blob: 178130880b908515a105eccf9fa428f7cf61719a [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>
Ingo Molnar6e84f312017-02-08 18:51:29 +01009#include <linux/sched/mm.h>
Ingo Molnar29930022017-02-08 18:51:36 +010010#include <linux/sched/task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/hugetlb.h>
12#include <linux/mman.h>
13#include <linux/slab.h>
14#include <linux/kernel_stat.h>
15#include <linux/swap.h>
16#include <linux/vmalloc.h>
17#include <linux/pagemap.h>
18#include <linux/namei.h>
Hugh Dickins072441e2011-06-27 16:18:02 -070019#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080021#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/writeback.h>
23#include <linux/proc_fs.h>
24#include <linux/seq_file.h>
25#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080026#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/rmap.h>
28#include <linux/security.h>
29#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080030#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080031#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070034#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070035#include <linux/oom.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060036#include <linux/frontswap.h>
37#include <linux/swapfile.h>
Mel Gormanf981c592012-07-31 16:44:47 -070038#include <linux/export.h>
Tim Chen67afa382017-02-22 15:45:39 -080039#include <linux/swap_slots.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/pgtable.h>
42#include <asm/tlbflush.h>
43#include <linux/swapops.h>
Johannes Weiner5d1ea48b2014-12-10 15:44:55 -080044#include <linux/swap_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Hugh Dickins570a335b2009-12-14 17:58:46 -080046static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
47 unsigned char);
48static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080049static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080050
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060051DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070052static unsigned int nr_swapfiles;
Shaohua Liec8acf22013-02-22 16:34:38 -080053atomic_long_t nr_swap_pages;
Chris Wilsonfb0fec52015-12-04 15:58:53 +000054/*
55 * Some modules use swappable objects and may try to swap them out under
56 * memory pressure (via the shrinker). Before doing so, they may wish to
57 * check to see if any swap space is available.
58 */
59EXPORT_SYMBOL_GPL(nr_swap_pages);
Shaohua Liec8acf22013-02-22 16:34:38 -080060/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070061long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070062static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064static const char Bad_file[] = "Bad swap file entry ";
65static const char Unused_file[] = "Unused swap file entry ";
66static const char Bad_offset[] = "Bad swap offset entry ";
67static const char Unused_offset[] = "Unused swap offset entry ";
68
Dan Streetmanadfab832014-06-04 16:09:53 -070069/*
70 * all active swap_info_structs
71 * protected with swap_lock, and ordered by priority.
72 */
Dan Streetman18ab4d42014-06-04 16:09:59 -070073PLIST_HEAD(swap_active_head);
74
75/*
76 * all available (active, not full) swap_info_structs
77 * protected with swap_avail_lock, ordered by priority.
78 * This is used by get_swap_page() instead of swap_active_head
79 * because swap_active_head includes all swap_info_structs,
80 * but get_swap_page() doesn't need to look at full ones.
81 * This uses its own lock instead of swap_lock because when a
82 * swap_info_struct changes between not-full/full, it needs to
83 * add/remove itself to/from this list, but the swap_info_struct->lock
84 * is held and the locking order requires swap_lock to be taken
85 * before any swap_info_struct->lock.
86 */
87static PLIST_HEAD(swap_avail_head);
88static DEFINE_SPINLOCK(swap_avail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060090struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Ingo Molnarfc0abb12006-01-18 17:42:33 -080092static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Kay Sievers66d7dd52010-10-26 14:22:06 -070094static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
95/* Activity counter to indicate that a swapon or swapoff has occurred */
96static atomic_t proc_poll_event = ATOMIC_INIT(0);
97
Hugh Dickins8d69aae2009-12-14 17:58:45 -080098static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070099{
Hugh Dickins570a335b2009-12-14 17:58:46 -0800100 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700101}
102
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800103/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700104static int
105__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
106{
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800107 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700108 struct page *page;
109 int ret = 0;
110
Huang Yingf6ab1f72016-10-07 17:00:21 -0700111 page = find_get_page(swap_address_space(entry), swp_offset(entry));
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700112 if (!page)
113 return 0;
114 /*
115 * This function is called from scan_swap_map() and it's called
116 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
117 * We have to use trylock for avoiding deadlock. This is a special
118 * case and you should use try_to_free_swap() with explicit lock_page()
119 * in usual operations.
120 */
121 if (trylock_page(page)) {
122 ret = try_to_free_swap(page);
123 unlock_page(page);
124 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300125 put_page(page);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700126 return ret;
127}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800130 * swapon tell device that all the old swap contents can be discarded,
131 * to allow the swap device to optimize its wear-levelling.
132 */
133static int discard_swap(struct swap_info_struct *si)
134{
135 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800136 sector_t start_block;
137 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800138 int err = 0;
139
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800140 /* Do not discard the swap header page! */
141 se = &si->first_swap_extent;
142 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
143 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
144 if (nr_blocks) {
145 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200146 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800147 if (err)
148 return err;
149 cond_resched();
150 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800151
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800152 list_for_each_entry(se, &si->first_swap_extent.list, list) {
153 start_block = se->start_block << (PAGE_SHIFT - 9);
154 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800155
156 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200157 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800158 if (err)
159 break;
160
161 cond_resched();
162 }
163 return err; /* That will often be -EOPNOTSUPP */
164}
165
Hugh Dickins7992fde2009-01-06 14:39:53 -0800166/*
167 * swap allocation tell device that a cluster of swap can now be discarded,
168 * to allow the swap device to optimize its wear-levelling.
169 */
170static void discard_swap_cluster(struct swap_info_struct *si,
171 pgoff_t start_page, pgoff_t nr_pages)
172{
173 struct swap_extent *se = si->curr_swap_extent;
174 int found_extent = 0;
175
176 while (nr_pages) {
Hugh Dickins7992fde2009-01-06 14:39:53 -0800177 if (se->start_page <= start_page &&
178 start_page < se->start_page + se->nr_pages) {
179 pgoff_t offset = start_page - se->start_page;
180 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800181 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800182
183 if (nr_blocks > nr_pages)
184 nr_blocks = nr_pages;
185 start_page += nr_blocks;
186 nr_pages -= nr_blocks;
187
188 if (!found_extent++)
189 si->curr_swap_extent = se;
190
191 start_block <<= PAGE_SHIFT - 9;
192 nr_blocks <<= PAGE_SHIFT - 9;
193 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200194 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800195 break;
196 }
197
Geliang Tanga8ae4992016-01-14 15:20:45 -0800198 se = list_next_entry(se, list);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800199 }
200}
201
Hugh Dickins048c27f2005-09-03 15:54:40 -0700202#define SWAPFILE_CLUSTER 256
203#define LATENCY_LIMIT 256
204
Shaohua Li2a8f9442013-09-11 14:20:28 -0700205static inline void cluster_set_flag(struct swap_cluster_info *info,
206 unsigned int flag)
207{
208 info->flags = flag;
209}
210
211static inline unsigned int cluster_count(struct swap_cluster_info *info)
212{
213 return info->data;
214}
215
216static inline void cluster_set_count(struct swap_cluster_info *info,
217 unsigned int c)
218{
219 info->data = c;
220}
221
222static inline void cluster_set_count_flag(struct swap_cluster_info *info,
223 unsigned int c, unsigned int f)
224{
225 info->flags = f;
226 info->data = c;
227}
228
229static inline unsigned int cluster_next(struct swap_cluster_info *info)
230{
231 return info->data;
232}
233
234static inline void cluster_set_next(struct swap_cluster_info *info,
235 unsigned int n)
236{
237 info->data = n;
238}
239
240static inline void cluster_set_next_flag(struct swap_cluster_info *info,
241 unsigned int n, unsigned int f)
242{
243 info->flags = f;
244 info->data = n;
245}
246
247static inline bool cluster_is_free(struct swap_cluster_info *info)
248{
249 return info->flags & CLUSTER_FLAG_FREE;
250}
251
252static inline bool cluster_is_null(struct swap_cluster_info *info)
253{
254 return info->flags & CLUSTER_FLAG_NEXT_NULL;
255}
256
257static inline void cluster_set_null(struct swap_cluster_info *info)
258{
259 info->flags = CLUSTER_FLAG_NEXT_NULL;
260 info->data = 0;
261}
262
Huang, Ying235b6212017-02-22 15:45:22 -0800263static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
264 unsigned long offset)
265{
266 struct swap_cluster_info *ci;
267
268 ci = si->cluster_info;
269 if (ci) {
270 ci += offset / SWAPFILE_CLUSTER;
271 spin_lock(&ci->lock);
272 }
273 return ci;
274}
275
276static inline void unlock_cluster(struct swap_cluster_info *ci)
277{
278 if (ci)
279 spin_unlock(&ci->lock);
280}
281
282static inline struct swap_cluster_info *lock_cluster_or_swap_info(
283 struct swap_info_struct *si,
284 unsigned long offset)
285{
286 struct swap_cluster_info *ci;
287
288 ci = lock_cluster(si, offset);
289 if (!ci)
290 spin_lock(&si->lock);
291
292 return ci;
293}
294
295static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
296 struct swap_cluster_info *ci)
297{
298 if (ci)
299 unlock_cluster(ci);
300 else
301 spin_unlock(&si->lock);
302}
303
Huang Ying6b534912016-10-07 16:58:42 -0700304static inline bool cluster_list_empty(struct swap_cluster_list *list)
305{
306 return cluster_is_null(&list->head);
307}
308
309static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
310{
311 return cluster_next(&list->head);
312}
313
314static void cluster_list_init(struct swap_cluster_list *list)
315{
316 cluster_set_null(&list->head);
317 cluster_set_null(&list->tail);
318}
319
320static void cluster_list_add_tail(struct swap_cluster_list *list,
321 struct swap_cluster_info *ci,
322 unsigned int idx)
323{
324 if (cluster_list_empty(list)) {
325 cluster_set_next_flag(&list->head, idx, 0);
326 cluster_set_next_flag(&list->tail, idx, 0);
327 } else {
Huang, Ying235b6212017-02-22 15:45:22 -0800328 struct swap_cluster_info *ci_tail;
Huang Ying6b534912016-10-07 16:58:42 -0700329 unsigned int tail = cluster_next(&list->tail);
330
Huang, Ying235b6212017-02-22 15:45:22 -0800331 /*
332 * Nested cluster lock, but both cluster locks are
333 * only acquired when we held swap_info_struct->lock
334 */
335 ci_tail = ci + tail;
336 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
337 cluster_set_next(ci_tail, idx);
338 unlock_cluster(ci_tail);
Huang Ying6b534912016-10-07 16:58:42 -0700339 cluster_set_next_flag(&list->tail, idx, 0);
340 }
341}
342
343static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
344 struct swap_cluster_info *ci)
345{
346 unsigned int idx;
347
348 idx = cluster_next(&list->head);
349 if (cluster_next(&list->tail) == idx) {
350 cluster_set_null(&list->head);
351 cluster_set_null(&list->tail);
352 } else
353 cluster_set_next_flag(&list->head,
354 cluster_next(&ci[idx]), 0);
355
356 return idx;
357}
358
Shaohua Li815c2c52013-09-11 14:20:30 -0700359/* Add a cluster to discard list and schedule it to do discard */
360static void swap_cluster_schedule_discard(struct swap_info_struct *si,
361 unsigned int idx)
362{
363 /*
364 * If scan_swap_map() can't find a free cluster, it will check
365 * si->swap_map directly. To make sure the discarding cluster isn't
366 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
367 * will be cleared after discard
368 */
369 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
370 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
371
Huang Ying6b534912016-10-07 16:58:42 -0700372 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
Shaohua Li815c2c52013-09-11 14:20:30 -0700373
374 schedule_work(&si->discard_work);
375}
376
377/*
378 * Doing discard actually. After a cluster discard is finished, the cluster
379 * will be added to free cluster list. caller should hold si->lock.
