blob: aebc2dd6e64975e2589429a0e22929f96c4d7532 [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;
51/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070052long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070053static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055static const char Bad_file[] = "Bad swap file entry ";
56static const char Unused_file[] = "Unused swap file entry ";
57static const char Bad_offset[] = "Bad swap offset entry ";
58static const char Unused_offset[] = "Unused swap offset entry ";
59
Dan Streetmanadfab832014-06-04 16:09:53 -070060/*
61 * all active swap_info_structs
62 * protected with swap_lock, and ordered by priority.
63 */
Dan Streetman18ab4d42014-06-04 16:09:59 -070064PLIST_HEAD(swap_active_head);
65
66/*
67 * all available (active, not full) swap_info_structs
68 * protected with swap_avail_lock, ordered by priority.
69 * This is used by get_swap_page() instead of swap_active_head
70 * because swap_active_head includes all swap_info_structs,
71 * but get_swap_page() doesn't need to look at full ones.
72 * This uses its own lock instead of swap_lock because when a
73 * swap_info_struct changes between not-full/full, it needs to
74 * add/remove itself to/from this list, but the swap_info_struct->lock
75 * is held and the locking order requires swap_lock to be taken
76 * before any swap_info_struct->lock.
77 */
78static PLIST_HEAD(swap_avail_head);
79static DEFINE_SPINLOCK(swap_avail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060081struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Ingo Molnarfc0abb12006-01-18 17:42:33 -080083static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Kay Sievers66d7dd52010-10-26 14:22:06 -070085static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
86/* Activity counter to indicate that a swapon or swapoff has occurred */
87static atomic_t proc_poll_event = ATOMIC_INIT(0);
88
Hugh Dickins8d69aae2009-12-14 17:58:45 -080089static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070090{
Hugh Dickins570a335b2009-12-14 17:58:46 -080091 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070092}
93
Hugh Dickinsefa90a92009-12-14 17:58:41 -080094/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070095static int
96__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
97{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080098 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070099 struct page *page;
100 int ret = 0;
101
Shaohua Li33806f02013-02-22 16:34:37 -0800102 page = find_get_page(swap_address_space(entry), entry.val);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700103 if (!page)
104 return 0;
105 /*
106 * This function is called from scan_swap_map() and it's called
107 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
108 * We have to use trylock for avoiding deadlock. This is a special
109 * case and you should use try_to_free_swap() with explicit lock_page()
110 * in usual operations.
111 */
112 if (trylock_page(page)) {
113 ret = try_to_free_swap(page);
114 unlock_page(page);
115 }
116 page_cache_release(page);
117 return ret;
118}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800121 * swapon tell device that all the old swap contents can be discarded,
122 * to allow the swap device to optimize its wear-levelling.
123 */
124static int discard_swap(struct swap_info_struct *si)
125{
126 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800127 sector_t start_block;
128 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800129 int err = 0;
130
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800131 /* Do not discard the swap header page! */
132 se = &si->first_swap_extent;
133 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
134 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
135 if (nr_blocks) {
136 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200137 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800138 if (err)
139 return err;
140 cond_resched();
141 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800142
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800143 list_for_each_entry(se, &si->first_swap_extent.list, list) {
144 start_block = se->start_block << (PAGE_SHIFT - 9);
145 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800146
147 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200148 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800149 if (err)
150 break;
151
152 cond_resched();
153 }
154 return err; /* That will often be -EOPNOTSUPP */
155}
156
Hugh Dickins7992fde2009-01-06 14:39:53 -0800157/*
158 * swap allocation tell device that a cluster of swap can now be discarded,
159 * to allow the swap device to optimize its wear-levelling.
160 */
161static void discard_swap_cluster(struct swap_info_struct *si,
162 pgoff_t start_page, pgoff_t nr_pages)
163{
164 struct swap_extent *se = si->curr_swap_extent;
165 int found_extent = 0;
166
167 while (nr_pages) {
168 struct list_head *lh;
169
170 if (se->start_page <= start_page &&
171 start_page < se->start_page + se->nr_pages) {
172 pgoff_t offset = start_page - se->start_page;
173 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800174 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800175
176 if (nr_blocks > nr_pages)
177 nr_blocks = nr_pages;
178 start_page += nr_blocks;
179 nr_pages -= nr_blocks;
180
181 if (!found_extent++)
182 si->curr_swap_extent = se;
183
184 start_block <<= PAGE_SHIFT - 9;
185 nr_blocks <<= PAGE_SHIFT - 9;
186 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200187 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800188 break;
189 }
190
191 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800192 se = list_entry(lh, struct swap_extent, list);
193 }
194}
195
Hugh Dickins048c27f2005-09-03 15:54:40 -0700196#define SWAPFILE_CLUSTER 256
197#define LATENCY_LIMIT 256
198
Shaohua Li2a8f9442013-09-11 14:20:28 -0700199static inline void cluster_set_flag(struct swap_cluster_info *info,
200 unsigned int flag)
201{
202 info->flags = flag;
203}
204
205static inline unsigned int cluster_count(struct swap_cluster_info *info)
206{
207 return info->data;
208}
209
210static inline void cluster_set_count(struct swap_cluster_info *info,
211 unsigned int c)
212{
213 info->data = c;
214}
215
216static inline void cluster_set_count_flag(struct swap_cluster_info *info,
217 unsigned int c, unsigned int f)
218{
219 info->flags = f;
220 info->data = c;
221}
222
223static inline unsigned int cluster_next(struct swap_cluster_info *info)
224{
225 return info->data;
226}
227
228static inline void cluster_set_next(struct swap_cluster_info *info,
229 unsigned int n)
230{
231 info->data = n;
232}
233
234static inline void cluster_set_next_flag(struct swap_cluster_info *info,
235 unsigned int n, unsigned int f)
236{
237 info->flags = f;
238 info->data = n;
239}
240
241static inline bool cluster_is_free(struct swap_cluster_info *info)
242{
243 return info->flags & CLUSTER_FLAG_FREE;
244}
245
246static inline bool cluster_is_null(struct swap_cluster_info *info)
247{
248 return info->flags & CLUSTER_FLAG_NEXT_NULL;
249}
250
251static inline void cluster_set_null(struct swap_cluster_info *info)
252{
253 info->flags = CLUSTER_FLAG_NEXT_NULL;
254 info->data = 0;
255}
256
Shaohua Li815c2c52013-09-11 14:20:30 -0700257/* Add a cluster to discard list and schedule it to do discard */
258static void swap_cluster_schedule_discard(struct swap_info_struct *si,
259 unsigned int idx)
260{
261 /*
262 * If scan_swap_map() can't find a free cluster, it will check
263 * si->swap_map directly. To make sure the discarding cluster isn't
264 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
265 * will be cleared after discard
266 */
267 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
268 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
269
270 if (cluster_is_null(&si->discard_cluster_head)) {
271 cluster_set_next_flag(&si->discard_cluster_head,
272 idx, 0);
273 cluster_set_next_flag(&si->discard_cluster_tail,
274 idx, 0);
275 } else {
276 unsigned int tail = cluster_next(&si->discard_cluster_tail);
277 cluster_set_next(&si->cluster_info[tail], idx);
278 cluster_set_next_flag(&si->discard_cluster_tail,
279 idx, 0);
280 }
281
282 schedule_work(&si->discard_work);
283}
284
285/*
286 * Doing discard actually. After a cluster discard is finished, the cluster
287 * will be added to free cluster list. caller should hold si->lock.
288*/
289static void swap_do_scheduled_discard(struct swap_info_struct *si)
290{
291 struct swap_cluster_info *info;
292 unsigned int idx;
293
294 info = si->cluster_info;
295
296 while (!cluster_is_null(&si->discard_cluster_head)) {
297 idx = cluster_next(&si->discard_cluster_head);
298
299 cluster_set_next_flag(&si->discard_cluster_head,
300 cluster_next(&info[idx]), 0);
301 if (cluster_next(&si->discard_cluster_tail) == idx) {
302 cluster_set_null(&si->discard_cluster_head);
303 cluster_set_null(&si->discard_cluster_tail);
304 }
305 spin_unlock(&si->lock);
306
307 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
308 SWAPFILE_CLUSTER);
309
310 spin_lock(&si->lock);
311 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
312 if (cluster_is_null(&si->free_cluster_head)) {
313 cluster_set_next_flag(&si->free_cluster_head,
314 idx, 0);
315 cluster_set_next_flag(&si->free_cluster_tail,
316 idx, 0);
317 } else {
318 unsigned int tail;
319
320 tail = cluster_next(&si->free_cluster_tail);
321 cluster_set_next(&info[tail], idx);
322 cluster_set_next_flag(&si->free_cluster_tail,
323 idx, 0);
324 }
325 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
326 0, SWAPFILE_CLUSTER);
327 }
328}
329
330static void swap_discard_work(struct work_struct *work)
331{
332 struct swap_info_struct *si;
333
334 si = container_of(work, struct swap_info_struct, discard_work);
335
336 spin_lock(&si->lock);
337 swap_do_scheduled_discard(si);
338 spin_unlock(&si->lock);
339}
340
Shaohua Li2a8f9442013-09-11 14:20:28 -0700341/*
342 * The cluster corresponding to page_nr will be used. The cluster will be
343 * removed from free cluster list and its usage counter will be increased.
