blob: de7c904e52e507079f5bae7a2b7854a9d7cf80b0 [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>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080041#include <linux/page_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;
Shaohua Liec8acf22013-02-22 16:34:38 -080054static atomic_t highest_priority_index = ATOMIC_INIT(-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static const char Bad_file[] = "Bad swap file entry ";
57static const char Unused_file[] = "Unused swap file entry ";
58static const char Bad_offset[] = "Bad swap offset entry ";
59static const char Unused_offset[] = "Unused swap offset entry ";
60
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060061struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060063struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Ingo Molnarfc0abb12006-01-18 17:42:33 -080065static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Kay Sievers66d7dd52010-10-26 14:22:06 -070067static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
68/* Activity counter to indicate that a swapon or swapoff has occurred */
69static atomic_t proc_poll_event = ATOMIC_INIT(0);
70
Hugh Dickins8d69aae2009-12-14 17:58:45 -080071static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070072{
Hugh Dickins570a335b2009-12-14 17:58:46 -080073 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070074}
75
Hugh Dickinsefa90a92009-12-14 17:58:41 -080076/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070077static int
78__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
79{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080080 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070081 struct page *page;
82 int ret = 0;
83
Shaohua Li33806f02013-02-22 16:34:37 -080084 page = find_get_page(swap_address_space(entry), entry.val);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070085 if (!page)
86 return 0;
87 /*
88 * This function is called from scan_swap_map() and it's called
89 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
90 * We have to use trylock for avoiding deadlock. This is a special
91 * case and you should use try_to_free_swap() with explicit lock_page()
92 * in usual operations.
93 */
94 if (trylock_page(page)) {
95 ret = try_to_free_swap(page);
96 unlock_page(page);
97 }
98 page_cache_release(page);
99 return ret;
100}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800103 * swapon tell device that all the old swap contents can be discarded,
104 * to allow the swap device to optimize its wear-levelling.
105 */
106static int discard_swap(struct swap_info_struct *si)
107{
108 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800109 sector_t start_block;
110 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800111 int err = 0;
112
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800113 /* Do not discard the swap header page! */
114 se = &si->first_swap_extent;
115 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
116 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
117 if (nr_blocks) {
118 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200119 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800120 if (err)
121 return err;
122 cond_resched();
123 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800124
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800125 list_for_each_entry(se, &si->first_swap_extent.list, list) {
126 start_block = se->start_block << (PAGE_SHIFT - 9);
127 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800128
129 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200130 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800131 if (err)
132 break;
133
134 cond_resched();
135 }
136 return err; /* That will often be -EOPNOTSUPP */
137}
138
Hugh Dickins7992fde2009-01-06 14:39:53 -0800139/*
140 * swap allocation tell device that a cluster of swap can now be discarded,
141 * to allow the swap device to optimize its wear-levelling.
142 */
143static void discard_swap_cluster(struct swap_info_struct *si,
144 pgoff_t start_page, pgoff_t nr_pages)
145{
146 struct swap_extent *se = si->curr_swap_extent;
147 int found_extent = 0;
148
149 while (nr_pages) {
150 struct list_head *lh;
151
152 if (se->start_page <= start_page &&
153 start_page < se->start_page + se->nr_pages) {
154 pgoff_t offset = start_page - se->start_page;
155 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800156 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800157
158 if (nr_blocks > nr_pages)
159 nr_blocks = nr_pages;
160 start_page += nr_blocks;
161 nr_pages -= nr_blocks;
162
163 if (!found_extent++)
164 si->curr_swap_extent = se;
165
166 start_block <<= PAGE_SHIFT - 9;
167 nr_blocks <<= PAGE_SHIFT - 9;
168 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200169 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800170 break;
171 }
172
173 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800174 se = list_entry(lh, struct swap_extent, list);
175 }
176}
177
Hugh Dickins048c27f2005-09-03 15:54:40 -0700178#define SWAPFILE_CLUSTER 256
179#define LATENCY_LIMIT 256
180
Shaohua Li2a8f9442013-09-11 14:20:28 -0700181static inline void cluster_set_flag(struct swap_cluster_info *info,
182 unsigned int flag)
183{
184 info->flags = flag;
185}
186
187static inline unsigned int cluster_count(struct swap_cluster_info *info)
188{
189 return info->data;
190}
191
192static inline void cluster_set_count(struct swap_cluster_info *info,
193 unsigned int c)
194{
195 info->data = c;
196}
197
198static inline void cluster_set_count_flag(struct swap_cluster_info *info,
199 unsigned int c, unsigned int f)
200{
201 info->flags = f;
202 info->data = c;
203}
204
205static inline unsigned int cluster_next(struct swap_cluster_info *info)
206{
207 return info->data;
208}
209
210static inline void cluster_set_next(struct swap_cluster_info *info,
211 unsigned int n)
212{
213 info->data = n;
214}
215
216static inline void cluster_set_next_flag(struct swap_cluster_info *info,
217 unsigned int n, unsigned int f)
218{
219 info->flags = f;
220 info->data = n;
221}
222
223static inline bool cluster_is_free(struct swap_cluster_info *info)
224{
225 return info->flags & CLUSTER_FLAG_FREE;
226}
227
228static inline bool cluster_is_null(struct swap_cluster_info *info)
229{
230 return info->flags & CLUSTER_FLAG_NEXT_NULL;
231}
232
233static inline void cluster_set_null(struct swap_cluster_info *info)
234{
235 info->flags = CLUSTER_FLAG_NEXT_NULL;
236 info->data = 0;
237}
238
Shaohua Li815c2c52013-09-11 14:20:30 -0700239/* Add a cluster to discard list and schedule it to do discard */
240static void swap_cluster_schedule_discard(struct swap_info_struct *si,
241 unsigned int idx)
242{
243 /*
244 * If scan_swap_map() can't find a free cluster, it will check
245 * si->swap_map directly. To make sure the discarding cluster isn't
246 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
247 * will be cleared after discard
248 */
249 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
250 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
251
252 if (cluster_is_null(&si->discard_cluster_head)) {
253 cluster_set_next_flag(&si->discard_cluster_head,
254 idx, 0);
255 cluster_set_next_flag(&si->discard_cluster_tail,
256 idx, 0);
257 } else {
258 unsigned int tail = cluster_next(&si->discard_cluster_tail);
259 cluster_set_next(&si->cluster_info[tail], idx);
260 cluster_set_next_flag(&si->discard_cluster_tail,
261 idx, 0);
262 }
263
264 schedule_work(&si->discard_work);
265}
266
267/*
268 * Doing discard actually. After a cluster discard is finished, the cluster
269 * will be added to free cluster list. caller should hold si->lock.