380*/
381static void swap_do_scheduled_discard(struct swap_info_struct *si)
382{
Huang, Ying235b6212017-02-22 15:45:22 -0800383 struct swap_cluster_info *info, *ci;
Shaohua Li815c2c52013-09-11 14:20:30 -0700384 unsigned int idx;
385
386 info = si->cluster_info;
387
Huang Ying6b534912016-10-07 16:58:42 -0700388 while (!cluster_list_empty(&si->discard_clusters)) {
389 idx = cluster_list_del_first(&si->discard_clusters, info);
Shaohua Li815c2c52013-09-11 14:20:30 -0700390 spin_unlock(&si->lock);
391
392 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
393 SWAPFILE_CLUSTER);
394
395 spin_lock(&si->lock);
Huang, Ying235b6212017-02-22 15:45:22 -0800396 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
397 cluster_set_flag(ci, CLUSTER_FLAG_FREE);
398 unlock_cluster(ci);
Huang Ying6b534912016-10-07 16:58:42 -0700399 cluster_list_add_tail(&si->free_clusters, info, idx);
Huang, Ying235b6212017-02-22 15:45:22 -0800400 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
Shaohua Li815c2c52013-09-11 14:20:30 -0700401 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
402 0, SWAPFILE_CLUSTER);
Huang, Ying235b6212017-02-22 15:45:22 -0800403 unlock_cluster(ci);
Shaohua Li815c2c52013-09-11 14:20:30 -0700404 }
405}
406
407static void swap_discard_work(struct work_struct *work)
408{
409 struct swap_info_struct *si;
410
411 si = container_of(work, struct swap_info_struct, discard_work);
412
413 spin_lock(&si->lock);
414 swap_do_scheduled_discard(si);
415 spin_unlock(&si->lock);
416}
417
Shaohua Li2a8f9442013-09-11 14:20:28 -0700418/*
419 * The cluster corresponding to page_nr will be used. The cluster will be
420 * removed from free cluster list and its usage counter will be increased.
421 */
422static void inc_cluster_info_page(struct swap_info_struct *p,
423 struct swap_cluster_info *cluster_info, unsigned long page_nr)
424{
425 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
426
427 if (!cluster_info)
428 return;
429 if (cluster_is_free(&cluster_info[idx])) {
Huang Ying6b534912016-10-07 16:58:42 -0700430 VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx);
431 cluster_list_del_first(&p->free_clusters, cluster_info);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700432 cluster_set_count_flag(&cluster_info[idx], 0, 0);
433 }
434
435 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
436 cluster_set_count(&cluster_info[idx],
437 cluster_count(&cluster_info[idx]) + 1);
438}
439
440/*
441 * The cluster corresponding to page_nr decreases one usage. If the usage
442 * counter becomes 0, which means no page in the cluster is in using, we can
443 * optionally discard the cluster and add it to free cluster list.
444 */
445static void dec_cluster_info_page(struct swap_info_struct *p,
446 struct swap_cluster_info *cluster_info, unsigned long page_nr)
447{
448 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
449
450 if (!cluster_info)
451 return;
452
453 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
454 cluster_set_count(&cluster_info[idx],
455 cluster_count(&cluster_info[idx]) - 1);
456
457 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700458 /*
459 * If the swap is discardable, prepare discard the cluster
460 * instead of free it immediately. The cluster will be freed
461 * after discard.
462 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700463 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
464 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700465 swap_cluster_schedule_discard(p, idx);
466 return;
467 }
468
Shaohua Li2a8f9442013-09-11 14:20:28 -0700469 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -0700470 cluster_list_add_tail(&p->free_clusters, cluster_info, idx);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700471 }
472}
473
474/*
475 * It's possible scan_swap_map() uses a free cluster in the middle of free
476 * cluster list. Avoiding such abuse to avoid list corruption.
477 */
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700478static bool
479scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700480 unsigned long offset)
481{
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700482 struct percpu_cluster *percpu_cluster;
483 bool conflict;
484
Shaohua Li2a8f9442013-09-11 14:20:28 -0700485 offset /= SWAPFILE_CLUSTER;
Huang Ying6b534912016-10-07 16:58:42 -0700486 conflict = !cluster_list_empty(&si->free_clusters) &&
487 offset != cluster_list_first(&si->free_clusters) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700488 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700489
490 if (!conflict)
491 return false;
492
493 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
494 cluster_set_null(&percpu_cluster->index);
495 return true;
496}
497
498/*
499 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
500 * might involve allocating a new cluster for current CPU too.
501 */
Tim Chen36005ba2017-02-22 15:45:33 -0800502static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700503 unsigned long *offset, unsigned long *scan_base)
504{
505 struct percpu_cluster *cluster;
Huang, Ying235b6212017-02-22 15:45:22 -0800506 struct swap_cluster_info *ci;
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700507 bool found_free;
Huang, Ying235b6212017-02-22 15:45:22 -0800508 unsigned long tmp, max;
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700509
510new_cluster:
511 cluster = this_cpu_ptr(si->percpu_cluster);
512 if (cluster_is_null(&cluster->index)) {
Huang Ying6b534912016-10-07 16:58:42 -0700513 if (!cluster_list_empty(&si->free_clusters)) {
514 cluster->index = si->free_clusters.head;
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700515 cluster->next = cluster_next(&cluster->index) *
516 SWAPFILE_CLUSTER;
Huang Ying6b534912016-10-07 16:58:42 -0700517 } else if (!cluster_list_empty(&si->discard_clusters)) {
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700518 /*
519 * we don't have free cluster but have some clusters in
520 * discarding, do discard now and reclaim them
521 */
522 swap_do_scheduled_discard(si);
523 *scan_base = *offset = si->cluster_next;
524 goto new_cluster;
525 } else
Tim Chen36005ba2017-02-22 15:45:33 -0800526 return false;
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700527 }
528
529 found_free = false;
530
531 /*
532 * Other CPUs can use our cluster if they can't find a free cluster,
533 * check if there is still free entry in the cluster
534 */
535 tmp = cluster->next;
Huang, Ying235b6212017-02-22 15:45:22 -0800536 max = min_t(unsigned long, si->max,
537 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
538 if (tmp >= max) {
539 cluster_set_null(&cluster->index);
540 goto new_cluster;
541 }
542 ci = lock_cluster(si, tmp);
543 while (tmp < max) {
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700544 if (!si->swap_map[tmp]) {
545 found_free = true;
546 break;
547 }
548 tmp++;
549 }
Huang, Ying235b6212017-02-22 15:45:22 -0800550 unlock_cluster(ci);
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700551 if (!found_free) {
552 cluster_set_null(&cluster->index);
553 goto new_cluster;
554 }
555 cluster->next = tmp + 1;
556 *offset = tmp;
557 *scan_base = tmp;
Tim Chen36005ba2017-02-22 15:45:33 -0800558 return found_free;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700559}
560
Tim Chen36005ba2017-02-22 15:45:33 -0800561static int scan_swap_map_slots(struct swap_info_struct *si,
562 unsigned char usage, int nr,
563 swp_entry_t slots[])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
Huang, Ying235b6212017-02-22 15:45:22 -0800565 struct swap_cluster_info *ci;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800566 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800567 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800568 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700569 int latency_ration = LATENCY_LIMIT;
Tim Chen36005ba2017-02-22 15:45:33 -0800570 int n_ret = 0;
571
572 if (nr > SWAP_BATCH)
573 nr = SWAP_BATCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800575 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700576 * We try to cluster swap pages by allocating them sequentially
577 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
578 * way, however, we resort to first-free allocation, starting
579 * a new cluster. This prevents us from scattering swap pages
580 * all over the entire swap partition, so that we reduce
581 * overall disk seek times between swap pages. -- sct
582 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800583 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700584 */
585
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700586 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800587 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800588
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700589 /* SSD algorithm */
590 if (si->cluster_info) {
Tim Chen36005ba2017-02-22 15:45:33 -0800591 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
592 goto checks;
593 else
594 goto scan;
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700595 }
596
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800597 if (unlikely(!si->cluster_nr--)) {
598 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
599 si->cluster_nr = SWAPFILE_CLUSTER - 1;
600 goto checks;
601 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700602
Shaohua Liec8acf22013-02-22 16:34:38 -0800603 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700604
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800605 /*
606 * If seek is expensive, start searching for new cluster from
607 * start of partition, to minimize the span of allocated swap.
Chen Yucong50088c42014-06-04 16:10:57 -0700608 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
609 * case, just handled by scan_swap_map_try_ssd_cluster() above.
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800610 */
Chen Yucong50088c42014-06-04 16:10:57 -0700611 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700612 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
613
614 /* Locate the first empty (unaligned) cluster */
615 for (; last_in_cluster <= si->highest_bit; offset++) {
616 if (si->swap_map[offset])
617 last_in_cluster = offset + SWAPFILE_CLUSTER;
618 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800619 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800620 offset -= SWAPFILE_CLUSTER - 1;
621 si->cluster_next = offset;
622 si->cluster_nr = SWAPFILE_CLUSTER - 1;
623 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700625 if (unlikely(--latency_ration < 0)) {
626 cond_resched();
627 latency_ration = LATENCY_LIMIT;
628 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700629 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800630
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800631 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800632 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800633 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700635
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800636checks:
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700637 if (si->cluster_info) {
Tim Chen36005ba2017-02-22 15:45:33 -0800638 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
639 /* take a break if we already got some slots */
640 if (n_ret)
641 goto done;
642 if (!scan_swap_map_try_ssd_cluster(si, &offset,
643 &scan_base))
644 goto scan;
645 }
Shaohua Liebc2a1a62013-09-11 14:20:32 -0700646 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800647 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700648 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700649 if (!si->highest_bit)
650 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800651 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800652 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700653
Huang, Ying235b6212017-02-22 15:45:22 -0800654 ci = lock_cluster(si, offset);
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700655 /* reuse swap entry of cache-only swap if not busy. */
656 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700657 int swap_was_freed;
Huang, Ying235b6212017-02-22 15:45:22 -0800658 unlock_cluster(ci);
Shaohua Liec8acf22013-02-22 16:34:38 -0800659 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700660 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800661 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700662 /* entry was freed successfully, try to use this again */
663 if (swap_was_freed)
664 goto checks;
665 goto scan; /* check next one */
666 }
667
Huang, Ying235b6212017-02-22 15:45:22 -0800668 if (si->swap_map[offset]) {
669 unlock_cluster(ci);
Tim Chen36005ba2017-02-22 15:45:33 -0800670 if (!n_ret)
671 goto scan;
672 else
673 goto done;
Huang, Ying235b6212017-02-22 15:45:22 -0800674 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700675
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800676 if (offset == si->lowest_bit)
677 si->lowest_bit++;
678 if (offset == si->highest_bit)
679 si->highest_bit--;
680 si->inuse_pages++;
681 if (si->inuse_pages == si->pages) {
682 si->lowest_bit = si->max;
683 si->highest_bit = 0;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700684 spin_lock(&swap_avail_lock);
685 plist_del(&si->avail_list, &swap_avail_head);
686 spin_unlock(&swap_avail_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800687 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800688 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700689 inc_cluster_info_page(si, si->cluster_info, offset);
Huang, Ying235b6212017-02-22 15:45:22 -0800690 unlock_cluster(ci);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800691 si->cluster_next = offset + 1;
Tim Chen36005ba2017-02-22 15:45:33 -0800692 slots[n_ret++] = swp_entry(si->type, offset);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800693
Tim Chen36005ba2017-02-22 15:45:33 -0800694 /* got enough slots or reach max slots? */
695 if ((n_ret == nr) || (offset >= si->highest_bit))
696 goto done;
697
698 /* search for next available slot */
699
700 /* time to take a break? */
701 if (unlikely(--latency_ration < 0)) {
702 if (n_ret)
703 goto done;
704 spin_unlock(&si->lock);
705 cond_resched();
706 spin_lock(&si->lock);
707 latency_ration = LATENCY_LIMIT;
708 }
709
710 /* try to get more slots in cluster */
711 if (si->cluster_info) {
712 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
713 goto checks;
714 else
715 goto done;
716 }
717 /* non-ssd case */
718 ++offset;
719
720 /* non-ssd case, still more slots in cluster? */
721 if (si->cluster_nr && !si->swap_map[offset]) {
722 --si->cluster_nr;
723 goto checks;
724 }
725
726done:
727 si->flags -= SWP_SCANNING;
728 return n_ret;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800729
730scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800731 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700732 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700733 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800734 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700735 goto checks;
736 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700737 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800738 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700739 goto checks;
740 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700741 if (unlikely(--latency_ration < 0)) {
742 cond_resched();
743 latency_ration = LATENCY_LIMIT;
744 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700745 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800746 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800747 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800748 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800749 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800750 goto checks;
751 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700752 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800753 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700754 goto checks;
755 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800756 if (unlikely(--latency_ration < 0)) {
757 cond_resched();
758 latency_ration = LATENCY_LIMIT;
759 }
Jamie Liua5998062014-01-23 15:53:40 -0800760 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800761 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800762 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700763
764no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700765 si->flags -= SWP_SCANNING;
Tim Chen36005ba2017-02-22 15:45:33 -0800766 return n_ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767}
768
Tim Chen36005ba2017-02-22 15:45:33 -0800769static unsigned long scan_swap_map(struct swap_info_struct *si,
770 unsigned char usage)
771{
772 swp_entry_t entry;
773 int n_ret;
774
775 n_ret = scan_swap_map_slots(si, usage, 1, &entry);
776
777 if (n_ret)
778 return swp_offset(entry);
779 else
780 return 0;
781
782}
783
784int get_swap_pages(int n_goal, swp_entry_t swp_entries[])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Dan Streetmanadfab832014-06-04 16:09:53 -0700786 struct swap_info_struct *si, *next;
Tim Chen36005ba2017-02-22 15:45:33 -0800787 long avail_pgs;
788 int n_ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Tim Chen36005ba2017-02-22 15:45:33 -0800790 avail_pgs = atomic_long_read(&nr_swap_pages);
791 if (avail_pgs <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700792 goto noswap;
Tim Chen36005ba2017-02-22 15:45:33 -0800793
794 if (n_goal > SWAP_BATCH)
795 n_goal = SWAP_BATCH;
796
797 if (n_goal > avail_pgs)
798 n_goal = avail_pgs;
799
800 atomic_long_sub(n_goal, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Dan Streetman18ab4d42014-06-04 16:09:59 -0700802 spin_lock(&swap_avail_lock);
803
804start_over:
805 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
806 /* requeue si to after same-priority siblings */
807 plist_requeue(&si->avail_list, &swap_avail_head);
808 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800809 spin_lock(&si->lock);
Dan Streetmanadfab832014-06-04 16:09:53 -0700810 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
Dan Streetman18ab4d42014-06-04 16:09:59 -0700811 spin_lock(&swap_avail_lock);
812 if (plist_node_empty(&si->avail_list)) {
813 spin_unlock(&si->lock);
814 goto nextsi;
815 }
816 WARN(!si->highest_bit,
817 "swap_info %d in list but !highest_bit\n",
818 si->type);
819 WARN(!(si->flags & SWP_WRITEOK),
820 "swap_info %d in list but !SWP_WRITEOK\n",
821 si->type);
822 plist_del(&si->avail_list, &swap_avail_head);
Shaohua Liec8acf22013-02-22 16:34:38 -0800823 spin_unlock(&si->lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700824 goto nextsi;
Shaohua Liec8acf22013-02-22 16:34:38 -0800825 }
Tim Chen36005ba2017-02-22 15:45:33 -0800826 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
827 n_goal, swp_entries);
Shaohua Liec8acf22013-02-22 16:34:38 -0800828 spin_unlock(&si->lock);
Tim Chen36005ba2017-02-22 15:45:33 -0800829 if (n_ret)
830 goto check_out;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700831 pr_debug("scan_swap_map of si %d failed to find offset\n",
Tim Chen36005ba2017-02-22 15:45:33 -0800832 si->type);
833
Dan Streetman18ab4d42014-06-04 16:09:59 -0700834 spin_lock(&swap_avail_lock);
835nextsi:
Dan Streetmanadfab832014-06-04 16:09:53 -0700836 /*
837 * if we got here, it's likely that si was almost full before,
838 * and since scan_swap_map() can drop the si->lock, multiple
839 * callers probably all tried to get a page from the same si
Dan Streetman18ab4d42014-06-04 16:09:59 -0700840 * and it filled up before we could get one; or, the si filled
841 * up between us dropping swap_avail_lock and taking si->lock.
