 break;
1416 }
1417
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001418 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 * Get a page for the entry, using the existing swap
1420 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001421 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 */
1423 swap_map = &si->swap_map[i];
1424 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001425 page = read_swap_cache_async(entry,
1426 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (!page) {
1428 /*
1429 * Either swap_duplicate() failed because entry
1430 * has been freed independently, and will not be
1431 * reused since sys_swapoff() already disabled
1432 * allocation from here, or alloc_page() failed.
1433 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001434 swcount = *swap_map;
1435 /*
1436 * We don't hold lock here, so the swap entry could be
1437 * SWAP_MAP_BAD (when the cluster is discarding).
1438 * Instead of fail out, We can just skip the swap
1439 * entry because swapoff will wait for discarding
1440 * finish anyway.
1441 */
1442 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 continue;
1444 retval = -ENOMEM;
1445 break;
1446 }
1447
1448 /*
1449 * Don't hold on to start_mm if it looks like exiting.
1450 */
1451 if (atomic_read(&start_mm->mm_users) == 1) {
1452 mmput(start_mm);
1453 start_mm = &init_mm;
1454 atomic_inc(&init_mm.mm_users);
1455 }
1456
1457 /*
1458 * Wait for and lock page. When do_swap_page races with
1459 * try_to_unuse, do_swap_page can handle the fault much
1460 * faster than try_to_unuse can locate the entry. This
1461 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1462 * defer to do_swap_page in such a case - in some tests,
1463 * do_swap_page and try_to_unuse repeatedly compete.
1464 */
1465 wait_on_page_locked(page);
1466 wait_on_page_writeback(page);
1467 lock_page(page);
1468 wait_on_page_writeback(page);
1469
1470 /*
1471 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001474 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1475 retval = shmem_unuse(entry, page);
1476 /* page has already been unlocked and released */
1477 if (retval < 0)
1478 break;
1479 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001481 if (swap_count(swcount) && start_mm != &init_mm)
1482 retval = unuse_mm(start_mm, entry, page);
1483
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001484 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 int set_start_mm = (*swap_map >= swcount);
1486 struct list_head *p = &start_mm->mmlist;
1487 struct mm_struct *new_start_mm = start_mm;
1488 struct mm_struct *prev_mm = start_mm;
1489 struct mm_struct *mm;
1490
1491 atomic_inc(&new_start_mm->mm_users);
1492 atomic_inc(&prev_mm->mm_users);
1493 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001494 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 (p = p->next) != &start_mm->mmlist) {
1496 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001497 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 spin_unlock(&mmlist_lock);
1500 mmput(prev_mm);
1501 prev_mm = mm;
1502
1503 cond_resched();
1504
1505 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001506 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001508 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001510 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001512
Bo Liu32c5fc12009-11-02 16:50:33 +00001513 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 mmput(new_start_mm);
1515 atomic_inc(&mm->mm_users);
1516 new_start_mm = mm;
1517 set_start_mm = 0;
1518 }
1519 spin_lock(&mmlist_lock);
1520 }
1521 spin_unlock(&mmlist_lock);
1522 mmput(prev_mm);
1523 mmput(start_mm);
1524 start_mm = new_start_mm;
1525 }
1526 if (retval) {
1527 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001528 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 break;
1530 }
1531
1532 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 * If a reference remains (rare), we would like to leave
1534 * the page in the swap cache; but try_to_unmap could
1535 * then re-duplicate the entry once we drop page lock,
1536 * so we might loop indefinitely; also, that page could
1537 * not be swapped out to other storage meanwhile. So:
1538 * delete from cache even if there's another reference,
1539 * after ensuring that the data has been saved to disk -
1540 * since if the reference remains (rarer), it will be
1541 * read from disk into another page. Splitting into two
1542 * pages would be incorrect if swap supported "shared
1543 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001544 *
1545 * Given how unuse_vma() targets one particular offset
1546 * in an anon_vma, once the anon_vma has been determined,
1547 * this splitting happens to be just what is needed to
1548 * handle where KSM pages have been swapped out: re-reading
1549 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001551 if (swap_count(*swap_map) &&
1552 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 struct writeback_control wbc = {
1554 .sync_mode = WB_SYNC_NONE,
1555 };
1556
1557 swap_writepage(page, &wbc);
1558 lock_page(page);
1559 wait_on_page_writeback(page);
1560 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001561
1562 /*
1563 * It is conceivable that a racing task removed this page from
1564 * swap cache just before we acquired the page lock at the top,
1565 * or while we dropped it in unuse_mm(). The page might even
1566 * be back in swap cache on another swap area: that we must not
1567 * delete, since it may not have been written out to swap yet.
1568 */
1569 if (PageSwapCache(page) &&
1570 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001571 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
1573 /*
1574 * So we could skip searching mms once swap count went
1575 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001576 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 */
1578 SetPageDirty(page);
1579 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001580 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 /*
1583 * Make sure that we aren't completely killing
1584 * interactive performance.