344 */
345static void inc_cluster_info_page(struct swap_info_struct *p,
346 struct swap_cluster_info *cluster_info, unsigned long page_nr)
347{
348 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
349
350 if (!cluster_info)
351 return;
352 if (cluster_is_free(&cluster_info[idx])) {
353 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
354 cluster_set_next_flag(&p->free_cluster_head,
355 cluster_next(&cluster_info[idx]), 0);
356 if (cluster_next(&p->free_cluster_tail) == idx) {
357 cluster_set_null(&p->free_cluster_tail);
358 cluster_set_null(&p->free_cluster_head);
359 }
360 cluster_set_count_flag(&cluster_info[idx], 0, 0);
361 }
362
363 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
364 cluster_set_count(&cluster_info[idx],
365 cluster_count(&cluster_info[idx]) + 1);
366}
367
368/*
369 * The cluster corresponding to page_nr decreases one usage. If the usage
370 * counter becomes 0, which means no page in the cluster is in using, we can
371 * optionally discard the cluster and add it to free cluster list.
372 */
373static void dec_cluster_info_page(struct swap_info_struct *p,
374 struct swap_cluster_info *cluster_info, unsigned long page_nr)
375{
376 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
377
378 if (!cluster_info)
379 return;
380
381 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
382 cluster_set_count(&cluster_info[idx],
383 cluster_count(&cluster_info[idx]) - 1);
384
385 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700386 /*
387 * If the swap is discardable, prepare discard the cluster
388 * instead of free it immediately. The cluster will be freed
389 * after discard.
390 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700391 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
392 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700393 swap_cluster_schedule_discard(p, idx);
394 return;
395 }
396
Shaohua Li2a8f9442013-09-11 14:20:28 -0700397 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
398 if (cluster_is_null(&p->free_cluster_head)) {
399 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
400 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
401 } else {
402 unsigned int tail = cluster_next(&p->free_cluster_tail);
403 cluster_set_next(&cluster_info[tail], idx);
404 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
405 }
406 }
407}
408
409/*
410 * It's possible scan_swap_map() uses a free cluster in the middle of free
411 * cluster list. Avoiding such abuse to avoid list corruption.
412 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700413static bool
414scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700415 unsigned long offset)
416{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700417 struct percpu_cluster *percpu_cluster;
418 bool conflict;
419
Shaohua Li2a8f9442013-09-11 14:20:28 -0700420 offset /= SWAPFILE_CLUSTER;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700421 conflict = !cluster_is_null(&si->free_cluster_head) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700422 offset != cluster_next(&si->free_cluster_head) &&
423 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700424
425 if (!conflict)
426 return false;
427
428 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
429 cluster_set_null(&percpu_cluster->index);
430 return true;
431}
432
433/*
434 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
435 * might involve allocating a new cluster for current CPU too.
436 */
437static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
438 unsigned long *offset, unsigned long *scan_base)
439{
440 struct percpu_cluster *cluster;
441 bool found_free;
442 unsigned long tmp;
443
444new_cluster:
445 cluster = this_cpu_ptr(si->percpu_cluster);
446 if (cluster_is_null(&cluster->index)) {
447 if (!cluster_is_null(&si->free_cluster_head)) {
448 cluster->index = si->free_cluster_head;
449 cluster->next = cluster_next(&cluster->index) *
450 SWAPFILE_CLUSTER;
451 } else if (!cluster_is_null(&si->discard_cluster_head)) {
452 /*
453 * we don't have free cluster but have some clusters in
454 * discarding, do discard now and reclaim them
455 */
456 swap_do_scheduled_discard(si);
457 *scan_base = *offset = si->cluster_next;
458 goto new_cluster;
459 } else
460 return;
461 }
462
463 found_free = false;
464
465 /*
466 * Other CPUs can use our cluster if they can't find a free cluster,
467 * check if there is still free entry in the cluster
468 */
469 tmp = cluster->next;
470 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
471 SWAPFILE_CLUSTER) {
472 if (!si->swap_map[tmp]) {
473 found_free = true;
474 break;
475 }
476 tmp++;
477 }
478 if (!found_free) {
479 cluster_set_null(&cluster->index);
480 goto new_cluster;
481 }
482 cluster->next = tmp + 1;
483 *offset = tmp;
484 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700485}
486
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700487static unsigned long scan_swap_map(struct swap_info_struct *si,
488 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800490 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800491 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800492 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700493 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800495 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700496 * We try to cluster swap pages by allocating them sequentially
497 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
498 * way, however, we resort to first-free allocation, starting
499 * a new cluster. This prevents us from scattering swap pages
500 * all over the entire swap partition, so that we reduce
501 * overall disk seek times between swap pages. -- sct
502 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800503 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700504 */
505
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700506 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800507 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800508
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700509 /* SSD algorithm */
510 if (si->cluster_info) {
511 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
512 goto checks;
513 }
514
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800515 if (unlikely(!si->cluster_nr--)) {
516 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
517 si->cluster_nr = SWAPFILE_CLUSTER - 1;
518 goto checks;
519 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700520
Shaohua Liec8acf22013-02-22 16:34:38 -0800521 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700522
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800523 /*
524 * If seek is expensive, start searching for new cluster from
525 * start of partition, to minimize the span of allocated swap.
Chen Yucong50088c42014-06-04 16:10:57 -0700526 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
527 * case, just handled by scan_swap_map_try_ssd_cluster() above.
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800528 */
Chen Yucong50088c42014-06-04 16:10:57 -0700529 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700530 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
531
532 /* Locate the first empty (unaligned) cluster */
533 for (; last_in_cluster <= si->highest_bit; offset++) {
534 if (si->swap_map[offset])
535 last_in_cluster = offset + SWAPFILE_CLUSTER;
536 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800537 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800538 offset -= SWAPFILE_CLUSTER - 1;
539 si->cluster_next = offset;
540 si->cluster_nr = SWAPFILE_CLUSTER - 1;
541 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700543 if (unlikely(--latency_ration < 0)) {
544 cond_resched();
545 latency_ration = LATENCY_LIMIT;
546 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700547 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800548
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800549 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800550 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800551 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700553
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800554checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700555 if (si->cluster_info) {
556 while (scan_swap_map_ssd_cluster_conflict(si, offset))
557 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
558 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800559 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700560 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700561 if (!si->highest_bit)
562 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800563 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800564 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700565
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700566 /* reuse swap entry of cache-only swap if not busy. */
567 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700568 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800569 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700570 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800571 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700572 /* entry was freed successfully, try to use this again */
573 if (swap_was_freed)
574 goto checks;
575 goto scan; /* check next one */
576 }
577
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800578 if (si->swap_map[offset])
579 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700580
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800581 if (offset == si->lowest_bit)
582 si->lowest_bit++;
583 if (offset == si->highest_bit)
584 si->highest_bit--;
585 si->inuse_pages++;
586 if (si->inuse_pages == si->pages) {
587 si->lowest_bit = si->max;
588 si->highest_bit = 0;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700589 spin_lock(&swap_avail_lock);
590 plist_del(&si->avail_list, &swap_avail_head);
591 spin_unlock(&swap_avail_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800592 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800593 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700594 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800595 si->cluster_next = offset + 1;
596 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800597
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800598 return offset;
599
600scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800601 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700602 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700603 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800604 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700605 goto checks;
606 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700607 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800608 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700609 goto checks;
610 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700611 if (unlikely(--latency_ration < 0)) {
612 cond_resched();
613 latency_ration = LATENCY_LIMIT;
614 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700615 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800616 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800617 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800618 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800619 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800620 goto checks;
621 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700622 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800623 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700624 goto checks;
625 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800626 if (unlikely(--latency_ration < 0)) {
627 cond_resched();
628 latency_ration = LATENCY_LIMIT;
629 }
Jamie Liua5998062014-01-23 15:53:40 -0800630 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800631 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800632 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700633
634no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700635 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 return 0;
637}
638
639swp_entry_t get_swap_page(void)
640{
Dan Streetmanadfab832014-06-04 16:09:53 -0700641 struct swap_info_struct *si, *next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700642 pgoff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Shaohua Liec8acf22013-02-22 16:34:38 -0800644 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700645 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800646 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Dan Streetman18ab4d42014-06-04 16:09:59 -0700648 spin_lock(&swap_avail_lock);
649
650start_over:
651 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
652 /* requeue si to after same-priority siblings */
653 plist_requeue(&si->avail_list, &swap_avail_head);
654 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800655 spin_lock(&si->lock);
Dan Streetmanadfab832014-06-04 16:09:53 -0700656 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
Dan Streetman18ab4d42014-06-04 16:09:59 -0700657 spin_lock(&swap_avail_lock);
658 if (plist_node_empty(&si->avail_list)) {
659 spin_unlock(&si->lock);
660 goto nextsi;
661 }
662 WARN(!si->highest_bit,
663 "swap_info %d in list but !highest_bit\n",
664 si->type);
665 WARN(!(si->flags & SWP_WRITEOK),
666 "swap_info %d in list but !SWP_WRITEOK\n",
667 si->type);
668 plist_del(&si->avail_list, &swap_avail_head);
Shaohua Liec8acf22013-02-22 16:34:38 -0800669 spin_unlock(&si->lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700670 goto nextsi;
Shaohua Liec8acf22013-02-22 16:34:38 -0800671 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700672
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700673 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800674 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800675 spin_unlock(&si->lock);
676 if (offset)
Dan Streetmanadfab832014-06-04 16:09:53 -0700677 return swp_entry(si->type, offset);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700678 pr_debug("scan_swap_map of si %d failed to find offset\n",
679 si->type);
680 spin_lock(&swap_avail_lock);
681nextsi:
Dan Streetmanadfab832014-06-04 16:09:53 -0700682 /*
683 * if we got here, it's likely that si was almost full before,
684 * and since scan_swap_map() can drop the si->lock, multiple
685 * callers probably all tried to get a page from the same si
Dan Streetman18ab4d42014-06-04 16:09:59 -0700686 * and it filled up before we could get one; or, the si filled
687 * up between us dropping swap_avail_lock and taking si->lock.
