blob: 9cf2595268a4bfda4615d4797e35f0ab22905529 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/mm.h>
9#include <linux/hugetlb.h>
10#include <linux/mman.h>
11#include <linux/slab.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/vmalloc.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
Hugh Dickins072441e2011-06-27 16:18:02 -070017#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080019#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080024#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070032#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070033#include <linux/oom.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060034#include <linux/frontswap.h>
35#include <linux/swapfile.h>
Mel Gormanf981c592012-07-31 16:44:47 -070036#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/pgtable.h>
39#include <asm/tlbflush.h>
40#include <linux/swapops.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080041#include <linux/swap_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Hugh Dickins570a335b2009-12-14 17:58:46 -080043static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
44 unsigned char);
45static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080046static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080047
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060048DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070049static unsigned int nr_swapfiles;
Shaohua Liec8acf22013-02-22 16:34:38 -080050atomic_long_t nr_swap_pages;
Chris Wilsonfb0fec52015-12-04 15:58:53 +000051/*
52 * Some modules use swappable objects and may try to swap them out under
53 * memory pressure (via the shrinker). Before doing so, they may wish to
54 * check to see if any swap space is available.
55 */
56EXPORT_SYMBOL_GPL(nr_swap_pages);
Shaohua Liec8acf22013-02-22 16:34:38 -080057/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070058long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070059static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061static const char Bad_file[] = "Bad swap file entry ";
62static const char Unused_file[] = "Unused swap file entry ";
63static const char Bad_offset[] = "Bad swap offset entry ";
64static const char Unused_offset[] = "Unused swap offset entry ";
65
Dan Streetmanadfab832014-06-04 16:09:53 -070066/*
67 * all active swap_info_structs
68 * protected with swap_lock, and ordered by priority.
69 */
Dan Streetman18ab4d42014-06-04 16:09:59 -070070PLIST_HEAD(swap_active_head);
71
72/*
73 * all available (active, not full) swap_info_structs
74 * protected with swap_avail_lock, ordered by priority.
75 * This is used by get_swap_page() instead of swap_active_head
76 * because swap_active_head includes all swap_info_structs,
77 * but get_swap_page() doesn't need to look at full ones.
78 * This uses its own lock instead of swap_lock because when a
79 * swap_info_struct changes between not-full/full, it needs to
80 * add/remove itself to/from this list, but the swap_info_struct->lock
81 * is held and the locking order requires swap_lock to be taken
82 * before any swap_info_struct->lock.
83 */
Vinayak Menonc1c6a992015-12-21 13:00:58 +053084PLIST_HEAD(swap_avail_head);
85DEFINE_SPINLOCK(swap_avail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060087struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Ingo Molnarfc0abb12006-01-18 17:42:33 -080089static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Kay Sievers66d7dd52010-10-26 14:22:06 -070091static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
92/* Activity counter to indicate that a swapon or swapoff has occurred */
93static atomic_t proc_poll_event = ATOMIC_INIT(0);
94
Hugh Dickins8d69aae2009-12-14 17:58:45 -080095static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070096{
Hugh Dickins570a335b2009-12-14 17:58:46 -080097 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070098}
99
Vinayak Menonbb730a12015-02-25 19:43:59 +0530100bool is_swap_fast(swp_entry_t entry)
101{
102 struct swap_info_struct *p;
103 unsigned long type;
104
105 if (non_swap_entry(entry))
106 return false;
107
108 type = swp_type(entry);
109 if (type >= nr_swapfiles)
110 return false;
111
112 p = swap_info[type];
113
114 if (p->flags & SWP_FAST)
115 return true;
116
117 return false;
118}
119
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800120/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700121static int
122__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
123{
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800124 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700125 struct page *page;
126 int ret = 0;
127
Huang Yingf6ab1f72016-10-07 17:00:21 -0700128 page = find_get_page(swap_address_space(entry), swp_offset(entry));
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700129 if (!page)
130 return 0;
131 /*
132 * This function is called from scan_swap_map() and it's called
133 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
134 * We have to use trylock for avoiding deadlock. This is a special
135 * case and you should use try_to_free_swap() with explicit lock_page()
136 * in usual operations.
137 */
138 if (trylock_page(page)) {
139 ret = try_to_free_swap(page);
140 unlock_page(page);
141 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300142 put_page(page);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700143 return ret;
144}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700145
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800147 * swapon tell device that all the old swap contents can be discarded,
148 * to allow the swap device to optimize its wear-levelling.
149 */
150static int discard_swap(struct swap_info_struct *si)
151{
152 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800153 sector_t start_block;
154 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800155 int err = 0;
156
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800157 /* Do not discard the swap header page! */
158 se = &si->first_swap_extent;
159 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
160 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
161 if (nr_blocks) {
162 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200163 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800164 if (err)
165 return err;
166 cond_resched();
167 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800168
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800169 list_for_each_entry(se, &si->first_swap_extent.list, list) {
170 start_block = se->start_block << (PAGE_SHIFT - 9);
171 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800172
173 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200174 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800175 if (err)
176 break;
177
178 cond_resched();
179 }
180 return err; /* That will often be -EOPNOTSUPP */
181}
182
Hugh Dickins7992fde2009-01-06 14:39:53 -0800183/*
184 * swap allocation tell device that a cluster of swap can now be discarded,
185 * to allow the swap device to optimize its wear-levelling.
186 */
187static void discard_swap_cluster(struct swap_info_struct *si,
188 pgoff_t start_page, pgoff_t nr_pages)
189{
190 struct swap_extent *se = si->curr_swap_extent;
191 int found_extent = 0;
192
193 while (nr_pages) {
Hugh Dickins7992fde2009-01-06 14:39:53 -0800194 if (se->start_page <= start_page &&
195 start_page < se->start_page + se->nr_pages) {
196 pgoff_t offset = start_page - se->start_page;
197 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800198 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800199
200 if (nr_blocks > nr_pages)
201 nr_blocks = nr_pages;
202 start_page += nr_blocks;
203 nr_pages -= nr_blocks;
204
205 if (!found_extent++)
206 si->curr_swap_extent = se;
207
208 start_block <<= PAGE_SHIFT - 9;
209 nr_blocks <<= PAGE_SHIFT - 9;
210 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200211 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800212 break;
213 }
214
Geliang Tanga8ae4992016-01-14 15:20:45 -0800215 se = list_next_entry(se, list);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800216 }
217}
218
Hugh Dickins048c27f2005-09-03 15:54:40 -0700219#define LATENCY_LIMIT 256
220
Shaohua Li2a8f9442013-09-11 14:20:28 -0700221static inline void cluster_set_flag(struct swap_cluster_info *info,
222 unsigned int flag)
223{
224 info->flags = flag;
225}
226
227static inline unsigned int cluster_count(struct swap_cluster_info *info)
228{
229 return info->data;
230}
231
232static inline void cluster_set_count(struct swap_cluster_info *info,
233 unsigned int c)
234{
235 info->data = c;
236}
237
238static inline void cluster_set_count_flag(struct swap_cluster_info *info,
239 unsigned int c, unsigned int f)
240{
241 info->flags = f;
242 info->data = c;
243}
244
245static inline unsigned int cluster_next(struct swap_cluster_info *info)
246{
247 return info->data;
248}
249
250static inline void cluster_set_next(struct swap_cluster_info *info,
251 unsigned int n)
252{
253 info->data = n;
254}
255
256static inline void cluster_set_next_flag(struct swap_cluster_info *info,
257 unsigned int n, unsigned int f)
258{
259 info->flags = f;
260 info->data = n;
261}
262
263static inline bool cluster_is_free(struct swap_cluster_info *info)
264{
265 return info->flags & CLUSTER_FLAG_FREE;
266}
267
268static inline bool cluster_is_null(struct swap_cluster_info *info)
269{
270 return info->flags & CLUSTER_FLAG_NEXT_NULL;
271}
272
273static inline void cluster_set_null(struct swap_cluster_info *info)
274{
275 info->flags = CLUSTER_FLAG_NEXT_NULL;
276 info->data = 0;
277}
278
Huang Ying6b534912016-10-07 16:58:42 -0700279static inline bool cluster_list_empty(struct swap_cluster_list *list)
280{
281 return cluster_is_null(&list->head);
282}
283
284static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
285{
286 return cluster_next(&list->head);
287}
288
289static void cluster_list_init(struct swap_cluster_list *list)
290{
291 cluster_set_null(&list->head);
292 cluster_set_null(&list->tail);
293}
294
295static void cluster_list_add_tail(struct swap_cluster_list *list,
296 struct swap_cluster_info *ci,
297 unsigned int idx)
298{
299 if (cluster_list_empty(list)) {
300 cluster_set_next_flag(&list->head, idx, 0);
301 cluster_set_next_flag(&list->tail, idx, 0);
302 } else {
303 unsigned int tail = cluster_next(&list->tail);
304
305 cluster_set_next(&ci[tail], idx);
306 cluster_set_next_flag(&list->tail, idx, 0);
307 }
308}
309
310static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
311 struct swap_cluster_info *ci)
312{
313 unsigned int idx;
314
315 idx = cluster_next(&list->head);
316 if (cluster_next(&list->tail) == idx) {
317 cluster_set_null(&list->head);
318 cluster_set_null(&list->tail);
319 } else
320 cluster_set_next_flag(&list->head,
321 cluster_next(&ci[idx]), 0);
322
323 return idx;
324}
325
Shaohua Li815c2c52013-09-11 14:20:30 -0700326/* Add a cluster to discard list and schedule it to do discard */
327static void swap_cluster_schedule_discard(struct swap_info_struct *si,
328 unsigned int idx)
329{
330 /*
331 * If scan_swap_map() can't find a free cluster, it will check
332 * si->swap_map directly. To make sure the discarding cluster isn't
333 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
334 * will be cleared after discard
335 */
336 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
337 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
338
Huang Ying6b534912016-10-07 16:58:42 -0700339 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
Shaohua Li815c2c52013-09-11 14:20:30 -0700340
341 schedule_work(&si->discard_work);
342}
343
344/*
345 * Doing discard actually. After a cluster discard is finished, the cluster
346 * will be added to free cluster list. caller should hold si->lock.