842 * Since we dropped the swap_avail_lock, the swap_avail_head
843 * list may have been modified; so if next is still in the
Tim Chen36005ba2017-02-22 15:45:33 -0800844 * swap_avail_head list then try it, otherwise start over
845 * if we have not gotten any slots.
Dan Streetmanadfab832014-06-04 16:09:53 -0700846 */
Dan Streetman18ab4d42014-06-04 16:09:59 -0700847 if (plist_node_empty(&next->avail_list))
848 goto start_over;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700850
Dan Streetman18ab4d42014-06-04 16:09:59 -0700851 spin_unlock(&swap_avail_lock);
852
Tim Chen36005ba2017-02-22 15:45:33 -0800853check_out:
854 if (n_ret < n_goal)
855 atomic_long_add((long) (n_goal-n_ret), &nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700856noswap:
Tim Chen36005ba2017-02-22 15:45:33 -0800857 return n_ret;
858}
859
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800860/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700861swp_entry_t get_swap_page_of_type(int type)
862{
863 struct swap_info_struct *si;
864 pgoff_t offset;
865
Hugh Dickins910321e2010-09-09 16:38:07 -0700866 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800867 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700868 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800869 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700870 /* This is called for allocating swap entry, not cache */
871 offset = scan_swap_map(si, 1);
872 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800873 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700874 return swp_entry(type, offset);
875 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800876 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700877 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800878 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700879 return (swp_entry_t) {0};
880}
881
Tim Chene8c26ab2017-02-22 15:45:29 -0800882static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800884 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 unsigned long offset, type;
886
887 if (!entry.val)
888 goto out;
889 type = swp_type(entry);
890 if (type >= nr_swapfiles)
891 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800892 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (!(p->flags & SWP_USED))
894 goto bad_device;
895 offset = swp_offset(entry);
896 if (offset >= p->max)
897 goto bad_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 return p;
899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900bad_offset:
Huang, Ying6a991fc2017-02-22 15:45:19 -0800901 pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 goto out;
903bad_device:
Huang, Ying6a991fc2017-02-22 15:45:19 -0800904 pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 goto out;
906bad_nofile:
Huang, Ying6a991fc2017-02-22 15:45:19 -0800907 pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908out:
909 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800910}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Tim Chene8c26ab2017-02-22 15:45:29 -0800912static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
913{
914 struct swap_info_struct *p;
915
916 p = __swap_info_get(entry);
917 if (!p)
918 goto out;
919 if (!p->swap_map[swp_offset(entry)])
920 goto bad_free;
921 return p;
922
923bad_free:
924 pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
925 goto out;
926out:
927 return NULL;
928}
929
Huang, Ying235b6212017-02-22 15:45:22 -0800930static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Huang, Ying235b6212017-02-22 15:45:22 -0800932 struct swap_info_struct *p;
933
934 p = _swap_info_get(entry);
935 if (p)
936 spin_lock(&p->lock);
937 return p;
938}
939
Tim Chen7c00baf2017-02-22 15:45:36 -0800940static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
941 struct swap_info_struct *q)
942{
943 struct swap_info_struct *p;
944
945 p = _swap_info_get(entry);
946
947 if (p != q) {
948 if (q != NULL)
949 spin_unlock(&q->lock);
950 if (p != NULL)
951 spin_lock(&p->lock);
952 }
953 return p;
954}
955
956static unsigned char __swap_entry_free(struct swap_info_struct *p,
957 swp_entry_t entry, unsigned char usage)
Huang, Ying235b6212017-02-22 15:45:22 -0800958{
959 struct swap_cluster_info *ci;
Hugh Dickins253d5532009-12-14 17:58:44 -0800960 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800961 unsigned char count;
962 unsigned char has_cache;
Huang, Ying235b6212017-02-22 15:45:22 -0800963
Tim Chen7c00baf2017-02-22 15:45:36 -0800964 ci = lock_cluster_or_swap_info(p, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700966 count = p->swap_map[offset];
Huang, Ying235b6212017-02-22 15:45:22 -0800967
Hugh Dickins253d5532009-12-14 17:58:44 -0800968 has_cache = count & SWAP_HAS_CACHE;
969 count &= ~SWAP_HAS_CACHE;
970
971 if (usage == SWAP_HAS_CACHE) {
972 VM_BUG_ON(!has_cache);
973 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800974 } else if (count == SWAP_MAP_SHMEM) {
975 /*
976 * Or we could insist on shmem.c using a special
977 * swap_shmem_free() and free_shmem_swap_and_cache()...
978 */
979 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800980 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
981 if (count == COUNT_CONTINUED) {
982 if (swap_count_continued(p, offset, count))
983 count = SWAP_MAP_MAX | COUNT_CONTINUED;
984 else
985 count = SWAP_MAP_MAX;
986 } else
987 count--;
988 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800989
Hugh Dickins253d5532009-12-14 17:58:44 -0800990 usage = count | has_cache;
Tim Chen7c00baf2017-02-22 15:45:36 -0800991 p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
Hugh Dickins253d5532009-12-14 17:58:44 -0800992
Tim Chen7c00baf2017-02-22 15:45:36 -0800993 unlock_cluster_or_swap_info(p, ci);
Huang, Ying235b6212017-02-22 15:45:22 -0800994
Hugh Dickins253d5532009-12-14 17:58:44 -0800995 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
997
Tim Chen7c00baf2017-02-22 15:45:36 -0800998static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
999{
1000 struct swap_cluster_info *ci;
1001 unsigned long offset = swp_offset(entry);
1002 unsigned char count;
1003
1004 ci = lock_cluster(p, offset);
1005 count = p->swap_map[offset];
1006 VM_BUG_ON(count != SWAP_HAS_CACHE);
1007 p->swap_map[offset] = 0;
1008 dec_cluster_info_page(p, p->cluster_info, offset);
1009 unlock_cluster(ci);
1010
1011 mem_cgroup_uncharge_swap(entry);
1012 if (offset < p->lowest_bit)
1013 p->lowest_bit = offset;
1014 if (offset > p->highest_bit) {
1015 bool was_full = !p->highest_bit;
1016
1017 p->highest_bit = offset;
1018 if (was_full && (p->flags & SWP_WRITEOK)) {
1019 spin_lock(&swap_avail_lock);
1020 WARN_ON(!plist_node_empty(&p->avail_list));
1021 if (plist_node_empty(&p->avail_list))
1022 plist_add(&p->avail_list,
1023 &swap_avail_head);
1024 spin_unlock(&swap_avail_lock);
1025 }
1026 }
1027 atomic_long_inc(&nr_swap_pages);
1028 p->inuse_pages--;
1029 frontswap_invalidate_page(p->type, offset);
1030 if (p->flags & SWP_BLKDEV) {
1031 struct gendisk *disk = p->bdev->bd_disk;
1032
1033 if (disk->fops->swap_slot_free_notify)
1034 disk->fops->swap_slot_free_notify(p->bdev,
1035 offset);
1036 }
1037}
1038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001040 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 * is still around or has not been recycled.
1042 */
1043void swap_free(swp_entry_t entry)
1044{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001045 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Huang, Ying235b6212017-02-22 15:45:22 -08001047 p = _swap_info_get(entry);
Tim Chen7c00baf2017-02-22 15:45:36 -08001048 if (p) {
1049 if (!__swap_entry_free(p, entry, 1))
Tim Chen67afa382017-02-22 15:45:39 -08001050 free_swap_slot(entry);
Tim Chen7c00baf2017-02-22 15:45:36 -08001051 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052}
1053
1054/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07001055 * Called after dropping swapcache to decrease refcnt to swap entries.
1056 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001057void swapcache_free(swp_entry_t entry)
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07001058{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001059 struct swap_info_struct *p;
1060
Huang, Ying235b6212017-02-22 15:45:22 -08001061 p = _swap_info_get(entry);
Tim Chen7c00baf2017-02-22 15:45:36 -08001062 if (p) {
1063 if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE))
Tim Chen67afa382017-02-22 15:45:39 -08001064 free_swap_slot(entry);
Tim Chen7c00baf2017-02-22 15:45:36 -08001065 }
1066}
1067
1068void swapcache_free_entries(swp_entry_t *entries, int n)
1069{
1070 struct swap_info_struct *p, *prev;
1071 int i;
1072
1073 if (n <= 0)
1074 return;
1075
1076 prev = NULL;
1077 p = NULL;
1078 for (i = 0; i < n; ++i) {
1079 p = swap_info_get_cont(entries[i], prev);
1080 if (p)
1081 swap_entry_free(p, entries[i]);
1082 else
1083 break;
1084 prev = p;
1085 }
Huang, Ying235b6212017-02-22 15:45:22 -08001086 if (p)
Tim Chen7c00baf2017-02-22 15:45:36 -08001087 spin_unlock(&p->lock);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07001088}
1089
1090/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001091 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -08001092 * This does not give an exact answer when swap count is continued,
1093 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001095int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001097 int count = 0;
1098 struct swap_info_struct *p;
Huang, Ying235b6212017-02-22 15:45:22 -08001099 struct swap_cluster_info *ci;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 swp_entry_t entry;
Huang, Ying235b6212017-02-22 15:45:22 -08001101 unsigned long offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001103 entry.val = page_private(page);
Huang, Ying235b6212017-02-22 15:45:22 -08001104 p = _swap_info_get(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 if (p) {
Huang, Ying235b6212017-02-22 15:45:22 -08001106 offset = swp_offset(entry);
1107 ci = lock_cluster_or_swap_info(p, offset);
1108 count = swap_count(p->swap_map[offset]);
1109 unlock_cluster_or_swap_info(p, ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001111 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112}
1113
1114/*
Minchan Kim8334b962015-09-08 15:00:24 -07001115 * How many references to @entry are currently swapped out?