270*/
271static void swap_do_scheduled_discard(struct swap_info_struct *si)
272{
273 struct swap_cluster_info *info;
274 unsigned int idx;
275
276 info = si->cluster_info;
277
278 while (!cluster_is_null(&si->discard_cluster_head)) {
279 idx = cluster_next(&si->discard_cluster_head);
280
281 cluster_set_next_flag(&si->discard_cluster_head,
282 cluster_next(&info[idx]), 0);
283 if (cluster_next(&si->discard_cluster_tail) == idx) {
284 cluster_set_null(&si->discard_cluster_head);
285 cluster_set_null(&si->discard_cluster_tail);
286 }
287 spin_unlock(&si->lock);
288
289 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
290 SWAPFILE_CLUSTER);
291
292 spin_lock(&si->lock);
293 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
294 if (cluster_is_null(&si->free_cluster_head)) {
295 cluster_set_next_flag(&si->free_cluster_head,
296 idx, 0);
297 cluster_set_next_flag(&si->free_cluster_tail,
298 idx, 0);
299 } else {
300 unsigned int tail;
301
302 tail = cluster_next(&si->free_cluster_tail);
303 cluster_set_next(&info[tail], idx);
304 cluster_set_next_flag(&si->free_cluster_tail,
305 idx, 0);
306 }
307 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
308 0, SWAPFILE_CLUSTER);
309 }
310}
311
312static void swap_discard_work(struct work_struct *work)
313{
314 struct swap_info_struct *si;
315
316 si = container_of(work, struct swap_info_struct, discard_work);
317
318 spin_lock(&si->lock);
319 swap_do_scheduled_discard(si);
320 spin_unlock(&si->lock);
321}
322
Shaohua Li2a8f9442013-09-11 14:20:28 -0700323/*
324 * The cluster corresponding to page_nr will be used. The cluster will be
325 * removed from free cluster list and its usage counter will be increased.
326 */
327static void inc_cluster_info_page(struct swap_info_struct *p,
328 struct swap_cluster_info *cluster_info, unsigned long page_nr)
329{
330 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
331
332 if (!cluster_info)
333 return;
334 if (cluster_is_free(&cluster_info[idx])) {
335 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
336 cluster_set_next_flag(&p->free_cluster_head,
337 cluster_next(&cluster_info[idx]), 0);
338 if (cluster_next(&p->free_cluster_tail) == idx) {
339 cluster_set_null(&p->free_cluster_tail);
340 cluster_set_null(&p->free_cluster_head);
341 }
342 cluster_set_count_flag(&cluster_info[idx], 0, 0);
343 }
344
345 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
346 cluster_set_count(&cluster_info[idx],
347 cluster_count(&cluster_info[idx]) + 1);
348}
349
350/*
351 * The cluster corresponding to page_nr decreases one usage. If the usage
352 * counter becomes 0, which means no page in the cluster is in using, we can
353 * optionally discard the cluster and add it to free cluster list.
354 */
355static void dec_cluster_info_page(struct swap_info_struct *p,
356 struct swap_cluster_info *cluster_info, unsigned long page_nr)
357{
358 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
359
360 if (!cluster_info)
361 return;
362
363 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
364 cluster_set_count(&cluster_info[idx],
365 cluster_count(&cluster_info[idx]) - 1);
366
367 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700368 /*
369 * If the swap is discardable, prepare discard the cluster
370 * instead of free it immediately. The cluster will be freed
371 * after discard.
372 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700373 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
374 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700375 swap_cluster_schedule_discard(p, idx);
376 return;
377 }
378
Shaohua Li2a8f9442013-09-11 14:20:28 -0700379 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
380 if (cluster_is_null(&p->free_cluster_head)) {
381 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
382 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
383 } else {
384 unsigned int tail = cluster_next(&p->free_cluster_tail);
385 cluster_set_next(&cluster_info[tail], idx);
386 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
387 }
388 }
389}
390
391/*
392 * It's possible scan_swap_map() uses a free cluster in the middle of free
393 * cluster list. Avoiding such abuse to avoid list corruption.
394 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700395static bool
396scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700397 unsigned long offset)
398{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700399 struct percpu_cluster *percpu_cluster;
400 bool conflict;
401
Shaohua Li2a8f9442013-09-11 14:20:28 -0700402 offset /= SWAPFILE_CLUSTER;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700403 conflict = !cluster_is_null(&si->free_cluster_head) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700404 offset != cluster_next(&si->free_cluster_head) &&
405 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700406
407 if (!conflict)
408 return false;
409
410 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
411 cluster_set_null(&percpu_cluster->index);
412 return true;
413}
414
415/*
416 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
417 * might involve allocating a new cluster for current CPU too.
