blob: 78cfa292a29aa0780e5e83efd7b7733ce958e605 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/mm.h>
9#include <linux/hugetlb.h>
10#include <linux/mman.h>
11#include <linux/slab.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/vmalloc.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
Hugh Dickins072441e2011-06-27 16:18:02 -070017#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080019#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080024#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070032#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070033#include <linux/oom.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060034#include <linux/frontswap.h>
35#include <linux/swapfile.h>
Mel Gormanf981c592012-07-31 16:44:47 -070036#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/pgtable.h>
39#include <asm/tlbflush.h>
40#include <linux/swapops.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080041#include <linux/swap_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Hugh Dickins570a335b2009-12-14 17:58:46 -080043static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
44 unsigned char);
45static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080046static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080047
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060048DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070049static unsigned int nr_swapfiles;
Shaohua Liec8acf22013-02-22 16:34:38 -080050atomic_long_t nr_swap_pages;
Chris Wilsonfb0fec52015-12-04 15:58:53 +000051/*
52 * Some modules use swappable objects and may try to swap them out under
53 * memory pressure (via the shrinker). Before doing so, they may wish to
54 * check to see if any swap space is available.
55 */
56EXPORT_SYMBOL_GPL(nr_swap_pages);
Shaohua Liec8acf22013-02-22 16:34:38 -080057/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070058long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070059static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061static const char Bad_file[] = "Bad swap file entry ";
62static const char Unused_file[] = "Unused swap file entry ";
63static const char Bad_offset[] = "Bad swap offset entry ";
64static const char Unused_offset[] = "Unused swap offset entry ";
65
Dan Streetmanadfab832014-06-04 16:09:53 -070066/*
67 * all active swap_info_structs
68 * protected with swap_lock, and ordered by priority.
69 */
Dan Streetman18ab4d42014-06-04 16:09:59 -070070PLIST_HEAD(swap_active_head);
71
72/*
73 * all available (active, not full) swap_info_structs
74 * protected with swap_avail_lock, ordered by priority.
75 * This is used by get_swap_page() instead of swap_active_head
76 * because swap_active_head includes all swap_info_structs,
77 * but get_swap_page() doesn't need to look at full ones.
78 * This uses its own lock instead of swap_lock because when a
79 * swap_info_struct changes between not-full/full, it needs to
80 * add/remove itself to/from this list, but the swap_info_struct->lock
81 * is held and the locking order requires swap_lock to be taken
82 * before any swap_info_struct->lock.
83 */
84static PLIST_HEAD(swap_avail_head);
85static DEFINE_SPINLOCK(swap_avail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060087struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Ingo Molnarfc0abb12006-01-18 17:42:33 -080089static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Kay Sievers66d7dd52010-10-26 14:22:06 -070091static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
92/* Activity counter to indicate that a swapon or swapoff has occurred */
93static atomic_t proc_poll_event = ATOMIC_INIT(0);
94
Hugh Dickins8d69aae2009-12-14 17:58:45 -080095static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070096{
Hugh Dickins570a335b2009-12-14 17:58:46 -080097 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070098}
99
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800100/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700101static int
102__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
103{
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800104 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700105 struct page *page;
106 int ret = 0;
107
Shaohua Li33806f02013-02-22 16:34:37 -0800108 page = find_get_page(swap_address_space(entry), entry.val);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700109 if (!page)
110 return 0;
111 /*
112 * This function is called from scan_swap_map() and it's called
113 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
114 * We have to use trylock for avoiding deadlock. This is a special
115 * case and you should use try_to_free_swap() with explicit lock_page()
116 * in usual operations.
117 */
118 if (trylock_page(page)) {
119 ret = try_to_free_swap(page);
120 unlock_page(page);
121 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300122 put_page(page);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700123 return ret;
124}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800127 * swapon tell device that all the old swap contents can be discarded,
128 * to allow the swap device to optimize its wear-levelling.
129 */
130static int discard_swap(struct swap_info_struct *si)
131{
132 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800133 sector_t start_block;
134 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800135 int err = 0;
136
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800137 /* Do not discard the swap header page! */
138 se = &si->first_swap_extent;
139 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
140 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
141 if (nr_blocks) {
142 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200143 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800144 if (err)
145 return err;
146 cond_resched();
147 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800148
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800149 list_for_each_entry(se, &si->first_swap_extent.list, list) {
150 start_block = se->start_block << (PAGE_SHIFT - 9);
151 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800152
153 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200154 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800155 if (err)
156 break;
157
158 cond_resched();
159 }
160 return err; /* That will often be -EOPNOTSUPP */
161}
162
Hugh Dickins7992fde2009-01-06 14:39:53 -0800163/*
164 * swap allocation tell device that a cluster of swap can now be discarded,
165 * to allow the swap device to optimize its wear-levelling.
166 */
167static void discard_swap_cluster(struct swap_info_struct *si,
168 pgoff_t start_page, pgoff_t nr_pages)
169{
170 struct swap_extent *se = si->curr_swap_extent;
171 int found_extent = 0;
172
173 while (nr_pages) {
Hugh Dickins7992fde2009-01-06 14:39:53 -0800174 if (se->start_page <= start_page &&
175 start_page < se->start_page + se->nr_pages) {
176 pgoff_t offset = start_page - se->start_page;
177 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800178 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800179
180 if (nr_blocks > nr_pages)
181 nr_blocks = nr_pages;
182 start_page += nr_blocks;
183 nr_pages -= nr_blocks;
184
185 if (!found_extent++)
186 si->curr_swap_extent = se;
187
188 start_block <<= PAGE_SHIFT - 9;
189 nr_blocks <<= PAGE_SHIFT - 9;
190 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200191 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800192 break;
193 }
194
Geliang Tanga8ae4992016-01-14 15:20:45 -0800195 se = list_next_entry(se, list);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800196 }
197}
198
Hugh Dickins048c27f2005-09-03 15:54:40 -0700199#define SWAPFILE_CLUSTER 256
200#define LATENCY_LIMIT 256
201
Shaohua Li2a8f9442013-09-11 14:20:28 -0700202static inline void cluster_set_flag(struct swap_cluster_info *info,
203 unsigned int flag)
204{
205 info->flags = flag;
206}
207
208static inline unsigned int cluster_count(struct swap_cluster_info *info)
209{
210 return info->data;
211}
212
213static inline void cluster_set_count(struct swap_cluster_info *info,
214 unsigned int c)
215{
216 info->data = c;
217}
218
219static inline void cluster_set_count_flag(struct swap_cluster_info *info,
220 unsigned int c, unsigned int f)
221{
222 info->flags = f;
223 info->data = c;
224}
225
226static inline unsigned int cluster_next(struct swap_cluster_info *info)
227{
228 return info->data;
229}
230
231static inline void cluster_set_next(struct swap_cluster_info *info,
232 unsigned int n)
233{
234 info->data = n;
235}
236
237static inline void cluster_set_next_flag(struct swap_cluster_info *info,
238 unsigned int n, unsigned int f)
239{
240 info->flags = f;
241 info->data = n;
242}
243
244static inline bool cluster_is_free(struct swap_cluster_info *info)
245{
246 return info->flags & CLUSTER_FLAG_FREE;
247}
248
249static inline bool cluster_is_null(struct swap_cluster_info *info)
250{
251 return info->flags & CLUSTER_FLAG_NEXT_NULL;
252}
253
254static inline void cluster_set_null(struct swap_cluster_info *info)
255{
256 info->flags = CLUSTER_FLAG_NEXT_NULL;
257 info->data = 0;
258}
259
Shaohua Li815c2c52013-09-11 14:20:30 -0700260/* Add a cluster to discard list and schedule it to do discard */
261static void swap_cluster_schedule_discard(struct swap_info_struct *si,
262 unsigned int idx)
263{
264 /*
265 * If scan_swap_map() can't find a free cluster, it will check
266 * si->swap_map directly. To make sure the discarding cluster isn't
267 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
268 * will be cleared after discard
269 */
270 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
271 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
272
273 if (cluster_is_null(&si->discard_cluster_head)) {
274 cluster_set_next_flag(&si->discard_cluster_head,
275 idx, 0);
276 cluster_set_next_flag(&si->discard_cluster_tail,
277 idx, 0);
278 } else {
279 unsigned int tail = cluster_next(&si->discard_cluster_tail);
280 cluster_set_next(&si->cluster_info[tail], idx);
281 cluster_set_next_flag(&si->discard_cluster_tail,
282 idx, 0);
283 }
284
285 schedule_work(&si->discard_work);
286}
287
288/*
289 * Doing discard actually. After a cluster discard is finished, the cluster
290 * will be added to free cluster list. caller should hold si->lock.
291*/
292static void swap_do_scheduled_discard(struct swap_info_struct *si)
293{
294 struct swap_cluster_info *info;
295 unsigned int idx;
296
297 info = si->cluster_info;
298
299 while (!cluster_is_null(&si->discard_cluster_head)) {
300 idx = cluster_next(&si->discard_cluster_head);
301
302 cluster_set_next_flag(&si->discard_cluster_head,
303 cluster_next(&info[idx]), 0);
304 if (cluster_next(&si->discard_cluster_tail) == idx) {
305 cluster_set_null(&si->discard_cluster_head);
306 cluster_set_null(&si->discard_cluster_tail);
307 }
308 spin_unlock(&si->lock);
309
310 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
311 SWAPFILE_CLUSTER);
312
313 spin_lock(&si->lock);
314 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
315 if (cluster_is_null(&si->free_cluster_head)) {
316 cluster_set_next_flag(&si->free_cluster_head,
317 idx, 0);
318 cluster_set_next_flag(&si->free_cluster_tail,
319 idx, 0);
320 } else {
321 unsigned int tail;
322
323 tail = cluster_next(&si->free_cluster_tail);
324 cluster_set_next(&info[tail], idx);
325 cluster_set_next_flag(&si->free_cluster_tail,
326 idx, 0);
327 }
328 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
329 0, SWAPFILE_CLUSTER);
330 }
331}
332
333static void swap_discard_work(struct work_struct *work)
334{
335 struct swap_info_struct *si;
336
337 si = container_of(work, struct swap_info_struct, discard_work);
338
339 spin_lock(&si->lock);
340 swap_do_scheduled_discard(si);
341 spin_unlock(&si->lock);
342}
343
Shaohua Li2a8f9442013-09-11 14:20:28 -0700344/*
345 * The cluster corresponding to page_nr will be used. The cluster will be
346 * removed from free cluster list and its usage counter will be increased.