1585 */
1586 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001587 if (frontswap && pages_to_unuse > 0) {
1588 if (!--pages_to_unuse)
1589 break;
1590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 }
1592
1593 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 return retval;
1595}
1596
1597/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001598 * After a successful try_to_unuse, if no swap is now in use, we know
1599 * we can empty the mmlist. swap_lock must be held on entry and exit.
1600 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 * added to the mmlist just after page_duplicate - before would be racy.
1602 */
1603static void drain_mmlist(void)
1604{
1605 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001606 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001608 for (type = 0; type < nr_swapfiles; type++)
1609 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 return;
1611 spin_lock(&mmlist_lock);
1612 list_for_each_safe(p, next, &init_mm.mmlist)
1613 list_del_init(p);
1614 spin_unlock(&mmlist_lock);
1615}
1616
1617/*
1618 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001619 * corresponds to page offset for the specified swap entry.
1620 * Note that the type of this function is sector_t, but it returns page offset
1621 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001623static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001625 struct swap_info_struct *sis;
1626 struct swap_extent *start_se;
1627 struct swap_extent *se;
1628 pgoff_t offset;
1629
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001630 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001631 *bdev = sis->bdev;
1632
1633 offset = swp_offset(entry);
1634 start_se = sis->curr_swap_extent;
1635 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 if (se->start_page <= offset &&
1639 offset < (se->start_page + se->nr_pages)) {
1640 return se->start_block + (offset - se->start_page);
1641 }
Geliang Tanga8ae4992016-01-14 15:20:45 -08001642 se = list_next_entry(se, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 sis->curr_swap_extent = se;
1644 BUG_ON(se == start_se); /* It *must* be present */
1645 }
1646}
1647
1648/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001649 * Returns the page offset into bdev for the specified page's swap entry.
1650 */
1651sector_t map_swap_page(struct page *page, struct block_device **bdev)
1652{
1653 swp_entry_t entry;
1654 entry.val = page_private(page);
1655 return map_swap_entry(entry, bdev);
1656}
1657
1658/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 * Free all of a swapdev's extent information
1660 */
1661static void destroy_swap_extents(struct swap_info_struct *sis)
1662{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001663 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 struct swap_extent *se;
1665
Geliang Tanga8ae4992016-01-14 15:20:45 -08001666 se = list_first_entry(&sis->first_swap_extent.list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 struct swap_extent, list);
1668 list_del(&se->list);
1669 kfree(se);
1670 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001671
1672 if (sis->flags & SWP_FILE) {
1673 struct file *swap_file = sis->swap_file;
1674 struct address_space *mapping = swap_file->f_mapping;
1675
1676 sis->flags &= ~SWP_FILE;
1677 mapping->a_ops->swap_deactivate(swap_file);
1678 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679}
1680
1681/*
1682 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001683 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001685 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 */
Mel Gormana509bc12012-07-31 16:44:57 -07001687int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1689 unsigned long nr_pages, sector_t start_block)
1690{
1691 struct swap_extent *se;
1692 struct swap_extent *new_se;
1693 struct list_head *lh;
1694
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001695 if (start_page == 0) {
1696 se = &sis->first_swap_extent;
1697 sis->curr_swap_extent = se;
1698 se->start_page = 0;
1699 se->nr_pages = nr_pages;
1700 se->start_block = start_block;
1701 return 1;
1702 } else {
1703 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001705 BUG_ON(se->start_page + se->nr_pages != start_page);
1706 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 /* Merge it */
1708 se->nr_pages += nr_pages;
1709 return 0;
1710 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 }
1712
1713 /*
1714 * No merge. Insert a new extent, preserving ordering.
1715 */
1716 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1717 if (new_se == NULL)
1718 return -ENOMEM;
1719 new_se->start_page = start_page;
1720 new_se->nr_pages = nr_pages;
1721 new_se->start_block = start_block;
1722
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001723 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001724 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725}
1726
1727/*
1728 * A `swap extent' is a simple thing which maps a contiguous range of pages
1729 * onto a contiguous range of disk blocks. An ordered list of swap extents
1730 * is built at swapon time and is then used at swap_writepage/swap_readpage
1731 * time for locating where on disk a page belongs.
1732 *
1733 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1734 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1735 * swap files identically.
1736 *
1737 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1738 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1739 * swapfiles are handled *identically* after swapon time.
1740 *
1741 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1742 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1743 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1744 * requirements, they are simply tossed out - we will never use those blocks
1745 * for swapping.
1746 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001747 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1749 * which will scribble on the fs.
1750 *
1751 * The amount of disk space which a single swap extent represents varies.
1752 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1753 * extents in the list. To avoid much list walking, we cache the previous
1754 * search location in `curr_swap_extent', and start new searches from there.