347*/
348static void swap_do_scheduled_discard(struct swap_info_struct *si)
349{
350 struct swap_cluster_info *info;
351 unsigned int idx;
352
353 info = si->cluster_info;
354
Huang Ying6b534912016-10-07 16:58:42 -0700355 while (!cluster_list_empty(&si->discard_clusters)) {
356 idx = cluster_list_del_first(&si->discard_clusters, info);
Shaohua Li815c2c52013-09-11 14:20:30 -0700357 spin_unlock(&si->lock);
358
359 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
360 SWAPFILE_CLUSTER);
361
362 spin_lock(&si->lock);
363 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -0700364 cluster_list_add_tail(&si->free_clusters, info, idx);
Shaohua Li815c2c52013-09-11 14:20:30 -0700365 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
366 0, SWAPFILE_CLUSTER);
367 }
368}
369
370static void swap_discard_work(struct work_struct *work)
371{
372 struct swap_info_struct *si;
373
374 si = container_of(work, struct swap_info_struct, discard_work);
375
376 spin_lock(&si->lock);
377 swap_do_scheduled_discard(si);
378 spin_unlock(&si->lock);
379}
380
Shaohua Li2a8f9442013-09-11 14:20:28 -0700381/*
382 * The cluster corresponding to page_nr will be used. The cluster will be
383 * removed from free cluster list and its usage counter will be increased.
384 */
385static void inc_cluster_info_page(struct swap_info_struct *p,
386 struct swap_cluster_info *cluster_info, unsigned long page_nr)
387{
388 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
389
390 if (!cluster_info)
391 return;
392 if (cluster_is_free(&cluster_info[idx])) {
Huang Ying6b534912016-10-07 16:58:42 -0700393 VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx);
394 cluster_list_del_first(&p->free_clusters, cluster_info);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700395 cluster_set_count_flag(&cluster_info[idx], 0, 0);
396 }
397
398 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
399 cluster_set_count(&cluster_info[idx],
400 cluster_count(&cluster_info[idx]) + 1);
401}
402
403/*
404 * The cluster corresponding to page_nr decreases one usage. If the usage
405 * counter becomes 0, which means no page in the cluster is in using, we can
406 * optionally discard the cluster and add it to free cluster list.
407 */
408static void dec_cluster_info_page(struct swap_info_struct *p,
409 struct swap_cluster_info *cluster_info, unsigned long page_nr)
410{
411 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
412
413 if (!cluster_info)
414 return;
415
416 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
417 cluster_set_count(&cluster_info[idx],
418 cluster_count(&cluster_info[idx]) - 1);
419
420 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700421 /*
422 * If the swap is discardable, prepare discard the cluster
423 * instead of free it immediately. The cluster will be freed
424 * after discard.
425 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700426 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
427 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700428 swap_cluster_schedule_discard(p, idx);
429 return;
430 }
431
Shaohua Li2a8f9442013-09-11 14:20:28 -0700432 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -0700433 cluster_list_add_tail(&p->free_clusters, cluster_info, idx);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700434 }
435}
436
437/*
438 * It's possible scan_swap_map() uses a free cluster in the middle of free
439 * cluster list. Avoiding such abuse to avoid list corruption.
440 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700441static bool
442scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700443 unsigned long offset)
444{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700445 struct percpu_cluster *percpu_cluster;
446 bool conflict;
447
Shaohua Li2a8f9442013-09-11 14:20:28 -0700448 offset /= SWAPFILE_CLUSTER;
Huang Ying6b534912016-10-07 16:58:42 -0700449 conflict = !cluster_list_empty(&si->free_clusters) &&
450 offset != cluster_list_first(&si->free_clusters) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700451 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700452
453 if (!conflict)
454 return false;
455
456 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
457 cluster_set_null(&percpu_cluster->index);
458 return true;
459}
460
461/*
462 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
463 * might involve allocating a new cluster for current CPU too.
464 */
465static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
466 unsigned long *offset, unsigned long *scan_base)
467{
468 struct percpu_cluster *cluster;
469 bool found_free;
470 unsigned long tmp;
471
472new_cluster:
473 cluster = this_cpu_ptr(si->percpu_cluster);
474 if (cluster_is_null(&cluster->index)) {
Huang Ying6b534912016-10-07 16:58:42 -0700475 if (!cluster_list_empty(&si->free_clusters)) {
476 cluster->index = si->free_clusters.head;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700477 cluster->next = cluster_next(&cluster->index) *
478 SWAPFILE_CLUSTER;
Huang Ying6b534912016-10-07 16:58:42 -0700479 } else if (!cluster_list_empty(&si->discard_clusters)) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700480 /*
481 * we don't have free cluster but have some clusters in
482 * discarding, do discard now and reclaim them
483 */
484 swap_do_scheduled_discard(si);
485 *scan_base = *offset = si->cluster_next;
486 goto new_cluster;
487 } else
488 return;
489 }
490
491 found_free = false;
492
493 /*
494 * Other CPUs can use our cluster if they can't find a free cluster,
495 * check if there is still free entry in the cluster
496 */
497 tmp = cluster->next;
498 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
499 SWAPFILE_CLUSTER) {
500 if (!si->swap_map[tmp]) {
501 found_free = true;
502 break;
503 }
504 tmp++;
505 }
506 if (!found_free) {
507 cluster_set_null(&cluster->index);
508 goto new_cluster;
509 }
510 cluster->next = tmp + 1;
511 *offset = tmp;
512 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700513}
514
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700515static unsigned long scan_swap_map(struct swap_info_struct *si,
516 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800518 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800519 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800520 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700521 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800523 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700524 * We try to cluster swap pages by allocating them sequentially
525 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
526 * way, however, we resort to first-free allocation, starting
527 * a new cluster. This prevents us from scattering swap pages
528 * all over the entire swap partition, so that we reduce
529 * overall disk seek times between swap pages. -- sct
530 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800531 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700532 */
533
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700534 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800535 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800536
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700537 /* SSD algorithm */
538 if (si->cluster_info) {
539 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
540 goto checks;
541 }
542
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800543 if (unlikely(!si->cluster_nr--)) {
544 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
545 si->cluster_nr = SWAPFILE_CLUSTER - 1;
546 goto checks;
547 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700548
Shaohua Liec8acf22013-02-22 16:34:38 -0800549 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700550
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800551 /*
552 * If seek is expensive, start searching for new cluster from
553 * start of partition, to minimize the span of allocated swap.
Chen Yucong50088c42014-06-04 16:10:57 -0700554 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
555 * case, just handled by scan_swap_map_try_ssd_cluster() above.
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800556 */
Chen Yucong50088c42014-06-04 16:10:57 -0700557 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700558 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
559
560 /* Locate the first empty (unaligned) cluster */
561 for (; last_in_cluster <= si->highest_bit; offset++) {
562 if (si->swap_map[offset])
563 last_in_cluster = offset + SWAPFILE_CLUSTER;
564 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800565 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800566 offset -= SWAPFILE_CLUSTER - 1;
567 si->cluster_next = offset;
568 si->cluster_nr = SWAPFILE_CLUSTER - 1;
569 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700571 if (unlikely(--latency_ration < 0)) {
572 cond_resched();
573 latency_ration = LATENCY_LIMIT;
574 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700575 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800576
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800577 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800578 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800579 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700581
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800582checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700583 if (si->cluster_info) {
584 while (scan_swap_map_ssd_cluster_conflict(si, offset))
585 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
586 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800587 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700588 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700589 if (!si->highest_bit)
590 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800591 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800592 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700593
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700594 /* reuse swap entry of cache-only swap if not busy. */
Vinayak Menonbb730a12015-02-25 19:43:59 +0530595 if (vm_swap_full(si) && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700596 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800597 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700598 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800599 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700600 /* entry was freed successfully, try to use this again */
601 if (swap_was_freed)
602 goto checks;
603 goto scan; /* check next one */
604 }
605
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800606 if (si->swap_map[offset])
607 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700608
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800609 if (offset == si->lowest_bit)
610 si->lowest_bit++;
611 if (offset == si->highest_bit)
612 si->highest_bit--;
613 si->inuse_pages++;
614 if (si->inuse_pages == si->pages) {
615 si->lowest_bit = si->max;
616 si->highest_bit = 0;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700617 spin_lock(&swap_avail_lock);
618 plist_del(&si->avail_list, &swap_avail_head);
619 spin_unlock(&swap_avail_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800620 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800621 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700622 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800623 si->cluster_next = offset + 1;
624 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800625
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800626 return offset;
627
628scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800629 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700630 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700631 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800632 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700633 goto checks;
634 }
Vinayak Menonbb730a12015-02-25 19:43:59 +0530635 if (vm_swap_full(si) &&
636 si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800637 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700638 goto checks;
639 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700640 if (unlikely(--latency_ration < 0)) {
641 cond_resched();
642 latency_ration = LATENCY_LIMIT;
643 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700644 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800645 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800646 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800647 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800648 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800649 goto checks;
650 }
Vinayak Menonbb730a12015-02-25 19:43:59 +0530651 if (vm_swap_full(si) &&
652 si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800653 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700654 goto checks;
655 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800656 if (unlikely(--latency_ration < 0)) {
657 cond_resched();
658 latency_ration = LATENCY_LIMIT;
659 }
Jamie Liua5998062014-01-23 15:53:40 -0800660 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800661 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800662 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700663
664no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700665 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return 0;
667}
668
669swp_entry_t get_swap_page(void)
670{
Dan Streetmanadfab832014-06-04 16:09:53 -0700671 struct swap_info_struct *si, *next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700672 pgoff_t offset;
Vinayak Menonc1c6a992015-12-21 13:00:58 +0530673 int swap_ratio_off = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Shaohua Liec8acf22013-02-22 16:34:38 -0800675 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700676 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800677 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Vinayak Menonc1c6a992015-12-21 13:00:58 +0530679lock_and_start:
Dan Streetman18ab4d42014-06-04 16:09:59 -0700680 spin_lock(&swap_avail_lock);
681
682start_over:
683 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
Vinayak Menonc1c6a992015-12-21 13:00:58 +0530684
685 if (sysctl_swap_ratio && !swap_ratio_off) {
686 int ret;
687
688 spin_unlock(&swap_avail_lock);
689 ret = swap_ratio(&si);
690 if (ret < 0) {
691 /*
692 * Error. Start again with swap
693 * ratio disabled.