688 * Since we dropped the swap_avail_lock, the swap_avail_head
689 * list may have been modified; so if next is still in the
690 * swap_avail_head list then try it, otherwise start over.
Dan Streetmanadfab832014-06-04 16:09:53 -0700691 */
Dan Streetman18ab4d42014-06-04 16:09:59 -0700692 if (plist_node_empty(&next->avail_list))
693 goto start_over;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700695
Dan Streetman18ab4d42014-06-04 16:09:59 -0700696 spin_unlock(&swap_avail_lock);
697
Shaohua Liec8acf22013-02-22 16:34:38 -0800698 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700699noswap:
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700700 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800703/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700704swp_entry_t get_swap_page_of_type(int type)
705{
706 struct swap_info_struct *si;
707 pgoff_t offset;
708
Hugh Dickins910321e2010-09-09 16:38:07 -0700709 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800710 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700711 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800712 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700713 /* This is called for allocating swap entry, not cache */
714 offset = scan_swap_map(si, 1);
715 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800716 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700717 return swp_entry(type, offset);
718 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800719 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700720 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800721 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700722 return (swp_entry_t) {0};
723}
724
Hugh Dickins73c34b62009-12-14 17:58:43 -0800725static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800727 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 unsigned long offset, type;
729
730 if (!entry.val)
731 goto out;
732 type = swp_type(entry);
733 if (type >= nr_swapfiles)
734 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800735 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 if (!(p->flags & SWP_USED))
737 goto bad_device;
738 offset = swp_offset(entry);
739 if (offset >= p->max)
740 goto bad_offset;
741 if (!p->swap_map[offset])
742 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800743 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 return p;
745
746bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700747 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 goto out;
749bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700750 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 goto out;
752bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700753 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 goto out;
755bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700756 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757out:
758 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800759}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800761static unsigned char swap_entry_free(struct swap_info_struct *p,
762 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763{
Hugh Dickins253d5532009-12-14 17:58:44 -0800764 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800765 unsigned char count;
766 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700768 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800769 has_cache = count & SWAP_HAS_CACHE;
770 count &= ~SWAP_HAS_CACHE;
771
772 if (usage == SWAP_HAS_CACHE) {
773 VM_BUG_ON(!has_cache);
774 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800775 } else if (count == SWAP_MAP_SHMEM) {
776 /*
777 * Or we could insist on shmem.c using a special
778 * swap_shmem_free() and free_shmem_swap_and_cache()...
779 */
780 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800781 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
782 if (count == COUNT_CONTINUED) {
783 if (swap_count_continued(p, offset, count))
784 count = SWAP_MAP_MAX | COUNT_CONTINUED;
785 else
786 count = SWAP_MAP_MAX;
787 } else
788 count--;
789 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800790
791 if (!count)
792 mem_cgroup_uncharge_swap(entry);
793
794 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) {
Shaohua Li2a8f9442013-09-11 14:20:28 -0700799 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700800 if (offset < p->lowest_bit)
801 p->lowest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700802 if (offset > p->highest_bit) {
803 bool was_full = !p->highest_bit;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700804 p->highest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700805 if (was_full && (p->flags & SWP_WRITEOK)) {
806 spin_lock(&swap_avail_lock);
807 WARN_ON(!plist_node_empty(&p->avail_list));
808 if (plist_node_empty(&p->avail_list))
809 plist_add(&p->avail_list,
810 &swap_avail_head);
811 spin_unlock(&swap_avail_lock);
812 }
813 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800814 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700815 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600816 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700817 if (p->flags & SWP_BLKDEV) {
818 struct gendisk *disk = p->bdev->bd_disk;
819 if (disk->fops->swap_slot_free_notify)
820 disk->fops->swap_slot_free_notify(p->bdev,
821 offset);
822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800824
825 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826}
827
828/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800829 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 * is still around or has not been recycled.
831 */
832void swap_free(swp_entry_t entry)
833{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800834 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836 p = swap_info_get(entry);
837 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800838 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800839 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
841}
842
843/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700844 * Called after dropping swapcache to decrease refcnt to swap entries.
845 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700846void swapcache_free(swp_entry_t entry)
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700847{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700848 struct swap_info_struct *p;
849
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700850 p = swap_info_get(entry);
851 if (p) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700852 swap_entry_free(p, entry, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800853 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700854 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700855}
856
857/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700858 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800859 * This does not give an exact answer when swap count is continued,
860 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700862int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700864 int count = 0;
865 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 swp_entry_t entry;
867
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700868 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 p = swap_info_get(entry);
870 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700871 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800872 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700874 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800878 * We can write to an anon page without COW if there are no other references
879 * to it. And as a side-effect, free up its swap: because the old content
880 * on disk will never be read, and seeking back there to write new content
881 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800883int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700885 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Sasha Levin309381fea2014-01-23 15:52:54 -0800887 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800888 if (unlikely(PageKsm(page)))
889 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700890 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800891 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700892 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800893 if (count == 1 && !PageWriteback(page)) {
894 delete_from_swap_cache(page);
895 SetPageDirty(page);
896 }
897 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800898 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
900
901/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800902 * If swap is getting full, or if there are no more mappings of this page,
903 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800905int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906{
Sasha Levin309381fea2014-01-23 15:52:54 -0800907 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 if (!PageSwapCache(page))
910 return 0;
911 if (PageWriteback(page))
912 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800913 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 return 0;
915
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700916 /*
917 * Once hibernation has begun to create its image of memory,
918 * there's a danger that one of the calls to try_to_free_swap()
919 * - most probably a call from __try_to_reclaim_swap() while
920 * hibernation is allocating its own swap pages for the image,
921 * but conceivably even a call from memory reclaim - will free
922 * the swap from a page which has already been recorded in the
923 * image as a clean swapcache page, and then reuse its swap for
924 * another page of the image. On waking from hibernation, the
925 * original page might be freed under memory pressure, then
926 * later read back in from swap, now with the wrong data.
927 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800928 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -0800929 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700930 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800931 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700932 return 0;
933
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800934 delete_from_swap_cache(page);
935 SetPageDirty(page);
936 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700937}
938
939/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 * Free the swap entry like above, but also try to
941 * free the page cache entry if it is the last user.
942 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800943int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800945 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 struct page *page = NULL;
947
Andi Kleena7420aa2009-09-16 11:50:05 +0200948 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800949 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 p = swap_info_get(entry);
952 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800953 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -0800954 page = find_get_page(swap_address_space(entry),
955 entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700956 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800957 page_cache_release(page);
958 page = NULL;
959 }
960 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800961 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 }
963 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800964 /*
965 * Not mapped elsewhere, or swap space full? Free it!
966 * Also recheck PageSwapCache now page is locked (above).
967 */
Nick Piggin93fac702006-03-31 02:29:56 -0800968 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800969 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 delete_from_swap_cache(page);
971 SetPageDirty(page);
972 }
973 unlock_page(page);
974 page_cache_release(page);
975 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800976 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977}
978
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200979#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800980/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800981 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800982 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800983 * @offset - number of the PAGE_SIZE-sized block of the device, starting
984 * from 0, in which the swap header is expected to be located.
985 *
986 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800987 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800988int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800989{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800990 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800991 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800992
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800993 if (device)
994 bdev = bdget(device);
995
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800996 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800997 for (type = 0; type < nr_swapfiles; type++) {
998 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800999
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001000 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001001 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001002
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001003 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001004 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001005 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001006
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001007 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001008 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001009 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001010 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001011 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001012
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001013 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001014 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001015 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001016
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001017 spin_unlock(&swap_lock);
1018 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001019 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001020 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001021 }
1022 }
1023 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001024 if (bdev)
1025 bdput(bdev);
1026
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001027 return -ENODEV;
1028}
1029
1030/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001031 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1032 * corresponding to given index in swap_info (swap type).
1033 */
1034sector_t swapdev_block(int type, pgoff_t offset)
1035{
1036 struct block_device *bdev;
1037
1038 if ((unsigned int)type >= nr_swapfiles)
1039 return 0;
1040 if (!(swap_info[type]->flags & SWP_WRITEOK))
1041 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001042 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001043}
1044
1045/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001046 * Return either the total number of swap pages of given type, or the number
1047 * of free pages of that type (depending on @free)
1048 *
1049 * This is needed for software suspend
1050 */
1051unsigned int count_swap_pages(int type, int free)
1052{
1053 unsigned int n = 0;
1054
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001055 spin_lock(&swap_lock);
1056 if ((unsigned int)type < nr_swapfiles) {
1057 struct swap_info_struct *sis = swap_info[type];
1058
Shaohua Liec8acf22013-02-22 16:34:38 -08001059 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001060 if (sis->flags & SWP_WRITEOK) {
1061 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001062 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001063 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001064 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001065 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001066 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001067 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001068 return n;
1069}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001070#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001071
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001072static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1073{
1074#ifdef CONFIG_MEM_SOFT_DIRTY
1075 /*
1076 * When pte keeps soft dirty bit the pte generated
1077 * from swap entry does not has it, still it's same
1078 * pte from logical point of view.