418 */
419static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
420 unsigned long *offset, unsigned long *scan_base)
421{
422 struct percpu_cluster *cluster;
423 bool found_free;
424 unsigned long tmp;
425
426new_cluster:
427 cluster = this_cpu_ptr(si->percpu_cluster);
428 if (cluster_is_null(&cluster->index)) {
429 if (!cluster_is_null(&si->free_cluster_head)) {
430 cluster->index = si->free_cluster_head;
431 cluster->next = cluster_next(&cluster->index) *
432 SWAPFILE_CLUSTER;
433 } else if (!cluster_is_null(&si->discard_cluster_head)) {
434 /*
435 * we don't have free cluster but have some clusters in
436 * discarding, do discard now and reclaim them
437 */
438 swap_do_scheduled_discard(si);
439 *scan_base = *offset = si->cluster_next;
440 goto new_cluster;
441 } else
442 return;
443 }
444
445 found_free = false;
446
447 /*
448 * Other CPUs can use our cluster if they can't find a free cluster,
449 * check if there is still free entry in the cluster
450 */
451 tmp = cluster->next;
452 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
453 SWAPFILE_CLUSTER) {
454 if (!si->swap_map[tmp]) {
455 found_free = true;
456 break;
457 }
458 tmp++;
459 }
460 if (!found_free) {
461 cluster_set_null(&cluster->index);
462 goto new_cluster;
463 }
464 cluster->next = tmp + 1;
465 *offset = tmp;
466 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700467}
468
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700469static unsigned long scan_swap_map(struct swap_info_struct *si,
470 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800472 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800473 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800474 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700475 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800477 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700478 * We try to cluster swap pages by allocating them sequentially
479 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
480 * way, however, we resort to first-free allocation, starting
481 * a new cluster. This prevents us from scattering swap pages
482 * all over the entire swap partition, so that we reduce
483 * overall disk seek times between swap pages. -- sct
484 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800485 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700486 */
487
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700488 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800489 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800490
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700491 /* SSD algorithm */
492 if (si->cluster_info) {
493 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
494 goto checks;
495 }
496
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800497 if (unlikely(!si->cluster_nr--)) {
498 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
499 si->cluster_nr = SWAPFILE_CLUSTER - 1;
500 goto checks;
501 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700502
Shaohua Liec8acf22013-02-22 16:34:38 -0800503 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700504
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800505 /*
506 * If seek is expensive, start searching for new cluster from
507 * start of partition, to minimize the span of allocated swap.
508 * But if seek is cheap, search from our current position, so
509 * that swap is allocated from all over the partition: if the
510 * Flash Translation Layer only remaps within limited zones,
511 * we don't want to wear out the first zone too quickly.
512 */
513 if (!(si->flags & SWP_SOLIDSTATE))
514 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700515 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
516
517 /* Locate the first empty (unaligned) cluster */
518 for (; last_in_cluster <= si->highest_bit; offset++) {
519 if (si->swap_map[offset])
520 last_in_cluster = offset + SWAPFILE_CLUSTER;
521 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800522 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800523 offset -= SWAPFILE_CLUSTER - 1;
524 si->cluster_next = offset;
525 si->cluster_nr = SWAPFILE_CLUSTER - 1;
526 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700528 if (unlikely(--latency_ration < 0)) {
529 cond_resched();
530 latency_ration = LATENCY_LIMIT;
531 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700532 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800533
534 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800535 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
536
537 /* Locate the first empty (unaligned) cluster */
538 for (; last_in_cluster < scan_base; offset++) {
539 if (si->swap_map[offset])
540 last_in_cluster = offset + SWAPFILE_CLUSTER;
541 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800542 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800543 offset -= SWAPFILE_CLUSTER - 1;
544 si->cluster_next = offset;
545 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800546 goto checks;
547 }
548 if (unlikely(--latency_ration < 0)) {
549 cond_resched();
550 latency_ration = LATENCY_LIMIT;
551 }
552 }
553
554 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800555 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800556 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700558
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800559checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700560 if (si->cluster_info) {
561 while (scan_swap_map_ssd_cluster_conflict(si, offset))
562 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
563 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800564 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700565 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700566 if (!si->highest_bit)
567 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800568 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800569 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700570
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700571 /* reuse swap entry of cache-only swap if not busy. */
572 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700573 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800574 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700575 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800576 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700577 /* entry was freed successfully, try to use this again */
578 if (swap_was_freed)
579 goto checks;
580 goto scan; /* check next one */
581 }
582
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800583 if (si->swap_map[offset])
584 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700585
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800586 if (offset == si->lowest_bit)
587 si->lowest_bit++;
588 if (offset == si->highest_bit)
589 si->highest_bit--;
590 si->inuse_pages++;
591 if (si->inuse_pages == si->pages) {
592 si->lowest_bit = si->max;
593 si->highest_bit = 0;
594 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800595 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700596 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800597 si->cluster_next = offset + 1;
598 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800599
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800600 return offset;
601
602scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800603 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700604 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700605 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800606 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700607 goto checks;
608 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700609 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800610 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700611 goto checks;
612 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700613 if (unlikely(--latency_ration < 0)) {
614 cond_resched();
615 latency_ration = LATENCY_LIMIT;
616 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700617 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800618 offset = si->lowest_bit;
619 while (++offset < scan_base) {
620 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800621 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800622 goto checks;
623 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700624 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800625 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700626 goto checks;
627 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800628 if (unlikely(--latency_ration < 0)) {
629 cond_resched();
630 latency_ration = LATENCY_LIMIT;
631 }
632 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800633 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700634
635no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700636 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 return 0;
638}
639
640swp_entry_t get_swap_page(void)
641{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700642 struct swap_info_struct *si;
643 pgoff_t offset;
644 int type, next;
645 int wrapped = 0;
Shaohua Liec8acf22013-02-22 16:34:38 -0800646 int hp_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Hugh Dickins5d337b92005-09-03 15:54:41 -0700648 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800649 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700650 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800651 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700653 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800654 hp_index = atomic_xchg(&highest_priority_index, -1);
655 /*
656 * highest_priority_index records current highest priority swap
657 * type which just frees swap entries. If its priority is
658 * higher than that of swap_list.next swap type, we use it. It
659 * isn't protected by swap_lock, so it can be an invalid value
660 * if the corresponding swap type is swapoff. We double check
661 * the flags here. It's even possible the swap type is swapoff
662 * and swapon again and its priority is changed. In such rare
663 * case, low prority swap type might be used, but eventually
664 * high priority swap will be used after several rounds of
665 * swap.