347 */
348static void inc_cluster_info_page(struct swap_info_struct *p,
349 struct swap_cluster_info *cluster_info, unsigned long page_nr)
350{
351 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
352
353 if (!cluster_info)
354 return;
355 if (cluster_is_free(&cluster_info[idx])) {
356 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
357 cluster_set_next_flag(&p->free_cluster_head,
358 cluster_next(&cluster_info[idx]), 0);
359 if (cluster_next(&p->free_cluster_tail) == idx) {
360 cluster_set_null(&p->free_cluster_tail);
361 cluster_set_null(&p->free_cluster_head);
362 }
363 cluster_set_count_flag(&cluster_info[idx], 0, 0);
364 }
365
366 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
367 cluster_set_count(&cluster_info[idx],
368 cluster_count(&cluster_info[idx]) + 1);
369}
370
371/*
372 * The cluster corresponding to page_nr decreases one usage. If the usage
373 * counter becomes 0, which means no page in the cluster is in using, we can
374 * optionally discard the cluster and add it to free cluster list.
375 */
376static void dec_cluster_info_page(struct swap_info_struct *p,
377 struct swap_cluster_info *cluster_info, unsigned long page_nr)
378{
379 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
380
381 if (!cluster_info)
382 return;
383
384 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
385 cluster_set_count(&cluster_info[idx],
386 cluster_count(&cluster_info[idx]) - 1);
387
388 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700389 /*
390 * If the swap is discardable, prepare discard the cluster
391 * instead of free it immediately. The cluster will be freed
392 * after discard.
393 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700394 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
395 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700396 swap_cluster_schedule_discard(p, idx);
397 return;
398 }
399
Shaohua Li2a8f9442013-09-11 14:20:28 -0700400 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
401 if (cluster_is_null(&p->free_cluster_head)) {
402 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
403 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
404 } else {
405 unsigned int tail = cluster_next(&p->free_cluster_tail);
406 cluster_set_next(&cluster_info[tail], idx);
407 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
408 }
409 }
410}
411
412/*
413 * It's possible scan_swap_map() uses a free cluster in the middle of free
414 * cluster list. Avoiding such abuse to avoid list corruption.
415 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700416static bool
417scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700418 unsigned long offset)
419{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700420 struct percpu_cluster *percpu_cluster;
421 bool conflict;
422
Shaohua Li2a8f9442013-09-11 14:20:28 -0700423 offset /= SWAPFILE_CLUSTER;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700424 conflict = !cluster_is_null(&si->free_cluster_head) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700425 offset != cluster_next(&si->free_cluster_head) &&
426 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700427
428 if (!conflict)
429 return false;
430
431 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
432 cluster_set_null(&percpu_cluster->index);
433 return true;
434}
435
436/*
437 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
438 * might involve allocating a new cluster for current CPU too.
439 */
440static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
441 unsigned long *offset, unsigned long *scan_base)
442{
443 struct percpu_cluster *cluster;
444 bool found_free;
445 unsigned long tmp;
446
447new_cluster:
448 cluster = this_cpu_ptr(si->percpu_cluster);
449 if (cluster_is_null(&cluster->index)) {
450 if (!cluster_is_null(&si->free_cluster_head)) {
451 cluster->index = si->free_cluster_head;
452 cluster->next = cluster_next(&cluster->index) *
453 SWAPFILE_CLUSTER;
454 } else if (!cluster_is_null(&si->discard_cluster_head)) {
455 /*
456 * we don't have free cluster but have some clusters in
457 * discarding, do discard now and reclaim them
458 */
459 swap_do_scheduled_discard(si);
460 *scan_base = *offset = si->cluster_next;
461 goto new_cluster;
462 } else
463 return;
464 }
465
466 found_free = false;
467
468 /*
469 * Other CPUs can use our cluster if they can't find a free cluster,
470 * check if there is still free entry in the cluster
471 */
472 tmp = cluster->next;
473 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
474 SWAPFILE_CLUSTER) {
475 if (!si->swap_map[tmp]) {
476 found_free = true;
477 break;
478 }
479 tmp++;
480 }
481 if (!found_free) {
482 cluster_set_null(&cluster->index);
483 goto new_cluster;
484 }
485 cluster->next = tmp + 1;
486 *offset = tmp;
487 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700488}
489
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700490static unsigned long scan_swap_map(struct swap_info_struct *si,
491 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800493 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800494 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800495 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700496 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800498 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700499 * We try to cluster swap pages by allocating them sequentially
500 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
501 * way, however, we resort to first-free allocation, starting
502 * a new cluster. This prevents us from scattering swap pages
503 * all over the entire swap partition, so that we reduce
504 * overall disk seek times between swap pages. -- sct
505 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800506 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700507 */
508
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700509 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800510 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800511
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700512 /* SSD algorithm */
513 if (si->cluster_info) {
514 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
515 goto checks;
516 }
517
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800518 if (unlikely(!si->cluster_nr--)) {
519 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
520 si->cluster_nr = SWAPFILE_CLUSTER - 1;
521 goto checks;
522 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700523
Shaohua Liec8acf22013-02-22 16:34:38 -0800524 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700525
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800526 /*
527 * If seek is expensive, start searching for new cluster from
528 * start of partition, to minimize the span of allocated swap.
Chen Yucong50088c42014-06-04 16:10:57 -0700529 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
530 * case, just handled by scan_swap_map_try_ssd_cluster() above.
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800531 */
Chen Yucong50088c42014-06-04 16:10:57 -0700532 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700533 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
534
535 /* Locate the first empty (unaligned) cluster */
536 for (; last_in_cluster <= si->highest_bit; offset++) {
537 if (si->swap_map[offset])
538 last_in_cluster = offset + SWAPFILE_CLUSTER;
539 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800540 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800541 offset -= SWAPFILE_CLUSTER - 1;
542 si->cluster_next = offset;
543 si->cluster_nr = SWAPFILE_CLUSTER - 1;
544 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700546 if (unlikely(--latency_ration < 0)) {
547 cond_resched();
548 latency_ration = LATENCY_LIMIT;
549 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700550 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800551
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800552 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800553 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800554 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700556
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800557checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700558 if (si->cluster_info) {
559 while (scan_swap_map_ssd_cluster_conflict(si, offset))
560 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
561 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800562 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700563 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700564 if (!si->highest_bit)
565 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800566 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800567 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700568
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700569 /* reuse swap entry of cache-only swap if not busy. */
570 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700571 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800572 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700573 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800574 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700575 /* entry was freed successfully, try to use this again */
576 if (swap_was_freed)
577 goto checks;
578 goto scan; /* check next one */
579 }
580
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800581 if (si->swap_map[offset])
582 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700583
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800584 if (offset == si->lowest_bit)
585 si->lowest_bit++;
586 if (offset == si->highest_bit)
587 si->highest_bit--;
588 si->inuse_pages++;
589 if (si->inuse_pages == si->pages) {
590 si->lowest_bit = si->max;
591 si->highest_bit = 0;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700592 spin_lock(&swap_avail_lock);
593 plist_del(&si->avail_list, &swap_avail_head);
594 spin_unlock(&swap_avail_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800595 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800596 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700597 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800598 si->cluster_next = offset + 1;
599 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800600
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800601 return offset;
602
603scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800604 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700605 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700606 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800607 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700608 goto checks;
609 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700610 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800611 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700612 goto checks;
613 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700614 if (unlikely(--latency_ration < 0)) {
615 cond_resched();
616 latency_ration = LATENCY_LIMIT;
617 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700618 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800619 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800620 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800621 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800622 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800623 goto checks;
624 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700625 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800626 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700627 goto checks;
628 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800629 if (unlikely(--latency_ration < 0)) {
630 cond_resched();
631 latency_ration = LATENCY_LIMIT;
632 }
Jamie Liua5998062014-01-23 15:53:40 -0800633 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800634 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800635 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700636
637no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700638 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 return 0;
640}
641
642swp_entry_t get_swap_page(void)
643{
Dan Streetmanadfab832014-06-04 16:09:53 -0700644 struct swap_info_struct *si, *next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700645 pgoff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Shaohua Liec8acf22013-02-22 16:34:38 -0800647 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700648 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800649 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Dan Streetman18ab4d42014-06-04 16:09:59 -0700651 spin_lock(&swap_avail_lock);
652
653start_over:
654 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
655 /* requeue si to after same-priority siblings */
656 plist_requeue(&si->avail_list, &swap_avail_head);
657 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800658 spin_lock(&si->lock);
Dan Streetmanadfab832014-06-04 16:09:53 -0700659 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
Dan Streetman18ab4d42014-06-04 16:09:59 -0700660 spin_lock(&swap_avail_lock);
661 if (plist_node_empty(&si->avail_list)) {
662 spin_unlock(&si->lock);
663 goto nextsi;
664 }
665 WARN(!si->highest_bit,
666 "swap_info %d in list but !highest_bit\n",
667 si->type);
668 WARN(!(si->flags & SWP_WRITEOK),
669 "swap_info %d in list but !SWP_WRITEOK\n",
670 si->type);
671 plist_del(&si->avail_list, &swap_avail_head);
Shaohua Liec8acf22013-02-22 16:34:38 -0800672 spin_unlock(&si->lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700673 goto nextsi;
Shaohua Liec8acf22013-02-22 16:34:38 -0800674 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700675
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700676 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800677 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800678 spin_unlock(&si->lock);
679 if (offset)
Dan Streetmanadfab832014-06-04 16:09:53 -0700680 return swp_entry(si->type, offset);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700681 pr_debug("scan_swap_map of si %d failed to find offset\n",
682 si->type);
683 spin_lock(&swap_avail_lock);
684nextsi:
Dan Streetmanadfab832014-06-04 16:09:53 -0700685 /*
686 * if we got here, it's likely that si was almost full before,
687 * and since scan_swap_map() can drop the si->lock, multiple
688 * callers probably all tried to get a page from the same si
Dan Streetman18ab4d42014-06-04 16:09:59 -0700689 * and it filled up before we could get one; or, the si filled
690 * up between us dropping swap_avail_lock and taking si->lock.