Tim Chene8c26ab2017-02-22 15:45:29 -08001116 * This does not give an exact answer when swap count is continued,
1117 * but does include the high COUNT_CONTINUED flag to allow for that.
1118 */
1119int __swp_swapcount(swp_entry_t entry)
1120{
1121 int count = 0;
1122 pgoff_t offset;
1123 struct swap_info_struct *si;
1124 struct swap_cluster_info *ci;
1125
1126 si = __swap_info_get(entry);
1127 if (si) {
1128 offset = swp_offset(entry);
1129 ci = lock_cluster_or_swap_info(si, offset);
1130 count = swap_count(si->swap_map[offset]);
1131 unlock_cluster_or_swap_info(si, ci);
1132 }
1133 return count;
1134}
1135
1136/*
1137 * How many references to @entry are currently swapped out?
Minchan Kim8334b962015-09-08 15:00:24 -07001138 * This considers COUNT_CONTINUED so it returns exact answer.
1139 */
1140int swp_swapcount(swp_entry_t entry)
1141{
1142 int count, tmp_count, n;
1143 struct swap_info_struct *p;
Huang, Ying235b6212017-02-22 15:45:22 -08001144 struct swap_cluster_info *ci;
Minchan Kim8334b962015-09-08 15:00:24 -07001145 struct page *page;
1146 pgoff_t offset;
1147 unsigned char *map;
1148
Huang, Ying235b6212017-02-22 15:45:22 -08001149 p = _swap_info_get(entry);
Minchan Kim8334b962015-09-08 15:00:24 -07001150 if (!p)
1151 return 0;
1152
Huang, Ying235b6212017-02-22 15:45:22 -08001153 offset = swp_offset(entry);
1154
1155 ci = lock_cluster_or_swap_info(p, offset);
1156
1157 count = swap_count(p->swap_map[offset]);
Minchan Kim8334b962015-09-08 15:00:24 -07001158 if (!(count & COUNT_CONTINUED))
1159 goto out;
1160
1161 count &= ~COUNT_CONTINUED;
1162 n = SWAP_MAP_MAX + 1;
1163
Minchan Kim8334b962015-09-08 15:00:24 -07001164 page = vmalloc_to_page(p->swap_map + offset);
1165 offset &= ~PAGE_MASK;
1166 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1167
1168 do {
Geliang Tanga8ae4992016-01-14 15:20:45 -08001169 page = list_next_entry(page, lru);
Minchan Kim8334b962015-09-08 15:00:24 -07001170 map = kmap_atomic(page);
1171 tmp_count = map[offset];
1172 kunmap_atomic(map);
1173
1174 count += (tmp_count & ~COUNT_CONTINUED) * n;
1175 n *= (SWAP_CONT_MAX + 1);
1176 } while (tmp_count & COUNT_CONTINUED);
1177out:
Huang, Ying235b6212017-02-22 15:45:22 -08001178 unlock_cluster_or_swap_info(p, ci);
Minchan Kim8334b962015-09-08 15:00:24 -07001179 return count;
1180}
1181
1182/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001183 * We can write to an anon page without COW if there are no other references
1184 * to it. And as a side-effect, free up its swap: because the old content
1185 * on disk will never be read, and seeking back there to write new content
1186 * later would only waste time away from clustering.
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07001187 *
1188 * NOTE: total_mapcount should not be relied upon by the caller if
1189 * reuse_swap_page() returns false, but it may be always overwritten
1190 * (see the other implementation for CONFIG_SWAP=n).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 */
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07001192bool reuse_swap_page(struct page *page, int *total_mapcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001194 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Sasha Levin309381fea2014-01-23 15:52:54 -08001196 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -08001197 if (unlikely(PageKsm(page)))
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07001198 return false;
1199 count = page_trans_huge_mapcount(page, total_mapcount);
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001200 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001201 count += page_swapcount(page);
Minchan Kimf0571422017-01-10 16:58:15 -08001202 if (count != 1)
1203 goto out;
1204 if (!PageWriteback(page)) {
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001205 delete_from_swap_cache(page);
1206 SetPageDirty(page);
Minchan Kimf0571422017-01-10 16:58:15 -08001207 } else {
1208 swp_entry_t entry;
1209 struct swap_info_struct *p;
1210
1211 entry.val = page_private(page);
1212 p = swap_info_get(entry);
1213 if (p->flags & SWP_STABLE_WRITES) {
1214 spin_unlock(&p->lock);
1215 return false;
1216 }
1217 spin_unlock(&p->lock);
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001218 }
1219 }
Minchan Kimf0571422017-01-10 16:58:15 -08001220out:
Hugh Dickins5ad64682009-12-14 17:59:24 -08001221 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223
1224/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001225 * If swap is getting full, or if there are no more mappings of this page,
1226 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001228int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229{
Sasha Levin309381fea2014-01-23 15:52:54 -08001230 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
1232 if (!PageSwapCache(page))
1233 return 0;
1234 if (PageWriteback(page))
1235 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001236 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 return 0;
1238
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -07001239 /*
1240 * Once hibernation has begun to create its image of memory,
1241 * there's a danger that one of the calls to try_to_free_swap()
1242 * - most probably a call from __try_to_reclaim_swap() while
1243 * hibernation is allocating its own swap pages for the image,
1244 * but conceivably even a call from memory reclaim - will free
1245 * the swap from a page which has already been recorded in the
1246 * image as a clean swapcache page, and then reuse its swap for
1247 * another page of the image. On waking from hibernation, the
1248 * original page might be freed under memory pressure, then
1249 * later read back in from swap, now with the wrong data.
1250 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001251 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -08001252 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -07001253 */
Mel Gormanf90ac392012-01-10 15:07:15 -08001254 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -07001255 return 0;
1256
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001257 delete_from_swap_cache(page);
1258 SetPageDirty(page);
1259 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -07001260}
1261
1262/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 * Free the swap entry like above, but also try to
1264 * free the page cache entry if it is the last user.
1265 */
Hugh Dickins2509ef22009-01-06 14:40:10 -08001266int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
Hugh Dickins2509ef22009-01-06 14:40:10 -08001268 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 struct page *page = NULL;
Tim Chen7c00baf2017-02-22 15:45:36 -08001270 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Andi Kleena7420aa2009-09-16 11:50:05 +02001272 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -08001273 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -07001274
Tim Chen7c00baf2017-02-22 15:45:36 -08001275 p = _swap_info_get(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (p) {
Tim Chen7c00baf2017-02-22 15:45:36 -08001277 count = __swap_entry_free(p, entry, 1);
1278 if (count == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -08001279 page = find_get_page(swap_address_space(entry),
Huang Yingf6ab1f72016-10-07 17:00:21 -07001280 swp_offset(entry));
Nick Piggin8413ac92008-10-18 20:26:59 -07001281 if (page && !trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001282 put_page(page);
Nick Piggin93fac702006-03-31 02:29:56 -08001283 page = NULL;
1284 }
Tim Chen7c00baf2017-02-22 15:45:36 -08001285 } else if (!count)
Tim Chen67afa382017-02-22 15:45:39 -08001286 free_swap_slot(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001289 /*
1290 * Not mapped elsewhere, or swap space full? Free it!
1291 * Also recheck PageSwapCache now page is locked (above).
1292 */
Nick Piggin93fac702006-03-31 02:29:56 -08001293 if (PageSwapCache(page) && !PageWriteback(page) &&
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08001294 (!page_mapped(page) || mem_cgroup_swap_full(page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 delete_from_swap_cache(page);
1296 SetPageDirty(page);
1297 }
1298 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001299 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 }
Hugh Dickins2509ef22009-01-06 14:40:10 -08001301 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001304#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001305/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001306 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001307 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001308 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1309 * from 0, in which the swap header is expected to be located.
1310 *
1311 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001312 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001313int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001314{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001315 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001316 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001317
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001318 if (device)
1319 bdev = bdget(device);
1320
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001321 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001322 for (type = 0; type < nr_swapfiles; type++) {
1323 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001324
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001325 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001326 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001327
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001328 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001329 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001330 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001331
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001332 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001333 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001334 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001335 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001336 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001337
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001338 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001339 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001340 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001341
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001342 spin_unlock(&swap_lock);
1343 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001344 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001345 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001346 }
1347 }
1348 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001349 if (bdev)
1350 bdput(bdev);
1351
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001352 return -ENODEV;
1353}
1354
1355/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001356 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1357 * corresponding to given index in swap_info (swap type).
1358 */
1359sector_t swapdev_block(int type, pgoff_t offset)
1360{
1361 struct block_device *bdev;
1362
1363 if ((unsigned int)type >= nr_swapfiles)
1364 return 0;
1365 if (!(swap_info[type]->flags & SWP_WRITEOK))
1366 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001367 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001368}
1369
1370/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001371 * Return either the total number of swap pages of given type, or the number
1372 * of free pages of that type (depending on @free)
1373 *
1374 * This is needed for software suspend
1375 */
1376unsigned int count_swap_pages(int type, int free)
1377{
1378 unsigned int n = 0;
1379
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001380 spin_lock(&swap_lock);
1381 if ((unsigned int)type < nr_swapfiles) {
1382 struct swap_info_struct *sis = swap_info[type];
1383
Shaohua Liec8acf22013-02-22 16:34:38 -08001384 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001385 if (sis->flags & SWP_WRITEOK) {
1386 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001387 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001388 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001389 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001390 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001391 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001392 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001393 return n;
1394}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001395#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001396
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001397static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001398{
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001399 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001400}
1401
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001403 * No need to decide whether this PTE shares the swap entry with others,
1404 * just let do_wp_page work it out if a write is requested later - to
1405 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001407static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 unsigned long addr, swp_entry_t entry, struct page *page)
1409{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001410 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001411 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001412 spinlock_t *ptl;
1413 pte_t *pte;
1414 int ret = 1;
1415
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001416 swapcache = page;
1417 page = ksm_might_need_to_copy(page, vma, addr);
1418 if (unlikely(!page))
1419 return -ENOMEM;
1420
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001421 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1422 &memcg, false)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001423 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001424 goto out_nolock;
1425 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001426
1427 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001428 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001429 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001430 ret = 0;
1431 goto out;
1432 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001433
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001434 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001435 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 get_page(page);
1437 set_pte_at(vma->vm_mm, addr, pte,
1438 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001439 if (page == swapcache) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001440 page_add_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001441 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001442 } else { /* ksm created a completely new copy */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001443 page_add_new_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001444 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001445 lru_cache_add_active_or_unevictable(page, vma);
1446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 swap_free(entry);
1448 /*
1449 * Move the page to the active list so it is not
1450 * immediately swapped out again after swapon.
1451 */
1452 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001453out:
1454 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001455out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001456 if (page != swapcache) {
1457 unlock_page(page);
1458 put_page(page);
1459 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001460 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462
1463static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1464 unsigned long addr, unsigned long end,
1465 swp_entry_t entry, struct page *page)
1466{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001468 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001469 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
Hugh Dickins044d66c2008-02-07 00:14:04 -08001471 /*
1472 * We don't actually need pte lock while scanning for swp_pte: since
1473 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1474 * page table while we're scanning; though it could get zapped, and on
1475 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1476 * of unmatched parts which look like swp_pte, so unuse_pte must
1477 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001478 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001479 */
1480 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 do {
1482 /*
1483 * swapoff spends a _lot_ of time in this loop!
1484 * Test inline before going to call unuse_pte.