694 */
695 swap_ratio_off = 1;
696 goto lock_and_start;
697 } else {
698 goto start;
699 }
700 }
701
Dan Streetman18ab4d42014-06-04 16:09:59 -0700702 /* requeue si to after same-priority siblings */
703 plist_requeue(&si->avail_list, &swap_avail_head);
704 spin_unlock(&swap_avail_lock);
Vinayak Menonc1c6a992015-12-21 13:00:58 +0530705start:
Shaohua Liec8acf22013-02-22 16:34:38 -0800706 spin_lock(&si->lock);
Dan Streetmanadfab832014-06-04 16:09:53 -0700707 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
Dan Streetman18ab4d42014-06-04 16:09:59 -0700708 spin_lock(&swap_avail_lock);
709 if (plist_node_empty(&si->avail_list)) {
710 spin_unlock(&si->lock);
711 goto nextsi;
712 }
713 WARN(!si->highest_bit,
714 "swap_info %d in list but !highest_bit\n",
715 si->type);
716 WARN(!(si->flags & SWP_WRITEOK),
717 "swap_info %d in list but !SWP_WRITEOK\n",
718 si->type);
719 plist_del(&si->avail_list, &swap_avail_head);
Shaohua Liec8acf22013-02-22 16:34:38 -0800720 spin_unlock(&si->lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700721 goto nextsi;
Shaohua Liec8acf22013-02-22 16:34:38 -0800722 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700723
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700724 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800725 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800726 spin_unlock(&si->lock);
727 if (offset)
Dan Streetmanadfab832014-06-04 16:09:53 -0700728 return swp_entry(si->type, offset);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700729 pr_debug("scan_swap_map of si %d failed to find offset\n",
730 si->type);
731 spin_lock(&swap_avail_lock);
732nextsi:
Dan Streetmanadfab832014-06-04 16:09:53 -0700733 /*
734 * if we got here, it's likely that si was almost full before,
735 * and since scan_swap_map() can drop the si->lock, multiple
736 * callers probably all tried to get a page from the same si
Dan Streetman18ab4d42014-06-04 16:09:59 -0700737 * and it filled up before we could get one; or, the si filled
738 * up between us dropping swap_avail_lock and taking si->lock.
739 * Since we dropped the swap_avail_lock, the swap_avail_head
740 * list may have been modified; so if next is still in the
741 * swap_avail_head list then try it, otherwise start over.
Dan Streetmanadfab832014-06-04 16:09:53 -0700742 */
Dan Streetman18ab4d42014-06-04 16:09:59 -0700743 if (plist_node_empty(&next->avail_list))
744 goto start_over;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700746
Dan Streetman18ab4d42014-06-04 16:09:59 -0700747 spin_unlock(&swap_avail_lock);
748
Shaohua Liec8acf22013-02-22 16:34:38 -0800749 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700750noswap:
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700751 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800754/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700755swp_entry_t get_swap_page_of_type(int type)
756{
757 struct swap_info_struct *si;
758 pgoff_t offset;
759
Hugh Dickins910321e2010-09-09 16:38:07 -0700760 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800761 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700762 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800763 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700764 /* This is called for allocating swap entry, not cache */
765 offset = scan_swap_map(si, 1);
766 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800767 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700768 return swp_entry(type, offset);
769 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800770 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700771 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800772 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700773 return (swp_entry_t) {0};
774}
775
Hugh Dickins73c34b62009-12-14 17:58:43 -0800776static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800778 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 unsigned long offset, type;
780
781 if (!entry.val)
782 goto out;
783 type = swp_type(entry);
784 if (type >= nr_swapfiles)
785 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800786 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 if (!(p->flags & SWP_USED))
788 goto bad_device;
789 offset = swp_offset(entry);
790 if (offset >= p->max)
791 goto bad_offset;
792 if (!p->swap_map[offset])
793 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800794 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 return p;
796
797bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700798 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 goto out;
800bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700801 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 goto out;
803bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700804 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 goto out;
806bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700807 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808out:
809 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800810}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800812static unsigned char swap_entry_free(struct swap_info_struct *p,
813 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
Hugh Dickins253d5532009-12-14 17:58:44 -0800815 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800816 unsigned char count;
817 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700819 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800820 has_cache = count & SWAP_HAS_CACHE;
821 count &= ~SWAP_HAS_CACHE;
822
823 if (usage == SWAP_HAS_CACHE) {
824 VM_BUG_ON(!has_cache);
825 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800826 } else if (count == SWAP_MAP_SHMEM) {
827 /*
828 * Or we could insist on shmem.c using a special
829 * swap_shmem_free() and free_shmem_swap_and_cache()...
830 */
831 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800832 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
833 if (count == COUNT_CONTINUED) {
834 if (swap_count_continued(p, offset, count))
835 count = SWAP_MAP_MAX | COUNT_CONTINUED;
836 else
837 count = SWAP_MAP_MAX;
838 } else
839 count--;
840 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800841
Hugh Dickins253d5532009-12-14 17:58:44 -0800842 usage = count | has_cache;
843 p->swap_map[offset] = usage;
844
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700845 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800846 if (!usage) {
Vladimir Davydov37e84352016-01-20 15:02:56 -0800847 mem_cgroup_uncharge_swap(entry);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700848 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700849 if (offset < p->lowest_bit)
850 p->lowest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700851 if (offset > p->highest_bit) {
852 bool was_full = !p->highest_bit;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700853 p->highest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700854 if (was_full && (p->flags & SWP_WRITEOK)) {
855 spin_lock(&swap_avail_lock);
856 WARN_ON(!plist_node_empty(&p->avail_list));
857 if (plist_node_empty(&p->avail_list))
858 plist_add(&p->avail_list,
859 &swap_avail_head);
860 spin_unlock(&swap_avail_lock);
861 }
862 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800863 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700864 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600865 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700866 if (p->flags & SWP_BLKDEV) {
867 struct gendisk *disk = p->bdev->bd_disk;
868 if (disk->fops->swap_slot_free_notify)
869 disk->fops->swap_slot_free_notify(p->bdev,
870 offset);
871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800873
874 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800878 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 * is still around or has not been recycled.
880 */
881void swap_free(swp_entry_t entry)
882{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800883 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885 p = swap_info_get(entry);
886 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800887 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800888 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890}
891
892/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700893 * Called after dropping swapcache to decrease refcnt to swap entries.
894 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700895void swapcache_free(swp_entry_t entry)
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700896{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700897 struct swap_info_struct *p;
898
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700899 p = swap_info_get(entry);
900 if (p) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700901 swap_entry_free(p, entry, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800902 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700903 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700904}
905
906/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700907 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800908 * This does not give an exact answer when swap count is continued,
909 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700911int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700913 int count = 0;
914 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 swp_entry_t entry;
916
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700917 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 p = swap_info_get(entry);
919 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700920 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800921 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700923 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924}
925
926/*
Minchan Kim8334b962015-09-08 15:00:24 -0700927 * How many references to @entry are currently swapped out?
928 * This considers COUNT_CONTINUED so it returns exact answer.
929 */
930int swp_swapcount(swp_entry_t entry)
931{
932 int count, tmp_count, n;
933 struct swap_info_struct *p;
934 struct page *page;
935 pgoff_t offset;
936 unsigned char *map;
937
938 p = swap_info_get(entry);
939 if (!p)
940 return 0;
941
942 count = swap_count(p->swap_map[swp_offset(entry)]);
943 if (!(count & COUNT_CONTINUED))
944 goto out;
945
946 count &= ~COUNT_CONTINUED;
947 n = SWAP_MAP_MAX + 1;
948
949 offset = swp_offset(entry);
950 page = vmalloc_to_page(p->swap_map + offset);
951 offset &= ~PAGE_MASK;
952 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
953
954 do {
Geliang Tanga8ae4992016-01-14 15:20:45 -0800955 page = list_next_entry(page, lru);
Minchan Kim8334b962015-09-08 15:00:24 -0700956 map = kmap_atomic(page);
957 tmp_count = map[offset];
958 kunmap_atomic(map);
959
960 count += (tmp_count & ~COUNT_CONTINUED) * n;
961 n *= (SWAP_CONT_MAX + 1);
962 } while (tmp_count & COUNT_CONTINUED);
963out:
964 spin_unlock(&p->lock);
965 return count;
966}
967
968/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800969 * We can write to an anon page without COW if there are no other references
970 * to it. And as a side-effect, free up its swap: because the old content
971 * on disk will never be read, and seeking back there to write new content
972 * later would only waste time away from clustering.
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700973 *
974 * NOTE: total_mapcount should not be relied upon by the caller if
975 * reuse_swap_page() returns false, but it may be always overwritten
976 * (see the other implementation for CONFIG_SWAP=n).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 */
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700978bool reuse_swap_page(struct page *page, int *total_mapcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700980 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Sasha Levin309381fea2014-01-23 15:52:54 -0800982 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800983 if (unlikely(PageKsm(page)))
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700984 return false;
985 count = page_trans_huge_mapcount(page, total_mapcount);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800986 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700987 count += page_swapcount(page);
Minchan Kim6ca29ee2017-01-10 16:58:15 -0800988 if (count != 1)
989 goto out;
990 if (!PageWriteback(page)) {
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800991 delete_from_swap_cache(page);
992 SetPageDirty(page);
Minchan Kim6ca29ee2017-01-10 16:58:15 -0800993 } else {
994 swp_entry_t entry;
995 struct swap_info_struct *p;
996
997 entry.val = page_private(page);
998 p = swap_info_get(entry);
999 if (p->flags & SWP_STABLE_WRITES) {
1000 spin_unlock(&p->lock);
1001 return false;
1002 }
1003 spin_unlock(&p->lock);
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001004 }
1005 }
Minchan Kim6ca29ee2017-01-10 16:58:15 -08001006out:
Hugh Dickins5ad64682009-12-14 17:59:24 -08001007 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008}
1009
1010/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001011 * If swap is getting full, or if there are no more mappings of this page,
1012 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001014int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015{
Sasha Levin309381fea2014-01-23 15:52:54 -08001016 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 if (!PageSwapCache(page))
1019 return 0;
1020 if (PageWriteback(page))
1021 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001022 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 return 0;
1024
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -07001025 /*
1026 * Once hibernation has begun to create its image of memory,
1027 * there's a danger that one of the calls to try_to_free_swap()
1028 * - most probably a call from __try_to_reclaim_swap() while
1029 * hibernation is allocating its own swap pages for the image,
1030 * but conceivably even a call from memory reclaim - will free
1031 * the swap from a page which has already been recorded in the
1032 * image as a clean swapcache page, and then reuse its swap for
1033 * another page of the image. On waking from hibernation, the
1034 * original page might be freed under memory pressure, then
1035 * later read back in from swap, now with the wrong data.