666 */
667 if (hp_index != -1 && hp_index != type &&
668 swap_info[type]->prio < swap_info[hp_index]->prio &&
669 (swap_info[hp_index]->flags & SWP_WRITEOK)) {
670 type = hp_index;
671 swap_list.next = type;
672 }
673
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800674 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700675 next = si->next;
676 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800677 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700678 next = swap_list.head;
679 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700681
Shaohua Liec8acf22013-02-22 16:34:38 -0800682 spin_lock(&si->lock);
683 if (!si->highest_bit) {
684 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700685 continue;
Shaohua Liec8acf22013-02-22 16:34:38 -0800686 }
687 if (!(si->flags & SWP_WRITEOK)) {
688 spin_unlock(&si->lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700689 continue;
Shaohua Liec8acf22013-02-22 16:34:38 -0800690 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700691
692 swap_list.next = next;
Shaohua Liec8acf22013-02-22 16:34:38 -0800693
694 spin_unlock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700695 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800696 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800697 spin_unlock(&si->lock);
698 if (offset)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700699 return swp_entry(type, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800700 spin_lock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700701 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700703
Shaohua Liec8acf22013-02-22 16:34:38 -0800704 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700705noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700706 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700707 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
Hugh Dickins910321e2010-09-09 16:38:07 -0700710/* The only caller of this function is now susupend routine */
711swp_entry_t get_swap_page_of_type(int type)
712{
713 struct swap_info_struct *si;
714 pgoff_t offset;
715
Hugh Dickins910321e2010-09-09 16:38:07 -0700716 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800717 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700718 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800719 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700720 /* This is called for allocating swap entry, not cache */
721 offset = scan_swap_map(si, 1);
722 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800723 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700724 return swp_entry(type, offset);
725 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800726 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700727 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800728 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700729 return (swp_entry_t) {0};
730}
731
Hugh Dickins73c34b62009-12-14 17:58:43 -0800732static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800734 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 unsigned long offset, type;
736
737 if (!entry.val)
738 goto out;
739 type = swp_type(entry);
740 if (type >= nr_swapfiles)
741 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800742 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 if (!(p->flags & SWP_USED))
744 goto bad_device;
745 offset = swp_offset(entry);
746 if (offset >= p->max)
747 goto bad_offset;
748 if (!p->swap_map[offset])
749 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800750 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 return p;
752
753bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700754 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 goto out;
756bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700757 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 goto out;
759bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700760 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 goto out;
762bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700763 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764out:
765 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800766}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Shaohua Liec8acf22013-02-22 16:34:38 -0800768/*
769 * This swap type frees swap entry, check if it is the highest priority swap
770 * type which just frees swap entry. get_swap_page() uses
771 * highest_priority_index to search highest priority swap type. The
772 * swap_info_struct.lock can't protect us if there are multiple swap types
773 * active, so we use atomic_cmpxchg.
774 */
775static void set_highest_priority_index(int type)
776{
777 int old_hp_index, new_hp_index;
778
779 do {
780 old_hp_index = atomic_read(&highest_priority_index);
781 if (old_hp_index != -1 &&
782 swap_info[old_hp_index]->prio >= swap_info[type]->prio)
783 break;
784 new_hp_index = type;
785 } while (atomic_cmpxchg(&highest_priority_index,
786 old_hp_index, new_hp_index) != old_hp_index);
787}
788
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800789static unsigned char swap_entry_free(struct swap_info_struct *p,
790 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
Hugh Dickins253d5532009-12-14 17:58:44 -0800792 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800793 unsigned char count;
794 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700796 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800797 has_cache = count & SWAP_HAS_CACHE;
798 count &= ~SWAP_HAS_CACHE;
799
800 if (usage == SWAP_HAS_CACHE) {
801 VM_BUG_ON(!has_cache);
802 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800803 } else if (count == SWAP_MAP_SHMEM) {
804 /*
805 * Or we could insist on shmem.c using a special
806 * swap_shmem_free() and free_shmem_swap_and_cache()...
807 */
808 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800809 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
810 if (count == COUNT_CONTINUED) {
811 if (swap_count_continued(p, offset, count))
812 count = SWAP_MAP_MAX | COUNT_CONTINUED;
813 else
814 count = SWAP_MAP_MAX;
815 } else
816 count--;
817 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800818
819 if (!count)
820 mem_cgroup_uncharge_swap(entry);
821
822 usage = count | has_cache;
823 p->swap_map[offset] = usage;
824
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700825 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800826 if (!usage) {
Shaohua Li2a8f9442013-09-11 14:20:28 -0700827 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700828 if (offset < p->lowest_bit)
829 p->lowest_bit = offset;
830 if (offset > p->highest_bit)
831 p->highest_bit = offset;
Shaohua Liec8acf22013-02-22 16:34:38 -0800832 set_highest_priority_index(p->type);
833 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700834 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600835 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700836 if (p->flags & SWP_BLKDEV) {
837 struct gendisk *disk = p->bdev->bd_disk;
838 if (disk->fops->swap_slot_free_notify)
839 disk->fops->swap_slot_free_notify(p->bdev,
840 offset);
841 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800843
844 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845}
846
847/*
848 * Caller has made sure that the swapdevice corresponding to entry
849 * is still around or has not been recycled.
850 */
851void swap_free(swp_entry_t entry)
852{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800853 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 p = swap_info_get(entry);
856 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800857 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800858 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 }
860}
861
862/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700863 * Called after dropping swapcache to decrease refcnt to swap entries.