1079 */
1080 pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1081 return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1082#else
1083 return pte_same(pte, swp_pte);
1084#endif
1085}
1086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001088 * No need to decide whether this PTE shares the swap entry with others,
1089 * just let do_wp_page work it out if a write is requested later - to
1090 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001092static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 unsigned long addr, swp_entry_t entry, struct page *page)
1094{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001095 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001096 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001097 spinlock_t *ptl;
1098 pte_t *pte;
1099 int ret = 1;
1100
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001101 swapcache = page;
1102 page = ksm_might_need_to_copy(page, vma, addr);
1103 if (unlikely(!page))
1104 return -ENOMEM;
1105
Johannes Weiner00501b52014-08-08 14:19:20 -07001106 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001107 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001108 goto out_nolock;
1109 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001110
1111 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001112 if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
Johannes Weiner00501b52014-08-08 14:19:20 -07001113 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001114 ret = 0;
1115 goto out;
1116 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001117
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001118 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001119 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 get_page(page);
1121 set_pte_at(vma->vm_mm, addr, pte,
1122 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001123 if (page == swapcache) {
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001124 page_add_anon_rmap(page, vma, addr);
Johannes Weiner00501b52014-08-08 14:19:20 -07001125 mem_cgroup_commit_charge(page, memcg, true);
1126 } else { /* ksm created a completely new copy */
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001127 page_add_new_anon_rmap(page, vma, addr);
Johannes Weiner00501b52014-08-08 14:19:20 -07001128 mem_cgroup_commit_charge(page, memcg, false);
1129 lru_cache_add_active_or_unevictable(page, vma);
1130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 swap_free(entry);
1132 /*
1133 * Move the page to the active list so it is not
1134 * immediately swapped out again after swapon.
1135 */
1136 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001137out:
1138 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001139out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001140 if (page != swapcache) {
1141 unlock_page(page);
1142 put_page(page);
1143 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001144 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145}
1146
1147static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1148 unsigned long addr, unsigned long end,
1149 swp_entry_t entry, struct page *page)
1150{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001152 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001153 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Hugh Dickins044d66c2008-02-07 00:14:04 -08001155 /*
1156 * We don't actually need pte lock while scanning for swp_pte: since
1157 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1158 * page table while we're scanning; though it could get zapped, and on
1159 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1160 * of unmatched parts which look like swp_pte, so unuse_pte must
1161 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001162 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001163 */
1164 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 do {
1166 /*
1167 * swapoff spends a _lot_ of time in this loop!
1168 * Test inline before going to call unuse_pte.
1169 */
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001170 if (unlikely(maybe_same_pte(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001171 pte_unmap(pte);
1172 ret = unuse_pte(vma, pmd, addr, entry, page);
1173 if (ret)
1174 goto out;
1175 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
1177 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001178 pte_unmap(pte - 1);
1179out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001180 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
1183static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1184 unsigned long addr, unsigned long end,
1185 swp_entry_t entry, struct page *page)
1186{
1187 pmd_t *pmd;
1188 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001189 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 pmd = pmd_offset(pud, addr);
1192 do {
1193 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001194 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001196 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1197 if (ret)
1198 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 } while (pmd++, addr = next, addr != end);
1200 return 0;
1201}
1202
1203static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1204 unsigned long addr, unsigned long end,
1205 swp_entry_t entry, struct page *page)
1206{
1207 pud_t *pud;
1208 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001209 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211 pud = pud_offset(pgd, addr);
1212 do {
1213 next = pud_addr_end(addr, end);
1214 if (pud_none_or_clear_bad(pud))
1215 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001216 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1217 if (ret)
1218 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 } while (pud++, addr = next, addr != end);
1220 return 0;
1221}
1222
1223static int unuse_vma(struct vm_area_struct *vma,
1224 swp_entry_t entry, struct page *page)
1225{
1226 pgd_t *pgd;
1227 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001228 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001230 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 addr = page_address_in_vma(page, vma);
1232 if (addr == -EFAULT)
1233 return 0;
1234 else
1235 end = addr + PAGE_SIZE;
1236 } else {
1237 addr = vma->vm_start;
1238 end = vma->vm_end;
1239 }
1240
1241 pgd = pgd_offset(vma->vm_mm, addr);
1242 do {
1243 next = pgd_addr_end(addr, end);
1244 if (pgd_none_or_clear_bad(pgd))
1245 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001246 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1247 if (ret)
1248 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 } while (pgd++, addr = next, addr != end);
1250 return 0;
1251}
1252
1253static int unuse_mm(struct mm_struct *mm,
1254 swp_entry_t entry, struct page *page)
1255{
1256 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001257 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 if (!down_read_trylock(&mm->mmap_sem)) {
1260 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001261 * Activate page so shrink_inactive_list is unlikely to unmap
1262 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001264 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 unlock_page(page);
1266 down_read(&mm->mmap_sem);
1267 lock_page(page);
1268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001270 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 break;
1272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001274 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275}
1276
1277/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001278 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1279 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 * Recycle to start on reaching the end, returning 0 when empty.
1281 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001282static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001283 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001285 unsigned int max = si->max;
1286 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001287 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
1289 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001290 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 * for whether an entry is in use, not modifying it; false
1292 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001293 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 */
1295 for (;;) {
1296 if (++i >= max) {
1297 if (!prev) {
1298 i = 0;
1299 break;
1300 }
1301 /*
1302 * No entries in use at top of swap_map,
1303 * loop back to start and recheck there.
1304 */
1305 max = prev + 1;
1306 prev = 0;
1307 i = 1;
1308 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001309 if (frontswap) {
1310 if (frontswap_test(si, i))
1311 break;
1312 else
1313 continue;
1314 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001315 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001316 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 break;
1318 }
1319 return i;
1320}
1321
1322/*
1323 * We completely avoid races by reading each swap page in advance,
1324 * and then search for the process using it. All the necessary
1325 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001326 *
1327 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1328 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001330int try_to_unuse(unsigned int type, bool frontswap,
1331 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001333 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001335 volatile unsigned char *swap_map; /* swap_map is accessed without
1336 * locking. Mark it as volatile
1337 * to prevent compiler doing
1338 * something odd.
1339 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001340 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 struct page *page;
1342 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001343 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 /*
1347 * When searching mms for an entry, a good strategy is to
1348 * start at the first mm we freed the previous entry from
1349 * (though actually we don't notice whether we or coincidence
1350 * freed the entry). Initialize this start_mm with a hold.
1351 *
1352 * A simpler strategy would be to start at the last mm we
1353 * freed the previous entry from; but that would take less
1354 * advantage of mmlist ordering, which clusters forked mms
1355 * together, child after parent. If we race with dup_mmap(), we
1356 * prefer to resolve parent before child, lest we miss entries
1357 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001358 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 */
1360 start_mm = &init_mm;
1361 atomic_inc(&init_mm.mm_users);
1362
1363 /*
1364 * Keep on scanning until all entries have gone. Usually,
1365 * one pass through swap_map is enough, but not necessarily:
1366 * there are races when an instance of an entry might be missed.
1367 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001368 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 if (signal_pending(current)) {
1370 retval = -EINTR;
1371 break;
1372 }
1373
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001374 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 * Get a page for the entry, using the existing swap
1376 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001377 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 */
1379 swap_map = &si->swap_map[i];
1380 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001381 page = read_swap_cache_async(entry,
1382 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 if (!page) {
1384 /*
1385 * Either swap_duplicate() failed because entry
1386 * has been freed independently, and will not be
1387 * reused since sys_swapoff() already disabled
1388 * allocation from here, or alloc_page() failed.
1389 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001390 swcount = *swap_map;
1391 /*
1392 * We don't hold lock here, so the swap entry could be
1393 * SWAP_MAP_BAD (when the cluster is discarding).
1394 * Instead of fail out, We can just skip the swap
1395 * entry because swapoff will wait for discarding
1396 * finish anyway.
1397 */
1398 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 continue;
1400 retval = -ENOMEM;
1401 break;
1402 }
1403
1404 /*
1405 * Don't hold on to start_mm if it looks like exiting.
1406 */
1407 if (atomic_read(&start_mm->mm_users) == 1) {
1408 mmput(start_mm);
1409 start_mm = &init_mm;
1410 atomic_inc(&init_mm.mm_users);
1411 }
1412
1413 /*
1414 * Wait for and lock page. When do_swap_page races with
1415 * try_to_unuse, do_swap_page can handle the fault much
1416 * faster than try_to_unuse can locate the entry. This
1417 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1418 * defer to do_swap_page in such a case - in some tests,
1419 * do_swap_page and try_to_unuse repeatedly compete.