1036 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001037 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -08001038 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -07001039 */
Mel Gormanf90ac392012-01-10 15:07:15 -08001040 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -07001041 return 0;
1042
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001043 delete_from_swap_cache(page);
1044 SetPageDirty(page);
1045 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -07001046}
1047
1048/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 * Free the swap entry like above, but also try to
1050 * free the page cache entry if it is the last user.
1051 */
Hugh Dickins2509ef22009-01-06 14:40:10 -08001052int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
Hugh Dickins2509ef22009-01-06 14:40:10 -08001054 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 struct page *page = NULL;
1056
Andi Kleena7420aa2009-09-16 11:50:05 +02001057 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -08001058 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -07001059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 p = swap_info_get(entry);
1061 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -08001062 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -08001063 page = find_get_page(swap_address_space(entry),
Huang Yingf6ab1f72016-10-07 17:00:21 -07001064 swp_offset(entry));
Nick Piggin8413ac92008-10-18 20:26:59 -07001065 if (page && !trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001066 put_page(page);
Nick Piggin93fac702006-03-31 02:29:56 -08001067 page = NULL;
1068 }
1069 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001070 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 }
1072 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001073 /*
1074 * Not mapped elsewhere, or swap space full? Free it!
1075 * Also recheck PageSwapCache now page is locked (above).
1076 */
Nick Piggin93fac702006-03-31 02:29:56 -08001077 if (PageSwapCache(page) && !PageWriteback(page) &&
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08001078 (!page_mapped(page) || mem_cgroup_swap_full(page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 delete_from_swap_cache(page);
1080 SetPageDirty(page);
1081 }
1082 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001083 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 }
Hugh Dickins2509ef22009-01-06 14:40:10 -08001085 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086}
1087
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001088#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001089/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001090 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001091 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001092 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1093 * from 0, in which the swap header is expected to be located.
1094 *
1095 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001096 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001097int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001098{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001099 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001100 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001101
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001102 if (device)
1103 bdev = bdget(device);
1104
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001105 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001106 for (type = 0; type < nr_swapfiles; type++) {
1107 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001108
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001109 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001110 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001111
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001112 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001113 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001114 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001115
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001116 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001117 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001118 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001119 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001120 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001121
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001122 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001123 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001124 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001125
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001126 spin_unlock(&swap_lock);
1127 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001128 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001129 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001130 }
1131 }
1132 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001133 if (bdev)
1134 bdput(bdev);
1135
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001136 return -ENODEV;
1137}
1138
1139/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001140 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1141 * corresponding to given index in swap_info (swap type).
1142 */
1143sector_t swapdev_block(int type, pgoff_t offset)
1144{
1145 struct block_device *bdev;
1146
1147 if ((unsigned int)type >= nr_swapfiles)
1148 return 0;
1149 if (!(swap_info[type]->flags & SWP_WRITEOK))
1150 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001151 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001152}
1153
1154/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001155 * Return either the total number of swap pages of given type, or the number
1156 * of free pages of that type (depending on @free)
1157 *
1158 * This is needed for software suspend
1159 */
1160unsigned int count_swap_pages(int type, int free)
1161{
1162 unsigned int n = 0;
1163
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001164 spin_lock(&swap_lock);
1165 if ((unsigned int)type < nr_swapfiles) {
1166 struct swap_info_struct *sis = swap_info[type];
1167
Shaohua Liec8acf22013-02-22 16:34:38 -08001168 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001169 if (sis->flags & SWP_WRITEOK) {
1170 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001171 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001172 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001173 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001174 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001175 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001176 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001177 return n;
1178}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001179#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001180
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001181static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001182{
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001183 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001184}
1185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001187 * No need to decide whether this PTE shares the swap entry with others,
1188 * just let do_wp_page work it out if a write is requested later - to
1189 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001191static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 unsigned long addr, swp_entry_t entry, struct page *page)
1193{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001194 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001195 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001196 spinlock_t *ptl;
1197 pte_t *pte;
1198 int ret = 1;
1199
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001200 swapcache = page;
1201 page = ksm_might_need_to_copy(page, vma, addr);
1202 if (unlikely(!page))
1203 return -ENOMEM;
1204
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001205 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1206 &memcg, false)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001207 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001208 goto out_nolock;
1209 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001210
1211 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001212 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001213 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001214 ret = 0;
1215 goto out;
1216 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001217
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001218 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001219 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 get_page(page);
1221 set_pte_at(vma->vm_mm, addr, pte,
1222 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001223 if (page == swapcache) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001224 page_add_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001225 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001226 } else { /* ksm created a completely new copy */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001227 page_add_new_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001228 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001229 lru_cache_add_active_or_unevictable(page, vma);
1230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 swap_free(entry);
1232 /*
1233 * Move the page to the active list so it is not
1234 * immediately swapped out again after swapon.
1235 */
1236 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001237out:
1238 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001239out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001240 if (page != swapcache) {
1241 unlock_page(page);
1242 put_page(page);
1243 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001244 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245}
1246
1247static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1248 unsigned long addr, unsigned long end,
1249 swp_entry_t entry, struct page *page)
1250{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001252 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001253 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Hugh Dickins044d66c2008-02-07 00:14:04 -08001255 /*
1256 * We don't actually need pte lock while scanning for swp_pte: since
1257 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1258 * page table while we're scanning; though it could get zapped, and on
1259 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1260 * of unmatched parts which look like swp_pte, so unuse_pte must
1261 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001262 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001263 */
1264 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 do {
1266 /*
1267 * swapoff spends a _lot_ of time in this loop!
1268 * Test inline before going to call unuse_pte.
1269 */
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001270 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001271 pte_unmap(pte);
1272 ret = unuse_pte(vma, pmd, addr, entry, page);
1273 if (ret)
1274 goto out;
1275 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
1277 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001278 pte_unmap(pte - 1);
1279out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001280 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281}
1282
1283static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1284 unsigned long addr, unsigned long end,
1285 swp_entry_t entry, struct page *page)
1286{
1287 pmd_t *pmd;
1288 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001289 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 pmd = pmd_offset(pud, addr);
1292 do {
1293 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001294 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001296 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1297 if (ret)
1298 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 } while (pmd++, addr = next, addr != end);
1300 return 0;
1301}
1302
1303static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1304 unsigned long addr, unsigned long end,
1305 swp_entry_t entry, struct page *page)
1306{
1307 pud_t *pud;
1308 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001309 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311 pud = pud_offset(pgd, addr);
1312 do {
1313 next = pud_addr_end(addr, end);
1314 if (pud_none_or_clear_bad(pud))
1315 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001316 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1317 if (ret)
1318 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 } while (pud++, addr = next, addr != end);
1320 return 0;
1321}
1322
1323static int unuse_vma(struct vm_area_struct *vma,
1324 swp_entry_t entry, struct page *page)
1325{
1326 pgd_t *pgd;
1327 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001328 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001330 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 addr = page_address_in_vma(page, vma);
1332 if (addr == -EFAULT)
1333 return 0;
1334 else
1335 end = addr + PAGE_SIZE;
1336 } else {
1337 addr = vma->vm_start;
1338 end = vma->vm_end;
1339 }
1340
1341 pgd = pgd_offset(vma->vm_mm, addr);
1342 do {
1343 next = pgd_addr_end(addr, end);
1344 if (pgd_none_or_clear_bad(pgd))
1345 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001346 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1347 if (ret)
1348 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 } while (pgd++, addr = next, addr != end);
1350 return 0;
1351}
1352
1353static int unuse_mm(struct mm_struct *mm,
1354 swp_entry_t entry, struct page *page)
1355{
1356 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001357 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 if (!down_read_trylock(&mm->mmap_sem)) {
1360 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001361 * Activate page so shrink_inactive_list is unlikely to unmap
1362 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001364 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 unlock_page(page);
1366 down_read(&mm->mmap_sem);
1367 lock_page(page);
1368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001370 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 break;
1372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001374 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
1377/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001378 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1379 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 * Recycle to start on reaching the end, returning 0 when empty.
1381 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001382static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001383 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001385 unsigned int max = si->max;
1386 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001387 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
1389 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001390 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 * for whether an entry is in use, not modifying it; false
1392 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001393 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 */
1395 for (;;) {
1396 if (++i >= max) {
1397 if (!prev) {
1398 i = 0;
1399 break;
1400 }
1401 /*
1402 * No entries in use at top of swap_map,
1403 * loop back to start and recheck there.
1404 */
1405 max = prev + 1;
1406 prev = 0;
1407 i = 1;
1408 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001409 if (frontswap) {
1410 if (frontswap_test(si, i))
1411 break;
1412 else
1413 continue;
1414 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001415 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001416 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 break;
1418 }
1419 return i;
1420}
1421
1422/*
1423 * We completely avoid races by reading each swap page in advance,
1424 * and then search for the process using it. All the necessary
1425 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001426 *
1427 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1428 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001430int try_to_unuse(unsigned int type, bool frontswap,
1431 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001433 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001435 volatile unsigned char *swap_map; /* swap_map is accessed without
1436 * locking. Mark it as volatile
1437 * to prevent compiler doing
1438 * something odd.
1439 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001440 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 struct page *page;
1442 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001443 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
1446 /*
1447 * When searching mms for an entry, a good strategy is to
1448 * start at the first mm we freed the previous entry from
1449 * (though actually we don't notice whether we or coincidence
1450 * freed the entry). Initialize this start_mm with a hold.