1485 */
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001486 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001487 pte_unmap(pte);
1488 ret = unuse_pte(vma, pmd, addr, entry, page);
1489 if (ret)
1490 goto out;
1491 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 }
1493 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001494 pte_unmap(pte - 1);
1495out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001496 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497}
1498
1499static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1500 unsigned long addr, unsigned long end,
1501 swp_entry_t entry, struct page *page)
1502{
1503 pmd_t *pmd;
1504 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001505 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
1507 pmd = pmd_offset(pud, addr);
1508 do {
Hugh Dickinsdc644a02016-12-12 16:44:44 -08001509 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001511 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001513 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1514 if (ret)
1515 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 } while (pmd++, addr = next, addr != end);
1517 return 0;
1518}
1519
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001520static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 unsigned long addr, unsigned long end,
1522 swp_entry_t entry, struct page *page)
1523{
1524 pud_t *pud;
1525 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001526 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001528 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 do {
1530 next = pud_addr_end(addr, end);
1531 if (pud_none_or_clear_bad(pud))
1532 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001533 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1534 if (ret)
1535 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 } while (pud++, addr = next, addr != end);
1537 return 0;
1538}
1539
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001540static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1541 unsigned long addr, unsigned long end,
1542 swp_entry_t entry, struct page *page)
1543{
1544 p4d_t *p4d;
1545 unsigned long next;
1546 int ret;
1547
1548 p4d = p4d_offset(pgd, addr);
1549 do {
1550 next = p4d_addr_end(addr, end);
1551 if (p4d_none_or_clear_bad(p4d))
1552 continue;
1553 ret = unuse_pud_range(vma, p4d, addr, next, entry, page);
1554 if (ret)
1555 return ret;
1556 } while (p4d++, addr = next, addr != end);
1557 return 0;
1558}
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560static int unuse_vma(struct vm_area_struct *vma,
1561 swp_entry_t entry, struct page *page)
1562{
1563 pgd_t *pgd;
1564 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001565 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001567 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 addr = page_address_in_vma(page, vma);
1569 if (addr == -EFAULT)
1570 return 0;
1571 else
1572 end = addr + PAGE_SIZE;
1573 } else {
1574 addr = vma->vm_start;
1575 end = vma->vm_end;
1576 }
1577
1578 pgd = pgd_offset(vma->vm_mm, addr);
1579 do {
1580 next = pgd_addr_end(addr, end);
1581 if (pgd_none_or_clear_bad(pgd))
1582 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001583 ret = unuse_p4d_range(vma, pgd, addr, next, entry, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001584 if (ret)
1585 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 } while (pgd++, addr = next, addr != end);
1587 return 0;
1588}
1589
1590static int unuse_mm(struct mm_struct *mm,
1591 swp_entry_t entry, struct page *page)
1592{
1593 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001594 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
1596 if (!down_read_trylock(&mm->mmap_sem)) {
1597 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001598 * Activate page so shrink_inactive_list is unlikely to unmap
1599 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001601 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 unlock_page(page);
1603 down_read(&mm->mmap_sem);
1604 lock_page(page);
1605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001607 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 break;
Hugh Dickinsdc644a02016-12-12 16:44:44 -08001609 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001612 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613}
1614
1615/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001616 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1617 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 * Recycle to start on reaching the end, returning 0 when empty.
1619 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001620static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001621 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001623 unsigned int max = si->max;
1624 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001625 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
1627 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001628 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 * for whether an entry is in use, not modifying it; false
1630 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001631 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 */
1633 for (;;) {
1634 if (++i >= max) {
1635 if (!prev) {
1636 i = 0;
1637 break;
1638 }
1639 /*
1640 * No entries in use at top of swap_map,
1641 * loop back to start and recheck there.
1642 */
1643 max = prev + 1;
1644 prev = 0;
1645 i = 1;
1646 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001647 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001648 if (count && swap_count(count) != SWAP_MAP_BAD)
Hugh Dickinsdc644a02016-12-12 16:44:44 -08001649 if (!frontswap || frontswap_test(si, i))
1650 break;
1651 if ((i % LATENCY_LIMIT) == 0)
1652 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 }
1654 return i;
1655}
1656
1657/*
1658 * We completely avoid races by reading each swap page in advance,
1659 * and then search for the process using it. All the necessary
1660 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001661 *
1662 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1663 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001665int try_to_unuse(unsigned int type, bool frontswap,
1666 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001668 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001670 volatile unsigned char *swap_map; /* swap_map is accessed without
1671 * locking. Mark it as volatile
1672 * to prevent compiler doing
1673 * something odd.
1674 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001675 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 struct page *page;
1677 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001678 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
1681 /*
1682 * When searching mms for an entry, a good strategy is to
1683 * start at the first mm we freed the previous entry from
1684 * (though actually we don't notice whether we or coincidence
1685 * freed the entry). Initialize this start_mm with a hold.
1686 *
1687 * A simpler strategy would be to start at the last mm we
1688 * freed the previous entry from; but that would take less
1689 * advantage of mmlist ordering, which clusters forked mms
1690 * together, child after parent. If we race with dup_mmap(), we
1691 * prefer to resolve parent before child, lest we miss entries
1692 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001693 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 */
1695 start_mm = &init_mm;
Vegard Nossum3fce3712017-02-27 14:30:10 -08001696 mmget(&init_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
1698 /*
1699 * Keep on scanning until all entries have gone. Usually,
1700 * one pass through swap_map is enough, but not necessarily:
1701 * there are races when an instance of an entry might be missed.
1702 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001703 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 if (signal_pending(current)) {
1705 retval = -EINTR;
1706 break;
1707 }
1708
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001709 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 * Get a page for the entry, using the existing swap
1711 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001712 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 */
1714 swap_map = &si->swap_map[i];
1715 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001716 page = read_swap_cache_async(entry,
1717 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 if (!page) {
1719 /*
1720 * Either swap_duplicate() failed because entry
1721 * has been freed independently, and will not be
1722 * reused since sys_swapoff() already disabled
1723 * allocation from here, or alloc_page() failed.
1724 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001725 swcount = *swap_map;
1726 /*
1727 * We don't hold lock here, so the swap entry could be
1728 * SWAP_MAP_BAD (when the cluster is discarding).
1729 * Instead of fail out, We can just skip the swap
1730 * entry because swapoff will wait for discarding
1731 * finish anyway.
1732 */
1733 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 continue;
1735 retval = -ENOMEM;
1736 break;
1737 }
1738
1739 /*
1740 * Don't hold on to start_mm if it looks like exiting.
1741 */
1742 if (atomic_read(&start_mm->mm_users) == 1) {
1743 mmput(start_mm);
1744 start_mm = &init_mm;
Vegard Nossum3fce3712017-02-27 14:30:10 -08001745 mmget(&init_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
1747
1748 /*
1749 * Wait for and lock page. When do_swap_page races with
1750 * try_to_unuse, do_swap_page can handle the fault much
1751 * faster than try_to_unuse can locate the entry. This
1752 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1753 * defer to do_swap_page in such a case - in some tests,
1754 * do_swap_page and try_to_unuse repeatedly compete.
1755 */
1756 wait_on_page_locked(page);
1757 wait_on_page_writeback(page);
1758 lock_page(page);
1759 wait_on_page_writeback(page);
1760
1761 /*
1762 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001765 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1766 retval = shmem_unuse(entry, page);
1767 /* page has already been unlocked and released */
1768 if (retval < 0)
1769 break;
1770 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001772 if (swap_count(swcount) && start_mm != &init_mm)
1773 retval = unuse_mm(start_mm, entry, page);
1774
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001775 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 int set_start_mm = (*swap_map >= swcount);
1777 struct list_head *p = &start_mm->mmlist;
1778 struct mm_struct *new_start_mm = start_mm;
1779 struct mm_struct *prev_mm = start_mm;
1780 struct mm_struct *mm;
1781
Vegard Nossum3fce3712017-02-27 14:30:10 -08001782 mmget(new_start_mm);
1783 mmget(prev_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001785 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 (p = p->next) != &start_mm->mmlist) {
1787 mm = list_entry(p, struct mm_struct, mmlist);
Vegard Nossum388f7932017-02-27 14:30:13 -08001788 if (!mmget_not_zero(mm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 spin_unlock(&mmlist_lock);
1791 mmput(prev_mm);
1792 prev_mm = mm;
1793
1794 cond_resched();
1795
1796 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001797 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001799 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001801 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001803
Bo Liu32c5fc12009-11-02 16:50:33 +00001804 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 mmput(new_start_mm);
Vegard Nossum3fce3712017-02-27 14:30:10 -08001806 mmget(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 new_start_mm = mm;
1808 set_start_mm = 0;
1809 }
1810 spin_lock(&mmlist_lock);
1811 }
1812 spin_unlock(&mmlist_lock);
1813 mmput(prev_mm);
1814 mmput(start_mm);
1815 start_mm = new_start_mm;
1816 }
1817 if (retval) {
1818 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001819 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 break;
1821 }
1822
1823 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 * If a reference remains (rare), we would like to leave
1825 * the page in the swap cache; but try_to_unmap could
1826 * then re-duplicate the entry once we drop page lock,
1827 * so we might loop indefinitely; also, that page could
1828 * not be swapped out to other storage meanwhile. So:
1829 * delete from cache even if there's another reference,
1830 * after ensuring that the data has been saved to disk -
1831 * since if the reference remains (rarer), it will be
1832 * read from disk into another page. Splitting into two
1833 * pages would be incorrect if swap supported "shared
1834 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001835 *
1836 * Given how unuse_vma() targets one particular offset
1837 * in an anon_vma, once the anon_vma has been determined,
1838 * this splitting happens to be just what is needed to
1839 * handle where KSM pages have been swapped out: re-reading
1840 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001842 if (swap_count(*swap_map) &&
1843 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 struct writeback_control wbc = {
1845 .sync_mode = WB_SYNC_NONE,
1846 };
1847
1848 swap_writepage(page, &wbc);
1849 lock_page(page);
1850 wait_on_page_writeback(page);
1851 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001852
1853 /*
1854 * It is conceivable that a racing task removed this page from
1855 * swap cache just before we acquired the page lock at the top,
1856 * or while we dropped it in unuse_mm(). The page might even
1857 * be back in swap cache on another swap area: that we must not
1858 * delete, since it may not have been written out to swap yet.
1859 */
1860 if (PageSwapCache(page) &&
1861 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001862 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
1864 /*
1865 * So we could skip searching mms once swap count went
1866 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001867 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 */
1869 SetPageDirty(page);
1870 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001871 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873 /*
1874 * Make sure that we aren't completely killing
1875 * interactive performance.
1876 */
1877 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001878 if (frontswap && pages_to_unuse > 0) {
1879 if (!--pages_to_unuse)
1880 break;
1881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 }
1883
1884 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 return retval;
1886}
1887
1888/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001889 * After a successful try_to_unuse, if no swap is now in use, we know
1890 * we can empty the mmlist. swap_lock must be held on entry and exit.
1891 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 * added to the mmlist just after page_duplicate - before would be racy.
1893 */
1894static void drain_mmlist(void)
1895{
1896 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001897 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001899 for (type = 0; type < nr_swapfiles; type++)
1900 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 return;
1902 spin_lock(&mmlist_lock);
1903 list_for_each_safe(p, next, &init_mm.mmlist)
1904 list_del_init(p);
1905 spin_unlock(&mmlist_lock);
1906}
1907
1908/*
1909 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001910 * corresponds to page offset for the specified swap entry.
1911 * Note that the type of this function is sector_t, but it returns page offset
1912 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001914static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001916 struct swap_info_struct *sis;
1917 struct swap_extent *start_se;
1918 struct swap_extent *se;
1919 pgoff_t offset;
1920
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001921 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001922 *bdev = sis->bdev;
1923
1924 offset = swp_offset(entry);
1925 start_se = sis->curr_swap_extent;
1926 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
1928 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 if (se->start_page <= offset &&
1930 offset < (se->start_page + se->nr_pages)) {
1931 return se->start_block + (offset - se->start_page);
1932 }
Geliang Tanga8ae4992016-01-14 15:20:45 -08001933 se = list_next_entry(se, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 sis->curr_swap_extent = se;
1935 BUG_ON(se == start_se); /* It *must* be present */
1936 }
1937}
1938
1939/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001940 * Returns the page offset into bdev for the specified page's swap entry.