864 */
865void swapcache_free(swp_entry_t entry, struct page *page)
866{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700867 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800868 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700869
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700870 p = swap_info_get(entry);
871 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800872 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
873 if (page)
874 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
Shaohua Liec8acf22013-02-22 16:34:38 -0800875 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700876 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700877}
878
879/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700880 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800881 * This does not give an exact answer when swap count is continued,
882 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700884int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700886 int count = 0;
887 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 swp_entry_t entry;
889
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700890 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 p = swap_info_get(entry);
892 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700893 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800894 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700896 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
899/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800900 * We can write to an anon page without COW if there are no other references
901 * to it. And as a side-effect, free up its swap: because the old content
902 * on disk will never be read, and seeking back there to write new content
903 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800905int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700907 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Hugh Dickins51726b12009-01-06 14:39:25 -0800909 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800910 if (unlikely(PageKsm(page)))
911 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700912 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800913 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700914 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800915 if (count == 1 && !PageWriteback(page)) {
916 delete_from_swap_cache(page);
917 SetPageDirty(page);
918 }
919 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800920 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921}
922
923/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800924 * If swap is getting full, or if there are no more mappings of this page,
925 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800927int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
Hugh Dickins51726b12009-01-06 14:39:25 -0800929 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
931 if (!PageSwapCache(page))
932 return 0;
933 if (PageWriteback(page))
934 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800935 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return 0;
937
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700938 /*
939 * Once hibernation has begun to create its image of memory,
940 * there's a danger that one of the calls to try_to_free_swap()
941 * - most probably a call from __try_to_reclaim_swap() while
942 * hibernation is allocating its own swap pages for the image,
943 * but conceivably even a call from memory reclaim - will free
944 * the swap from a page which has already been recorded in the
945 * image as a clean swapcache page, and then reuse its swap for
946 * another page of the image. On waking from hibernation, the
947 * original page might be freed under memory pressure, then
948 * later read back in from swap, now with the wrong data.
949 *
Mel Gormanf90ac392012-01-10 15:07:15 -0800950 * Hibration suspends storage while it is writing the image
951 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700952 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800953 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700954 return 0;
955
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800956 delete_from_swap_cache(page);
957 SetPageDirty(page);
958 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700959}
960
961/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 * Free the swap entry like above, but also try to
963 * free the page cache entry if it is the last user.
964 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800965int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800967 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 struct page *page = NULL;
969
Andi Kleena7420aa2009-09-16 11:50:05 +0200970 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800971 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700972
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 p = swap_info_get(entry);
974 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800975 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -0800976 page = find_get_page(swap_address_space(entry),
977 entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700978 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800979 page_cache_release(page);
980 page = NULL;
981 }
982 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800983 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 }
985 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800986 /*
987 * Not mapped elsewhere, or swap space full? Free it!
988 * Also recheck PageSwapCache now page is locked (above).
989 */
Nick Piggin93fac702006-03-31 02:29:56 -0800990 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800991 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 delete_from_swap_cache(page);
993 SetPageDirty(page);
994 }
995 unlock_page(page);
996 page_cache_release(page);
997 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800998 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999}
1000
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001001#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001002/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001003 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001004 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001005 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1006 * from 0, in which the swap header is expected to be located.
1007 *
1008 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001009 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001010int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001011{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001012 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001013 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001014
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001015 if (device)
1016 bdev = bdget(device);
1017
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001018 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001019 for (type = 0; type < nr_swapfiles; type++) {
1020 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001021
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001022 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001023 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001024
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001025 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001026 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001027 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001028
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001029 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001030 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001031 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001032 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001033 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001034
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001035 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001036 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001037 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001038
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001039 spin_unlock(&swap_lock);
1040 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001041 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001042 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001043 }
1044 }
1045 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001046 if (bdev)
1047 bdput(bdev);
1048
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001049 return -ENODEV;
1050}
1051
1052/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001053 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1054 * corresponding to given index in swap_info (swap type).
1055 */
1056sector_t swapdev_block(int type, pgoff_t offset)
1057{
1058 struct block_device *bdev;
1059
1060 if ((unsigned int)type >= nr_swapfiles)
1061 return 0;
1062 if (!(swap_info[type]->flags & SWP_WRITEOK))
1063 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001064 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001065}
1066
1067/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001068 * Return either the total number of swap pages of given type, or the number
1069 * of free pages of that type (depending on @free)
1070 *
1071 * This is needed for software suspend
1072 */
1073unsigned int count_swap_pages(int type, int free)
1074{
1075 unsigned int n = 0;
1076
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001077 spin_lock(&swap_lock);
1078 if ((unsigned int)type < nr_swapfiles) {
1079 struct swap_info_struct *sis = swap_info[type];
1080
Shaohua Liec8acf22013-02-22 16:34:38 -08001081 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001082 if (sis->flags & SWP_WRITEOK) {
1083 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001084 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001085 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001086 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001087 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001088 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001089 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001090 return n;
1091}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001092#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001093
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001094static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1095{
1096#ifdef CONFIG_MEM_SOFT_DIRTY
1097 /*
1098 * When pte keeps soft dirty bit the pte generated
1099 * from swap entry does not has it, still it's same
1100 * pte from logical point of view.
1101 */
1102 pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1103 return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1104#else
1105 return pte_same(pte, swp_pte);
1106#endif
1107}
1108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001110 * No need to decide whether this PTE shares the swap entry with others,
1111 * just let do_wp_page work it out if a write is requested later - to
1112 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001114static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 unsigned long addr, swp_entry_t entry, struct page *page)
1116{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001117 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001118 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001119 spinlock_t *ptl;
1120 pte_t *pte;
1121 int ret = 1;
1122
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001123 swapcache = page;
1124 page = ksm_might_need_to_copy(page, vma, addr);
1125 if (unlikely(!page))
1126 return -ENOMEM;
1127
Johannes Weiner72835c82012-01-12 17:18:32 -08001128 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
1129 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001130 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001131 goto out_nolock;
1132 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001133
1134 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001135 if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
Johannes Weiner5d84c772012-07-31 16:45:28 -07001136 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001137 ret = 0;
1138 goto out;
1139 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001140
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001141 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001142 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 get_page(page);
1144 set_pte_at(vma->vm_mm, addr, pte,
1145 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001146 if (page == swapcache)
1147 page_add_anon_rmap(page, vma, addr);
1148 else /* ksm created a completely new copy */
1149 page_add_new_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -08001150 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 swap_free(entry);
1152 /*
1153 * Move the page to the active list so it is not
1154 * immediately swapped out again after swapon.