1755 * This is extremely effective. The average number of iterations in
1756 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1757 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001758static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759{
Mel Gorman62c230b2012-07-31 16:44:55 -07001760 struct file *swap_file = sis->swap_file;
1761 struct address_space *mapping = swap_file->f_mapping;
1762 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 int ret;
1764
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 if (S_ISBLK(inode->i_mode)) {
1766 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001767 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001768 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 }
1770
Mel Gorman62c230b2012-07-31 16:44:55 -07001771 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001772 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001773 if (!ret) {
1774 sis->flags |= SWP_FILE;
1775 ret = add_swap_extent(sis, 0, sis->max, 0);
1776 *span = sis->pages;
1777 }
Mel Gormana509bc12012-07-31 16:44:57 -07001778 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001779 }
1780
Mel Gormana509bc12012-07-31 16:44:57 -07001781 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782}
1783
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001784static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001785 unsigned char *swap_map,
1786 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001787{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001788 if (prio >= 0)
1789 p->prio = prio;
1790 else
1791 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001792 /*
1793 * the plist prio is negated because plist ordering is
1794 * low-to-high, while swap ordering is high-to-low
1795 */
1796 p->list.prio = -p->prio;
1797 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001798 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001799 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001800 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001801 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001802 total_swap_pages += p->pages;
1803
Dan Streetmanadfab832014-06-04 16:09:53 -07001804 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001805 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07001806 * both lists are plists, and thus priority ordered.
1807 * swap_active_head needs to be priority ordered for swapoff(),
1808 * which on removal of any swap_info_struct with an auto-assigned
1809 * (i.e. negative) priority increments the auto-assigned priority
1810 * of any lower-priority swap_info_structs.
1811 * swap_avail_head needs to be priority ordered for get_swap_page(),
1812 * which allocates swap pages from the highest available priority
1813 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07001814 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07001815 plist_add(&p->list, &swap_active_head);
1816 spin_lock(&swap_avail_lock);
1817 plist_add(&p->avail_list, &swap_avail_head);
1818 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001819}
1820
1821static void enable_swap_info(struct swap_info_struct *p, int prio,
1822 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001823 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001824 unsigned long *frontswap_map)
1825{
Minchan Kim4f898492013-04-30 15:26:54 -07001826 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001827 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001828 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001829 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001830 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001831 spin_unlock(&swap_lock);
1832}
1833
1834static void reinsert_swap_info(struct swap_info_struct *p)
1835{
1836 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001837 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001838 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001839 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001840 spin_unlock(&swap_lock);
1841}
1842
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001843SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001845 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001846 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001847 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001848 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 struct file *swap_file, *victim;
1850 struct address_space *mapping;
1851 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001852 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07001853 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001854 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 if (!capable(CAP_SYS_ADMIN))
1857 return -EPERM;
1858
Al Viro191c5422012-02-13 03:58:52 +00001859 BUG_ON(!current->mm);
1860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001863 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Jeff Layton669abf42012-10-10 16:43:10 -04001865 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 err = PTR_ERR(victim);
1867 if (IS_ERR(victim))
1868 goto out;
1869
1870 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001871 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07001872 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8fd2009-01-06 14:39:48 -08001873 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001874 if (p->swap_file->f_mapping == mapping) {
1875 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07001877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 }
Dan Streetmanadfab832014-06-04 16:09:53 -07001880 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001882 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 goto out_dput;
1884 }
Al Viro191c5422012-02-13 03:58:52 +00001885 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 vm_unacct_memory(p->pages);
1887 else {
1888 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001889 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 goto out_dput;
1891 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001892 spin_lock(&swap_avail_lock);
1893 plist_del(&p->avail_list, &swap_avail_head);
1894 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001895 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001896 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001897 struct swap_info_struct *si = p;
1898
Dan Streetman18ab4d42014-06-04 16:09:59 -07001899 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001900 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001901 si->list.prio--;
1902 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07001903 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001904 least_priority++;
1905 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001906 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08001907 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 total_swap_pages -= p->pages;
1909 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001910 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001911 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001912
David Rientjese1e12d22012-12-11 16:02:56 -08001913 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07001914 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08001915 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 if (err) {
1918 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001919 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 goto out_dput;
1921 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001922
Shaohua Li815c2c52013-09-11 14:20:30 -07001923 flush_work(&p->discard_work);
1924
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001925 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001926 if (p->flags & SWP_CONTINUED)
1927 free_swap_count_continuations(p);
1928
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001929 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001930 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001931 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001933
1934 /* wait for anyone still in scan_swap_map */
1935 p->highest_bit = 0; /* cuts scans short */
1936 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001937 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001938 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001939 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001940 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001941 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001942 }
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001945 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 p->swap_file = NULL;
1947 p->max = 0;
1948 swap_map = p->swap_map;
1949 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001950 cluster_info = p->cluster_info;
1951 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07001952 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08001953 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001954 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001955 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08001956 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001957 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001958 free_percpu(p->percpu_cluster);
1959 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001961 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001962 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001963 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07001964 swap_cgroup_swapoff(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001965
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 inode = mapping->host;
1967 if (S_ISBLK(inode->i_mode)) {
1968 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001969 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001970 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 } else {
Al Viro59551022016-01-22 15:40:57 -05001972 inode_lock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 inode->i_flags &= ~S_SWAPFILE;
Al Viro59551022016-01-22 15:40:57 -05001974 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 }
1976 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08001977
1978 /*
1979 * Clear the SWP_USED flag after all resources are freed so that swapon
1980 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1981 * not hold p->lock after we cleared its SWP_WRITEOK.