1941 */
1942sector_t map_swap_page(struct page *page, struct block_device **bdev)
1943{
1944 swp_entry_t entry;
1945 entry.val = page_private(page);
1946 return map_swap_entry(entry, bdev);
1947}
1948
1949/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 * Free all of a swapdev's extent information
1951 */
1952static void destroy_swap_extents(struct swap_info_struct *sis)
1953{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001954 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 struct swap_extent *se;
1956
Geliang Tanga8ae4992016-01-14 15:20:45 -08001957 se = list_first_entry(&sis->first_swap_extent.list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 struct swap_extent, list);
1959 list_del(&se->list);
1960 kfree(se);
1961 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001962
1963 if (sis->flags & SWP_FILE) {
1964 struct file *swap_file = sis->swap_file;
1965 struct address_space *mapping = swap_file->f_mapping;
1966
1967 sis->flags &= ~SWP_FILE;
1968 mapping->a_ops->swap_deactivate(swap_file);
1969 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970}
1971
1972/*
1973 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001974 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001976 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 */
Mel Gormana509bc12012-07-31 16:44:57 -07001978int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1980 unsigned long nr_pages, sector_t start_block)
1981{
1982 struct swap_extent *se;
1983 struct swap_extent *new_se;
1984 struct list_head *lh;
1985
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001986 if (start_page == 0) {
1987 se = &sis->first_swap_extent;
1988 sis->curr_swap_extent = se;
1989 se->start_page = 0;
1990 se->nr_pages = nr_pages;
1991 se->start_block = start_block;
1992 return 1;
1993 } else {
1994 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001996 BUG_ON(se->start_page + se->nr_pages != start_page);
1997 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 /* Merge it */
1999 se->nr_pages += nr_pages;
2000 return 0;
2001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 }
2003
2004 /*
2005 * No merge. Insert a new extent, preserving ordering.
2006 */
2007 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2008 if (new_se == NULL)
2009 return -ENOMEM;
2010 new_se->start_page = start_page;
2011 new_se->nr_pages = nr_pages;
2012 new_se->start_block = start_block;
2013
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002014 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07002015 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016}
2017
2018/*
2019 * A `swap extent' is a simple thing which maps a contiguous range of pages
2020 * onto a contiguous range of disk blocks. An ordered list of swap extents
2021 * is built at swapon time and is then used at swap_writepage/swap_readpage
2022 * time for locating where on disk a page belongs.
2023 *
2024 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2025 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2026 * swap files identically.
2027 *
2028 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2029 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2030 * swapfiles are handled *identically* after swapon time.
2031 *
2032 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2033 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2034 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2035 * requirements, they are simply tossed out - we will never use those blocks
2036 * for swapping.
2037 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07002038 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2040 * which will scribble on the fs.
2041 *
2042 * The amount of disk space which a single swap extent represents varies.
2043 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2044 * extents in the list. To avoid much list walking, we cache the previous
2045 * search location in `curr_swap_extent', and start new searches from there.
2046 * This is extremely effective. The average number of iterations in
2047 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2048 */
Hugh Dickins53092a72005-09-03 15:54:34 -07002049static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050{
Mel Gorman62c230b2012-07-31 16:44:55 -07002051 struct file *swap_file = sis->swap_file;
2052 struct address_space *mapping = swap_file->f_mapping;
2053 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 int ret;
2055
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (S_ISBLK(inode->i_mode)) {
2057 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07002058 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07002059 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 }
2061
Mel Gorman62c230b2012-07-31 16:44:55 -07002062 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07002063 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07002064 if (!ret) {
2065 sis->flags |= SWP_FILE;
2066 ret = add_swap_extent(sis, 0, sis->max, 0);
2067 *span = sis->pages;
2068 }
Mel Gormana509bc12012-07-31 16:44:57 -07002069 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07002070 }
2071
Mel Gormana509bc12012-07-31 16:44:57 -07002072 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073}
2074
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08002075static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002076 unsigned char *swap_map,
2077 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002078{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002079 if (prio >= 0)
2080 p->prio = prio;
2081 else
2082 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07002083 /*
2084 * the plist prio is negated because plist ordering is
2085 * low-to-high, while swap ordering is high-to-low
2086 */
2087 p->list.prio = -p->prio;
2088 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002089 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002090 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002091 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08002092 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002093 total_swap_pages += p->pages;
2094
Dan Streetmanadfab832014-06-04 16:09:53 -07002095 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07002096 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07002097 * both lists are plists, and thus priority ordered.
2098 * swap_active_head needs to be priority ordered for swapoff(),
2099 * which on removal of any swap_info_struct with an auto-assigned
2100 * (i.e. negative) priority increments the auto-assigned priority
2101 * of any lower-priority swap_info_structs.
2102 * swap_avail_head needs to be priority ordered for get_swap_page(),
2103 * which allocates swap pages from the highest available priority
2104 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07002105 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07002106 plist_add(&p->list, &swap_active_head);
2107 spin_lock(&swap_avail_lock);
2108 plist_add(&p->avail_list, &swap_avail_head);
2109 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08002110}
2111
2112static void enable_swap_info(struct swap_info_struct *p, int prio,
2113 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002114 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08002115 unsigned long *frontswap_map)
2116{
Minchan Kim4f898492013-04-30 15:26:54 -07002117 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08002118 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002119 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002120 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08002121 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08002122 spin_unlock(&swap_lock);
2123}
2124
2125static void reinsert_swap_info(struct swap_info_struct *p)
2126{
2127 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002128 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002129 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08002130 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002131 spin_unlock(&swap_lock);
2132}
2133
Tim Chen67afa382017-02-22 15:45:39 -08002134bool has_usable_swap(void)
2135{
2136 bool ret = true;
2137
2138 spin_lock(&swap_lock);
2139 if (plist_head_empty(&swap_active_head))
2140 ret = false;
2141 spin_unlock(&swap_lock);
2142 return ret;
2143}
2144
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01002145SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002147 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002148 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002149 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07002150 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 struct file *swap_file, *victim;
2152 struct address_space *mapping;
2153 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04002154 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07002155 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07002156 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 if (!capable(CAP_SYS_ADMIN))
2159 return -EPERM;
2160
Al Viro191c5422012-02-13 03:58:52 +00002161 BUG_ON(!current->mm);
2162
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08002165 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
Jeff Layton669abf42012-10-10 16:43:10 -04002167 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 err = PTR_ERR(victim);
2169 if (IS_ERR(victim))
2170 goto out;
2171
2172 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002173 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002174 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08002175 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07002176 if (p->swap_file->f_mapping == mapping) {
2177 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07002179 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 }
Dan Streetmanadfab832014-06-04 16:09:53 -07002182 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002184 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 goto out_dput;
2186 }
Al Viro191c5422012-02-13 03:58:52 +00002187 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 vm_unacct_memory(p->pages);
2189 else {
2190 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002191 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 goto out_dput;
2193 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07002194 spin_lock(&swap_avail_lock);
2195 plist_del(&p->avail_list, &swap_avail_head);
2196 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002197 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07002198 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07002199 struct swap_info_struct *si = p;
2200
Dan Streetman18ab4d42014-06-04 16:09:59 -07002201 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07002202 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07002203 si->list.prio--;
2204 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07002205 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07002206 least_priority++;
2207 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07002208 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08002209 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 total_swap_pages -= p->pages;
2211 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08002212 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002213 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07002214
Tim Chen039939a2017-02-22 15:45:43 -08002215 disable_swap_slots_cache_lock();
2216
David Rientjese1e12d22012-12-11 16:02:56 -08002217 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07002218 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08002219 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 if (err) {
2222 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08002223 reinsert_swap_info(p);
Tim Chen039939a2017-02-22 15:45:43 -08002224 reenable_swap_slots_cache_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 goto out_dput;
2226 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07002227
Tim Chen039939a2017-02-22 15:45:43 -08002228 reenable_swap_slots_cache_unlock();
2229
Shaohua Li815c2c52013-09-11 14:20:30 -07002230 flush_work(&p->discard_work);
2231
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002232 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002233 if (p->flags & SWP_CONTINUED)
2234 free_swap_count_continuations(p);
2235
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002236 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002237 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002238 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07002240
2241 /* wait for anyone still in scan_swap_map */
2242 p->highest_bit = 0; /* cuts scans short */
2243 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002244 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002245 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07002246 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002247 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002248 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002249 }
2250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07002252 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 p->swap_file = NULL;
2254 p->max = 0;
2255 swap_map = p->swap_map;
2256 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002257 cluster_info = p->cluster_info;
2258 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07002259 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08002260 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002261 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07002262 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08002263 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002264 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a62013-09-11 14:20:32 -07002265 free_percpu(p->percpu_cluster);
2266 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002268 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07002269 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002270 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07002271 swap_cgroup_swapoff(p->type);
Huang, Ying4b3ef9d2017-02-22 15:45:26 -08002272 exit_swap_address_space(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 inode = mapping->host;
2275 if (S_ISBLK(inode->i_mode)) {
2276 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07002277 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01002278 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 } else {
Al Viro59551022016-01-22 15:40:57 -05002280 inode_lock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 inode->i_flags &= ~S_SWAPFILE;
Al Viro59551022016-01-22 15:40:57 -05002282 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 }
2284 filp_close(swap_file, NULL);
Weijie Yangf893ab412014-02-06 12:04:23 -08002285
2286 /*
2287 * Clear the SWP_USED flag after all resources are freed so that swapon
2288 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2289 * not hold p->lock after we cleared its SWP_WRITEOK.
2290 */
2291 spin_lock(&swap_lock);
2292 p->flags = 0;
2293 spin_unlock(&swap_lock);
2294
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002296 atomic_inc(&proc_poll_event);
2297 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299out_dput:
2300 filp_close(victim, NULL);
2301out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08002302 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 return err;
2304}
2305
2306#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07002307static unsigned swaps_poll(struct file *file, poll_table *wait)
2308{
Kay Sieversf1514632011-07-12 20:48:39 +02002309 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002310
2311 poll_wait(file, &proc_poll_wait, wait);
2312
Kay Sieversf1514632011-07-12 20:48:39 +02002313 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2314 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002315 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2316 }
2317
2318 return POLLIN | POLLRDNORM;
2319}
2320
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321/* iterator */
2322static void *swap_start(struct seq_file *swap, loff_t *pos)
2323{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002324 struct swap_info_struct *si;
2325 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 loff_t l = *pos;
2327
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002328 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002330 if (!l)
2331 return SEQ_START_TOKEN;
2332
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002333 for (type = 0; type < nr_swapfiles; type++) {
2334 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2335 si = swap_info[type];
2336 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002338 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002339 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 }
2341
2342 return NULL;
2343}
2344
2345static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2346{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002347 struct swap_info_struct *si = v;
2348 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002350 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002351 type = 0;
2352 else
2353 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002354
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002355 for (; type < nr_swapfiles; type++) {
2356 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2357 si = swap_info[type];
2358 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 continue;
2360 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002361 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 }
2363
2364 return NULL;
2365}
2366
2367static void swap_stop(struct seq_file *swap, void *v)
2368{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002369 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370}
2371
2372static int swap_show(struct seq_file *swap, void *v)
2373{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002374 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 struct file *file;
2376 int len;
2377
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002378 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002379 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2380 return 0;
2381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002383 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002384 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002385 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002386 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002387 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002389 si->pages << (PAGE_SHIFT - 10),
2390 si->inuse_pages << (PAGE_SHIFT - 10),
2391 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 return 0;
2393}
2394
Helge Deller15ad7cd2006-12-06 20:40:36 -08002395static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 .start = swap_start,
2397 .next = swap_next,
2398 .stop = swap_stop,
2399 .show = swap_show
2400};
2401
2402static int swaps_open(struct inode *inode, struct file *file)
2403{
Kay Sieversf1514632011-07-12 20:48:39 +02002404 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002405 int ret;
2406
Kay Sievers66d7dd52010-10-26 14:22:06 -07002407 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002408 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002409 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002410
Kay Sieversf1514632011-07-12 20:48:39 +02002411 seq = file->private_data;
2412 seq->poll_event = atomic_read(&proc_poll_event);
2413 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414}
2415
Helge Deller15ad7cd2006-12-06 20:40:36 -08002416static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 .open = swaps_open,
2418 .read = seq_read,
2419 .llseek = seq_lseek,
2420 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002421 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422};
2423
2424static int __init procswaps_init(void)
2425{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002426 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 return 0;
2428}
2429__initcall(procswaps_init);
2430#endif /* CONFIG_PROC_FS */
2431
Jan Beulich17963162008-12-16 11:35:24 +00002432#ifdef MAX_SWAPFILES_CHECK
2433static int __init max_swapfiles_check(void)
2434{
2435 MAX_SWAPFILES_CHECK();
2436 return 0;
2437}
2438late_initcall(max_swapfiles_check);
2439#endif
2440
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002441static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002443 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002445
2446 p = kzalloc(sizeof(*p), GFP_KERNEL);
2447 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002448 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002449
Hugh Dickins5d337b92005-09-03 15:54:41 -07002450 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002451 for (type = 0; type < nr_swapfiles; type++) {
2452 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002454 }
Christoph Lameter06972122006-06-23 02:03:35 -07002455 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002456 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002457 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002458 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002460 if (type >= nr_swapfiles) {
2461 p->type = type;
2462 swap_info[type] = p;
2463 /*
2464 * Write swap_info[type] before nr_swapfiles, in case a
2465 * racing procfs swap_start() or swap_next() is reading them.