1451 *
1452 * A simpler strategy would be to start at the last mm we
1453 * freed the previous entry from; but that would take less
1454 * advantage of mmlist ordering, which clusters forked mms
1455 * together, child after parent. If we race with dup_mmap(), we
1456 * prefer to resolve parent before child, lest we miss entries
1457 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001458 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 */
1460 start_mm = &init_mm;
1461 atomic_inc(&init_mm.mm_users);
1462
1463 /*
1464 * Keep on scanning until all entries have gone. Usually,
1465 * one pass through swap_map is enough, but not necessarily:
1466 * there are races when an instance of an entry might be missed.
1467 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001468 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (signal_pending(current)) {
1470 retval = -EINTR;
1471 break;
1472 }
1473
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001474 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 * Get a page for the entry, using the existing swap
1476 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001477 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 */
1479 swap_map = &si->swap_map[i];
1480 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001481 page = read_swap_cache_async(entry,
1482 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 if (!page) {
1484 /*
1485 * Either swap_duplicate() failed because entry
1486 * has been freed independently, and will not be
1487 * reused since sys_swapoff() already disabled
1488 * allocation from here, or alloc_page() failed.
1489 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001490 swcount = *swap_map;
1491 /*
1492 * We don't hold lock here, so the swap entry could be
1493 * SWAP_MAP_BAD (when the cluster is discarding).
1494 * Instead of fail out, We can just skip the swap
1495 * entry because swapoff will wait for discarding
1496 * finish anyway.
1497 */
1498 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 continue;
1500 retval = -ENOMEM;
1501 break;
1502 }
1503
1504 /*
1505 * Don't hold on to start_mm if it looks like exiting.
1506 */
1507 if (atomic_read(&start_mm->mm_users) == 1) {
1508 mmput(start_mm);
1509 start_mm = &init_mm;
1510 atomic_inc(&init_mm.mm_users);
1511 }
1512
1513 /*
1514 * Wait for and lock page. When do_swap_page races with
1515 * try_to_unuse, do_swap_page can handle the fault much
1516 * faster than try_to_unuse can locate the entry. This
1517 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1518 * defer to do_swap_page in such a case - in some tests,
1519 * do_swap_page and try_to_unuse repeatedly compete.
1520 */
1521 wait_on_page_locked(page);
1522 wait_on_page_writeback(page);
1523 lock_page(page);
1524 wait_on_page_writeback(page);
1525
1526 /*
1527 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001530 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1531 retval = shmem_unuse(entry, page);
1532 /* page has already been unlocked and released */
1533 if (retval < 0)
1534 break;
1535 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001537 if (swap_count(swcount) && start_mm != &init_mm)
1538 retval = unuse_mm(start_mm, entry, page);
1539
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001540 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 int set_start_mm = (*swap_map >= swcount);
1542 struct list_head *p = &start_mm->mmlist;
1543 struct mm_struct *new_start_mm = start_mm;
1544 struct mm_struct *prev_mm = start_mm;
1545 struct mm_struct *mm;
1546
1547 atomic_inc(&new_start_mm->mm_users);
1548 atomic_inc(&prev_mm->mm_users);
1549 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001550 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 (p = p->next) != &start_mm->mmlist) {
1552 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001553 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 spin_unlock(&mmlist_lock);
1556 mmput(prev_mm);
1557 prev_mm = mm;
1558
1559 cond_resched();
1560
1561 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001562 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001564 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001566 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001568
Bo Liu32c5fc12009-11-02 16:50:33 +00001569 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 mmput(new_start_mm);
1571 atomic_inc(&mm->mm_users);
1572 new_start_mm = mm;
1573 set_start_mm = 0;
1574 }
1575 spin_lock(&mmlist_lock);
1576 }
1577 spin_unlock(&mmlist_lock);
1578 mmput(prev_mm);
1579 mmput(start_mm);
1580 start_mm = new_start_mm;
1581 }
1582 if (retval) {
1583 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001584 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 break;
1586 }
1587
1588 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 * If a reference remains (rare), we would like to leave
1590 * the page in the swap cache; but try_to_unmap could
1591 * then re-duplicate the entry once we drop page lock,
1592 * so we might loop indefinitely; also, that page could
1593 * not be swapped out to other storage meanwhile. So:
1594 * delete from cache even if there's another reference,
1595 * after ensuring that the data has been saved to disk -
1596 * since if the reference remains (rarer), it will be
1597 * read from disk into another page. Splitting into two
1598 * pages would be incorrect if swap supported "shared
1599 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001600 *
1601 * Given how unuse_vma() targets one particular offset
1602 * in an anon_vma, once the anon_vma has been determined,
1603 * this splitting happens to be just what is needed to
1604 * handle where KSM pages have been swapped out: re-reading
1605 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001607 if (swap_count(*swap_map) &&
1608 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 struct writeback_control wbc = {
1610 .sync_mode = WB_SYNC_NONE,
1611 };
1612
1613 swap_writepage(page, &wbc);
1614 lock_page(page);
1615 wait_on_page_writeback(page);
1616 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001617
1618 /*
1619 * It is conceivable that a racing task removed this page from
1620 * swap cache just before we acquired the page lock at the top,
1621 * or while we dropped it in unuse_mm(). The page might even
1622 * be back in swap cache on another swap area: that we must not
1623 * delete, since it may not have been written out to swap yet.
1624 */
1625 if (PageSwapCache(page) &&
1626 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001627 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 /*
1630 * So we could skip searching mms once swap count went
1631 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001632 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 */
1634 SetPageDirty(page);
1635 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001636 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
1638 /*
1639 * Make sure that we aren't completely killing
1640 * interactive performance.
1641 */
1642 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001643 if (frontswap && pages_to_unuse > 0) {
1644 if (!--pages_to_unuse)
1645 break;
1646 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 }
1648
1649 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 return retval;
1651}
1652
1653/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001654 * After a successful try_to_unuse, if no swap is now in use, we know
1655 * we can empty the mmlist. swap_lock must be held on entry and exit.
1656 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 * added to the mmlist just after page_duplicate - before would be racy.
1658 */
1659static void drain_mmlist(void)
1660{
1661 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001662 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001664 for (type = 0; type < nr_swapfiles; type++)
1665 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 return;
1667 spin_lock(&mmlist_lock);
1668 list_for_each_safe(p, next, &init_mm.mmlist)
1669 list_del_init(p);
1670 spin_unlock(&mmlist_lock);
1671}
1672
1673/*
1674 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001675 * corresponds to page offset for the specified swap entry.
1676 * Note that the type of this function is sector_t, but it returns page offset
1677 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001679static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001681 struct swap_info_struct *sis;
1682 struct swap_extent *start_se;
1683 struct swap_extent *se;
1684 pgoff_t offset;
1685
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001686 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001687 *bdev = sis->bdev;
1688
1689 offset = swp_offset(entry);
1690 start_se = sis->curr_swap_extent;
1691 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 if (se->start_page <= offset &&
1695 offset < (se->start_page + se->nr_pages)) {
1696 return se->start_block + (offset - se->start_page);
1697 }
Geliang Tanga8ae4992016-01-14 15:20:45 -08001698 se = list_next_entry(se, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 sis->curr_swap_extent = se;
1700 BUG_ON(se == start_se); /* It *must* be present */
1701 }
1702}
1703
1704/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001705 * Returns the page offset into bdev for the specified page's swap entry.
1706 */
1707sector_t map_swap_page(struct page *page, struct block_device **bdev)
1708{
1709 swp_entry_t entry;
1710 entry.val = page_private(page);
1711 return map_swap_entry(entry, bdev);
1712}
1713
1714/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 * Free all of a swapdev's extent information
1716 */
1717static void destroy_swap_extents(struct swap_info_struct *sis)
1718{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001719 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 struct swap_extent *se;
1721
Geliang Tanga8ae4992016-01-14 15:20:45 -08001722 se = list_first_entry(&sis->first_swap_extent.list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 struct swap_extent, list);
1724 list_del(&se->list);
1725 kfree(se);
1726 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001727
1728 if (sis->flags & SWP_FILE) {
1729 struct file *swap_file = sis->swap_file;
1730 struct address_space *mapping = swap_file->f_mapping;
1731
1732 sis->flags &= ~SWP_FILE;
1733 mapping->a_ops->swap_deactivate(swap_file);
1734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735}
1736
1737/*
1738 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001739 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001741 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 */
Mel Gormana509bc12012-07-31 16:44:57 -07001743int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1745 unsigned long nr_pages, sector_t start_block)
1746{
1747 struct swap_extent *se;
1748 struct swap_extent *new_se;
1749 struct list_head *lh;
1750
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001751 if (start_page == 0) {
1752 se = &sis->first_swap_extent;
1753 sis->curr_swap_extent = se;
1754 se->start_page = 0;
1755 se->nr_pages = nr_pages;
1756 se->start_block = start_block;
1757 return 1;
1758 } else {
1759 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001761 BUG_ON(se->start_page + se->nr_pages != start_page);
1762 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 /* Merge it */
1764 se->nr_pages += nr_pages;
1765 return 0;
1766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 }
1768
1769 /*
1770 * No merge. Insert a new extent, preserving ordering.
1771 */
1772 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1773 if (new_se == NULL)
1774 return -ENOMEM;
1775 new_se->start_page = start_page;
1776 new_se->nr_pages = nr_pages;
1777 new_se->start_block = start_block;
1778
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001779 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001780 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781}
1782
1783/*
1784 * A `swap extent' is a simple thing which maps a contiguous range of pages
1785 * onto a contiguous range of disk blocks. An ordered list of swap extents
1786 * is built at swapon time and is then used at swap_writepage/swap_readpage
1787 * time for locating where on disk a page belongs.
1788 *
1789 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1790 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1791 * swap files identically.
1792 *
1793 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1794 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1795 * swapfiles are handled *identically* after swapon time.
1796 *
1797 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1798 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1799 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1800 * requirements, they are simply tossed out - we will never use those blocks
1801 * for swapping.
1802 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001803 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1805 * which will scribble on the fs.
1806 *
1807 * The amount of disk space which a single swap extent represents varies.
1808 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1809 * extents in the list. To avoid much list walking, we cache the previous
1810 * search location in `curr_swap_extent', and start new searches from there.