1420 */
1421 wait_on_page_locked(page);
1422 wait_on_page_writeback(page);
1423 lock_page(page);
1424 wait_on_page_writeback(page);
1425
1426 /*
1427 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001430 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1431 retval = shmem_unuse(entry, page);
1432 /* page has already been unlocked and released */
1433 if (retval < 0)
1434 break;
1435 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001437 if (swap_count(swcount) && start_mm != &init_mm)
1438 retval = unuse_mm(start_mm, entry, page);
1439
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001440 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 int set_start_mm = (*swap_map >= swcount);
1442 struct list_head *p = &start_mm->mmlist;
1443 struct mm_struct *new_start_mm = start_mm;
1444 struct mm_struct *prev_mm = start_mm;
1445 struct mm_struct *mm;
1446
1447 atomic_inc(&new_start_mm->mm_users);
1448 atomic_inc(&prev_mm->mm_users);
1449 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001450 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 (p = p->next) != &start_mm->mmlist) {
1452 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001453 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 spin_unlock(&mmlist_lock);
1456 mmput(prev_mm);
1457 prev_mm = mm;
1458
1459 cond_resched();
1460
1461 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001462 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001464 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001466 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001468
Bo Liu32c5fc12009-11-02 16:50:33 +00001469 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 mmput(new_start_mm);
1471 atomic_inc(&mm->mm_users);
1472 new_start_mm = mm;
1473 set_start_mm = 0;
1474 }
1475 spin_lock(&mmlist_lock);
1476 }
1477 spin_unlock(&mmlist_lock);
1478 mmput(prev_mm);
1479 mmput(start_mm);
1480 start_mm = new_start_mm;
1481 }
1482 if (retval) {
1483 unlock_page(page);
1484 page_cache_release(page);
1485 break;
1486 }
1487
1488 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 * If a reference remains (rare), we would like to leave
1490 * the page in the swap cache; but try_to_unmap could
1491 * then re-duplicate the entry once we drop page lock,
1492 * so we might loop indefinitely; also, that page could
1493 * not be swapped out to other storage meanwhile. So:
1494 * delete from cache even if there's another reference,
1495 * after ensuring that the data has been saved to disk -
1496 * since if the reference remains (rarer), it will be
1497 * read from disk into another page. Splitting into two
1498 * pages would be incorrect if swap supported "shared
1499 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001500 *
1501 * Given how unuse_vma() targets one particular offset
1502 * in an anon_vma, once the anon_vma has been determined,
1503 * this splitting happens to be just what is needed to
1504 * handle where KSM pages have been swapped out: re-reading
1505 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001507 if (swap_count(*swap_map) &&
1508 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 struct writeback_control wbc = {
1510 .sync_mode = WB_SYNC_NONE,
1511 };
1512
1513 swap_writepage(page, &wbc);
1514 lock_page(page);
1515 wait_on_page_writeback(page);
1516 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001517
1518 /*
1519 * It is conceivable that a racing task removed this page from
1520 * swap cache just before we acquired the page lock at the top,
1521 * or while we dropped it in unuse_mm(). The page might even
1522 * be back in swap cache on another swap area: that we must not
1523 * delete, since it may not have been written out to swap yet.
1524 */
1525 if (PageSwapCache(page) &&
1526 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001527 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529 /*
1530 * So we could skip searching mms once swap count went
1531 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001532 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 */
1534 SetPageDirty(page);
1535 unlock_page(page);
1536 page_cache_release(page);
1537
1538 /*
1539 * Make sure that we aren't completely killing
1540 * interactive performance.
1541 */
1542 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001543 if (frontswap && pages_to_unuse > 0) {
1544 if (!--pages_to_unuse)
1545 break;
1546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 }
1548
1549 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 return retval;
1551}
1552
1553/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001554 * After a successful try_to_unuse, if no swap is now in use, we know
1555 * we can empty the mmlist. swap_lock must be held on entry and exit.
1556 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 * added to the mmlist just after page_duplicate - before would be racy.
1558 */
1559static void drain_mmlist(void)
1560{
1561 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001562 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001564 for (type = 0; type < nr_swapfiles; type++)
1565 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 return;
1567 spin_lock(&mmlist_lock);
1568 list_for_each_safe(p, next, &init_mm.mmlist)
1569 list_del_init(p);
1570 spin_unlock(&mmlist_lock);
1571}
1572
1573/*
1574 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001575 * corresponds to page offset for the specified swap entry.
1576 * Note that the type of this function is sector_t, but it returns page offset
1577 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001579static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001581 struct swap_info_struct *sis;
1582 struct swap_extent *start_se;
1583 struct swap_extent *se;
1584 pgoff_t offset;
1585
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001586 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001587 *bdev = sis->bdev;
1588
1589 offset = swp_offset(entry);
1590 start_se = sis->curr_swap_extent;
1591 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
1593 for ( ; ; ) {
1594 struct list_head *lh;
1595
1596 if (se->start_page <= offset &&
1597 offset < (se->start_page + se->nr_pages)) {
1598 return se->start_block + (offset - se->start_page);
1599 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001600 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 se = list_entry(lh, struct swap_extent, list);
1602 sis->curr_swap_extent = se;
1603 BUG_ON(se == start_se); /* It *must* be present */
1604 }
1605}
1606
1607/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001608 * Returns the page offset into bdev for the specified page's swap entry.
1609 */
1610sector_t map_swap_page(struct page *page, struct block_device **bdev)
1611{
1612 swp_entry_t entry;
1613 entry.val = page_private(page);
1614 return map_swap_entry(entry, bdev);
1615}
1616
1617/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 * Free all of a swapdev's extent information
1619 */
1620static void destroy_swap_extents(struct swap_info_struct *sis)
1621{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001622 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 struct swap_extent *se;
1624
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001625 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 struct swap_extent, list);
1627 list_del(&se->list);
1628 kfree(se);
1629 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001630
1631 if (sis->flags & SWP_FILE) {
1632 struct file *swap_file = sis->swap_file;
1633 struct address_space *mapping = swap_file->f_mapping;
1634
1635 sis->flags &= ~SWP_FILE;
1636 mapping->a_ops->swap_deactivate(swap_file);
1637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638}
1639
1640/*
1641 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001642 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001644 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 */
Mel Gormana509bc12012-07-31 16:44:57 -07001646int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1648 unsigned long nr_pages, sector_t start_block)
1649{
1650 struct swap_extent *se;
1651 struct swap_extent *new_se;
1652 struct list_head *lh;
1653
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001654 if (start_page == 0) {
1655 se = &sis->first_swap_extent;
1656 sis->curr_swap_extent = se;
1657 se->start_page = 0;
1658 se->nr_pages = nr_pages;
1659 se->start_block = start_block;
1660 return 1;
1661 } else {
1662 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001664 BUG_ON(se->start_page + se->nr_pages != start_page);
1665 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 /* Merge it */
1667 se->nr_pages += nr_pages;
1668 return 0;
1669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
1671
1672 /*
1673 * No merge. Insert a new extent, preserving ordering.
1674 */
1675 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1676 if (new_se == NULL)
1677 return -ENOMEM;
1678 new_se->start_page = start_page;
1679 new_se->nr_pages = nr_pages;
1680 new_se->start_block = start_block;
1681
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001682 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001683 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684}
1685
1686/*
1687 * A `swap extent' is a simple thing which maps a contiguous range of pages
1688 * onto a contiguous range of disk blocks. An ordered list of swap extents
1689 * is built at swapon time and is then used at swap_writepage/swap_readpage
1690 * time for locating where on disk a page belongs.
1691 *
1692 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1693 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1694 * swap files identically.
1695 *
1696 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1697 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1698 * swapfiles are handled *identically* after swapon time.
1699 *
1700 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1701 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1702 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1703 * requirements, they are simply tossed out - we will never use those blocks
1704 * for swapping.
1705 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001706 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1708 * which will scribble on the fs.
1709 *
1710 * The amount of disk space which a single swap extent represents varies.
1711 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1712 * extents in the list. To avoid much list walking, we cache the previous
1713 * search location in `curr_swap_extent', and start new searches from there.
1714 * This is extremely effective. The average number of iterations in
1715 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1716 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001717static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718{
Mel Gorman62c230b2012-07-31 16:44:55 -07001719 struct file *swap_file = sis->swap_file;
1720 struct address_space *mapping = swap_file->f_mapping;
1721 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 int ret;
1723
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 if (S_ISBLK(inode->i_mode)) {
1725 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001726 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001727 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 }
1729
Mel Gorman62c230b2012-07-31 16:44:55 -07001730 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001731 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001732 if (!ret) {
1733 sis->flags |= SWP_FILE;
1734 ret = add_swap_extent(sis, 0, sis->max, 0);
1735 *span = sis->pages;
1736 }
Mel Gormana509bc12012-07-31 16:44:57 -07001737 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001738 }
1739
Mel Gormana509bc12012-07-31 16:44:57 -07001740 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741}
1742
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001743static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001744 unsigned char *swap_map,
1745 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001746{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001747 if (prio >= 0)
1748 p->prio = prio;
1749 else
1750 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001751 /*
1752 * the plist prio is negated because plist ordering is
1753 * low-to-high, while swap ordering is high-to-low
1754 */
1755 p->list.prio = -p->prio;
1756 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001757 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001758 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001759 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001760 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001761 total_swap_pages += p->pages;
1762
Dan Streetmanadfab832014-06-04 16:09:53 -07001763 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001764 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07001765 * both lists are plists, and thus priority ordered.
1766 * swap_active_head needs to be priority ordered for swapoff(),
1767 * which on removal of any swap_info_struct with an auto-assigned
1768 * (i.e. negative) priority increments the auto-assigned priority
1769 * of any lower-priority swap_info_structs.