691 * Since we dropped the swap_avail_lock, the swap_avail_head
692 * list may have been modified; so if next is still in the
693 * swap_avail_head list then try it, otherwise start over.
Dan Streetmanadfab832014-06-04 16:09:53 -0700694 */
Dan Streetman18ab4d42014-06-04 16:09:59 -0700695 if (plist_node_empty(&next->avail_list))
696 goto start_over;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700698
Dan Streetman18ab4d42014-06-04 16:09:59 -0700699 spin_unlock(&swap_avail_lock);
700
Shaohua Liec8acf22013-02-22 16:34:38 -0800701 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700702noswap:
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700703 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800706/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700707swp_entry_t get_swap_page_of_type(int type)
708{
709 struct swap_info_struct *si;
710 pgoff_t offset;
711
Hugh Dickins910321e2010-09-09 16:38:07 -0700712 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800713 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700714 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800715 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700716 /* This is called for allocating swap entry, not cache */
717 offset = scan_swap_map(si, 1);
718 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800719 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700720 return swp_entry(type, offset);
721 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800722 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700723 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800724 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700725 return (swp_entry_t) {0};
726}
727
Hugh Dickins73c34b62009-12-14 17:58:43 -0800728static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800730 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 unsigned long offset, type;
732
733 if (!entry.val)
734 goto out;
735 type = swp_type(entry);
736 if (type >= nr_swapfiles)
737 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800738 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (!(p->flags & SWP_USED))
740 goto bad_device;
741 offset = swp_offset(entry);
742 if (offset >= p->max)
743 goto bad_offset;
744 if (!p->swap_map[offset])
745 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800746 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 return p;
748
749bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700750 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 goto out;
752bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700753 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 goto out;
755bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700756 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 goto out;
758bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700759 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760out:
761 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800762}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800764static unsigned char swap_entry_free(struct swap_info_struct *p,
765 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
Hugh Dickins253d5532009-12-14 17:58:44 -0800767 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800768 unsigned char count;
769 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700771 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800772 has_cache = count & SWAP_HAS_CACHE;
773 count &= ~SWAP_HAS_CACHE;
774
775 if (usage == SWAP_HAS_CACHE) {
776 VM_BUG_ON(!has_cache);
777 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800778 } else if (count == SWAP_MAP_SHMEM) {
779 /*
780 * Or we could insist on shmem.c using a special
781 * swap_shmem_free() and free_shmem_swap_and_cache()...
782 */
783 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800784 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
785 if (count == COUNT_CONTINUED) {
786 if (swap_count_continued(p, offset, count))
787 count = SWAP_MAP_MAX | COUNT_CONTINUED;
788 else
789 count = SWAP_MAP_MAX;
790 } else
791 count--;
792 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800793
Hugh Dickins253d5532009-12-14 17:58:44 -0800794 usage = count | has_cache;
795 p->swap_map[offset] = usage;
796
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700797 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800798 if (!usage) {
Vladimir Davydov37e84352016-01-20 15:02:56 -0800799 mem_cgroup_uncharge_swap(entry);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700800 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700801 if (offset < p->lowest_bit)
802 p->lowest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700803 if (offset > p->highest_bit) {
804 bool was_full = !p->highest_bit;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700805 p->highest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700806 if (was_full && (p->flags & SWP_WRITEOK)) {
807 spin_lock(&swap_avail_lock);
808 WARN_ON(!plist_node_empty(&p->avail_list));
809 if (plist_node_empty(&p->avail_list))
810 plist_add(&p->avail_list,
811 &swap_avail_head);
812 spin_unlock(&swap_avail_lock);
813 }
814 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800815 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700816 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600817 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700818 if (p->flags & SWP_BLKDEV) {
819 struct gendisk *disk = p->bdev->bd_disk;
820 if (disk->fops->swap_slot_free_notify)
821 disk->fops->swap_slot_free_notify(p->bdev,
822 offset);
823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800825
826 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
828
829/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800830 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 * is still around or has not been recycled.
832 */
833void swap_free(swp_entry_t entry)
834{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800835 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
837 p = swap_info_get(entry);
838 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800839 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800840 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 }
842}
843
844/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700845 * Called after dropping swapcache to decrease refcnt to swap entries.
846 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700847void swapcache_free(swp_entry_t entry)
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700848{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700849 struct swap_info_struct *p;
850
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700851 p = swap_info_get(entry);
852 if (p) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700853 swap_entry_free(p, entry, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800854 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700855 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700856}
857
858/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700859 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800860 * This does not give an exact answer when swap count is continued,
861 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700863int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700865 int count = 0;
866 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 swp_entry_t entry;
868
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700869 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 p = swap_info_get(entry);
871 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700872 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800873 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700875 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
877
878/*
Minchan Kim8334b962015-09-08 15:00:24 -0700879 * How many references to @entry are currently swapped out?
880 * This considers COUNT_CONTINUED so it returns exact answer.
881 */
882int swp_swapcount(swp_entry_t entry)
883{
884 int count, tmp_count, n;
885 struct swap_info_struct *p;
886 struct page *page;
887 pgoff_t offset;
888 unsigned char *map;
889
890 p = swap_info_get(entry);
891 if (!p)
892 return 0;
893
894 count = swap_count(p->swap_map[swp_offset(entry)]);
895 if (!(count & COUNT_CONTINUED))
896 goto out;
897
898 count &= ~COUNT_CONTINUED;
899 n = SWAP_MAP_MAX + 1;
900
901 offset = swp_offset(entry);
902 page = vmalloc_to_page(p->swap_map + offset);
903 offset &= ~PAGE_MASK;
904 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
905
906 do {
Geliang Tanga8ae4992016-01-14 15:20:45 -0800907 page = list_next_entry(page, lru);
Minchan Kim8334b962015-09-08 15:00:24 -0700908 map = kmap_atomic(page);
909 tmp_count = map[offset];
910 kunmap_atomic(map);
911
912 count += (tmp_count & ~COUNT_CONTINUED) * n;
913 n *= (SWAP_CONT_MAX + 1);
914 } while (tmp_count & COUNT_CONTINUED);
915out:
916 spin_unlock(&p->lock);
917 return count;
918}
919
920/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800921 * We can write to an anon page without COW if there are no other references
922 * to it. And as a side-effect, free up its swap: because the old content
923 * on disk will never be read, and seeking back there to write new content
924 * later would only waste time away from clustering.
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700925 *
926 * NOTE: total_mapcount should not be relied upon by the caller if
927 * reuse_swap_page() returns false, but it may be always overwritten
928 * (see the other implementation for CONFIG_SWAP=n).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 */
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700930bool reuse_swap_page(struct page *page, int *total_mapcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700932 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Sasha Levin309381fea2014-01-23 15:52:54 -0800934 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800935 if (unlikely(PageKsm(page)))
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -0700936 return false;
937 count = page_trans_huge_mapcount(page, total_mapcount);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800938 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700939 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800940 if (count == 1 && !PageWriteback(page)) {
941 delete_from_swap_cache(page);
942 SetPageDirty(page);
943 }
944 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800945 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}
947
948/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800949 * If swap is getting full, or if there are no more mappings of this page,
950 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800952int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
Sasha Levin309381fea2014-01-23 15:52:54 -0800954 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
956 if (!PageSwapCache(page))
957 return 0;
958 if (PageWriteback(page))
959 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800960 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 return 0;
962
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700963 /*
964 * Once hibernation has begun to create its image of memory,
965 * there's a danger that one of the calls to try_to_free_swap()
966 * - most probably a call from __try_to_reclaim_swap() while
967 * hibernation is allocating its own swap pages for the image,
968 * but conceivably even a call from memory reclaim - will free
969 * the swap from a page which has already been recorded in the
970 * image as a clean swapcache page, and then reuse its swap for
971 * another page of the image. On waking from hibernation, the
972 * original page might be freed under memory pressure, then
973 * later read back in from swap, now with the wrong data.
974 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800975 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -0800976 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700977 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800978 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700979 return 0;
980
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800981 delete_from_swap_cache(page);
982 SetPageDirty(page);
983 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700984}
985
986/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 * Free the swap entry like above, but also try to
988 * free the page cache entry if it is the last user.
989 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800990int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800992 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 struct page *page = NULL;
994
Andi Kleena7420aa2009-09-16 11:50:05 +0200995 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800996 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 p = swap_info_get(entry);
999 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -08001000 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -08001001 page = find_get_page(swap_address_space(entry),
1002 entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -07001003 if (page && !trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001004 put_page(page);
Nick Piggin93fac702006-03-31 02:29:56 -08001005 page = NULL;
1006 }
1007 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001008 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001011 /*
1012 * Not mapped elsewhere, or swap space full? Free it!
1013 * Also recheck PageSwapCache now page is locked (above).