1982 */
1983 spin_lock(&swap_lock);
1984 p->flags = 0;
1985 spin_unlock(&swap_lock);
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001988 atomic_inc(&proc_poll_event);
1989 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
1991out_dput:
1992 filp_close(victim, NULL);
1993out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001994 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 return err;
1996}
1997
1998#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001999static unsigned swaps_poll(struct file *file, poll_table *wait)
2000{
Kay Sieversf1514632011-07-12 20:48:39 +02002001 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002002
2003 poll_wait(file, &proc_poll_wait, wait);
2004
Kay Sieversf1514632011-07-12 20:48:39 +02002005 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2006 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002007 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2008 }
2009
2010 return POLLIN | POLLRDNORM;
2011}
2012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013/* iterator */
2014static void *swap_start(struct seq_file *swap, loff_t *pos)
2015{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002016 struct swap_info_struct *si;
2017 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 loff_t l = *pos;
2019
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002020 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002022 if (!l)
2023 return SEQ_START_TOKEN;
2024
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002025 for (type = 0; type < nr_swapfiles; type++) {
2026 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2027 si = swap_info[type];
2028 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002030 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002031 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 }
2033
2034 return NULL;
2035}
2036
2037static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2038{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002039 struct swap_info_struct *si = v;
2040 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002042 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002043 type = 0;
2044 else
2045 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002046
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002047 for (; type < nr_swapfiles; type++) {
2048 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2049 si = swap_info[type];
2050 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 continue;
2052 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002053 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 }
2055
2056 return NULL;
2057}
2058
2059static void swap_stop(struct seq_file *swap, void *v)
2060{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002061 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062}
2063
2064static int swap_show(struct seq_file *swap, void *v)
2065{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002066 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 struct file *file;
2068 int len;
2069
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002070 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002071 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2072 return 0;
2073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002075 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002076 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002077 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002078 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002079 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002081 si->pages << (PAGE_SHIFT - 10),
2082 si->inuse_pages << (PAGE_SHIFT - 10),
2083 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 return 0;
2085}
2086
Helge Deller15ad7cd2006-12-06 20:40:36 -08002087static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 .start = swap_start,
2089 .next = swap_next,
2090 .stop = swap_stop,
2091 .show = swap_show
2092};
2093
2094static int swaps_open(struct inode *inode, struct file *file)
2095{
Kay Sieversf1514632011-07-12 20:48:39 +02002096 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002097 int ret;
2098
Kay Sievers66d7dd52010-10-26 14:22:06 -07002099 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002100 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002101 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002102
Kay Sieversf1514632011-07-12 20:48:39 +02002103 seq = file->private_data;
2104 seq->poll_event = atomic_read(&proc_poll_event);
2105 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106}
2107
Helge Deller15ad7cd2006-12-06 20:40:36 -08002108static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 .open = swaps_open,
2110 .read = seq_read,
2111 .llseek = seq_lseek,
2112 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002113 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114};
2115
2116static int __init procswaps_init(void)
2117{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002118 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 return 0;
2120}
2121__initcall(procswaps_init);
2122#endif /* CONFIG_PROC_FS */
2123
Jan Beulich17963162008-12-16 11:35:24 +00002124#ifdef MAX_SWAPFILES_CHECK
2125static int __init max_swapfiles_check(void)
2126{
2127 MAX_SWAPFILES_CHECK();
2128 return 0;
2129}
2130late_initcall(max_swapfiles_check);
2131#endif
2132
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002133static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002135 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002137
2138 p = kzalloc(sizeof(*p), GFP_KERNEL);
2139 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002140 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002141
Hugh Dickins5d337b92005-09-03 15:54:41 -07002142 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002143 for (type = 0; type < nr_swapfiles; type++) {
2144 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002146 }
Christoph Lameter06972122006-06-23 02:03:35 -07002147 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002148 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002149 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002150 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002152 if (type >= nr_swapfiles) {
2153 p->type = type;
2154 swap_info[type] = p;
2155 /*
2156 * Write swap_info[type] before nr_swapfiles, in case a
2157 * racing procfs swap_start() or swap_next() is reading them.
2158 * (We never shrink nr_swapfiles, we never free this entry.)
2159 */
2160 smp_wmb();
2161 nr_swapfiles++;
2162 } else {
2163 kfree(p);
2164 p = swap_info[type];
2165 /*
2166 * Do not memset this entry: a racing procfs swap_next()
2167 * would be relying on p->type to remain valid.