1155 */
1156 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001157out:
1158 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001159out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001160 if (page != swapcache) {
1161 unlock_page(page);
1162 put_page(page);
1163 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001164 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165}
1166
1167static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1168 unsigned long addr, unsigned long end,
1169 swp_entry_t entry, struct page *page)
1170{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001172 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001173 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
Hugh Dickins044d66c2008-02-07 00:14:04 -08001175 /*
1176 * We don't actually need pte lock while scanning for swp_pte: since
1177 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1178 * page table while we're scanning; though it could get zapped, and on
1179 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1180 * of unmatched parts which look like swp_pte, so unuse_pte must
1181 * recheck under pte lock. Scanning without pte lock lets it be
1182 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1183 */
1184 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 do {
1186 /*
1187 * swapoff spends a _lot_ of time in this loop!
1188 * Test inline before going to call unuse_pte.
1189 */
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001190 if (unlikely(maybe_same_pte(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001191 pte_unmap(pte);
1192 ret = unuse_pte(vma, pmd, addr, entry, page);
1193 if (ret)
1194 goto out;
1195 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 }
1197 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001198 pte_unmap(pte - 1);
1199out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001200 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
1202
1203static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1204 unsigned long addr, unsigned long end,
1205 swp_entry_t entry, struct page *page)
1206{
1207 pmd_t *pmd;
1208 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001209 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211 pmd = pmd_offset(pud, addr);
1212 do {
1213 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001214 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001216 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1217 if (ret)
1218 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 } while (pmd++, addr = next, addr != end);
1220 return 0;
1221}
1222
1223static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1224 unsigned long addr, unsigned long end,
1225 swp_entry_t entry, struct page *page)
1226{
1227 pud_t *pud;
1228 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001229 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
1231 pud = pud_offset(pgd, addr);
1232 do {
1233 next = pud_addr_end(addr, end);
1234 if (pud_none_or_clear_bad(pud))
1235 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001236 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1237 if (ret)
1238 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 } while (pud++, addr = next, addr != end);
1240 return 0;
1241}
1242
1243static int unuse_vma(struct vm_area_struct *vma,
1244 swp_entry_t entry, struct page *page)
1245{
1246 pgd_t *pgd;
1247 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001248 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001250 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 addr = page_address_in_vma(page, vma);
1252 if (addr == -EFAULT)
1253 return 0;
1254 else
1255 end = addr + PAGE_SIZE;
1256 } else {
1257 addr = vma->vm_start;
1258 end = vma->vm_end;
1259 }
1260
1261 pgd = pgd_offset(vma->vm_mm, addr);
1262 do {
1263 next = pgd_addr_end(addr, end);
1264 if (pgd_none_or_clear_bad(pgd))
1265 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001266 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1267 if (ret)
1268 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 } while (pgd++, addr = next, addr != end);
1270 return 0;
1271}
1272
1273static int unuse_mm(struct mm_struct *mm,
1274 swp_entry_t entry, struct page *page)
1275{
1276 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001277 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
1279 if (!down_read_trylock(&mm->mmap_sem)) {
1280 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001281 * Activate page so shrink_inactive_list is unlikely to unmap
1282 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001284 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 unlock_page(page);
1286 down_read(&mm->mmap_sem);
1287 lock_page(page);
1288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001290 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 break;
1292 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001294 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295}
1296
1297/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001298 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1299 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 * Recycle to start on reaching the end, returning 0 when empty.
1301 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001302static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001303 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001305 unsigned int max = si->max;
1306 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001307 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
1309 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001310 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 * for whether an entry is in use, not modifying it; false
1312 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001313 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 */
1315 for (;;) {
1316 if (++i >= max) {
1317 if (!prev) {
1318 i = 0;
1319 break;
1320 }
1321 /*
1322 * No entries in use at top of swap_map,
1323 * loop back to start and recheck there.
1324 */
1325 max = prev + 1;
1326 prev = 0;
1327 i = 1;
1328 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001329 if (frontswap) {
1330 if (frontswap_test(si, i))
1331 break;
1332 else
1333 continue;
1334 }
Shaohua Liedfe23d2013-09-11 14:20:31 -07001335 count = ACCESS_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001336 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 break;
1338 }
1339 return i;
1340}
1341
1342/*
1343 * We completely avoid races by reading each swap page in advance,
1344 * and then search for the process using it. All the necessary
1345 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001346 *
1347 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1348 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001350int try_to_unuse(unsigned int type, bool frontswap,
1351 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001353 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001355 volatile unsigned char *swap_map; /* swap_map is accessed without
1356 * locking. Mark it as volatile
1357 * to prevent compiler doing
1358 * something odd.
1359 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001360 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 struct page *page;
1362 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001363 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
1366 /*
1367 * When searching mms for an entry, a good strategy is to
1368 * start at the first mm we freed the previous entry from
1369 * (though actually we don't notice whether we or coincidence
1370 * freed the entry). Initialize this start_mm with a hold.
1371 *
1372 * A simpler strategy would be to start at the last mm we
1373 * freed the previous entry from; but that would take less
1374 * advantage of mmlist ordering, which clusters forked mms
1375 * together, child after parent. If we race with dup_mmap(), we
1376 * prefer to resolve parent before child, lest we miss entries
1377 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001378 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 */
1380 start_mm = &init_mm;
1381 atomic_inc(&init_mm.mm_users);
1382
1383 /*
1384 * Keep on scanning until all entries have gone. Usually,
1385 * one pass through swap_map is enough, but not necessarily:
1386 * there are races when an instance of an entry might be missed.