1014 */
Nick Piggin93fac702006-03-31 02:29:56 -08001015 if (PageSwapCache(page) && !PageWriteback(page) &&
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08001016 (!page_mapped(page) || mem_cgroup_swap_full(page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 delete_from_swap_cache(page);
1018 SetPageDirty(page);
1019 }
1020 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001021 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 }
Hugh Dickins2509ef22009-01-06 14:40:10 -08001023 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024}
1025
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001026#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001027/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001028 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001029 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001030 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1031 * from 0, in which the swap header is expected to be located.
1032 *
1033 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001034 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001035int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001036{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001037 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001038 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001039
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001040 if (device)
1041 bdev = bdget(device);
1042
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001043 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001044 for (type = 0; type < nr_swapfiles; type++) {
1045 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001046
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001047 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001048 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001049
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001050 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001051 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001052 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001053
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001054 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001055 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001056 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001057 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001058 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001059
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001060 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001061 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001062 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001063
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001064 spin_unlock(&swap_lock);
1065 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001066 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001067 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001068 }
1069 }
1070 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001071 if (bdev)
1072 bdput(bdev);
1073
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001074 return -ENODEV;
1075}
1076
1077/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001078 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1079 * corresponding to given index in swap_info (swap type).
1080 */
1081sector_t swapdev_block(int type, pgoff_t offset)
1082{
1083 struct block_device *bdev;
1084
1085 if ((unsigned int)type >= nr_swapfiles)
1086 return 0;
1087 if (!(swap_info[type]->flags & SWP_WRITEOK))
1088 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001089 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001090}
1091
1092/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001093 * Return either the total number of swap pages of given type, or the number
1094 * of free pages of that type (depending on @free)
1095 *
1096 * This is needed for software suspend
1097 */
1098unsigned int count_swap_pages(int type, int free)
1099{
1100 unsigned int n = 0;
1101
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001102 spin_lock(&swap_lock);
1103 if ((unsigned int)type < nr_swapfiles) {
1104 struct swap_info_struct *sis = swap_info[type];
1105
Shaohua Liec8acf22013-02-22 16:34:38 -08001106 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001107 if (sis->flags & SWP_WRITEOK) {
1108 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001109 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001110 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001111 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001112 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001113 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001114 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001115 return n;
1116}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001117#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001118
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001119static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001120{
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001121 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001122}
1123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001125 * No need to decide whether this PTE shares the swap entry with others,
1126 * just let do_wp_page work it out if a write is requested later - to
1127 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001129static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 unsigned long addr, swp_entry_t entry, struct page *page)
1131{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001132 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001133 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001134 spinlock_t *ptl;
1135 pte_t *pte;
1136 int ret = 1;
1137
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001138 swapcache = page;
1139 page = ksm_might_need_to_copy(page, vma, addr);
1140 if (unlikely(!page))
1141 return -ENOMEM;
1142
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001143 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1144 &memcg, false)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001145 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001146 goto out_nolock;
1147 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001148
1149 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001150 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001151 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001152 ret = 0;
1153 goto out;
1154 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001155
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001156 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001157 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 get_page(page);
1159 set_pte_at(vma->vm_mm, addr, pte,
1160 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001161 if (page == swapcache) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001162 page_add_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001163 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001164 } else { /* ksm created a completely new copy */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001165 page_add_new_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001166 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001167 lru_cache_add_active_or_unevictable(page, vma);
1168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 swap_free(entry);
1170 /*
1171 * Move the page to the active list so it is not
1172 * immediately swapped out again after swapon.
1173 */
1174 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001175out:
1176 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001177out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001178 if (page != swapcache) {
1179 unlock_page(page);
1180 put_page(page);
1181 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001182 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183}
1184
1185static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1186 unsigned long addr, unsigned long end,
1187 swp_entry_t entry, struct page *page)
1188{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001190 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001191 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Hugh Dickins044d66c2008-02-07 00:14:04 -08001193 /*
1194 * We don't actually need pte lock while scanning for swp_pte: since
1195 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1196 * page table while we're scanning; though it could get zapped, and on
1197 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1198 * of unmatched parts which look like swp_pte, so unuse_pte must
1199 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001200 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001201 */
1202 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 do {
1204 /*
1205 * swapoff spends a _lot_ of time in this loop!
1206 * Test inline before going to call unuse_pte.
1207 */
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001208 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001209 pte_unmap(pte);
1210 ret = unuse_pte(vma, pmd, addr, entry, page);
1211 if (ret)
1212 goto out;
1213 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001216 pte_unmap(pte - 1);
1217out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001218 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
1221static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1222 unsigned long addr, unsigned long end,
1223 swp_entry_t entry, struct page *page)
1224{
1225 pmd_t *pmd;
1226 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001227 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
1229 pmd = pmd_offset(pud, addr);
1230 do {
1231 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001232 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001234 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1235 if (ret)
1236 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 } while (pmd++, addr = next, addr != end);
1238 return 0;
1239}
1240
1241static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1242 unsigned long addr, unsigned long end,
1243 swp_entry_t entry, struct page *page)
1244{
1245 pud_t *pud;
1246 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001247 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 pud = pud_offset(pgd, addr);
1250 do {
1251 next = pud_addr_end(addr, end);
1252 if (pud_none_or_clear_bad(pud))
1253 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001254 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1255 if (ret)
1256 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 } while (pud++, addr = next, addr != end);
1258 return 0;
1259}
1260
1261static int unuse_vma(struct vm_area_struct *vma,
1262 swp_entry_t entry, struct page *page)
1263{
1264 pgd_t *pgd;
1265 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001266 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001268 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 addr = page_address_in_vma(page, vma);
1270 if (addr == -EFAULT)
1271 return 0;
1272 else
1273 end = addr + PAGE_SIZE;
1274 } else {
1275 addr = vma->vm_start;
1276 end = vma->vm_end;
1277 }
1278
1279 pgd = pgd_offset(vma->vm_mm, addr);
1280 do {
1281 next = pgd_addr_end(addr, end);
1282 if (pgd_none_or_clear_bad(pgd))
1283 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001284 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1285 if (ret)
1286 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 } while (pgd++, addr = next, addr != end);
1288 return 0;
1289}
1290
1291static int unuse_mm(struct mm_struct *mm,
1292 swp_entry_t entry, struct page *page)
1293{
1294 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001295 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297 if (!down_read_trylock(&mm->mmap_sem)) {
1298 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001299 * Activate page so shrink_inactive_list is unlikely to unmap
1300 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001302 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 unlock_page(page);
1304 down_read(&mm->mmap_sem);
1305 lock_page(page);
1306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001308 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 break;
1310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001312 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313}
1314
1315/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001316 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1317 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 * Recycle to start on reaching the end, returning 0 when empty.
1319 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001320static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001321 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001323 unsigned int max = si->max;
1324 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001325 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
1327 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001328 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 * for whether an entry is in use, not modifying it; false
1330 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001331 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 */
1333 for (;;) {
1334 if (++i >= max) {
1335 if (!prev) {
1336 i = 0;
1337 break;
1338 }
1339 /*
1340 * No entries in use at top of swap_map,
1341 * loop back to start and recheck there.
1342 */
1343 max = prev + 1;
1344 prev = 0;
1345 i = 1;
1346 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001347 if (frontswap) {
1348 if (frontswap_test(si, i))
1349 break;
1350 else
1351 continue;
1352 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001353 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001354 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 break;
1356 }
1357 return i;
1358}
1359
1360/*
1361 * We completely avoid races by reading each swap page in advance,
1362 * and then search for the process using it. All the necessary
1363 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001364 *
1365 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1366 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001368int try_to_unuse(unsigned int type, bool frontswap,
1369 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001371 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001373 volatile unsigned char *swap_map; /* swap_map is accessed without
1374 * locking. Mark it as volatile
1375 * to prevent compiler doing
1376 * something odd.
1377 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001378 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 struct page *page;
1380 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001381 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384 /*
1385 * When searching mms for an entry, a good strategy is to
1386 * start at the first mm we freed the previous entry from
1387 * (though actually we don't notice whether we or coincidence
1388 * freed the entry). Initialize this start_mm with a hold.
1389 *
1390 * A simpler strategy would be to start at the last mm we
1391 * freed the previous entry from; but that would take less
1392 * advantage of mmlist ordering, which clusters forked mms
1393 * together, child after parent. If we race with dup_mmap(), we
1394 * prefer to resolve parent before child, lest we miss entries
1395 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001396 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 */
1398 start_mm = &init_mm;
1399 atomic_inc(&init_mm.mm_users);
1400
1401 /*
1402 * Keep on scanning until all entries have gone. Usually,
1403 * one pass through swap_map is enough, but not necessarily:
1404 * there are races when an instance of an entry might be missed.
1405 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001406 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (signal_pending(current)) {
1408 retval = -EINTR;
1409 break;
1410 }
1411
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001412 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 * Get a page for the entry, using the existing swap
1414 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001415 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 */
1417 swap_map = &si->swap_map[i];
1418 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001419 page = read_swap_cache_async(entry,
1420 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 if (!page) {
1422 /*
1423 * Either swap_duplicate() failed because entry
1424 * has been freed independently, and will not be
1425 * reused since sys_swapoff() already disabled
1426 * allocation from here, or alloc_page() failed.
1427 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001428 swcount = *swap_map;
1429 /*
1430 * We don't hold lock here, so the swap entry could be
1431 * SWAP_MAP_BAD (when the cluster is discarding).
1432 * Instead of fail out, We can just skip the swap
1433 * entry because swapoff will wait for discarding
1434 * finish anyway.
1435 */
1436 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 continue;
1438 retval = -ENOMEM;
1439 break;
1440 }
1441
1442 /*
1443 * Don't hold on to start_mm if it looks like exiting.