2466 * (We never shrink nr_swapfiles, we never free this entry.)
2467 */
2468 smp_wmb();
2469 nr_swapfiles++;
2470 } else {
2471 kfree(p);
2472 p = swap_info[type];
2473 /*
2474 * Do not memset this entry: a racing procfs swap_next()
2475 * would be relying on p->type to remain valid.
2476 */
2477 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002478 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002479 plist_node_init(&p->list, 0);
2480 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002482 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002483 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002484
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002485 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002486}
2487
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002488static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2489{
2490 int error;
2491
2492 if (S_ISBLK(inode->i_mode)) {
2493 p->bdev = bdgrab(I_BDEV(inode));
2494 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002495 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002496 if (error < 0) {
2497 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002498 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002499 }
2500 p->old_block_size = block_size(p->bdev);
2501 error = set_blocksize(p->bdev, PAGE_SIZE);
2502 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002503 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002504 p->flags |= SWP_BLKDEV;
2505 } else if (S_ISREG(inode->i_mode)) {
2506 p->bdev = inode->i_sb->s_bdev;
Al Viro59551022016-01-22 15:40:57 -05002507 inode_lock(inode);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002508 if (IS_SWAPFILE(inode))
2509 return -EBUSY;
2510 } else
2511 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002512
2513 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002514}
2515
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002516static unsigned long read_swap_header(struct swap_info_struct *p,
2517 union swap_header *swap_header,
2518 struct inode *inode)
2519{
2520 int i;
2521 unsigned long maxpages;
2522 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002523 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002524
2525 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002526 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002527 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002528 }
2529
2530 /* swap partition endianess hack... */
2531 if (swab32(swap_header->info.version) == 1) {
2532 swab32s(&swap_header->info.version);
2533 swab32s(&swap_header->info.last_page);
2534 swab32s(&swap_header->info.nr_badpages);
Jann Horndd111be2016-11-10 10:46:19 -08002535 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2536 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002537 for (i = 0; i < swap_header->info.nr_badpages; i++)
2538 swab32s(&swap_header->info.badpages[i]);
2539 }
2540 /* Check the swap header's sub-version */
2541 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002542 pr_warn("Unable to handle swap header version %d\n",
2543 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002544 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002545 }
2546
2547 p->lowest_bit = 1;
2548 p->cluster_next = 1;
2549 p->cluster_nr = 0;
2550
2551 /*
2552 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002553 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002554 * of bits for the swap offset in the swp_entry_t type, and
2555 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002556 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002557 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002558 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002559 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002560 * offset is extracted. This will mask all the bits from
2561 * the initial ~0UL mask that can't be encoded in either
2562 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002563 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002564 */
2565 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002566 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002567 last_page = swap_header->info.last_page;
2568 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002569 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002570 maxpages << (PAGE_SHIFT - 10),
2571 last_page << (PAGE_SHIFT - 10));
2572 }
2573 if (maxpages > last_page) {
2574 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002575 /* p->max is an unsigned int: don't overflow it */
2576 if ((unsigned int)maxpages == 0)
2577 maxpages = UINT_MAX;
2578 }
2579 p->highest_bit = maxpages - 1;
2580
2581 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002582 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002583 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2584 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002585 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002586 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002587 }
2588 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002589 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002590 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002591 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002592
2593 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002594}
2595
Huang, Ying4b3ef9d2017-02-22 15:45:26 -08002596#define SWAP_CLUSTER_INFO_COLS \
Huang, Ying235b6212017-02-22 15:45:22 -08002597 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
Huang, Ying4b3ef9d2017-02-22 15:45:26 -08002598#define SWAP_CLUSTER_SPACE_COLS \
2599 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2600#define SWAP_CLUSTER_COLS \
2601 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
Huang, Ying235b6212017-02-22 15:45:22 -08002602
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002603static int setup_swap_map_and_extents(struct swap_info_struct *p,
2604 union swap_header *swap_header,
2605 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002606 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002607 unsigned long maxpages,
2608 sector_t *span)
2609{
Huang, Ying235b6212017-02-22 15:45:22 -08002610 unsigned int j, k;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002611 unsigned int nr_good_pages;
2612 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002613 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
Huang, Ying235b6212017-02-22 15:45:22 -08002614 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2615 unsigned long i, idx;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002616
2617 nr_good_pages = maxpages - 1; /* omit header page */
2618
Huang Ying6b534912016-10-07 16:58:42 -07002619 cluster_list_init(&p->free_clusters);
2620 cluster_list_init(&p->discard_clusters);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002621
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002622 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2623 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002624 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2625 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002626 if (page_nr < maxpages) {
2627 swap_map[page_nr] = SWAP_MAP_BAD;
2628 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002629 /*
2630 * Haven't marked the cluster free yet, no list
2631 * operation involved
2632 */
2633 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002634 }
2635 }
2636
Shaohua Li2a8f9442013-09-11 14:20:28 -07002637 /* Haven't marked the cluster free yet, no list operation involved */
2638 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2639 inc_cluster_info_page(p, cluster_info, i);
2640
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002641 if (nr_good_pages) {
2642 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002643 /*
2644 * Not mark the cluster free yet, no list
2645 * operation involved
2646 */
2647 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002648 p->max = maxpages;
2649 p->pages = nr_good_pages;
2650 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002651 if (nr_extents < 0)
2652 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002653 nr_good_pages = p->pages;
2654 }
2655 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002656 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002657 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002658 }
2659
Shaohua Li2a8f9442013-09-11 14:20:28 -07002660 if (!cluster_info)
2661 return nr_extents;
2662
Huang, Ying235b6212017-02-22 15:45:22 -08002663
Huang, Ying4b3ef9d2017-02-22 15:45:26 -08002664 /*
2665 * Reduce false cache line sharing between cluster_info and
2666 * sharing same address space.
2667 */
Huang, Ying235b6212017-02-22 15:45:22 -08002668 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2669 j = (k + col) % SWAP_CLUSTER_COLS;
2670 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2671 idx = i * SWAP_CLUSTER_COLS + j;
2672 if (idx >= nr_clusters)
2673 continue;
2674 if (cluster_count(&cluster_info[idx]))
2675 continue;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002676 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -07002677 cluster_list_add_tail(&p->free_clusters, cluster_info,
2678 idx);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002679 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002680 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002681 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002682}
2683
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002684/*
2685 * Helper to sys_swapon determining if a given swap
2686 * backing device queue supports DISCARD operations.
2687 */
2688static bool swap_discardable(struct swap_info_struct *si)
2689{
2690 struct request_queue *q = bdev_get_queue(si->bdev);
2691
2692 if (!q || !blk_queue_discard(q))
2693 return false;
2694
2695 return true;
2696}
2697
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002698SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2699{
2700 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002701 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002702 struct file *swap_file = NULL;
2703 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002704 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002705 int error;
2706 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002707 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002708 sector_t span;
2709 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002710 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002711 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002712 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002713 struct page *page = NULL;
2714 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002715
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002716 if (swap_flags & ~SWAP_FLAGS_VALID)
2717 return -EINVAL;
2718
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002719 if (!capable(CAP_SYS_ADMIN))
2720 return -EPERM;
2721
2722 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002723 if (IS_ERR(p))
2724 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002725
Shaohua Li815c2c52013-09-11 14:20:30 -07002726 INIT_WORK(&p->discard_work, swap_discard_work);
2727
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002730 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002732 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 }
Jeff Layton669abf42012-10-10 16:43:10 -04002734 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002736 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002738 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 }
2740
2741 p->swap_file = swap_file;
2742 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002743 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002744
Al Viro59551022016-01-22 15:40:57 -05002745 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002746 error = claim_swapfile(p, inode);
2747 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 /*
2751 * Read the swap header.
2752 */
2753 if (!mapping->a_ops->readpage) {
2754 error = -EINVAL;
2755 goto bad_swap;
2756 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002757 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 if (IS_ERR(page)) {
2759 error = PTR_ERR(page);
2760 goto bad_swap;
2761 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002762 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002764 maxpages = read_swap_header(p, swap_header, inode);
2765 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 error = -EINVAL;
2767 goto bad_swap;
2768 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002769
Hugh Dickins81e33972009-01-06 14:39:49 -08002770 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002771 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002772 if (!swap_map) {
2773 error = -ENOMEM;
2774 goto bad_swap;
2775 }
Minchan Kimf0571422017-01-10 16:58:15 -08002776
2777 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
2778 p->flags |= SWP_STABLE_WRITES;
2779
Shaohua Li2a8f9442013-09-11 14:20:28 -07002780 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002781 int cpu;
Huang, Ying235b6212017-02-22 15:45:22 -08002782 unsigned long ci, nr_cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002783
Shaohua Li2a8f9442013-09-11 14:20:28 -07002784 p->flags |= SWP_SOLIDSTATE;
2785 /*
2786 * select a random position to start with to help wear leveling
2787 * SSD
2788 */
2789 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
Huang, Ying235b6212017-02-22 15:45:22 -08002790 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002791
Huang, Ying235b6212017-02-22 15:45:22 -08002792 cluster_info = vzalloc(nr_cluster * sizeof(*cluster_info));
Shaohua Li2a8f9442013-09-11 14:20:28 -07002793 if (!cluster_info) {
2794 error = -ENOMEM;
2795 goto bad_swap;
2796 }
Huang, Ying235b6212017-02-22 15:45:22 -08002797
2798 for (ci = 0; ci < nr_cluster; ci++)
2799 spin_lock_init(&((cluster_info + ci)->lock));
2800
Shaohua Liebc2a1a62013-09-11 14:20:32 -07002801 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2802 if (!p->percpu_cluster) {
2803 error = -ENOMEM;
2804 goto bad_swap;
2805 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002806 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a62013-09-11 14:20:32 -07002807 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002808 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a62013-09-11 14:20:32 -07002809 cluster_set_null(&cluster->index);
2810 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002811 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002812
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002813 error = swap_cgroup_swapon(p->type, maxpages);
2814 if (error)
2815 goto bad_swap;
2816
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002817 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002818 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002819 if (unlikely(nr_extents < 0)) {
2820 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 goto bad_swap;
2822 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002823 /* frontswap enabled? set up bit-per-page map for frontswap */
Vlastimil Babka8ea1d2a2016-07-26 15:24:42 -07002824 if (IS_ENABLED(CONFIG_FRONTSWAP))
Akinobu Mita7b579762013-06-12 14:05:08 -07002825 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
Shaohua Li2a8f9442013-09-11 14:20:28 -07002827 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2828 /*
2829 * When discard is enabled for swap with no particular
2830 * policy flagged, we set all swap discard flags here in
2831 * order to sustain backward compatibility with older
2832 * swapon(8) releases.
2833 */
2834 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2835 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002836
Shaohua Li2a8f9442013-09-11 14:20:28 -07002837 /*
2838 * By flagging sys_swapon, a sysadmin can tell us to
2839 * either do single-time area discards only, or to just
2840 * perform discards for released swap page-clusters.
2841 * Now it's time to adjust the p->flags accordingly.