1811 * This is extremely effective. The average number of iterations in
1812 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1813 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001814static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815{
Mel Gorman62c230b2012-07-31 16:44:55 -07001816 struct file *swap_file = sis->swap_file;
1817 struct address_space *mapping = swap_file->f_mapping;
1818 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 int ret;
1820
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 if (S_ISBLK(inode->i_mode)) {
1822 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001823 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001824 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 }
1826
Mel Gorman62c230b2012-07-31 16:44:55 -07001827 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001828 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001829 if (!ret) {
1830 sis->flags |= SWP_FILE;
1831 ret = add_swap_extent(sis, 0, sis->max, 0);
1832 *span = sis->pages;
1833 }
Mel Gormana509bc12012-07-31 16:44:57 -07001834 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001835 }
1836
Mel Gormana509bc12012-07-31 16:44:57 -07001837 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838}
1839
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001840static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001841 unsigned char *swap_map,
1842 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001843{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001844 if (prio >= 0)
1845 p->prio = prio;
1846 else
1847 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001848 /*
1849 * the plist prio is negated because plist ordering is
1850 * low-to-high, while swap ordering is high-to-low
1851 */
1852 p->list.prio = -p->prio;
1853 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001854 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001855 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001856 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001857 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001858 total_swap_pages += p->pages;
1859
Dan Streetmanadfab832014-06-04 16:09:53 -07001860 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001861 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07001862 * both lists are plists, and thus priority ordered.
1863 * swap_active_head needs to be priority ordered for swapoff(),
1864 * which on removal of any swap_info_struct with an auto-assigned
1865 * (i.e. negative) priority increments the auto-assigned priority
1866 * of any lower-priority swap_info_structs.
1867 * swap_avail_head needs to be priority ordered for get_swap_page(),
1868 * which allocates swap pages from the highest available priority
1869 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07001870 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07001871 plist_add(&p->list, &swap_active_head);
1872 spin_lock(&swap_avail_lock);
1873 plist_add(&p->avail_list, &swap_avail_head);
1874 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001875}
1876
1877static void enable_swap_info(struct swap_info_struct *p, int prio,
1878 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001879 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001880 unsigned long *frontswap_map)
1881{
Minchan Kim4f898492013-04-30 15:26:54 -07001882 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001883 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001884 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001885 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001886 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001887 spin_unlock(&swap_lock);
1888}
1889
1890static void reinsert_swap_info(struct swap_info_struct *p)
1891{
1892 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001893 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001894 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001895 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001896 spin_unlock(&swap_lock);
1897}
1898
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001899SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001901 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001902 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001903 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001904 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 struct file *swap_file, *victim;
1906 struct address_space *mapping;
1907 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001908 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07001909 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001910 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001911
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 if (!capable(CAP_SYS_ADMIN))
1913 return -EPERM;
1914
Al Viro191c5422012-02-13 03:58:52 +00001915 BUG_ON(!current->mm);
1916
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001919 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
Jeff Layton669abf42012-10-10 16:43:10 -04001921 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 err = PTR_ERR(victim);
1923 if (IS_ERR(victim))
1924 goto out;
1925
1926 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001927 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07001928 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001929 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001930 if (p->swap_file->f_mapping == mapping) {
1931 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07001933 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 }
Dan Streetmanadfab832014-06-04 16:09:53 -07001936 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001938 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 goto out_dput;
1940 }
Al Viro191c5422012-02-13 03:58:52 +00001941 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 vm_unacct_memory(p->pages);
1943 else {
1944 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001945 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 goto out_dput;
1947 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001948 spin_lock(&swap_avail_lock);
1949 plist_del(&p->avail_list, &swap_avail_head);
1950 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001951 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001952 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001953 struct swap_info_struct *si = p;
1954
Dan Streetman18ab4d42014-06-04 16:09:59 -07001955 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001956 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001957 si->list.prio--;
1958 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07001959 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001960 least_priority++;
1961 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001962 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08001963 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 total_swap_pages -= p->pages;
1965 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001966 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001967 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001968
David Rientjese1e12d22012-12-11 16:02:56 -08001969 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07001970 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08001971 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 if (err) {
1974 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001975 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 goto out_dput;
1977 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001978
Shaohua Li815c2c52013-09-11 14:20:30 -07001979 flush_work(&p->discard_work);
1980
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001981 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001982 if (p->flags & SWP_CONTINUED)
1983 free_swap_count_continuations(p);
1984
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001985 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001986 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001987 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001989
1990 /* wait for anyone still in scan_swap_map */
1991 p->highest_bit = 0; /* cuts scans short */
1992 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001993 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001994 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001995 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001996 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001997 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001998 }
1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07002001 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 p->swap_file = NULL;
2003 p->max = 0;
2004 swap_map = p->swap_map;
2005 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002006 cluster_info = p->cluster_info;
2007 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07002008 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08002009 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002010 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07002011 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08002012 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002013 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002014 free_percpu(p->percpu_cluster);
2015 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002017 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07002018 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002019 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07002020 swap_cgroup_swapoff(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002021
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 inode = mapping->host;
2023 if (S_ISBLK(inode->i_mode)) {
2024 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07002025 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01002026 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 } else {
Al Viro59551022016-01-22 15:40:57 -05002028 inode_lock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 inode->i_flags &= ~S_SWAPFILE;
Al Viro59551022016-01-22 15:40:57 -05002030 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 }
2032 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08002033
2034 /*
2035 * Clear the SWP_USED flag after all resources are freed so that swapon
2036 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2037 * not hold p->lock after we cleared its SWP_WRITEOK.
2038 */
2039 spin_lock(&swap_lock);
2040 p->flags = 0;
2041 spin_unlock(&swap_lock);
2042
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002044 atomic_inc(&proc_poll_event);
2045 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047out_dput:
2048 filp_close(victim, NULL);
2049out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08002050 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 return err;
2052}
2053
2054#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07002055static unsigned swaps_poll(struct file *file, poll_table *wait)
2056{
Kay Sieversf1514632011-07-12 20:48:39 +02002057 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002058
2059 poll_wait(file, &proc_poll_wait, wait);
2060
Kay Sieversf1514632011-07-12 20:48:39 +02002061 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2062 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002063 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2064 }
2065
2066 return POLLIN | POLLRDNORM;
2067}
2068
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069/* iterator */
2070static void *swap_start(struct seq_file *swap, loff_t *pos)
2071{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002072 struct swap_info_struct *si;
2073 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 loff_t l = *pos;
2075
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002076 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002078 if (!l)
2079 return SEQ_START_TOKEN;
2080
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002081 for (type = 0; type < nr_swapfiles; type++) {
2082 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2083 si = swap_info[type];
2084 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002086 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002087 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 }
2089
2090 return NULL;
2091}
2092
2093static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2094{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002095 struct swap_info_struct *si = v;
2096 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002098 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002099 type = 0;
2100 else
2101 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002102
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002103 for (; type < nr_swapfiles; type++) {
2104 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2105 si = swap_info[type];
2106 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 continue;
2108 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002109 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 }
2111
2112 return NULL;
2113}
2114
2115static void swap_stop(struct seq_file *swap, void *v)
2116{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002117 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118}
2119
2120static int swap_show(struct seq_file *swap, void *v)
2121{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002122 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 struct file *file;
2124 int len;
2125
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002126 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002127 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2128 return 0;
2129 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002131 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002132 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002133 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002134 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002135 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002137 si->pages << (PAGE_SHIFT - 10),
2138 si->inuse_pages << (PAGE_SHIFT - 10),
2139 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 return 0;
2141}
2142
Helge Deller15ad7cd2006-12-06 20:40:36 -08002143static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 .start = swap_start,
2145 .next = swap_next,
2146 .stop = swap_stop,
2147 .show = swap_show
2148};
2149
2150static int swaps_open(struct inode *inode, struct file *file)
2151{
Kay Sieversf1514632011-07-12 20:48:39 +02002152 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002153 int ret;
2154
Kay Sievers66d7dd52010-10-26 14:22:06 -07002155 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002156 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002157 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002158
Kay Sieversf1514632011-07-12 20:48:39 +02002159 seq = file->private_data;
2160 seq->poll_event = atomic_read(&proc_poll_event);
2161 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162}
2163
Helge Deller15ad7cd2006-12-06 20:40:36 -08002164static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 .open = swaps_open,
2166 .read = seq_read,
2167 .llseek = seq_lseek,
2168 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002169 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170};
2171
2172static int __init procswaps_init(void)
2173{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002174 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 return 0;
2176}
2177__initcall(procswaps_init);
2178#endif /* CONFIG_PROC_FS */
2179
Jan Beulich17963162008-12-16 11:35:24 +00002180#ifdef MAX_SWAPFILES_CHECK
2181static int __init max_swapfiles_check(void)
2182{
2183 MAX_SWAPFILES_CHECK();
2184 return 0;
2185}
2186late_initcall(max_swapfiles_check);
2187#endif
2188
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002189static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002191 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002193
2194 p = kzalloc(sizeof(*p), GFP_KERNEL);
2195 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002196 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002197
Hugh Dickins5d337b92005-09-03 15:54:41 -07002198 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002199 for (type = 0; type < nr_swapfiles; type++) {
2200 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002202 }
Christoph Lameter06972122006-06-23 02:03:35 -07002203 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002204 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002205 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002206 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002208 if (type >= nr_swapfiles) {
2209 p->type = type;
2210 swap_info[type] = p;
2211 /*
2212 * Write swap_info[type] before nr_swapfiles, in case a
2213 * racing procfs swap_start() or swap_next() is reading them.
2214 * (We never shrink nr_swapfiles, we never free this entry.)
2215 */
2216 smp_wmb();
2217 nr_swapfiles++;
2218 } else {
2219 kfree(p);
2220 p = swap_info[type];
2221 /*
2222 * Do not memset this entry: a racing procfs swap_next()
2223 * would be relying on p->type to remain valid.