1444 */
1445 if (atomic_read(&start_mm->mm_users) == 1) {
1446 mmput(start_mm);
1447 start_mm = &init_mm;
1448 atomic_inc(&init_mm.mm_users);
1449 }
1450
1451 /*
1452 * Wait for and lock page. When do_swap_page races with
1453 * try_to_unuse, do_swap_page can handle the fault much
1454 * faster than try_to_unuse can locate the entry. This
1455 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1456 * defer to do_swap_page in such a case - in some tests,
1457 * do_swap_page and try_to_unuse repeatedly compete.
1458 */
1459 wait_on_page_locked(page);
1460 wait_on_page_writeback(page);
1461 lock_page(page);
1462 wait_on_page_writeback(page);
1463
1464 /*
1465 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001468 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1469 retval = shmem_unuse(entry, page);
1470 /* page has already been unlocked and released */
1471 if (retval < 0)
1472 break;
1473 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001475 if (swap_count(swcount) && start_mm != &init_mm)
1476 retval = unuse_mm(start_mm, entry, page);
1477
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001478 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 int set_start_mm = (*swap_map >= swcount);
1480 struct list_head *p = &start_mm->mmlist;
1481 struct mm_struct *new_start_mm = start_mm;
1482 struct mm_struct *prev_mm = start_mm;
1483 struct mm_struct *mm;
1484
1485 atomic_inc(&new_start_mm->mm_users);
1486 atomic_inc(&prev_mm->mm_users);
1487 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001488 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 (p = p->next) != &start_mm->mmlist) {
1490 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001491 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 spin_unlock(&mmlist_lock);
1494 mmput(prev_mm);
1495 prev_mm = mm;
1496
1497 cond_resched();
1498
1499 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001500 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001502 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001504 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001506
Bo Liu32c5fc12009-11-02 16:50:33 +00001507 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 mmput(new_start_mm);
1509 atomic_inc(&mm->mm_users);
1510 new_start_mm = mm;
1511 set_start_mm = 0;
1512 }
1513 spin_lock(&mmlist_lock);
1514 }
1515 spin_unlock(&mmlist_lock);
1516 mmput(prev_mm);
1517 mmput(start_mm);
1518 start_mm = new_start_mm;
1519 }
1520 if (retval) {
1521 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001522 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 break;
1524 }
1525
1526 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 * If a reference remains (rare), we would like to leave
1528 * the page in the swap cache; but try_to_unmap could
1529 * then re-duplicate the entry once we drop page lock,
1530 * so we might loop indefinitely; also, that page could
1531 * not be swapped out to other storage meanwhile. So:
1532 * delete from cache even if there's another reference,
1533 * after ensuring that the data has been saved to disk -
1534 * since if the reference remains (rarer), it will be
1535 * read from disk into another page. Splitting into two
1536 * pages would be incorrect if swap supported "shared
1537 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001538 *
1539 * Given how unuse_vma() targets one particular offset
1540 * in an anon_vma, once the anon_vma has been determined,
1541 * this splitting happens to be just what is needed to
1542 * handle where KSM pages have been swapped out: re-reading
1543 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001545 if (swap_count(*swap_map) &&
1546 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 struct writeback_control wbc = {
1548 .sync_mode = WB_SYNC_NONE,
1549 };
1550
1551 swap_writepage(page, &wbc);
1552 lock_page(page);
1553 wait_on_page_writeback(page);
1554 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001555
1556 /*
1557 * It is conceivable that a racing task removed this page from
1558 * swap cache just before we acquired the page lock at the top,
1559 * or while we dropped it in unuse_mm(). The page might even
1560 * be back in swap cache on another swap area: that we must not
1561 * delete, since it may not have been written out to swap yet.
1562 */
1563 if (PageSwapCache(page) &&
1564 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001565 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
1567 /*
1568 * So we could skip searching mms once swap count went
1569 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001570 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 */
1572 SetPageDirty(page);
1573 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001574 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 /*
1577 * Make sure that we aren't completely killing
1578 * interactive performance.
1579 */
1580 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001581 if (frontswap && pages_to_unuse > 0) {
1582 if (!--pages_to_unuse)
1583 break;
1584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 }
1586
1587 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return retval;
1589}
1590
1591/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001592 * After a successful try_to_unuse, if no swap is now in use, we know
1593 * we can empty the mmlist. swap_lock must be held on entry and exit.
1594 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 * added to the mmlist just after page_duplicate - before would be racy.
1596 */
1597static void drain_mmlist(void)
1598{
1599 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001600 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001602 for (type = 0; type < nr_swapfiles; type++)
1603 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 return;
1605 spin_lock(&mmlist_lock);
1606 list_for_each_safe(p, next, &init_mm.mmlist)
1607 list_del_init(p);
1608 spin_unlock(&mmlist_lock);
1609}
1610
1611/*
1612 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001613 * corresponds to page offset for the specified swap entry.
1614 * Note that the type of this function is sector_t, but it returns page offset
1615 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001617static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001619 struct swap_info_struct *sis;
1620 struct swap_extent *start_se;
1621 struct swap_extent *se;
1622 pgoff_t offset;
1623
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001624 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001625 *bdev = sis->bdev;
1626
1627 offset = swp_offset(entry);
1628 start_se = sis->curr_swap_extent;
1629 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if (se->start_page <= offset &&
1633 offset < (se->start_page + se->nr_pages)) {
1634 return se->start_block + (offset - se->start_page);
1635 }
Geliang Tanga8ae4992016-01-14 15:20:45 -08001636 se = list_next_entry(se, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 sis->curr_swap_extent = se;
1638 BUG_ON(se == start_se); /* It *must* be present */
1639 }
1640}
1641
1642/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001643 * Returns the page offset into bdev for the specified page's swap entry.
1644 */
1645sector_t map_swap_page(struct page *page, struct block_device **bdev)
1646{
1647 swp_entry_t entry;
1648 entry.val = page_private(page);
1649 return map_swap_entry(entry, bdev);
1650}
1651
1652/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 * Free all of a swapdev's extent information
1654 */
1655static void destroy_swap_extents(struct swap_info_struct *sis)
1656{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001657 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 struct swap_extent *se;
1659
Geliang Tanga8ae4992016-01-14 15:20:45 -08001660 se = list_first_entry(&sis->first_swap_extent.list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 struct swap_extent, list);
1662 list_del(&se->list);
1663 kfree(se);
1664 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001665
1666 if (sis->flags & SWP_FILE) {
1667 struct file *swap_file = sis->swap_file;
1668 struct address_space *mapping = swap_file->f_mapping;
1669
1670 sis->flags &= ~SWP_FILE;
1671 mapping->a_ops->swap_deactivate(swap_file);
1672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
1674
1675/*
1676 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001677 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001679 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 */
Mel Gormana509bc12012-07-31 16:44:57 -07001681int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1683 unsigned long nr_pages, sector_t start_block)
1684{
1685 struct swap_extent *se;
1686 struct swap_extent *new_se;
1687 struct list_head *lh;
1688
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001689 if (start_page == 0) {
1690 se = &sis->first_swap_extent;
1691 sis->curr_swap_extent = se;
1692 se->start_page = 0;
1693 se->nr_pages = nr_pages;
1694 se->start_block = start_block;
1695 return 1;
1696 } else {
1697 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001699 BUG_ON(se->start_page + se->nr_pages != start_page);
1700 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 /* Merge it */
1702 se->nr_pages += nr_pages;
1703 return 0;
1704 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 }
1706
1707 /*
1708 * No merge. Insert a new extent, preserving ordering.
1709 */
1710 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1711 if (new_se == NULL)
1712 return -ENOMEM;
1713 new_se->start_page = start_page;
1714 new_se->nr_pages = nr_pages;
1715 new_se->start_block = start_block;
1716
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001717 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001718 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719}
1720
1721/*
1722 * A `swap extent' is a simple thing which maps a contiguous range of pages
1723 * onto a contiguous range of disk blocks. An ordered list of swap extents
1724 * is built at swapon time and is then used at swap_writepage/swap_readpage
1725 * time for locating where on disk a page belongs.
1726 *
1727 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1728 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1729 * swap files identically.
1730 *
1731 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1732 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1733 * swapfiles are handled *identically* after swapon time.
1734 *
1735 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1736 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1737 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1738 * requirements, they are simply tossed out - we will never use those blocks
1739 * for swapping.
1740 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001741 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1743 * which will scribble on the fs.
1744 *
1745 * The amount of disk space which a single swap extent represents varies.
1746 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1747 * extents in the list. To avoid much list walking, we cache the previous
1748 * search location in `curr_swap_extent', and start new searches from there.
1749 * This is extremely effective. The average number of iterations in
1750 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1751 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001752static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Mel Gorman62c230b2012-07-31 16:44:55 -07001754 struct file *swap_file = sis->swap_file;
1755 struct address_space *mapping = swap_file->f_mapping;
1756 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 int ret;
1758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (S_ISBLK(inode->i_mode)) {
1760 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001761 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001762 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
1764
Mel Gorman62c230b2012-07-31 16:44:55 -07001765 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001766 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001767 if (!ret) {
1768 sis->flags |= SWP_FILE;
1769 ret = add_swap_extent(sis, 0, sis->max, 0);
1770 *span = sis->pages;
1771 }
Mel Gormana509bc12012-07-31 16:44:57 -07001772 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001773 }
1774
Mel Gormana509bc12012-07-31 16:44:57 -07001775 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776}
1777
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001778static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001779 unsigned char *swap_map,
1780 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001781{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001782 if (prio >= 0)
1783 p->prio = prio;
1784 else
1785 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001786 /*
1787 * the plist prio is negated because plist ordering is
1788 * low-to-high, while swap ordering is high-to-low
1789 */
1790 p->list.prio = -p->prio;
1791 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001792 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001793 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001794 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001795 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001796 total_swap_pages += p->pages;
1797
Dan Streetmanadfab832014-06-04 16:09:53 -07001798 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001799 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07001800 * both lists are plists, and thus priority ordered.