2168 */
2169 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002170 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002171 plist_node_init(&p->list, 0);
2172 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002174 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002175 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002176
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002177 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002178}
2179
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002180static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2181{
2182 int error;
2183
2184 if (S_ISBLK(inode->i_mode)) {
2185 p->bdev = bdgrab(I_BDEV(inode));
2186 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002187 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002188 if (error < 0) {
2189 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002190 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002191 }
2192 p->old_block_size = block_size(p->bdev);
2193 error = set_blocksize(p->bdev, PAGE_SIZE);
2194 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002195 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002196 p->flags |= SWP_BLKDEV;
2197 } else if (S_ISREG(inode->i_mode)) {
2198 p->bdev = inode->i_sb->s_bdev;
Al Viro59551022016-01-22 15:40:57 -05002199 inode_lock(inode);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002200 if (IS_SWAPFILE(inode))
2201 return -EBUSY;
2202 } else
2203 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002204
2205 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002206}
2207
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002208static unsigned long read_swap_header(struct swap_info_struct *p,
2209 union swap_header *swap_header,
2210 struct inode *inode)
2211{
2212 int i;
2213 unsigned long maxpages;
2214 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002215 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002216
2217 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002218 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002219 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002220 }
2221
2222 /* swap partition endianess hack... */
2223 if (swab32(swap_header->info.version) == 1) {
2224 swab32s(&swap_header->info.version);
2225 swab32s(&swap_header->info.last_page);
2226 swab32s(&swap_header->info.nr_badpages);
2227 for (i = 0; i < swap_header->info.nr_badpages; i++)
2228 swab32s(&swap_header->info.badpages[i]);
2229 }
2230 /* Check the swap header's sub-version */
2231 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002232 pr_warn("Unable to handle swap header version %d\n",
2233 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002234 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002235 }
2236
2237 p->lowest_bit = 1;
2238 p->cluster_next = 1;
2239 p->cluster_nr = 0;
2240
2241 /*
2242 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002243 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002244 * of bits for the swap offset in the swp_entry_t type, and
2245 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002246 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002247 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002248 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002249 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002250 * offset is extracted. This will mask all the bits from
2251 * the initial ~0UL mask that can't be encoded in either
2252 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002253 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002254 */
2255 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002256 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002257 last_page = swap_header->info.last_page;
2258 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002259 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002260 maxpages << (PAGE_SHIFT - 10),
2261 last_page << (PAGE_SHIFT - 10));
2262 }
2263 if (maxpages > last_page) {
2264 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002265 /* p->max is an unsigned int: don't overflow it */
2266 if ((unsigned int)maxpages == 0)
2267 maxpages = UINT_MAX;
2268 }
2269 p->highest_bit = maxpages - 1;
2270
2271 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002272 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002273 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2274 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002275 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002276 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002277 }
2278 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002279 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002280 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002281 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002282
2283 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002284}
2285
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002286static int setup_swap_map_and_extents(struct swap_info_struct *p,
2287 union swap_header *swap_header,
2288 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002289 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002290 unsigned long maxpages,
2291 sector_t *span)
2292{
2293 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002294 unsigned int nr_good_pages;
2295 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002296 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2297 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002298
2299 nr_good_pages = maxpages - 1; /* omit header page */
2300
Huang Ying6b534912016-10-07 16:58:42 -07002301 cluster_list_init(&p->free_clusters);
2302 cluster_list_init(&p->discard_clusters);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002303
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002304 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2305 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002306 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2307 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002308 if (page_nr < maxpages) {
2309 swap_map[page_nr] = SWAP_MAP_BAD;
2310 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002311 /*
2312 * Haven't marked the cluster free yet, no list
2313 * operation involved
2314 */
2315 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002316 }
2317 }
2318
Shaohua Li2a8f9442013-09-11 14:20:28 -07002319 /* Haven't marked the cluster free yet, no list operation involved */
2320 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2321 inc_cluster_info_page(p, cluster_info, i);
2322
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002323 if (nr_good_pages) {
2324 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002325 /*
2326 * Not mark the cluster free yet, no list
2327 * operation involved
2328 */
2329 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002330 p->max = maxpages;
2331 p->pages = nr_good_pages;
2332 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002333 if (nr_extents < 0)
2334 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002335 nr_good_pages = p->pages;
2336 }
2337 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002338 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002339 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002340 }
2341
Shaohua Li2a8f9442013-09-11 14:20:28 -07002342 if (!cluster_info)
2343 return nr_extents;
2344
2345 for (i = 0; i < nr_clusters; i++) {
2346 if (!cluster_count(&cluster_info[idx])) {
2347 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -07002348 cluster_list_add_tail(&p->free_clusters, cluster_info,
2349 idx);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002350 }
2351 idx++;
2352 if (idx == nr_clusters)
2353 idx = 0;
2354 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002355 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002356}
2357
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002358/*
2359 * Helper to sys_swapon determining if a given swap
2360 * backing device queue supports DISCARD operations.