2842 */
2843 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2844 p->flags &= ~SWP_PAGE_DISCARD;
2845 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2846 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002847
Shaohua Li2a8f9442013-09-11 14:20:28 -07002848 /* issue a swapon-time discard if it's still required */
2849 if (p->flags & SWP_AREA_DISCARD) {
2850 int err = discard_swap(p);
2851 if (unlikely(err))
2852 pr_err("swapon: discard_swap(%p): %d\n",
2853 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002854 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002855 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002856
Huang, Ying4b3ef9d2017-02-22 15:45:26 -08002857 error = init_swap_address_space(p->type, maxpages);
2858 if (error)
2859 goto bad_swap;
2860
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002861 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002862 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002863 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002864 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002865 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002866 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002867
Joe Perches756a0252016-03-17 14:19:47 -07002868 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 -04002869 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002870 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2871 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002872 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002873 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2874 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002875 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002876
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002877 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002878 atomic_inc(&proc_poll_event);
2879 wake_up_interruptible(&proc_poll_wait);
2880
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002881 if (S_ISREG(inode->i_mode))
2882 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 error = 0;
2884 goto out;
2885bad_swap:
Shaohua Liebc2a1a62013-09-11 14:20:32 -07002886 free_percpu(p->percpu_cluster);
2887 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002888 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002889 set_blocksize(p->bdev, p->old_block_size);
2890 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002892 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002893 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002894 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002897 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002899 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002900 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002901 if (inode && S_ISREG(inode->i_mode)) {
Al Viro59551022016-01-22 15:40:57 -05002902 inode_unlock(inode);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002903 inode = NULL;
2904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907out:
2908 if (page && !IS_ERR(page)) {
2909 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002910 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 }
2912 if (name)
2913 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002914 if (inode && S_ISREG(inode->i_mode))
Al Viro59551022016-01-22 15:40:57 -05002915 inode_unlock(inode);
Tim Chen039939a2017-02-22 15:45:43 -08002916 if (!error)
2917 enable_swap_slots_cache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 return error;
2919}
2920
2921void si_swapinfo(struct sysinfo *val)
2922{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002923 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 unsigned long nr_to_be_unused = 0;
2925
Hugh Dickins5d337b92005-09-03 15:54:41 -07002926 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002927 for (type = 0; type < nr_swapfiles; type++) {
2928 struct swap_info_struct *si = swap_info[type];
2929
2930 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2931 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002933 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002935 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936}
2937
2938/*
2939 * Verify that a swap entry is valid and increment its swap map count.
2940 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002941 * Returns error code in following case.
2942 * - success -> 0
2943 * - swp_entry is invalid -> EINVAL
2944 * - swp_entry is migration entry -> EINVAL
2945 * - swap-cache reference is requested but there is already one. -> EEXIST
2946 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002947 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002949static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002951 struct swap_info_struct *p;
Huang, Ying235b6212017-02-22 15:45:22 -08002952 struct swap_cluster_info *ci;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002954 unsigned char count;
2955 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002956 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957
Andi Kleena7420aa2009-09-16 11:50:05 +02002958 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002959 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002960
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 type = swp_type(entry);
2962 if (type >= nr_swapfiles)
2963 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002964 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 offset = swp_offset(entry);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002966 if (unlikely(offset >= p->max))
Huang, Ying235b6212017-02-22 15:45:22 -08002967 goto out;
2968
2969 ci = lock_cluster_or_swap_info(p, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002970
Hugh Dickins253d5532009-12-14 17:58:44 -08002971 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002972
2973 /*
2974 * swapin_readahead() doesn't check if a swap entry is valid, so the
2975 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2976 */
2977 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2978 err = -ENOENT;
2979 goto unlock_out;
2980 }
2981
Hugh Dickins253d5532009-12-14 17:58:44 -08002982 has_cache = count & SWAP_HAS_CACHE;
2983 count &= ~SWAP_HAS_CACHE;
2984 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002985
Hugh Dickins253d5532009-12-14 17:58:44 -08002986 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002987
2988 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002989 if (!has_cache && count)
2990 has_cache = SWAP_HAS_CACHE;
2991 else if (has_cache) /* someone else added cache */
2992 err = -EEXIST;
2993 else /* no users remaining */
2994 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002995
2996 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002997
Hugh Dickins570a335b2009-12-14 17:58:46 -08002998 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2999 count += usage;
3000 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08003001 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003002 else if (swap_count_continued(p, offset, count))
3003 count = COUNT_CONTINUED;
3004 else
3005 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003006 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08003007 err = -ENOENT; /* unused swap entry */
3008
3009 p->swap_map[offset] = count | has_cache;
3010
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003011unlock_out:
Huang, Ying235b6212017-02-22 15:45:22 -08003012 unlock_cluster_or_swap_info(p, ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013out:
Hugh Dickins253d5532009-12-14 17:58:44 -08003014 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015
3016bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07003017 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 goto out;
3019}
Hugh Dickins253d5532009-12-14 17:58:44 -08003020
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003021/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08003022 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3023 * (in which case its reference count is never incremented).
3024 */
3025void swap_shmem_alloc(swp_entry_t entry)
3026{
3027 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3028}
3029
3030/*
Hugh Dickins08259d52010-03-05 13:42:25 -08003031 * Increase reference count of swap entry by 1.
3032 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3033 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3034 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3035 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003036 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08003037int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003038{
Hugh Dickins570a335b2009-12-14 17:58:46 -08003039 int err = 0;
3040
3041 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3042 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3043 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003044}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07003046/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003047 * @entry: swap entry for which we allocate swap cache.
3048 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08003049 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07003050 * This can return error codes. Returns 0 at success.
3051 * -EBUSY means there is a swap cache.
3052 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07003053 */
3054int swapcache_prepare(swp_entry_t entry)
3055{
Hugh Dickins253d5532009-12-14 17:58:44 -08003056 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07003057}
3058
Mel Gormanf981c592012-07-31 16:44:47 -07003059struct swap_info_struct *page_swap_info(struct page *page)
3060{
3061 swp_entry_t swap = { .val = page_private(page) };
Mel Gormanf981c592012-07-31 16:44:47 -07003062 return swap_info[swp_type(swap)];
3063}
3064
3065/*
3066 * out-of-line __page_file_ methods to avoid include hell.
3067 */
3068struct address_space *__page_file_mapping(struct page *page)
3069{
Sasha Levin309381fea2014-01-23 15:52:54 -08003070 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07003071 return page_swap_info(page)->swap_file->f_mapping;
3072}
3073EXPORT_SYMBOL_GPL(__page_file_mapping);
3074
3075pgoff_t __page_file_index(struct page *page)
3076{
3077 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08003078 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07003079 return swp_offset(swap);
3080}
3081EXPORT_SYMBOL_GPL(__page_file_index);
3082
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08003084 * add_swap_count_continuation - called when a swap count is duplicated
3085 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3086 * page of the original vmalloc'ed swap_map, to hold the continuation count
3087 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3088 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3089 *
3090 * These continuation pages are seldom referenced: the common paths all work
3091 * on the original swap_map, only referring to a continuation page when the
3092 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3093 *
3094 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3095 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3096 * can be called after dropping locks.
3097 */
3098int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3099{
3100 struct swap_info_struct *si;
Huang, Ying235b6212017-02-22 15:45:22 -08003101 struct swap_cluster_info *ci;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003102 struct page *head;
3103 struct page *page;
3104 struct page *list_page;
3105 pgoff_t offset;
3106 unsigned char count;
3107
3108 /*
3109 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3110 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3111 */
3112 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3113
3114 si = swap_info_get(entry);
3115 if (!si) {
3116 /*
3117 * An acceptable race has occurred since the failing
3118 * __swap_duplicate(): the swap entry has been freed,
3119 * perhaps even the whole swap_map cleared for swapoff.
3120 */
3121 goto outer;
3122 }
3123
3124 offset = swp_offset(entry);
Huang, Ying235b6212017-02-22 15:45:22 -08003125
3126 ci = lock_cluster(si, offset);
3127
Hugh Dickins570a335b2009-12-14 17:58:46 -08003128 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3129
3130 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3131 /*
3132 * The higher the swap count, the more likely it is that tasks
3133 * will race to add swap count continuation: we need to avoid
3134 * over-provisioning.
3135 */
3136 goto out;
3137 }
3138
3139 if (!page) {
Huang, Ying235b6212017-02-22 15:45:22 -08003140 unlock_cluster(ci);
Shaohua Liec8acf22013-02-22 16:34:38 -08003141 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003142 return -ENOMEM;
3143 }
3144
3145 /*
3146 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08003147 * no architecture is using highmem pages for kernel page tables: so it
3148 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08003149 */
3150 head = vmalloc_to_page(si->swap_map + offset);
3151 offset &= ~PAGE_MASK;
3152
3153 /*
3154 * Page allocation does not initialize the page's lru field,
3155 * but it does always reset its private field.
3156 */
3157 if (!page_private(head)) {
3158 BUG_ON(count & COUNT_CONTINUED);
3159 INIT_LIST_HEAD(&head->lru);
3160 set_page_private(head, SWP_CONTINUED);
3161 si->flags |= SWP_CONTINUED;
3162 }
3163
3164 list_for_each_entry(list_page, &head->lru, lru) {
3165 unsigned char *map;
3166
3167 /*
3168 * If the previous map said no continuation, but we've found
3169 * a continuation page, free our allocation and use this one.
3170 */
3171 if (!(count & COUNT_CONTINUED))
3172 goto out;
3173
Cong Wang9b04c5f2011-11-25 23:14:39 +08003174 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003175 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08003176 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003177
3178 /*
3179 * If this continuation count now has some space in it,
3180 * free our allocation and use this one.
3181 */
3182 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3183 goto out;
3184 }
3185
3186 list_add_tail(&page->lru, &head->lru);
3187 page = NULL; /* now it's attached, don't free it */
3188out:
Huang, Ying235b6212017-02-22 15:45:22 -08003189 unlock_cluster(ci);
Shaohua Liec8acf22013-02-22 16:34:38 -08003190 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003191outer:
3192 if (page)
3193 __free_page(page);
3194 return 0;
3195}
3196
3197/*
3198 * swap_count_continued - when the original swap_map count is incremented
3199 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3200 * into, carry if so, or else fail until a new continuation page is allocated;
3201 * when the original swap_map count is decremented from 0 with continuation,
3202 * borrow from the continuation and report whether it still holds more.
Huang, Ying235b6212017-02-22 15:45:22 -08003203 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3204 * lock.
Hugh Dickins570a335b2009-12-14 17:58:46 -08003205 */
3206static bool swap_count_continued(struct swap_info_struct *si,
3207 pgoff_t offset, unsigned char count)
3208{
3209 struct page *head;
3210 struct page *page;
3211 unsigned char *map;
3212
3213 head = vmalloc_to_page(si->swap_map + offset);
3214 if (page_private(head) != SWP_CONTINUED) {
3215 BUG_ON(count & COUNT_CONTINUED);
3216 return false; /* need to add count continuation */
3217 }
3218
3219 offset &= ~PAGE_MASK;
3220 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08003221 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003222
3223 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3224 goto init_map; /* jump over SWAP_CONT_MAX checks */
3225
3226 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3227 /*
3228 * Think of how you add 1 to 999
3229 */
3230 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08003231 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003232 page = list_entry(page->lru.next, struct page, lru);
3233 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08003234 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003235 }
3236 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08003237 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003238 page = list_entry(page->lru.next, struct page, lru);
3239 if (page == head)
3240 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08003241 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003242init_map: *map = 0; /* we didn't zero the page */
3243 }
3244 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08003245 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003246 page = list_entry(page->lru.prev, struct page, lru);
3247 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08003248 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003249 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08003250 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003251 page = list_entry(page->lru.prev, struct page, lru);
3252 }
3253 return true; /* incremented */
3254
3255 } else { /* decrementing */
3256 /*
3257 * Think of how you subtract 1 from 1000
3258 */
3259 BUG_ON(count != COUNT_CONTINUED);
3260 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08003261 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003262 page = list_entry(page->lru.next, struct page, lru);
3263 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08003264 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003265 }
3266 BUG_ON(*map == 0);
3267 *map -= 1;
3268 if (*map == 0)
3269 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08003270 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003271 page = list_entry(page->lru.prev, struct page, lru);
3272 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08003273 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08003274 *map = SWAP_CONT_MAX | count;
3275 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08003276 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003277 page = list_entry(page->lru.prev, struct page, lru);
3278 }
3279 return count == COUNT_CONTINUED;
3280 }
3281}
3282
3283/*
3284 * free_swap_count_continuations - swapoff free all the continuation pages
3285 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3286 */
3287static void free_swap_count_continuations(struct swap_info_struct *si)
3288{
3289 pgoff_t offset;
3290
3291 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3292 struct page *head;
3293 head = vmalloc_to_page(si->swap_map + offset);
3294 if (page_private(head)) {
Geliang Tang0d576d22016-01-14 15:21:49 -08003295 struct page *page, *next;
3296
3297 list_for_each_entry_safe(page, next, &head->lru, lru) {
3298 list_del(&page->lru);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003299 __free_page(page);
3300 }
3301 }
3302 }
3303}