1770 * swap_avail_head needs to be priority ordered for get_swap_page(),
1771 * which allocates swap pages from the highest available priority
1772 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07001773 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07001774 plist_add(&p->list, &swap_active_head);
1775 spin_lock(&swap_avail_lock);
1776 plist_add(&p->avail_list, &swap_avail_head);
1777 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001778}
1779
1780static void enable_swap_info(struct swap_info_struct *p, int prio,
1781 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001782 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001783 unsigned long *frontswap_map)
1784{
Minchan Kim4f898492013-04-30 15:26:54 -07001785 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001786 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001787 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001788 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001789 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001790 spin_unlock(&swap_lock);
1791}
1792
1793static void reinsert_swap_info(struct swap_info_struct *p)
1794{
1795 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001796 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001797 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001798 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001799 spin_unlock(&swap_lock);
1800}
1801
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001802SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001804 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001805 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001806 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001807 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 struct file *swap_file, *victim;
1809 struct address_space *mapping;
1810 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001811 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07001812 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001813 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001814
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 if (!capable(CAP_SYS_ADMIN))
1816 return -EPERM;
1817
Al Viro191c5422012-02-13 03:58:52 +00001818 BUG_ON(!current->mm);
1819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001822 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Jeff Layton669abf42012-10-10 16:43:10 -04001824 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 err = PTR_ERR(victim);
1826 if (IS_ERR(victim))
1827 goto out;
1828
1829 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001830 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07001831 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001832 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001833 if (p->swap_file->f_mapping == mapping) {
1834 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07001836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 }
Dan Streetmanadfab832014-06-04 16:09:53 -07001839 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001841 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 goto out_dput;
1843 }
Al Viro191c5422012-02-13 03:58:52 +00001844 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 vm_unacct_memory(p->pages);
1846 else {
1847 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001848 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 goto out_dput;
1850 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001851 spin_lock(&swap_avail_lock);
1852 plist_del(&p->avail_list, &swap_avail_head);
1853 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001854 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001855 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001856 struct swap_info_struct *si = p;
1857
Dan Streetman18ab4d42014-06-04 16:09:59 -07001858 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001859 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001860 si->list.prio--;
1861 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07001862 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001863 least_priority++;
1864 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001865 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08001866 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 total_swap_pages -= p->pages;
1868 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001869 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001870 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001871
David Rientjese1e12d22012-12-11 16:02:56 -08001872 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07001873 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08001874 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 if (err) {
1877 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001878 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 goto out_dput;
1880 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001881
Shaohua Li815c2c52013-09-11 14:20:30 -07001882 flush_work(&p->discard_work);
1883
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001884 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001885 if (p->flags & SWP_CONTINUED)
1886 free_swap_count_continuations(p);
1887
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001888 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001889 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001890 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001892
1893 /* wait for anyone still in scan_swap_map */
1894 p->highest_bit = 0; /* cuts scans short */
1895 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001896 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001897 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001898 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001899 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001900 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001901 }
1902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001904 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 p->swap_file = NULL;
1906 p->max = 0;
1907 swap_map = p->swap_map;
1908 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001909 cluster_info = p->cluster_info;
1910 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07001911 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08001912 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001913 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001914 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08001915 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001916 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001917 free_percpu(p->percpu_cluster);
1918 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001920 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001921 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001922 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07001923 swap_cgroup_swapoff(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001924
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 inode = mapping->host;
1926 if (S_ISBLK(inode->i_mode)) {
1927 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001928 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001929 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001931 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001933 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 }
1935 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08001936
1937 /*
1938 * Clear the SWP_USED flag after all resources are freed so that swapon
1939 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1940 * not hold p->lock after we cleared its SWP_WRITEOK.
1941 */
1942 spin_lock(&swap_lock);
1943 p->flags = 0;
1944 spin_unlock(&swap_lock);
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001947 atomic_inc(&proc_poll_event);
1948 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
1950out_dput:
1951 filp_close(victim, NULL);
1952out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001953 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return err;
1955}
1956
1957#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001958static unsigned swaps_poll(struct file *file, poll_table *wait)
1959{
Kay Sieversf1514632011-07-12 20:48:39 +02001960 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001961
1962 poll_wait(file, &proc_poll_wait, wait);
1963
Kay Sieversf1514632011-07-12 20:48:39 +02001964 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1965 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001966 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1967 }
1968
1969 return POLLIN | POLLRDNORM;
1970}
1971
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972/* iterator */
1973static void *swap_start(struct seq_file *swap, loff_t *pos)
1974{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001975 struct swap_info_struct *si;
1976 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 loff_t l = *pos;
1978
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001979 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001981 if (!l)
1982 return SEQ_START_TOKEN;
1983
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001984 for (type = 0; type < nr_swapfiles; type++) {
1985 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1986 si = swap_info[type];
1987 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001989 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001990 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 }
1992
1993 return NULL;
1994}
1995
1996static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1997{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001998 struct swap_info_struct *si = v;
1999 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002001 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002002 type = 0;
2003 else
2004 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002005
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002006 for (; type < nr_swapfiles; type++) {
2007 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2008 si = swap_info[type];
2009 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 continue;
2011 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002012 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 }
2014
2015 return NULL;
2016}
2017
2018static void swap_stop(struct seq_file *swap, void *v)
2019{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002020 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021}
2022
2023static int swap_show(struct seq_file *swap, void *v)
2024{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002025 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 struct file *file;
2027 int len;
2028
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002029 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002030 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2031 return 0;
2032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002034 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002035 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002036 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002037 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002038 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002040 si->pages << (PAGE_SHIFT - 10),
2041 si->inuse_pages << (PAGE_SHIFT - 10),
2042 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return 0;
2044}
2045
Helge Deller15ad7cd2006-12-06 20:40:36 -08002046static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 .start = swap_start,
2048 .next = swap_next,
2049 .stop = swap_stop,
2050 .show = swap_show
2051};
2052
2053static int swaps_open(struct inode *inode, struct file *file)
2054{
Kay Sieversf1514632011-07-12 20:48:39 +02002055 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002056 int ret;
2057
Kay Sievers66d7dd52010-10-26 14:22:06 -07002058 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002059 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002060 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002061
Kay Sieversf1514632011-07-12 20:48:39 +02002062 seq = file->private_data;
2063 seq->poll_event = atomic_read(&proc_poll_event);
2064 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065}
2066
Helge Deller15ad7cd2006-12-06 20:40:36 -08002067static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 .open = swaps_open,
2069 .read = seq_read,
2070 .llseek = seq_lseek,
2071 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002072 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073};
2074
2075static int __init procswaps_init(void)
2076{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002077 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 return 0;
2079}
2080__initcall(procswaps_init);
2081#endif /* CONFIG_PROC_FS */
2082
Jan Beulich17963162008-12-16 11:35:24 +00002083#ifdef MAX_SWAPFILES_CHECK
2084static int __init max_swapfiles_check(void)
2085{
2086 MAX_SWAPFILES_CHECK();
2087 return 0;
2088}
2089late_initcall(max_swapfiles_check);
2090#endif
2091
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002092static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002094 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002096
2097 p = kzalloc(sizeof(*p), GFP_KERNEL);
2098 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002099 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002100
Hugh Dickins5d337b92005-09-03 15:54:41 -07002101 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002102 for (type = 0; type < nr_swapfiles; type++) {
2103 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002105 }
Christoph Lameter06972122006-06-23 02:03:35 -07002106 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002107 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002108 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002109 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002111 if (type >= nr_swapfiles) {
2112 p->type = type;
2113 swap_info[type] = p;
2114 /*
2115 * Write swap_info[type] before nr_swapfiles, in case a
2116 * racing procfs swap_start() or swap_next() is reading them.
2117 * (We never shrink nr_swapfiles, we never free this entry.)
2118 */
2119 smp_wmb();
2120 nr_swapfiles++;
2121 } else {
2122 kfree(p);
2123 p = swap_info[type];
2124 /*
2125 * Do not memset this entry: a racing procfs swap_next()
2126 * would be relying on p->type to remain valid.