2361 */
2362static bool swap_discardable(struct swap_info_struct *si)
2363{
2364 struct request_queue *q = bdev_get_queue(si->bdev);
2365
2366 if (!q || !blk_queue_discard(q))
2367 return false;
2368
2369 return true;
2370}
2371
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002372SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2373{
2374 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002375 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002376 struct file *swap_file = NULL;
2377 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002378 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002379 int error;
2380 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002381 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002382 sector_t span;
2383 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002384 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002385 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002386 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002387 struct page *page = NULL;
2388 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002389
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002390 if (swap_flags & ~SWAP_FLAGS_VALID)
2391 return -EINVAL;
2392
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002393 if (!capable(CAP_SYS_ADMIN))
2394 return -EPERM;
2395
2396 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002397 if (IS_ERR(p))
2398 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002399
Shaohua Li815c2c52013-09-11 14:20:30 -07002400 INIT_WORK(&p->discard_work, swap_discard_work);
2401
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002404 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002406 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 }
Jeff Layton669abf42012-10-10 16:43:10 -04002408 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002410 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002412 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
2414
2415 p->swap_file = swap_file;
2416 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002417 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002418
Al Viro59551022016-01-22 15:40:57 -05002419 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002420 error = claim_swapfile(p, inode);
2421 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 /*
2425 * Read the swap header.
2426 */
2427 if (!mapping->a_ops->readpage) {
2428 error = -EINVAL;
2429 goto bad_swap;
2430 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002431 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 if (IS_ERR(page)) {
2433 error = PTR_ERR(page);
2434 goto bad_swap;
2435 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002436 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002438 maxpages = read_swap_header(p, swap_header, inode);
2439 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 error = -EINVAL;
2441 goto bad_swap;
2442 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002443
Hugh Dickins81e33972009-01-06 14:39:49 -08002444 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002445 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002446 if (!swap_map) {
2447 error = -ENOMEM;
2448 goto bad_swap;
2449 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002450 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002451 int cpu;
2452
Shaohua Li2a8f9442013-09-11 14:20:28 -07002453 p->flags |= SWP_SOLIDSTATE;
2454 /*
2455 * select a random position to start with to help wear leveling
2456 * SSD
2457 */
2458 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2459
2460 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2461 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2462 if (!cluster_info) {
2463 error = -ENOMEM;
2464 goto bad_swap;
2465 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002466 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2467 if (!p->percpu_cluster) {
2468 error = -ENOMEM;
2469 goto bad_swap;
2470 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002471 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002472 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002473 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002474 cluster_set_null(&cluster->index);
2475 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002476 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002477
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002478 error = swap_cgroup_swapon(p->type, maxpages);
2479 if (error)
2480 goto bad_swap;
2481
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002482 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002483 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002484 if (unlikely(nr_extents < 0)) {
2485 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 goto bad_swap;
2487 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002488 /* frontswap enabled? set up bit-per-page map for frontswap */
Vlastimil Babka8ea1d2a2016-07-26 15:24:42 -07002489 if (IS_ENABLED(CONFIG_FRONTSWAP))
Akinobu Mita7b579762013-06-12 14:05:08 -07002490 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Shaohua Li2a8f9442013-09-11 14:20:28 -07002492 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2493 /*
2494 * When discard is enabled for swap with no particular
2495 * policy flagged, we set all swap discard flags here in
2496 * order to sustain backward compatibility with older
2497 * swapon(8) releases.
2498 */
2499 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2500 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002501
Shaohua Li2a8f9442013-09-11 14:20:28 -07002502 /*
2503 * By flagging sys_swapon, a sysadmin can tell us to
2504 * either do single-time area discards only, or to just
2505 * perform discards for released swap page-clusters.
2506 * Now it's time to adjust the p->flags accordingly.
2507 */
2508 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2509 p->flags &= ~SWP_PAGE_DISCARD;
2510 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2511 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002512
Shaohua Li2a8f9442013-09-11 14:20:28 -07002513 /* issue a swapon-time discard if it's still required */
2514 if (p->flags & SWP_AREA_DISCARD) {
2515 int err = discard_swap(p);
2516 if (unlikely(err))
2517 pr_err("swapon: discard_swap(%p): %d\n",
2518 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002519 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002520 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002521
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002522 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002523 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002524 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002525 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002526 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002527 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002528
Joe Perches756a0252016-03-17 14:19:47 -07002529 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 -04002530 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002531 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2532 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002533 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002534 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2535 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002536 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002537
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002538 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002539 atomic_inc(&proc_poll_event);
2540 wake_up_interruptible(&proc_poll_wait);
2541
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002542 if (S_ISREG(inode->i_mode))
2543 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 error = 0;
2545 goto out;
2546bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002547 free_percpu(p->percpu_cluster);
2548 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002549 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002550 set_blocksize(p->bdev, p->old_block_size);
2551 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002553 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002554 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002555 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002558 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002560 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002561 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002562 if (inode && S_ISREG(inode->i_mode)) {
Al Viro59551022016-01-22 15:40:57 -05002563 inode_unlock(inode);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002564 inode = NULL;
2565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568out:
2569 if (page && !IS_ERR(page)) {
2570 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002571 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 }
2573 if (name)
2574 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002575 if (inode && S_ISREG(inode->i_mode))
Al Viro59551022016-01-22 15:40:57 -05002576 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 return error;
2578}
2579
2580void si_swapinfo(struct sysinfo *val)
2581{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002582 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 unsigned long nr_to_be_unused = 0;
2584
Hugh Dickins5d337b92005-09-03 15:54:41 -07002585 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002586 for (type = 0; type < nr_swapfiles; type++) {
2587 struct swap_info_struct *si = swap_info[type];
2588
2589 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2590 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002592 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002594 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595}
2596
2597/*
2598 * Verify that a swap entry is valid and increment its swap map count.