1387 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001388 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (signal_pending(current)) {
1390 retval = -EINTR;
1391 break;
1392 }
1393
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001394 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 * Get a page for the entry, using the existing swap
1396 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001397 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 */
1399 swap_map = &si->swap_map[i];
1400 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001401 page = read_swap_cache_async(entry,
1402 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 if (!page) {
1404 /*
1405 * Either swap_duplicate() failed because entry
1406 * has been freed independently, and will not be
1407 * reused since sys_swapoff() already disabled
1408 * allocation from here, or alloc_page() failed.
1409 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001410 swcount = *swap_map;
1411 /*
1412 * We don't hold lock here, so the swap entry could be
1413 * SWAP_MAP_BAD (when the cluster is discarding).
1414 * Instead of fail out, We can just skip the swap
1415 * entry because swapoff will wait for discarding
1416 * finish anyway.
1417 */
1418 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 continue;
1420 retval = -ENOMEM;
1421 break;
1422 }
1423
1424 /*
1425 * Don't hold on to start_mm if it looks like exiting.
1426 */
1427 if (atomic_read(&start_mm->mm_users) == 1) {
1428 mmput(start_mm);
1429 start_mm = &init_mm;
1430 atomic_inc(&init_mm.mm_users);
1431 }
1432
1433 /*
1434 * Wait for and lock page. When do_swap_page races with
1435 * try_to_unuse, do_swap_page can handle the fault much
1436 * faster than try_to_unuse can locate the entry. This
1437 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1438 * defer to do_swap_page in such a case - in some tests,
1439 * do_swap_page and try_to_unuse repeatedly compete.
1440 */
1441 wait_on_page_locked(page);
1442 wait_on_page_writeback(page);
1443 lock_page(page);
1444 wait_on_page_writeback(page);
1445
1446 /*
1447 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001450 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1451 retval = shmem_unuse(entry, page);
1452 /* page has already been unlocked and released */
1453 if (retval < 0)
1454 break;
1455 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001457 if (swap_count(swcount) && start_mm != &init_mm)
1458 retval = unuse_mm(start_mm, entry, page);
1459
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001460 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 int set_start_mm = (*swap_map >= swcount);
1462 struct list_head *p = &start_mm->mmlist;
1463 struct mm_struct *new_start_mm = start_mm;
1464 struct mm_struct *prev_mm = start_mm;
1465 struct mm_struct *mm;
1466
1467 atomic_inc(&new_start_mm->mm_users);
1468 atomic_inc(&prev_mm->mm_users);
1469 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001470 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 (p = p->next) != &start_mm->mmlist) {
1472 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001473 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 spin_unlock(&mmlist_lock);
1476 mmput(prev_mm);
1477 prev_mm = mm;
1478
1479 cond_resched();
1480
1481 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001482 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001484 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001486 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001488
Bo Liu32c5fc12009-11-02 16:50:33 +00001489 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 mmput(new_start_mm);
1491 atomic_inc(&mm->mm_users);
1492 new_start_mm = mm;
1493 set_start_mm = 0;
1494 }
1495 spin_lock(&mmlist_lock);
1496 }
1497 spin_unlock(&mmlist_lock);
1498 mmput(prev_mm);
1499 mmput(start_mm);
1500 start_mm = new_start_mm;
1501 }
1502 if (retval) {
1503 unlock_page(page);
1504 page_cache_release(page);
1505 break;
1506 }
1507
1508 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 * If a reference remains (rare), we would like to leave
1510 * the page in the swap cache; but try_to_unmap could
1511 * then re-duplicate the entry once we drop page lock,
1512 * so we might loop indefinitely; also, that page could
1513 * not be swapped out to other storage meanwhile. So:
1514 * delete from cache even if there's another reference,
1515 * after ensuring that the data has been saved to disk -
1516 * since if the reference remains (rarer), it will be
1517 * read from disk into another page. Splitting into two
1518 * pages would be incorrect if swap supported "shared
1519 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001520 *
1521 * Given how unuse_vma() targets one particular offset
1522 * in an anon_vma, once the anon_vma has been determined,
1523 * this splitting happens to be just what is needed to
1524 * handle where KSM pages have been swapped out: re-reading
1525 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001527 if (swap_count(*swap_map) &&
1528 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 struct writeback_control wbc = {
1530 .sync_mode = WB_SYNC_NONE,
1531 };
1532
1533 swap_writepage(page, &wbc);
1534 lock_page(page);
1535 wait_on_page_writeback(page);
1536 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001537
1538 /*
1539 * It is conceivable that a racing task removed this page from
1540 * swap cache just before we acquired the page lock at the top,
1541 * or while we dropped it in unuse_mm(). The page might even
1542 * be back in swap cache on another swap area: that we must not
1543 * delete, since it may not have been written out to swap yet.
1544 */
1545 if (PageSwapCache(page) &&
1546 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001547 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 /*
1550 * So we could skip searching mms once swap count went
1551 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001552 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
1554 SetPageDirty(page);
1555 unlock_page(page);
1556 page_cache_release(page);
1557
1558 /*
1559 * Make sure that we aren't completely killing
1560 * interactive performance.
1561 */
1562 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001563 if (frontswap && pages_to_unuse > 0) {
1564 if (!--pages_to_unuse)
1565 break;
1566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 }
1568
1569 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return retval;
1571}
1572
1573/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001574 * After a successful try_to_unuse, if no swap is now in use, we know
1575 * we can empty the mmlist. swap_lock must be held on entry and exit.
1576 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 * added to the mmlist just after page_duplicate - before would be racy.
1578 */
1579static void drain_mmlist(void)
1580{
1581 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001582 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001584 for (type = 0; type < nr_swapfiles; type++)
1585 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 return;
1587 spin_lock(&mmlist_lock);
1588 list_for_each_safe(p, next, &init_mm.mmlist)
1589 list_del_init(p);
1590 spin_unlock(&mmlist_lock);
1591}
1592
1593/*
1594 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001595 * corresponds to page offset for the specified swap entry.