1801 * swap_active_head needs to be priority ordered for swapoff(),
1802 * which on removal of any swap_info_struct with an auto-assigned
1803 * (i.e. negative) priority increments the auto-assigned priority
1804 * of any lower-priority swap_info_structs.
1805 * swap_avail_head needs to be priority ordered for get_swap_page(),
1806 * which allocates swap pages from the highest available priority
1807 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07001808 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07001809 plist_add(&p->list, &swap_active_head);
1810 spin_lock(&swap_avail_lock);
1811 plist_add(&p->avail_list, &swap_avail_head);
1812 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001813}
1814
1815static void enable_swap_info(struct swap_info_struct *p, int prio,
1816 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001817 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001818 unsigned long *frontswap_map)
1819{
Minchan Kim4f898492013-04-30 15:26:54 -07001820 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001821 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001822 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001823 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001824 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001825 spin_unlock(&swap_lock);
1826}
1827
1828static void reinsert_swap_info(struct swap_info_struct *p)
1829{
1830 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001831 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001832 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001833 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001834 spin_unlock(&swap_lock);
1835}
1836
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001837SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001839 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001840 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001841 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001842 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 struct file *swap_file, *victim;
1844 struct address_space *mapping;
1845 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001846 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07001847 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001848 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 if (!capable(CAP_SYS_ADMIN))
1851 return -EPERM;
1852
Al Viro191c5422012-02-13 03:58:52 +00001853 BUG_ON(!current->mm);
1854
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001857 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
Jeff Layton669abf42012-10-10 16:43:10 -04001859 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 err = PTR_ERR(victim);
1861 if (IS_ERR(victim))
1862 goto out;
1863
1864 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001865 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07001866 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8fd2009-01-06 14:39:48 -08001867 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001868 if (p->swap_file->f_mapping == mapping) {
1869 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07001871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 }
Dan Streetmanadfab832014-06-04 16:09:53 -07001874 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001876 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 goto out_dput;
1878 }
Al Viro191c5422012-02-13 03:58:52 +00001879 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 vm_unacct_memory(p->pages);
1881 else {
1882 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001883 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 goto out_dput;
1885 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001886 spin_lock(&swap_avail_lock);
1887 plist_del(&p->avail_list, &swap_avail_head);
1888 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001889 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001890 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001891 struct swap_info_struct *si = p;
1892
Dan Streetman18ab4d42014-06-04 16:09:59 -07001893 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001894 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001895 si->list.prio--;
1896 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07001897 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001898 least_priority++;
1899 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001900 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08001901 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 total_swap_pages -= p->pages;
1903 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001904 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001905 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001906
David Rientjese1e12d22012-12-11 16:02:56 -08001907 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07001908 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08001909 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 if (err) {
1912 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001913 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 goto out_dput;
1915 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001916
Shaohua Li815c2c52013-09-11 14:20:30 -07001917 flush_work(&p->discard_work);
1918
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001919 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001920 if (p->flags & SWP_CONTINUED)
1921 free_swap_count_continuations(p);
1922
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001923 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001924 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001925 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001927
1928 /* wait for anyone still in scan_swap_map */
1929 p->highest_bit = 0; /* cuts scans short */
1930 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001931 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001932 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001933 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001934 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001935 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001936 }
1937
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001939 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 p->swap_file = NULL;
1941 p->max = 0;
1942 swap_map = p->swap_map;
1943 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001944 cluster_info = p->cluster_info;
1945 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07001946 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08001947 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001948 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001949 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08001950 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001951 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001952 free_percpu(p->percpu_cluster);
1953 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001955 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001956 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001957 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07001958 swap_cgroup_swapoff(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 inode = mapping->host;
1961 if (S_ISBLK(inode->i_mode)) {
1962 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001963 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001964 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 } else {
Al Viro59551022016-01-22 15:40:57 -05001966 inode_lock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 inode->i_flags &= ~S_SWAPFILE;
Al Viro59551022016-01-22 15:40:57 -05001968 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 }
1970 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08001971
1972 /*
1973 * Clear the SWP_USED flag after all resources are freed so that swapon
1974 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1975 * not hold p->lock after we cleared its SWP_WRITEOK.
1976 */
1977 spin_lock(&swap_lock);
1978 p->flags = 0;
1979 spin_unlock(&swap_lock);
1980
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001982 atomic_inc(&proc_poll_event);
1983 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
1985out_dput:
1986 filp_close(victim, NULL);
1987out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001988 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 return err;
1990}
1991
1992#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001993static unsigned swaps_poll(struct file *file, poll_table *wait)
1994{
Kay Sieversf1514632011-07-12 20:48:39 +02001995 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001996
1997 poll_wait(file, &proc_poll_wait, wait);
1998
Kay Sieversf1514632011-07-12 20:48:39 +02001999 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2000 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002001 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2002 }
2003
2004 return POLLIN | POLLRDNORM;
2005}
2006
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007/* iterator */
2008static void *swap_start(struct seq_file *swap, loff_t *pos)
2009{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002010 struct swap_info_struct *si;
2011 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 loff_t l = *pos;
2013
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002014 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002016 if (!l)
2017 return SEQ_START_TOKEN;
2018
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002019 for (type = 0; type < nr_swapfiles; type++) {
2020 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2021 si = swap_info[type];
2022 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002024 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002025 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 }
2027
2028 return NULL;
2029}
2030
2031static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2032{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002033 struct swap_info_struct *si = v;
2034 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002036 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002037 type = 0;
2038 else
2039 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002040
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002041 for (; type < nr_swapfiles; type++) {
2042 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2043 si = swap_info[type];
2044 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 continue;
2046 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002047 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 }
2049
2050 return NULL;
2051}
2052
2053static void swap_stop(struct seq_file *swap, void *v)
2054{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002055 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056}
2057
2058static int swap_show(struct seq_file *swap, void *v)
2059{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002060 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 struct file *file;
2062 int len;
2063
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002064 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002065 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2066 return 0;
2067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002069 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002070 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002071 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002072 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002073 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002075 si->pages << (PAGE_SHIFT - 10),
2076 si->inuse_pages << (PAGE_SHIFT - 10),
2077 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 return 0;
2079}
2080
Helge Deller15ad7cd2006-12-06 20:40:36 -08002081static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 .start = swap_start,
2083 .next = swap_next,
2084 .stop = swap_stop,
2085 .show = swap_show
2086};
2087
2088static int swaps_open(struct inode *inode, struct file *file)
2089{
Kay Sieversf1514632011-07-12 20:48:39 +02002090 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002091 int ret;
2092
Kay Sievers66d7dd52010-10-26 14:22:06 -07002093 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002094 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002095 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002096
Kay Sieversf1514632011-07-12 20:48:39 +02002097 seq = file->private_data;
2098 seq->poll_event = atomic_read(&proc_poll_event);
2099 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100}
2101
Helge Deller15ad7cd2006-12-06 20:40:36 -08002102static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 .open = swaps_open,
2104 .read = seq_read,
2105 .llseek = seq_lseek,
2106 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002107 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108};
2109
2110static int __init procswaps_init(void)
2111{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002112 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 return 0;
2114}
2115__initcall(procswaps_init);
2116#endif /* CONFIG_PROC_FS */
2117
Jan Beulich17963162008-12-16 11:35:24 +00002118#ifdef MAX_SWAPFILES_CHECK
2119static int __init max_swapfiles_check(void)
2120{
2121 MAX_SWAPFILES_CHECK();
2122 return 0;
2123}
2124late_initcall(max_swapfiles_check);
2125#endif
2126
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002127static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002129 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002131
2132 p = kzalloc(sizeof(*p), GFP_KERNEL);
2133 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002134 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002135
Hugh Dickins5d337b92005-09-03 15:54:41 -07002136 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002137 for (type = 0; type < nr_swapfiles; type++) {
2138 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002140 }
Christoph Lameter06972122006-06-23 02:03:35 -07002141 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002142 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002143 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002144 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002146 if (type >= nr_swapfiles) {
2147 p->type = type;
2148 swap_info[type] = p;
2149 /*
2150 * Write swap_info[type] before nr_swapfiles, in case a
2151 * racing procfs swap_start() or swap_next() is reading them.
2152 * (We never shrink nr_swapfiles, we never free this entry.)
2153 */
2154 smp_wmb();
2155 nr_swapfiles++;
2156 } else {
2157 kfree(p);
2158 p = swap_info[type];
2159 /*
2160 * Do not memset this entry: a racing procfs swap_next()
2161 * would be relying on p->type to remain valid.