2599 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002600 * Returns error code in following case.
2601 * - success -> 0
2602 * - swp_entry is invalid -> EINVAL
2603 * - swp_entry is migration entry -> EINVAL
2604 * - swap-cache reference is requested but there is already one. -> EEXIST
2605 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002606 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002608static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002610 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002612 unsigned char count;
2613 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002614 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
Andi Kleena7420aa2009-09-16 11:50:05 +02002616 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002617 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 type = swp_type(entry);
2620 if (type >= nr_swapfiles)
2621 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002622 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 offset = swp_offset(entry);
2624
Shaohua Liec8acf22013-02-22 16:34:38 -08002625 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002626 if (unlikely(offset >= p->max))
2627 goto unlock_out;
2628
Hugh Dickins253d5532009-12-14 17:58:44 -08002629 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002630
2631 /*
2632 * swapin_readahead() doesn't check if a swap entry is valid, so the
2633 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2634 */
2635 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2636 err = -ENOENT;
2637 goto unlock_out;
2638 }
2639
Hugh Dickins253d5532009-12-14 17:58:44 -08002640 has_cache = count & SWAP_HAS_CACHE;
2641 count &= ~SWAP_HAS_CACHE;
2642 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002643
Hugh Dickins253d5532009-12-14 17:58:44 -08002644 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002645
2646 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002647 if (!has_cache && count)
2648 has_cache = SWAP_HAS_CACHE;
2649 else if (has_cache) /* someone else added cache */
2650 err = -EEXIST;
2651 else /* no users remaining */
2652 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002653
2654 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002655
Hugh Dickins570a335b2009-12-14 17:58:46 -08002656 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2657 count += usage;
2658 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002659 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002660 else if (swap_count_continued(p, offset, count))
2661 count = COUNT_CONTINUED;
2662 else
2663 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002664 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002665 err = -ENOENT; /* unused swap entry */
2666
2667 p->swap_map[offset] = count | has_cache;
2668
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002669unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002670 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002672 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673
2674bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002675 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 goto out;
2677}
Hugh Dickins253d5532009-12-14 17:58:44 -08002678
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002679/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002680 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2681 * (in which case its reference count is never incremented).
2682 */
2683void swap_shmem_alloc(swp_entry_t entry)
2684{
2685 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2686}
2687
2688/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002689 * Increase reference count of swap entry by 1.
2690 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2691 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2692 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2693 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002694 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002695int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002696{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002697 int err = 0;
2698
2699 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2700 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2701 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002702}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002704/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002705 * @entry: swap entry for which we allocate swap cache.
2706 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002707 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002708 * This can return error codes. Returns 0 at success.
2709 * -EBUSY means there is a swap cache.
2710 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002711 */
2712int swapcache_prepare(swp_entry_t entry)
2713{
Hugh Dickins253d5532009-12-14 17:58:44 -08002714 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002715}
2716
Mel Gormanf981c592012-07-31 16:44:47 -07002717struct swap_info_struct *page_swap_info(struct page *page)
2718{
2719 swp_entry_t swap = { .val = page_private(page) };
Mel Gormanf981c592012-07-31 16:44:47 -07002720 return swap_info[swp_type(swap)];
2721}
2722
2723/*
2724 * out-of-line __page_file_ methods to avoid include hell.
2725 */
2726struct address_space *__page_file_mapping(struct page *page)
2727{
Sasha Levin309381fea2014-01-23 15:52:54 -08002728 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002729 return page_swap_info(page)->swap_file->f_mapping;
2730}
2731EXPORT_SYMBOL_GPL(__page_file_mapping);
2732
2733pgoff_t __page_file_index(struct page *page)
2734{
2735 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002736 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002737 return swp_offset(swap);
2738}
2739EXPORT_SYMBOL_GPL(__page_file_index);
2740
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002742 * add_swap_count_continuation - called when a swap count is duplicated
2743 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2744 * page of the original vmalloc'ed swap_map, to hold the continuation count
2745 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2746 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2747 *
2748 * These continuation pages are seldom referenced: the common paths all work
2749 * on the original swap_map, only referring to a continuation page when the
2750 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2751 *
2752 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2753 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2754 * can be called after dropping locks.
2755 */
2756int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2757{
2758 struct swap_info_struct *si;
2759 struct page *head;
2760 struct page *page;
2761 struct page *list_page;
2762 pgoff_t offset;
2763 unsigned char count;
2764
2765 /*
2766 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2767 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2768 */
2769 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2770
2771 si = swap_info_get(entry);
2772 if (!si) {
2773 /*
2774 * An acceptable race has occurred since the failing
2775 * __swap_duplicate(): the swap entry has been freed,
2776 * perhaps even the whole swap_map cleared for swapoff.