2127 */
2128 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002129 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002130 plist_node_init(&p->list, 0);
2131 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002133 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002134 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002135
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002136 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002137}
2138
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002139static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2140{
2141 int error;
2142
2143 if (S_ISBLK(inode->i_mode)) {
2144 p->bdev = bdgrab(I_BDEV(inode));
2145 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002146 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002147 if (error < 0) {
2148 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002149 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002150 }
2151 p->old_block_size = block_size(p->bdev);
2152 error = set_blocksize(p->bdev, PAGE_SIZE);
2153 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002154 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002155 p->flags |= SWP_BLKDEV;
2156 } else if (S_ISREG(inode->i_mode)) {
2157 p->bdev = inode->i_sb->s_bdev;
2158 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002159 if (IS_SWAPFILE(inode))
2160 return -EBUSY;
2161 } else
2162 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002163
2164 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002165}
2166
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002167static unsigned long read_swap_header(struct swap_info_struct *p,
2168 union swap_header *swap_header,
2169 struct inode *inode)
2170{
2171 int i;
2172 unsigned long maxpages;
2173 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002174 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002175
2176 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002177 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002178 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002179 }
2180
2181 /* swap partition endianess hack... */
2182 if (swab32(swap_header->info.version) == 1) {
2183 swab32s(&swap_header->info.version);
2184 swab32s(&swap_header->info.last_page);
2185 swab32s(&swap_header->info.nr_badpages);
2186 for (i = 0; i < swap_header->info.nr_badpages; i++)
2187 swab32s(&swap_header->info.badpages[i]);
2188 }
2189 /* Check the swap header's sub-version */
2190 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002191 pr_warn("Unable to handle swap header version %d\n",
2192 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002193 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002194 }
2195
2196 p->lowest_bit = 1;
2197 p->cluster_next = 1;
2198 p->cluster_nr = 0;
2199
2200 /*
2201 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002202 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002203 * of bits for the swap offset in the swp_entry_t type, and
2204 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002205 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002206 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002207 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002208 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002209 * offset is extracted. This will mask all the bits from
2210 * the initial ~0UL mask that can't be encoded in either
2211 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002212 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002213 */
2214 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002215 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002216 last_page = swap_header->info.last_page;
2217 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002218 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002219 maxpages << (PAGE_SHIFT - 10),
2220 last_page << (PAGE_SHIFT - 10));
2221 }
2222 if (maxpages > last_page) {
2223 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002224 /* p->max is an unsigned int: don't overflow it */
2225 if ((unsigned int)maxpages == 0)
2226 maxpages = UINT_MAX;
2227 }
2228 p->highest_bit = maxpages - 1;
2229
2230 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002231 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002232 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2233 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002234 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002235 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002236 }
2237 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002238 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002239 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002240 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002241
2242 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002243}
2244
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002245static int setup_swap_map_and_extents(struct swap_info_struct *p,
2246 union swap_header *swap_header,
2247 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002248 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002249 unsigned long maxpages,
2250 sector_t *span)
2251{
2252 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002253 unsigned int nr_good_pages;
2254 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002255 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2256 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002257
2258 nr_good_pages = maxpages - 1; /* omit header page */
2259
Shaohua Li2a8f9442013-09-11 14:20:28 -07002260 cluster_set_null(&p->free_cluster_head);
2261 cluster_set_null(&p->free_cluster_tail);
Shaohua Li815c2c52013-09-11 14:20:30 -07002262 cluster_set_null(&p->discard_cluster_head);
2263 cluster_set_null(&p->discard_cluster_tail);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002264
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002265 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2266 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002267 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2268 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002269 if (page_nr < maxpages) {
2270 swap_map[page_nr] = SWAP_MAP_BAD;
2271 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002272 /*
2273 * Haven't marked the cluster free yet, no list
2274 * operation involved
2275 */
2276 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002277 }
2278 }
2279
Shaohua Li2a8f9442013-09-11 14:20:28 -07002280 /* Haven't marked the cluster free yet, no list operation involved */
2281 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2282 inc_cluster_info_page(p, cluster_info, i);
2283
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002284 if (nr_good_pages) {
2285 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002286 /*
2287 * Not mark the cluster free yet, no list
2288 * operation involved
2289 */
2290 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002291 p->max = maxpages;
2292 p->pages = nr_good_pages;
2293 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002294 if (nr_extents < 0)
2295 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002296 nr_good_pages = p->pages;
2297 }
2298 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002299 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002300 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002301 }
2302
Shaohua Li2a8f9442013-09-11 14:20:28 -07002303 if (!cluster_info)
2304 return nr_extents;
2305
2306 for (i = 0; i < nr_clusters; i++) {
2307 if (!cluster_count(&cluster_info[idx])) {
2308 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2309 if (cluster_is_null(&p->free_cluster_head)) {
2310 cluster_set_next_flag(&p->free_cluster_head,
2311 idx, 0);
2312 cluster_set_next_flag(&p->free_cluster_tail,
2313 idx, 0);
2314 } else {
2315 unsigned int tail;
2316
2317 tail = cluster_next(&p->free_cluster_tail);
2318 cluster_set_next(&cluster_info[tail], idx);
2319 cluster_set_next_flag(&p->free_cluster_tail,
2320 idx, 0);
2321 }
2322 }
2323 idx++;
2324 if (idx == nr_clusters)
2325 idx = 0;
2326 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002327 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002328}
2329
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002330/*
2331 * Helper to sys_swapon determining if a given swap
2332 * backing device queue supports DISCARD operations.
2333 */
2334static bool swap_discardable(struct swap_info_struct *si)
2335{
2336 struct request_queue *q = bdev_get_queue(si->bdev);
2337
2338 if (!q || !blk_queue_discard(q))
2339 return false;
2340
2341 return true;
2342}
2343
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002344SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2345{
2346 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002347 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002348 struct file *swap_file = NULL;
2349 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002350 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002351 int error;
2352 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002353 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002354 sector_t span;
2355 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002356 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002357 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002358 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002359 struct page *page = NULL;
2360 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002361
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002362 if (swap_flags & ~SWAP_FLAGS_VALID)
2363 return -EINVAL;
2364
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002365 if (!capable(CAP_SYS_ADMIN))
2366 return -EPERM;
2367
2368 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002369 if (IS_ERR(p))
2370 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002371
Shaohua Li815c2c52013-09-11 14:20:30 -07002372 INIT_WORK(&p->discard_work, swap_discard_work);
2373
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002376 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002378 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 }
Jeff Layton669abf42012-10-10 16:43:10 -04002380 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002382 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002384 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 }
2386
2387 p->swap_file = swap_file;
2388 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002389 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002390
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002391 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002392 error = claim_swapfile(p, inode);
2393 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 /*
2397 * Read the swap header.
2398 */
2399 if (!mapping->a_ops->readpage) {
2400 error = -EINVAL;
2401 goto bad_swap;
2402 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002403 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 if (IS_ERR(page)) {
2405 error = PTR_ERR(page);
2406 goto bad_swap;
2407 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002408 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002410 maxpages = read_swap_header(p, swap_header, inode);
2411 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 error = -EINVAL;
2413 goto bad_swap;
2414 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002415
Hugh Dickins81e33972009-01-06 14:39:49 -08002416 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002417 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002418 if (!swap_map) {
2419 error = -ENOMEM;
2420 goto bad_swap;
2421 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002422 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002423 int cpu;
2424
Shaohua Li2a8f9442013-09-11 14:20:28 -07002425 p->flags |= SWP_SOLIDSTATE;
2426 /*
2427 * select a random position to start with to help wear leveling
2428 * SSD
2429 */
2430 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2431
2432 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2433 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2434 if (!cluster_info) {
2435 error = -ENOMEM;
2436 goto bad_swap;
2437 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002438 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2439 if (!p->percpu_cluster) {
2440 error = -ENOMEM;
2441 goto bad_swap;
2442 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002443 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002444 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002445 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002446 cluster_set_null(&cluster->index);
2447 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002448 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002449
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002450 error = swap_cgroup_swapon(p->type, maxpages);
2451 if (error)
2452 goto bad_swap;
2453
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002454 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002455 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002456 if (unlikely(nr_extents < 0)) {
2457 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 goto bad_swap;
2459 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002460 /* frontswap enabled? set up bit-per-page map for frontswap */
2461 if (frontswap_enabled)
Akinobu Mita7b579762013-06-12 14:05:08 -07002462 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Shaohua Li2a8f9442013-09-11 14:20:28 -07002464 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2465 /*
2466 * When discard is enabled for swap with no particular
2467 * policy flagged, we set all swap discard flags here in
2468 * order to sustain backward compatibility with older
2469 * swapon(8) releases.
2470 */
2471 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2472 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002473
Shaohua Li2a8f9442013-09-11 14:20:28 -07002474 /*
2475 * By flagging sys_swapon, a sysadmin can tell us to
2476 * either do single-time area discards only, or to just
2477 * perform discards for released swap page-clusters.
2478 * Now it's time to adjust the p->flags accordingly.
2479 */
2480 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2481 p->flags &= ~SWP_PAGE_DISCARD;
2482 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2483 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002484
Shaohua Li2a8f9442013-09-11 14:20:28 -07002485 /* issue a swapon-time discard if it's still required */
2486 if (p->flags & SWP_AREA_DISCARD) {
2487 int err = discard_swap(p);
2488 if (unlikely(err))
2489 pr_err("swapon: discard_swap(%p): %d\n",
2490 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002491 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002492 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002493
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002494 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002495 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002496 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002497 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002498 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002499 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002500
Andrew Morton465c47f2013-09-11 14:20:17 -07002501 pr_info("Adding %uk swap on %s. "
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002502 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002503 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002504 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2505 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002506 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002507 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2508 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002509 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002510
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002511 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002512 atomic_inc(&proc_poll_event);
2513 wake_up_interruptible(&proc_poll_wait);
2514
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002515 if (S_ISREG(inode->i_mode))
2516 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 error = 0;
2518 goto out;
2519bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002520 free_percpu(p->percpu_cluster);
2521 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002522 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002523 set_blocksize(p->bdev, p->old_block_size);
2524 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002526 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002527 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002528 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002531 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002533 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002534 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002535 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002536 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002537 inode = NULL;
2538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541out:
2542 if (page && !IS_ERR(page)) {
2543 kunmap(page);
2544 page_cache_release(page);
2545 }
2546 if (name)
2547 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002548 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002549 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 return error;
2551}
2552
2553void si_swapinfo(struct sysinfo *val)
2554{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002555 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 unsigned long nr_to_be_unused = 0;
2557
Hugh Dickins5d337b92005-09-03 15:54:41 -07002558 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002559 for (type = 0; type < nr_swapfiles; type++) {
2560 struct swap_info_struct *si = swap_info[type];
2561
2562 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2563 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002565 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002567 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568}
2569
2570/*
2571 * Verify that a swap entry is valid and increment its swap map count.