2162 */
2163 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002164 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002165 plist_node_init(&p->list, 0);
2166 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002168 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002169 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002170
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002171 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002172}
2173
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002174static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2175{
2176 int error;
2177
2178 if (S_ISBLK(inode->i_mode)) {
2179 p->bdev = bdgrab(I_BDEV(inode));
2180 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002181 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002182 if (error < 0) {
2183 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002184 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002185 }
2186 p->old_block_size = block_size(p->bdev);
2187 error = set_blocksize(p->bdev, PAGE_SIZE);
2188 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002189 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002190 p->flags |= SWP_BLKDEV;
2191 } else if (S_ISREG(inode->i_mode)) {
2192 p->bdev = inode->i_sb->s_bdev;
Al Viro59551022016-01-22 15:40:57 -05002193 inode_lock(inode);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002194 if (IS_SWAPFILE(inode))
2195 return -EBUSY;
2196 } else
2197 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002198
2199 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002200}
2201
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002202static unsigned long read_swap_header(struct swap_info_struct *p,
2203 union swap_header *swap_header,
2204 struct inode *inode)
2205{
2206 int i;
2207 unsigned long maxpages;
2208 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002209 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002210
2211 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002212 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002213 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002214 }
2215
2216 /* swap partition endianess hack... */
2217 if (swab32(swap_header->info.version) == 1) {
2218 swab32s(&swap_header->info.version);
2219 swab32s(&swap_header->info.last_page);
2220 swab32s(&swap_header->info.nr_badpages);
2221 for (i = 0; i < swap_header->info.nr_badpages; i++)
2222 swab32s(&swap_header->info.badpages[i]);
2223 }
2224 /* Check the swap header's sub-version */
2225 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002226 pr_warn("Unable to handle swap header version %d\n",
2227 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002228 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002229 }
2230
2231 p->lowest_bit = 1;
2232 p->cluster_next = 1;
2233 p->cluster_nr = 0;
2234
2235 /*
2236 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002237 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002238 * of bits for the swap offset in the swp_entry_t type, and
2239 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002240 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002241 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002242 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002243 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002244 * offset is extracted. This will mask all the bits from
2245 * the initial ~0UL mask that can't be encoded in either
2246 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002247 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002248 */
2249 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002250 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002251 last_page = swap_header->info.last_page;
2252 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002253 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002254 maxpages << (PAGE_SHIFT - 10),
2255 last_page << (PAGE_SHIFT - 10));
2256 }
2257 if (maxpages > last_page) {
2258 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002259 /* p->max is an unsigned int: don't overflow it */
2260 if ((unsigned int)maxpages == 0)
2261 maxpages = UINT_MAX;
2262 }
2263 p->highest_bit = maxpages - 1;
2264
2265 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002266 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002267 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2268 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002269 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002270 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002271 }
2272 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002273 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002274 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002275 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002276
2277 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002278}
2279
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002280static int setup_swap_map_and_extents(struct swap_info_struct *p,
2281 union swap_header *swap_header,
2282 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002283 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002284 unsigned long maxpages,
2285 sector_t *span)
2286{
2287 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002288 unsigned int nr_good_pages;
2289 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002290 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2291 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002292
2293 nr_good_pages = maxpages - 1; /* omit header page */
2294
Shaohua Li2a8f9442013-09-11 14:20:28 -07002295 cluster_set_null(&p->free_cluster_head);
2296 cluster_set_null(&p->free_cluster_tail);
Shaohua Li815c2c52013-09-11 14:20:30 -07002297 cluster_set_null(&p->discard_cluster_head);
2298 cluster_set_null(&p->discard_cluster_tail);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002299
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002300 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2301 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002302 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2303 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002304 if (page_nr < maxpages) {
2305 swap_map[page_nr] = SWAP_MAP_BAD;
2306 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002307 /*
2308 * Haven't marked the cluster free yet, no list
2309 * operation involved
2310 */
2311 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002312 }
2313 }
2314
Shaohua Li2a8f9442013-09-11 14:20:28 -07002315 /* Haven't marked the cluster free yet, no list operation involved */
2316 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2317 inc_cluster_info_page(p, cluster_info, i);
2318
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002319 if (nr_good_pages) {
2320 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002321 /*
2322 * Not mark the cluster free yet, no list
2323 * operation involved
2324 */
2325 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002326 p->max = maxpages;
2327 p->pages = nr_good_pages;
2328 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002329 if (nr_extents < 0)
2330 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002331 nr_good_pages = p->pages;
2332 }
2333 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002334 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002335 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002336 }
2337
Shaohua Li2a8f9442013-09-11 14:20:28 -07002338 if (!cluster_info)
2339 return nr_extents;
2340
2341 for (i = 0; i < nr_clusters; i++) {
2342 if (!cluster_count(&cluster_info[idx])) {
2343 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2344 if (cluster_is_null(&p->free_cluster_head)) {
2345 cluster_set_next_flag(&p->free_cluster_head,
2346 idx, 0);
2347 cluster_set_next_flag(&p->free_cluster_tail,
2348 idx, 0);
2349 } else {
2350 unsigned int tail;
2351
2352 tail = cluster_next(&p->free_cluster_tail);
2353 cluster_set_next(&cluster_info[tail], idx);
2354 cluster_set_next_flag(&p->free_cluster_tail,
2355 idx, 0);
2356 }
2357 }
2358 idx++;
2359 if (idx == nr_clusters)
2360 idx = 0;
2361 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002362 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002363}
2364
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002365/*
2366 * Helper to sys_swapon determining if a given swap
2367 * backing device queue supports DISCARD operations.
2368 */
2369static bool swap_discardable(struct swap_info_struct *si)
2370{
2371 struct request_queue *q = bdev_get_queue(si->bdev);
2372
2373 if (!q || !blk_queue_discard(q))
2374 return false;
2375
2376 return true;
2377}
2378
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002379SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2380{
2381 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002382 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002383 struct file *swap_file = NULL;
2384 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002385 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002386 int error;
2387 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002388 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002389 sector_t span;
2390 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002391 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002392 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002393 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002394 struct page *page = NULL;
2395 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002396
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002397 if (swap_flags & ~SWAP_FLAGS_VALID)
2398 return -EINVAL;
2399
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002400 if (!capable(CAP_SYS_ADMIN))
2401 return -EPERM;
2402
2403 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002404 if (IS_ERR(p))
2405 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002406
Shaohua Li815c2c52013-09-11 14:20:30 -07002407 INIT_WORK(&p->discard_work, swap_discard_work);
2408
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002411 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002413 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 }
Jeff Layton669abf42012-10-10 16:43:10 -04002415 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002417 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002419 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 }
2421
2422 p->swap_file = swap_file;
2423 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002424 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002425
Al Viro59551022016-01-22 15:40:57 -05002426 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002427 error = claim_swapfile(p, inode);
2428 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 /*
2432 * Read the swap header.
2433 */
2434 if (!mapping->a_ops->readpage) {
2435 error = -EINVAL;
2436 goto bad_swap;
2437 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002438 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 if (IS_ERR(page)) {
2440 error = PTR_ERR(page);
2441 goto bad_swap;
2442 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002443 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002445 maxpages = read_swap_header(p, swap_header, inode);
2446 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 error = -EINVAL;
2448 goto bad_swap;
2449 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002450
Hugh Dickins81e33972009-01-06 14:39:49 -08002451 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002452 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002453 if (!swap_map) {
2454 error = -ENOMEM;
2455 goto bad_swap;
2456 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002457 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002458 int cpu;
2459
Shaohua Li2a8f9442013-09-11 14:20:28 -07002460 p->flags |= SWP_SOLIDSTATE;
2461 /*
2462 * select a random position to start with to help wear leveling
2463 * SSD
2464 */
2465 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2466
2467 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2468 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2469 if (!cluster_info) {
2470 error = -ENOMEM;
2471 goto bad_swap;
2472 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002473 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2474 if (!p->percpu_cluster) {
2475 error = -ENOMEM;
2476 goto bad_swap;
2477 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002478 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002479 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002480 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002481 cluster_set_null(&cluster->index);
2482 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002483 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002484
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002485 error = swap_cgroup_swapon(p->type, maxpages);
2486 if (error)
2487 goto bad_swap;
2488
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002489 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002490 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002491 if (unlikely(nr_extents < 0)) {
2492 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 goto bad_swap;
2494 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002495 /* frontswap enabled? set up bit-per-page map for frontswap */
Vlastimil Babka8ea1d2a2016-07-26 15:24:42 -07002496 if (IS_ENABLED(CONFIG_FRONTSWAP))
Akinobu Mita7b579762013-06-12 14:05:08 -07002497 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498
Shaohua Li2a8f9442013-09-11 14:20:28 -07002499 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2500 /*
2501 * When discard is enabled for swap with no particular
2502 * policy flagged, we set all swap discard flags here in
2503 * order to sustain backward compatibility with older
2504 * swapon(8) releases.
2505 */
2506 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2507 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002508
Shaohua Li2a8f9442013-09-11 14:20:28 -07002509 /*
2510 * By flagging sys_swapon, a sysadmin can tell us to
2511 * either do single-time area discards only, or to just
2512 * perform discards for released swap page-clusters.
2513 * Now it's time to adjust the p->flags accordingly.
2514 */
2515 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2516 p->flags &= ~SWP_PAGE_DISCARD;
2517 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2518 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002519
Shaohua Li2a8f9442013-09-11 14:20:28 -07002520 /* issue a swapon-time discard if it's still required */
2521 if (p->flags & SWP_AREA_DISCARD) {
2522 int err = discard_swap(p);
2523 if (unlikely(err))
2524 pr_err("swapon: discard_swap(%p): %d\n",
2525 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002526 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002527 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002528
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002529 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002530 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002531 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002532 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002533 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002534 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002535
Joe Perches756a0252016-03-17 14:19:47 -07002536 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 -04002537 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002538 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2539 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002540 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002541 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2542 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002543 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002544
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002545 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002546 atomic_inc(&proc_poll_event);
2547 wake_up_interruptible(&proc_poll_wait);
2548
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002549 if (S_ISREG(inode->i_mode))
2550 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 error = 0;
2552 goto out;
2553bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002554 free_percpu(p->percpu_cluster);
2555 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002556 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002557 set_blocksize(p->bdev, p->old_block_size);
2558 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002560 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002561 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002562 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002565 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002567 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002568 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002569 if (inode && S_ISREG(inode->i_mode)) {
Al Viro59551022016-01-22 15:40:57 -05002570 inode_unlock(inode);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002571 inode = NULL;
2572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575out:
2576 if (page && !IS_ERR(page)) {
2577 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002578 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 }
2580 if (name)
2581 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002582 if (inode && S_ISREG(inode->i_mode))
Al Viro59551022016-01-22 15:40:57 -05002583 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 return error;
2585}
2586
2587void si_swapinfo(struct sysinfo *val)
2588{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002589 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 unsigned long nr_to_be_unused = 0;
2591
Hugh Dickins5d337b92005-09-03 15:54:41 -07002592 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002593 for (type = 0; type < nr_swapfiles; type++) {
2594 struct swap_info_struct *si = swap_info[type];
2595
2596 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2597 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002599 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002601 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602}
2603
2604/*
2605 * Verify that a swap entry is valid and increment its swap map count.