2777 */
2778 goto outer;
2779 }
2780
2781 offset = swp_offset(entry);
2782 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2783
2784 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2785 /*
2786 * The higher the swap count, the more likely it is that tasks
2787 * will race to add swap count continuation: we need to avoid
2788 * over-provisioning.
2789 */
2790 goto out;
2791 }
2792
2793 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002794 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002795 return -ENOMEM;
2796 }
2797
2798 /*
2799 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002800 * no architecture is using highmem pages for kernel page tables: so it
2801 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002802 */
2803 head = vmalloc_to_page(si->swap_map + offset);
2804 offset &= ~PAGE_MASK;
2805
2806 /*
2807 * Page allocation does not initialize the page's lru field,
2808 * but it does always reset its private field.
2809 */
2810 if (!page_private(head)) {
2811 BUG_ON(count & COUNT_CONTINUED);
2812 INIT_LIST_HEAD(&head->lru);
2813 set_page_private(head, SWP_CONTINUED);
2814 si->flags |= SWP_CONTINUED;
2815 }
2816
2817 list_for_each_entry(list_page, &head->lru, lru) {
2818 unsigned char *map;
2819
2820 /*
2821 * If the previous map said no continuation, but we've found
2822 * a continuation page, free our allocation and use this one.
2823 */
2824 if (!(count & COUNT_CONTINUED))
2825 goto out;
2826
Cong Wang9b04c5f2011-11-25 23:14:39 +08002827 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002828 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002829 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002830
2831 /*
2832 * If this continuation count now has some space in it,
2833 * free our allocation and use this one.
2834 */
2835 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2836 goto out;
2837 }
2838
2839 list_add_tail(&page->lru, &head->lru);
2840 page = NULL; /* now it's attached, don't free it */
2841out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002842 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002843outer:
2844 if (page)
2845 __free_page(page);
2846 return 0;
2847}
2848
2849/*
2850 * swap_count_continued - when the original swap_map count is incremented
2851 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2852 * into, carry if so, or else fail until a new continuation page is allocated;
2853 * when the original swap_map count is decremented from 0 with continuation,
2854 * borrow from the continuation and report whether it still holds more.
2855 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2856 */
2857static bool swap_count_continued(struct swap_info_struct *si,
2858 pgoff_t offset, unsigned char count)
2859{
2860 struct page *head;
2861 struct page *page;
2862 unsigned char *map;
2863
2864 head = vmalloc_to_page(si->swap_map + offset);
2865 if (page_private(head) != SWP_CONTINUED) {
2866 BUG_ON(count & COUNT_CONTINUED);
2867 return false; /* need to add count continuation */
2868 }
2869
2870 offset &= ~PAGE_MASK;
2871 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002872 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002873
2874 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2875 goto init_map; /* jump over SWAP_CONT_MAX checks */
2876
2877 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2878 /*
2879 * Think of how you add 1 to 999
2880 */
2881 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002882 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002883 page = list_entry(page->lru.next, struct page, lru);
2884 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002885 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002886 }
2887 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002888 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002889 page = list_entry(page->lru.next, struct page, lru);
2890 if (page == head)
2891 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002892 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002893init_map: *map = 0; /* we didn't zero the page */
2894 }
2895 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002896 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002897 page = list_entry(page->lru.prev, struct page, lru);
2898 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002899 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002900 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002901 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002902 page = list_entry(page->lru.prev, struct page, lru);
2903 }
2904 return true; /* incremented */
2905
2906 } else { /* decrementing */
2907 /*
2908 * Think of how you subtract 1 from 1000
2909 */
2910 BUG_ON(count != COUNT_CONTINUED);
2911 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002912 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002913 page = list_entry(page->lru.next, struct page, lru);
2914 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002915 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002916 }
2917 BUG_ON(*map == 0);
2918 *map -= 1;
2919 if (*map == 0)
2920 count = 0;
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 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002924 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002925 *map = SWAP_CONT_MAX | count;
2926 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002927 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002928 page = list_entry(page->lru.prev, struct page, lru);
2929 }
2930 return count == COUNT_CONTINUED;
2931 }
2932}
2933
2934/*
2935 * free_swap_count_continuations - swapoff free all the continuation pages
2936 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2937 */
2938static void free_swap_count_continuations(struct swap_info_struct *si)
2939{
2940 pgoff_t offset;
2941
2942 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2943 struct page *head;
2944 head = vmalloc_to_page(si->swap_map + offset);
2945 if (page_private(head)) {
Geliang Tang0d576d22016-01-14 15:21:49 -08002946 struct page *page, *next;
2947
2948 list_for_each_entry_safe(page, next, &head->lru, lru) {
2949 list_del(&page->lru);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002950 __free_page(page);
2951 }
2952 }
2953 }
2954}