2572 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002573 * Returns error code in following case.
2574 * - success -> 0
2575 * - swp_entry is invalid -> EINVAL
2576 * - swp_entry is migration entry -> EINVAL
2577 * - swap-cache reference is requested but there is already one. -> EEXIST
2578 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002579 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002581static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002583 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002585 unsigned char count;
2586 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002587 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
Andi Kleena7420aa2009-09-16 11:50:05 +02002589 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002590 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002591
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 type = swp_type(entry);
2593 if (type >= nr_swapfiles)
2594 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002595 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 offset = swp_offset(entry);
2597
Shaohua Liec8acf22013-02-22 16:34:38 -08002598 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002599 if (unlikely(offset >= p->max))
2600 goto unlock_out;
2601
Hugh Dickins253d5532009-12-14 17:58:44 -08002602 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002603
2604 /*
2605 * swapin_readahead() doesn't check if a swap entry is valid, so the
2606 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2607 */
2608 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2609 err = -ENOENT;
2610 goto unlock_out;
2611 }
2612
Hugh Dickins253d5532009-12-14 17:58:44 -08002613 has_cache = count & SWAP_HAS_CACHE;
2614 count &= ~SWAP_HAS_CACHE;
2615 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002616
Hugh Dickins253d5532009-12-14 17:58:44 -08002617 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002618
2619 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002620 if (!has_cache && count)
2621 has_cache = SWAP_HAS_CACHE;
2622 else if (has_cache) /* someone else added cache */
2623 err = -EEXIST;
2624 else /* no users remaining */
2625 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002626
2627 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002628
Hugh Dickins570a335b2009-12-14 17:58:46 -08002629 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2630 count += usage;
2631 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002632 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002633 else if (swap_count_continued(p, offset, count))
2634 count = COUNT_CONTINUED;
2635 else
2636 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002637 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002638 err = -ENOENT; /* unused swap entry */
2639
2640 p->swap_map[offset] = count | has_cache;
2641
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002642unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002643 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002645 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
2647bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002648 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 goto out;
2650}
Hugh Dickins253d5532009-12-14 17:58:44 -08002651
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002652/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002653 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2654 * (in which case its reference count is never incremented).
2655 */
2656void swap_shmem_alloc(swp_entry_t entry)
2657{
2658 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2659}
2660
2661/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002662 * Increase reference count of swap entry by 1.
2663 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2664 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2665 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2666 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002667 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002668int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002669{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002670 int err = 0;
2671
2672 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2673 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2674 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002675}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002677/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002678 * @entry: swap entry for which we allocate swap cache.
2679 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002680 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002681 * This can return error codes. Returns 0 at success.
2682 * -EBUSY means there is a swap cache.
2683 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002684 */
2685int swapcache_prepare(swp_entry_t entry)
2686{
Hugh Dickins253d5532009-12-14 17:58:44 -08002687 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002688}
2689
Mel Gormanf981c592012-07-31 16:44:47 -07002690struct swap_info_struct *page_swap_info(struct page *page)
2691{
2692 swp_entry_t swap = { .val = page_private(page) };
2693 BUG_ON(!PageSwapCache(page));
2694 return swap_info[swp_type(swap)];
2695}
2696
2697/*
2698 * out-of-line __page_file_ methods to avoid include hell.
2699 */
2700struct address_space *__page_file_mapping(struct page *page)
2701{
Sasha Levin309381fea2014-01-23 15:52:54 -08002702 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002703 return page_swap_info(page)->swap_file->f_mapping;
2704}
2705EXPORT_SYMBOL_GPL(__page_file_mapping);
2706
2707pgoff_t __page_file_index(struct page *page)
2708{
2709 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002710 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002711 return swp_offset(swap);
2712}
2713EXPORT_SYMBOL_GPL(__page_file_index);
2714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002716 * add_swap_count_continuation - called when a swap count is duplicated
2717 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2718 * page of the original vmalloc'ed swap_map, to hold the continuation count
2719 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2720 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2721 *
2722 * These continuation pages are seldom referenced: the common paths all work
2723 * on the original swap_map, only referring to a continuation page when the
2724 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2725 *
2726 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2727 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2728 * can be called after dropping locks.
2729 */
2730int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2731{
2732 struct swap_info_struct *si;
2733 struct page *head;
2734 struct page *page;
2735 struct page *list_page;
2736 pgoff_t offset;
2737 unsigned char count;
2738
2739 /*
2740 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2741 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2742 */
2743 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2744
2745 si = swap_info_get(entry);
2746 if (!si) {
2747 /*
2748 * An acceptable race has occurred since the failing
2749 * __swap_duplicate(): the swap entry has been freed,
2750 * perhaps even the whole swap_map cleared for swapoff.
2751 */
2752 goto outer;
2753 }
2754
2755 offset = swp_offset(entry);
2756 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2757
2758 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2759 /*
2760 * The higher the swap count, the more likely it is that tasks
2761 * will race to add swap count continuation: we need to avoid
2762 * over-provisioning.
2763 */
2764 goto out;
2765 }
2766
2767 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002768 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002769 return -ENOMEM;
2770 }
2771
2772 /*
2773 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002774 * no architecture is using highmem pages for kernel page tables: so it
2775 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002776 */
2777 head = vmalloc_to_page(si->swap_map + offset);
2778 offset &= ~PAGE_MASK;
2779
2780 /*
2781 * Page allocation does not initialize the page's lru field,
2782 * but it does always reset its private field.
2783 */
2784 if (!page_private(head)) {
2785 BUG_ON(count & COUNT_CONTINUED);
2786 INIT_LIST_HEAD(&head->lru);
2787 set_page_private(head, SWP_CONTINUED);
2788 si->flags |= SWP_CONTINUED;
2789 }
2790
2791 list_for_each_entry(list_page, &head->lru, lru) {
2792 unsigned char *map;
2793
2794 /*
2795 * If the previous map said no continuation, but we've found
2796 * a continuation page, free our allocation and use this one.
2797 */
2798 if (!(count & COUNT_CONTINUED))
2799 goto out;
2800
Cong Wang9b04c5f2011-11-25 23:14:39 +08002801 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002802 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002803 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002804
2805 /*
2806 * If this continuation count now has some space in it,
2807 * free our allocation and use this one.
2808 */
2809 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2810 goto out;
2811 }
2812
2813 list_add_tail(&page->lru, &head->lru);
2814 page = NULL; /* now it's attached, don't free it */
2815out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002816 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002817outer:
2818 if (page)
2819 __free_page(page);
2820 return 0;
2821}
2822
2823/*
2824 * swap_count_continued - when the original swap_map count is incremented
2825 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2826 * into, carry if so, or else fail until a new continuation page is allocated;
2827 * when the original swap_map count is decremented from 0 with continuation,
2828 * borrow from the continuation and report whether it still holds more.
2829 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2830 */
2831static bool swap_count_continued(struct swap_info_struct *si,
2832 pgoff_t offset, unsigned char count)
2833{
2834 struct page *head;
2835 struct page *page;
2836 unsigned char *map;
2837
2838 head = vmalloc_to_page(si->swap_map + offset);
2839 if (page_private(head) != SWP_CONTINUED) {
2840 BUG_ON(count & COUNT_CONTINUED);
2841 return false; /* need to add count continuation */
2842 }
2843
2844 offset &= ~PAGE_MASK;
2845 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002846 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002847
2848 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2849 goto init_map; /* jump over SWAP_CONT_MAX checks */
2850
2851 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2852 /*
2853 * Think of how you add 1 to 999
2854 */
2855 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002856 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002857 page = list_entry(page->lru.next, struct page, lru);
2858 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002859 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002860 }
2861 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002862 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002863 page = list_entry(page->lru.next, struct page, lru);
2864 if (page == head)
2865 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002866 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002867init_map: *map = 0; /* we didn't zero the page */
2868 }
2869 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002870 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002871 page = list_entry(page->lru.prev, struct page, lru);
2872 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002873 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002874 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002875 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002876 page = list_entry(page->lru.prev, struct page, lru);
2877 }
2878 return true; /* incremented */
2879
2880 } else { /* decrementing */
2881 /*
2882 * Think of how you subtract 1 from 1000
2883 */
2884 BUG_ON(count != COUNT_CONTINUED);
2885 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002886 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002887 page = list_entry(page->lru.next, struct page, lru);
2888 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002889 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002890 }
2891 BUG_ON(*map == 0);
2892 *map -= 1;
2893 if (*map == 0)
2894 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002895 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002896 page = list_entry(page->lru.prev, struct page, lru);
2897 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002898 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002899 *map = SWAP_CONT_MAX | count;
2900 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002901 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002902 page = list_entry(page->lru.prev, struct page, lru);
2903 }
2904 return count == COUNT_CONTINUED;
2905 }
2906}
2907
2908/*
2909 * free_swap_count_continuations - swapoff free all the continuation pages
2910 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2911 */
2912static void free_swap_count_continuations(struct swap_info_struct *si)
2913{
2914 pgoff_t offset;
2915
2916 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2917 struct page *head;
2918 head = vmalloc_to_page(si->swap_map + offset);
2919 if (page_private(head)) {
2920 struct list_head *this, *next;
2921 list_for_each_safe(this, next, &head->lru) {
2922 struct page *page;
2923 page = list_entry(this, struct page, lru);
2924 list_del(this);
2925 __free_page(page);
2926 }
2927 }
2928 }
2929}