2606 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002607 * Returns error code in following case.
2608 * - success -> 0
2609 * - swp_entry is invalid -> EINVAL
2610 * - swp_entry is migration entry -> EINVAL
2611 * - swap-cache reference is requested but there is already one. -> EEXIST
2612 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002613 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002615static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002617 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002619 unsigned char count;
2620 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002621 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
Andi Kleena7420aa2009-09-16 11:50:05 +02002623 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002624 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002625
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 type = swp_type(entry);
2627 if (type >= nr_swapfiles)
2628 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002629 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 offset = swp_offset(entry);
2631
Shaohua Liec8acf22013-02-22 16:34:38 -08002632 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002633 if (unlikely(offset >= p->max))
2634 goto unlock_out;
2635
Hugh Dickins253d5532009-12-14 17:58:44 -08002636 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002637
2638 /*
2639 * swapin_readahead() doesn't check if a swap entry is valid, so the
2640 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2641 */
2642 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2643 err = -ENOENT;
2644 goto unlock_out;
2645 }
2646
Hugh Dickins253d5532009-12-14 17:58:44 -08002647 has_cache = count & SWAP_HAS_CACHE;
2648 count &= ~SWAP_HAS_CACHE;
2649 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002650
Hugh Dickins253d5532009-12-14 17:58:44 -08002651 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002652
2653 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002654 if (!has_cache && count)
2655 has_cache = SWAP_HAS_CACHE;
2656 else if (has_cache) /* someone else added cache */
2657 err = -EEXIST;
2658 else /* no users remaining */
2659 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002660
2661 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002662
Hugh Dickins570a335b2009-12-14 17:58:46 -08002663 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2664 count += usage;
2665 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002666 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002667 else if (swap_count_continued(p, offset, count))
2668 count = COUNT_CONTINUED;
2669 else
2670 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002671 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002672 err = -ENOENT; /* unused swap entry */
2673
2674 p->swap_map[offset] = count | has_cache;
2675
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002676unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002677 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002679 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680
2681bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002682 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 goto out;
2684}
Hugh Dickins253d5532009-12-14 17:58:44 -08002685
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002686/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002687 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2688 * (in which case its reference count is never incremented).
2689 */
2690void swap_shmem_alloc(swp_entry_t entry)
2691{
2692 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2693}
2694
2695/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002696 * Increase reference count of swap entry by 1.
2697 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2698 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2699 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2700 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002701 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002702int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002703{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002704 int err = 0;
2705
2706 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2707 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2708 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002709}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002711/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002712 * @entry: swap entry for which we allocate swap cache.
2713 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002714 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002715 * This can return error codes. Returns 0 at success.
2716 * -EBUSY means there is a swap cache.
2717 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002718 */
2719int swapcache_prepare(swp_entry_t entry)
2720{
Hugh Dickins253d5532009-12-14 17:58:44 -08002721 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002722}
2723
Mel Gormanf981c592012-07-31 16:44:47 -07002724struct swap_info_struct *page_swap_info(struct page *page)
2725{
2726 swp_entry_t swap = { .val = page_private(page) };
2727 BUG_ON(!PageSwapCache(page));
2728 return swap_info[swp_type(swap)];
2729}
2730
2731/*
2732 * out-of-line __page_file_ methods to avoid include hell.
2733 */
2734struct address_space *__page_file_mapping(struct page *page)
2735{
Sasha Levin309381fea2014-01-23 15:52:54 -08002736 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002737 return page_swap_info(page)->swap_file->f_mapping;
2738}
2739EXPORT_SYMBOL_GPL(__page_file_mapping);
2740
2741pgoff_t __page_file_index(struct page *page)
2742{
2743 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002744 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002745 return swp_offset(swap);
2746}
2747EXPORT_SYMBOL_GPL(__page_file_index);
2748
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002750 * add_swap_count_continuation - called when a swap count is duplicated
2751 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2752 * page of the original vmalloc'ed swap_map, to hold the continuation count
2753 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2754 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2755 *
2756 * These continuation pages are seldom referenced: the common paths all work
2757 * on the original swap_map, only referring to a continuation page when the
2758 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2759 *
2760 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2761 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2762 * can be called after dropping locks.
2763 */
2764int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2765{
2766 struct swap_info_struct *si;
2767 struct page *head;
2768 struct page *page;
2769 struct page *list_page;
2770 pgoff_t offset;
2771 unsigned char count;
2772
2773 /*
2774 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2775 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2776 */
2777 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2778
2779 si = swap_info_get(entry);
2780 if (!si) {
2781 /*
2782 * An acceptable race has occurred since the failing
2783 * __swap_duplicate(): the swap entry has been freed,
2784 * perhaps even the whole swap_map cleared for swapoff.
2785 */
2786 goto outer;
2787 }
2788
2789 offset = swp_offset(entry);
2790 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2791
2792 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2793 /*
2794 * The higher the swap count, the more likely it is that tasks
2795 * will race to add swap count continuation: we need to avoid
2796 * over-provisioning.
2797 */
2798 goto out;
2799 }
2800
2801 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002802 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002803 return -ENOMEM;
2804 }
2805
2806 /*
2807 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002808 * no architecture is using highmem pages for kernel page tables: so it
2809 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002810 */
2811 head = vmalloc_to_page(si->swap_map + offset);
2812 offset &= ~PAGE_MASK;
2813
2814 /*
2815 * Page allocation does not initialize the page's lru field,
2816 * but it does always reset its private field.
2817 */
2818 if (!page_private(head)) {
2819 BUG_ON(count & COUNT_CONTINUED);
2820 INIT_LIST_HEAD(&head->lru);
2821 set_page_private(head, SWP_CONTINUED);
2822 si->flags |= SWP_CONTINUED;
2823 }
2824
2825 list_for_each_entry(list_page, &head->lru, lru) {
2826 unsigned char *map;
2827
2828 /*
2829 * If the previous map said no continuation, but we've found
2830 * a continuation page, free our allocation and use this one.
2831 */
2832 if (!(count & COUNT_CONTINUED))
2833 goto out;
2834
Cong Wang9b04c5f2011-11-25 23:14:39 +08002835 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002836 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002837 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002838
2839 /*
2840 * If this continuation count now has some space in it,
2841 * free our allocation and use this one.
2842 */
2843 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2844 goto out;
2845 }
2846
2847 list_add_tail(&page->lru, &head->lru);
2848 page = NULL; /* now it's attached, don't free it */
2849out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002850 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002851outer:
2852 if (page)
2853 __free_page(page);
2854 return 0;
2855}
2856
2857/*
2858 * swap_count_continued - when the original swap_map count is incremented
2859 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2860 * into, carry if so, or else fail until a new continuation page is allocated;
2861 * when the original swap_map count is decremented from 0 with continuation,
2862 * borrow from the continuation and report whether it still holds more.
2863 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2864 */
2865static bool swap_count_continued(struct swap_info_struct *si,
2866 pgoff_t offset, unsigned char count)
2867{
2868 struct page *head;
2869 struct page *page;
2870 unsigned char *map;
2871
2872 head = vmalloc_to_page(si->swap_map + offset);
2873 if (page_private(head) != SWP_CONTINUED) {
2874 BUG_ON(count & COUNT_CONTINUED);
2875 return false; /* need to add count continuation */
2876 }
2877
2878 offset &= ~PAGE_MASK;
2879 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002880 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002881
2882 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2883 goto init_map; /* jump over SWAP_CONT_MAX checks */
2884
2885 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2886 /*
2887 * Think of how you add 1 to 999
2888 */
2889 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002890 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002891 page = list_entry(page->lru.next, struct page, lru);
2892 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002893 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002894 }
2895 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002896 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002897 page = list_entry(page->lru.next, struct page, lru);
2898 if (page == head)
2899 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002900 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002901init_map: *map = 0; /* we didn't zero the page */
2902 }
2903 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002904 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002905 page = list_entry(page->lru.prev, struct page, lru);
2906 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002907 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002908 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002909 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002910 page = list_entry(page->lru.prev, struct page, lru);
2911 }
2912 return true; /* incremented */
2913
2914 } else { /* decrementing */
2915 /*
2916 * Think of how you subtract 1 from 1000
2917 */
2918 BUG_ON(count != COUNT_CONTINUED);
2919 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002920 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002921 page = list_entry(page->lru.next, struct page, lru);
2922 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002923 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002924 }
2925 BUG_ON(*map == 0);
2926 *map -= 1;
2927 if (*map == 0)
2928 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002929 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002930 page = list_entry(page->lru.prev, struct page, lru);
2931 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002932 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002933 *map = SWAP_CONT_MAX | count;
2934 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002935 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002936 page = list_entry(page->lru.prev, struct page, lru);
2937 }
2938 return count == COUNT_CONTINUED;
2939 }
2940}
2941
2942/*
2943 * free_swap_count_continuations - swapoff free all the continuation pages
2944 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2945 */
2946static void free_swap_count_continuations(struct swap_info_struct *si)
2947{
2948 pgoff_t offset;
2949
2950 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2951 struct page *head;
2952 head = vmalloc_to_page(si->swap_map + offset);
2953 if (page_private(head)) {
Geliang Tang0d576d22016-01-14 15:21:49 -08002954 struct page *page, *next;
2955
2956 list_for_each_entry_safe(page, next, &head->lru, lru) {
2957 list_del(&page->lru);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002958 __free_page(page);
2959 }
2960 }
2961 }
2962}