2224 */
2225 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002226 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002227 plist_node_init(&p->list, 0);
2228 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002230 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002231 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002232
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002233 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002234}
2235
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002236static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2237{
2238 int error;
2239
2240 if (S_ISBLK(inode->i_mode)) {
2241 p->bdev = bdgrab(I_BDEV(inode));
2242 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002243 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002244 if (error < 0) {
2245 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002246 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002247 }
2248 p->old_block_size = block_size(p->bdev);
2249 error = set_blocksize(p->bdev, PAGE_SIZE);
2250 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002251 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002252 p->flags |= SWP_BLKDEV;
2253 } else if (S_ISREG(inode->i_mode)) {
2254 p->bdev = inode->i_sb->s_bdev;
Al Viro59551022016-01-22 15:40:57 -05002255 inode_lock(inode);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002256 if (IS_SWAPFILE(inode))
2257 return -EBUSY;
2258 } else
2259 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002260
2261 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002262}
2263
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002264static unsigned long read_swap_header(struct swap_info_struct *p,
2265 union swap_header *swap_header,
2266 struct inode *inode)
2267{
2268 int i;
2269 unsigned long maxpages;
2270 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002271 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002272
2273 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002274 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002275 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002276 }
2277
2278 /* swap partition endianess hack... */
2279 if (swab32(swap_header->info.version) == 1) {
2280 swab32s(&swap_header->info.version);
2281 swab32s(&swap_header->info.last_page);
2282 swab32s(&swap_header->info.nr_badpages);
Jann Horndd111be2016-11-10 10:46:19 -08002283 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2284 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002285 for (i = 0; i < swap_header->info.nr_badpages; i++)
2286 swab32s(&swap_header->info.badpages[i]);
2287 }
2288 /* Check the swap header's sub-version */
2289 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002290 pr_warn("Unable to handle swap header version %d\n",
2291 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002292 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002293 }
2294
2295 p->lowest_bit = 1;
2296 p->cluster_next = 1;
2297 p->cluster_nr = 0;
2298
2299 /*
2300 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002301 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002302 * of bits for the swap offset in the swp_entry_t type, and
2303 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002304 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002305 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002306 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002307 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002308 * offset is extracted. This will mask all the bits from
2309 * the initial ~0UL mask that can't be encoded in either
2310 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002311 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002312 */
2313 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002314 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002315 last_page = swap_header->info.last_page;
2316 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002317 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002318 maxpages << (PAGE_SHIFT - 10),
2319 last_page << (PAGE_SHIFT - 10));
2320 }
2321 if (maxpages > last_page) {
2322 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002323 /* p->max is an unsigned int: don't overflow it */
2324 if ((unsigned int)maxpages == 0)
2325 maxpages = UINT_MAX;
2326 }
2327 p->highest_bit = maxpages - 1;
2328
2329 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002330 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002331 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2332 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002333 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002334 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002335 }
2336 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002337 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002338 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002339 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002340
2341 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002342}
2343
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002344static int setup_swap_map_and_extents(struct swap_info_struct *p,
2345 union swap_header *swap_header,
2346 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002347 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002348 unsigned long maxpages,
2349 sector_t *span)
2350{
2351 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002352 unsigned int nr_good_pages;
2353 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002354 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2355 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002356
2357 nr_good_pages = maxpages - 1; /* omit header page */
2358
Huang Ying6b534912016-10-07 16:58:42 -07002359 cluster_list_init(&p->free_clusters);
2360 cluster_list_init(&p->discard_clusters);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002361
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002362 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2363 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002364 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2365 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002366 if (page_nr < maxpages) {
2367 swap_map[page_nr] = SWAP_MAP_BAD;
2368 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002369 /*
2370 * Haven't marked the cluster free yet, no list
2371 * operation involved
2372 */
2373 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002374 }
2375 }
2376
Shaohua Li2a8f9442013-09-11 14:20:28 -07002377 /* Haven't marked the cluster free yet, no list operation involved */
2378 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2379 inc_cluster_info_page(p, cluster_info, i);
2380
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002381 if (nr_good_pages) {
2382 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002383 /*
2384 * Not mark the cluster free yet, no list
2385 * operation involved
2386 */
2387 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002388 p->max = maxpages;
2389 p->pages = nr_good_pages;
2390 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002391 if (nr_extents < 0)
2392 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002393 nr_good_pages = p->pages;
2394 }
2395 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002396 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002397 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002398 }
2399
Shaohua Li2a8f9442013-09-11 14:20:28 -07002400 if (!cluster_info)
2401 return nr_extents;
2402
2403 for (i = 0; i < nr_clusters; i++) {
2404 if (!cluster_count(&cluster_info[idx])) {
2405 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
Huang Ying6b534912016-10-07 16:58:42 -07002406 cluster_list_add_tail(&p->free_clusters, cluster_info,
2407 idx);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002408 }
2409 idx++;
2410 if (idx == nr_clusters)
2411 idx = 0;
2412 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002413 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002414}
2415
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002416/*
2417 * Helper to sys_swapon determining if a given swap
2418 * backing device queue supports DISCARD operations.
2419 */
2420static bool swap_discardable(struct swap_info_struct *si)
2421{
2422 struct request_queue *q = bdev_get_queue(si->bdev);
2423
2424 if (!q || !blk_queue_discard(q))
2425 return false;
2426
2427 return true;
2428}
2429
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002430SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2431{
2432 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002433 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002434 struct file *swap_file = NULL;
2435 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002436 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002437 int error;
2438 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002439 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002440 sector_t span;
2441 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002442 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002443 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002444 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002445 struct page *page = NULL;
2446 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002447
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002448 if (swap_flags & ~SWAP_FLAGS_VALID)
2449 return -EINVAL;
2450
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002451 if (!capable(CAP_SYS_ADMIN))
2452 return -EPERM;
2453
2454 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002455 if (IS_ERR(p))
2456 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002457
Shaohua Li815c2c52013-09-11 14:20:30 -07002458 INIT_WORK(&p->discard_work, swap_discard_work);
2459
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002462 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002464 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 }
Jeff Layton669abf42012-10-10 16:43:10 -04002466 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002468 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002470 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 }
2472
2473 p->swap_file = swap_file;
2474 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002475 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002476
Al Viro59551022016-01-22 15:40:57 -05002477 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002478 error = claim_swapfile(p, inode);
2479 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 /*
2483 * Read the swap header.
2484 */
2485 if (!mapping->a_ops->readpage) {
2486 error = -EINVAL;
2487 goto bad_swap;
2488 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002489 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 if (IS_ERR(page)) {
2491 error = PTR_ERR(page);
2492 goto bad_swap;
2493 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002494 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002496 maxpages = read_swap_header(p, swap_header, inode);
2497 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 error = -EINVAL;
2499 goto bad_swap;
2500 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002501
Hugh Dickins81e33972009-01-06 14:39:49 -08002502 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002503 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002504 if (!swap_map) {
2505 error = -ENOMEM;
2506 goto bad_swap;
2507 }
Minchan Kim6ca29ee2017-01-10 16:58:15 -08002508
2509 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
2510 p->flags |= SWP_STABLE_WRITES;
2511
Shaohua Li2a8f9442013-09-11 14:20:28 -07002512 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002513 int cpu;
2514
Shaohua Li2a8f9442013-09-11 14:20:28 -07002515 p->flags |= SWP_SOLIDSTATE;
2516 /*
2517 * select a random position to start with to help wear leveling
2518 * SSD
2519 */
2520 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2521
2522 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2523 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2524 if (!cluster_info) {
2525 error = -ENOMEM;
2526 goto bad_swap;
2527 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002528 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2529 if (!p->percpu_cluster) {
2530 error = -ENOMEM;
2531 goto bad_swap;
2532 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002533 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002534 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002535 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002536 cluster_set_null(&cluster->index);
2537 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002538 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002539
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002540 error = swap_cgroup_swapon(p->type, maxpages);
2541 if (error)
2542 goto bad_swap;
2543
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002544 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002545 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002546 if (unlikely(nr_extents < 0)) {
2547 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 goto bad_swap;
2549 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002550 /* frontswap enabled? set up bit-per-page map for frontswap */
Vlastimil Babka8ea1d2a2016-07-26 15:24:42 -07002551 if (IS_ENABLED(CONFIG_FRONTSWAP))
Akinobu Mita7b579762013-06-12 14:05:08 -07002552 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
Shaohua Li2a8f9442013-09-11 14:20:28 -07002554 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2555 /*
2556 * When discard is enabled for swap with no particular
2557 * policy flagged, we set all swap discard flags here in
2558 * order to sustain backward compatibility with older
2559 * swapon(8) releases.
2560 */
2561 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2562 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002563
Shaohua Li2a8f9442013-09-11 14:20:28 -07002564 /*
2565 * By flagging sys_swapon, a sysadmin can tell us to
2566 * either do single-time area discards only, or to just
2567 * perform discards for released swap page-clusters.
2568 * Now it's time to adjust the p->flags accordingly.
2569 */
2570 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2571 p->flags &= ~SWP_PAGE_DISCARD;
2572 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2573 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002574
Shaohua Li2a8f9442013-09-11 14:20:28 -07002575 /* issue a swapon-time discard if it's still required */
2576 if (p->flags & SWP_AREA_DISCARD) {
2577 int err = discard_swap(p);
2578 if (unlikely(err))
2579 pr_err("swapon: discard_swap(%p): %d\n",
2580 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002581 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002582 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002583
Vinayak Menonee081c232015-11-20 13:37:08 +05302584 if (p->bdev && blk_queue_fast(bdev_get_queue(p->bdev)))
2585 p->flags |= SWP_FAST;
2586
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002587 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002588 prio = -1;
Vinayak Menonc1c6a992015-12-21 13:00:58 +05302589 if (swap_flags & SWAP_FLAG_PREFER) {
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002590 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002591 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Vinayak Menonc1c6a992015-12-21 13:00:58 +05302592 setup_swap_ratio(p, prio);
2593 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002594 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002595
Joe Perches756a025f02016-03-17 14:19:47 -07002596 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 -04002597 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002598 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2599 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002600 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002601 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2602 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002603 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002604
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002605 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002606 atomic_inc(&proc_poll_event);
2607 wake_up_interruptible(&proc_poll_wait);
2608
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002609 if (S_ISREG(inode->i_mode))
2610 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 error = 0;
2612 goto out;
2613bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002614 free_percpu(p->percpu_cluster);
2615 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002616 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002617 set_blocksize(p->bdev, p->old_block_size);
2618 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002620 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002621 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002622 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002625 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002627 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002628 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002629 if (inode && S_ISREG(inode->i_mode)) {
Al Viro59551022016-01-22 15:40:57 -05002630 inode_unlock(inode);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002631 inode = NULL;
2632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635out:
2636 if (page && !IS_ERR(page)) {
2637 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002638 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 }
2640 if (name)
2641 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002642 if (inode && S_ISREG(inode->i_mode))
Al Viro59551022016-01-22 15:40:57 -05002643 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 return error;
2645}
2646
2647void si_swapinfo(struct sysinfo *val)
2648{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002649 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 unsigned long nr_to_be_unused = 0;
2651
Hugh Dickins5d337b92005-09-03 15:54:41 -07002652 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002653 for (type = 0; type < nr_swapfiles; type++) {
2654 struct swap_info_struct *si = swap_info[type];
2655
2656 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2657 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002659 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002661 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662}
2663
2664/*
2665 * Verify that a swap entry is valid and increment its swap map count.
2666 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002667 * Returns error code in following case.
2668 * - success -> 0
2669 * - swp_entry is invalid -> EINVAL
2670 * - swp_entry is migration entry -> EINVAL
2671 * - swap-cache reference is requested but there is already one. -> EEXIST
2672 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002673 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002675static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002677 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002679 unsigned char count;
2680 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002681 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682
Andi Kleena7420aa2009-09-16 11:50:05 +02002683 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002684 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002685
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 type = swp_type(entry);
2687 if (type >= nr_swapfiles)
2688 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002689 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 offset = swp_offset(entry);
2691
Shaohua Liec8acf22013-02-22 16:34:38 -08002692 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002693 if (unlikely(offset >= p->max))
2694 goto unlock_out;
2695
Hugh Dickins253d5532009-12-14 17:58:44 -08002696 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002697
2698 /*
2699 * swapin_readahead() doesn't check if a swap entry is valid, so the
2700 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2701 */
2702 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2703 err = -ENOENT;
2704 goto unlock_out;
2705 }
2706
Hugh Dickins253d5532009-12-14 17:58:44 -08002707 has_cache = count & SWAP_HAS_CACHE;
2708 count &= ~SWAP_HAS_CACHE;
2709 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002710
Hugh Dickins253d5532009-12-14 17:58:44 -08002711 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002712
2713 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002714 if (!has_cache && count)
2715 has_cache = SWAP_HAS_CACHE;
2716 else if (has_cache) /* someone else added cache */
2717 err = -EEXIST;
2718 else /* no users remaining */
2719 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002720
2721 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002722
Hugh Dickins570a335b2009-12-14 17:58:46 -08002723 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2724 count += usage;
2725 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002726 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002727 else if (swap_count_continued(p, offset, count))
2728 count = COUNT_CONTINUED;
2729 else
2730 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002731 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002732 err = -ENOENT; /* unused swap entry */
2733
2734 p->swap_map[offset] = count | has_cache;
2735
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002736unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002737 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002739 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740
2741bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002742 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 goto out;
2744}
Hugh Dickins253d5532009-12-14 17:58:44 -08002745
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002746/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002747 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2748 * (in which case its reference count is never incremented).
2749 */
2750void swap_shmem_alloc(swp_entry_t entry)
2751{
2752 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2753}
2754
2755/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002756 * Increase reference count of swap entry by 1.
2757 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2758 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2759 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2760 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002761 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002762int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002763{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002764 int err = 0;
2765
2766 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2767 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2768 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002769}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002771/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002772 * @entry: swap entry for which we allocate swap cache.
2773 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002774 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002775 * This can return error codes. Returns 0 at success.
2776 * -EBUSY means there is a swap cache.
2777 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002778 */
2779int swapcache_prepare(swp_entry_t entry)
2780{
Hugh Dickins253d5532009-12-14 17:58:44 -08002781 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002782}
2783
Mel Gormanf981c592012-07-31 16:44:47 -07002784struct swap_info_struct *page_swap_info(struct page *page)
2785{
2786 swp_entry_t swap = { .val = page_private(page) };
Mel Gormanf981c592012-07-31 16:44:47 -07002787 return swap_info[swp_type(swap)];
2788}
2789
2790/*
2791 * out-of-line __page_file_ methods to avoid include hell.
2792 */
2793struct address_space *__page_file_mapping(struct page *page)
2794{
Sasha Levin309381fea2014-01-23 15:52:54 -08002795 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002796 return page_swap_info(page)->swap_file->f_mapping;
2797}
2798EXPORT_SYMBOL_GPL(__page_file_mapping);
2799
2800pgoff_t __page_file_index(struct page *page)
2801{
2802 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002803 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002804 return swp_offset(swap);
2805}
2806EXPORT_SYMBOL_GPL(__page_file_index);
2807
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002809 * add_swap_count_continuation - called when a swap count is duplicated
2810 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2811 * page of the original vmalloc'ed swap_map, to hold the continuation count
2812 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2813 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2814 *
2815 * These continuation pages are seldom referenced: the common paths all work
2816 * on the original swap_map, only referring to a continuation page when the
2817 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2818 *
2819 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2820 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2821 * can be called after dropping locks.
2822 */
2823int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2824{
2825 struct swap_info_struct *si;
2826 struct page *head;
2827 struct page *page;
2828 struct page *list_page;
2829 pgoff_t offset;
2830 unsigned char count;
2831
2832 /*
2833 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2834 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2835 */
2836 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2837
2838 si = swap_info_get(entry);
2839 if (!si) {
2840 /*
2841 * An acceptable race has occurred since the failing
2842 * __swap_duplicate(): the swap entry has been freed,
2843 * perhaps even the whole swap_map cleared for swapoff.
2844 */
2845 goto outer;
2846 }
2847
2848 offset = swp_offset(entry);
2849 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2850
2851 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2852 /*
2853 * The higher the swap count, the more likely it is that tasks
2854 * will race to add swap count continuation: we need to avoid
2855 * over-provisioning.
2856 */
2857 goto out;
2858 }
2859
2860 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002861 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002862 return -ENOMEM;
2863 }
2864
2865 /*
2866 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002867 * no architecture is using highmem pages for kernel page tables: so it
2868 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002869 */
2870 head = vmalloc_to_page(si->swap_map + offset);
2871 offset &= ~PAGE_MASK;
2872
2873 /*
2874 * Page allocation does not initialize the page's lru field,
2875 * but it does always reset its private field.
2876 */
2877 if (!page_private(head)) {
2878 BUG_ON(count & COUNT_CONTINUED);
2879 INIT_LIST_HEAD(&head->lru);
2880 set_page_private(head, SWP_CONTINUED);
2881 si->flags |= SWP_CONTINUED;
2882 }
2883
2884 list_for_each_entry(list_page, &head->lru, lru) {
2885 unsigned char *map;
2886
2887 /*
2888 * If the previous map said no continuation, but we've found
2889 * a continuation page, free our allocation and use this one.
2890 */
2891 if (!(count & COUNT_CONTINUED))
2892 goto out;
2893
Cong Wang9b04c5f2011-11-25 23:14:39 +08002894 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002895 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002896 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002897
2898 /*
2899 * If this continuation count now has some space in it,
2900 * free our allocation and use this one.
2901 */
2902 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2903 goto out;
2904 }
2905
2906 list_add_tail(&page->lru, &head->lru);
2907 page = NULL; /* now it's attached, don't free it */
2908out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002909 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002910outer:
2911 if (page)
2912 __free_page(page);
2913 return 0;
2914}
2915
2916/*
2917 * swap_count_continued - when the original swap_map count is incremented
2918 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2919 * into, carry if so, or else fail until a new continuation page is allocated;
2920 * when the original swap_map count is decremented from 0 with continuation,
2921 * borrow from the continuation and report whether it still holds more.
2922 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2923 */
2924static bool swap_count_continued(struct swap_info_struct *si,
2925 pgoff_t offset, unsigned char count)
2926{
2927 struct page *head;
2928 struct page *page;
2929 unsigned char *map;
2930
2931 head = vmalloc_to_page(si->swap_map + offset);
2932 if (page_private(head) != SWP_CONTINUED) {
2933 BUG_ON(count & COUNT_CONTINUED);
2934 return false; /* need to add count continuation */
2935 }
2936
2937 offset &= ~PAGE_MASK;
2938 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002939 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002940
2941 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2942 goto init_map; /* jump over SWAP_CONT_MAX checks */
2943
2944 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2945 /*
2946 * Think of how you add 1 to 999
2947 */
2948 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002949 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002950 page = list_entry(page->lru.next, struct page, lru);
2951 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002952 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002953 }
2954 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002955 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002956 page = list_entry(page->lru.next, struct page, lru);
2957 if (page == head)
2958 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002959 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002960init_map: *map = 0; /* we didn't zero the page */
2961 }
2962 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002963 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002964 page = list_entry(page->lru.prev, struct page, lru);
2965 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002966 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002967 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002968 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002969 page = list_entry(page->lru.prev, struct page, lru);
2970 }
2971 return true; /* incremented */
2972
2973 } else { /* decrementing */
2974 /*
2975 * Think of how you subtract 1 from 1000
2976 */
2977 BUG_ON(count != COUNT_CONTINUED);
2978 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002979 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002980 page = list_entry(page->lru.next, struct page, lru);
2981 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002982 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002983 }
2984 BUG_ON(*map == 0);
2985 *map -= 1;
2986 if (*map == 0)
2987 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002988 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002989 page = list_entry(page->lru.prev, struct page, lru);
2990 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002991 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002992 *map = SWAP_CONT_MAX | count;
2993 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002994 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002995 page = list_entry(page->lru.prev, struct page, lru);
2996 }
2997 return count == COUNT_CONTINUED;
2998 }
2999}
3000
3001/*
3002 * free_swap_count_continuations - swapoff free all the continuation pages
3003 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3004 */
3005static void free_swap_count_continuations(struct swap_info_struct *si)
3006{
3007 pgoff_t offset;
3008
3009 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3010 struct page *head;
3011 head = vmalloc_to_page(si->swap_map + offset);
3012 if (page_private(head)) {
Geliang Tang0d576d22016-01-14 15:21:49 -08003013 struct page *page, *next;
3014
3015 list_for_each_entry_safe(page, next, &head->lru, lru) {
3016 list_del(&page->lru);
Hugh Dickins570a335b2009-12-14 17:58:46 -08003017 __free_page(page);
3018 }
3019 }
3020 }
3021}