1596 * Note that the type of this function is sector_t, but it returns page offset
1597 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001599static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001601 struct swap_info_struct *sis;
1602 struct swap_extent *start_se;
1603 struct swap_extent *se;
1604 pgoff_t offset;
1605
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001606 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001607 *bdev = sis->bdev;
1608
1609 offset = swp_offset(entry);
1610 start_se = sis->curr_swap_extent;
1611 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
1613 for ( ; ; ) {
1614 struct list_head *lh;
1615
1616 if (se->start_page <= offset &&
1617 offset < (se->start_page + se->nr_pages)) {
1618 return se->start_block + (offset - se->start_page);
1619 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001620 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 se = list_entry(lh, struct swap_extent, list);
1622 sis->curr_swap_extent = se;
1623 BUG_ON(se == start_se); /* It *must* be present */
1624 }
1625}
1626
1627/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001628 * Returns the page offset into bdev for the specified page's swap entry.
1629 */
1630sector_t map_swap_page(struct page *page, struct block_device **bdev)
1631{
1632 swp_entry_t entry;
1633 entry.val = page_private(page);
1634 return map_swap_entry(entry, bdev);
1635}
1636
1637/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 * Free all of a swapdev's extent information
1639 */
1640static void destroy_swap_extents(struct swap_info_struct *sis)
1641{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001642 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 struct swap_extent *se;
1644
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001645 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 struct swap_extent, list);
1647 list_del(&se->list);
1648 kfree(se);
1649 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001650
1651 if (sis->flags & SWP_FILE) {
1652 struct file *swap_file = sis->swap_file;
1653 struct address_space *mapping = swap_file->f_mapping;
1654
1655 sis->flags &= ~SWP_FILE;
1656 mapping->a_ops->swap_deactivate(swap_file);
1657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658}
1659
1660/*
1661 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001662 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001664 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 */
Mel Gormana509bc12012-07-31 16:44:57 -07001666int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1668 unsigned long nr_pages, sector_t start_block)
1669{
1670 struct swap_extent *se;
1671 struct swap_extent *new_se;
1672 struct list_head *lh;
1673
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001674 if (start_page == 0) {
1675 se = &sis->first_swap_extent;
1676 sis->curr_swap_extent = se;
1677 se->start_page = 0;
1678 se->nr_pages = nr_pages;
1679 se->start_block = start_block;
1680 return 1;
1681 } else {
1682 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001684 BUG_ON(se->start_page + se->nr_pages != start_page);
1685 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 /* Merge it */
1687 se->nr_pages += nr_pages;
1688 return 0;
1689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 }
1691
1692 /*
1693 * No merge. Insert a new extent, preserving ordering.
1694 */
1695 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1696 if (new_se == NULL)
1697 return -ENOMEM;
1698 new_se->start_page = start_page;
1699 new_se->nr_pages = nr_pages;
1700 new_se->start_block = start_block;
1701
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001702 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001703 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704}
1705
1706/*
1707 * A `swap extent' is a simple thing which maps a contiguous range of pages
1708 * onto a contiguous range of disk blocks. An ordered list of swap extents
1709 * is built at swapon time and is then used at swap_writepage/swap_readpage
1710 * time for locating where on disk a page belongs.
1711 *
1712 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1713 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1714 * swap files identically.
1715 *
1716 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1717 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1718 * swapfiles are handled *identically* after swapon time.
1719 *
1720 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1721 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1722 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1723 * requirements, they are simply tossed out - we will never use those blocks
1724 * for swapping.
1725 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001726 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1728 * which will scribble on the fs.
1729 *
1730 * The amount of disk space which a single swap extent represents varies.
1731 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1732 * extents in the list. To avoid much list walking, we cache the previous
1733 * search location in `curr_swap_extent', and start new searches from there.
1734 * This is extremely effective. The average number of iterations in
1735 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1736 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001737static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738{
Mel Gorman62c230b2012-07-31 16:44:55 -07001739 struct file *swap_file = sis->swap_file;
1740 struct address_space *mapping = swap_file->f_mapping;
1741 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 int ret;
1743
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 if (S_ISBLK(inode->i_mode)) {
1745 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001746 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001747 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 }
1749
Mel Gorman62c230b2012-07-31 16:44:55 -07001750 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001751 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001752 if (!ret) {
1753 sis->flags |= SWP_FILE;
1754 ret = add_swap_extent(sis, 0, sis->max, 0);
1755 *span = sis->pages;
1756 }
Mel Gormana509bc12012-07-31 16:44:57 -07001757 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001758 }
1759
Mel Gormana509bc12012-07-31 16:44:57 -07001760 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761}
1762
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001763static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001764 unsigned char *swap_map,
1765 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001766{
1767 int i, prev;
1768
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001769 if (prio >= 0)
1770 p->prio = prio;
1771 else
1772 p->prio = --least_priority;
1773 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001774 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001775 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001776 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001777 total_swap_pages += p->pages;
1778
1779 /* insert swap space into swap_list: */
1780 prev = -1;
1781 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1782 if (p->prio >= swap_info[i]->prio)
1783 break;
1784 prev = i;
1785 }
1786 p->next = i;
1787 if (prev < 0)
1788 swap_list.head = swap_list.next = p->type;
1789 else
1790 swap_info[prev]->next = p->type;
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001791}
1792
1793static void enable_swap_info(struct swap_info_struct *p, int prio,
1794 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001795 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001796 unsigned long *frontswap_map)
1797{
Minchan Kim4f898492013-04-30 15:26:54 -07001798 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001799 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001800 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001801 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001802 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001803 spin_unlock(&swap_lock);
1804}
1805
1806static void reinsert_swap_info(struct swap_info_struct *p)
1807{
1808 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001809 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001810 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001811 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001812 spin_unlock(&swap_lock);
1813}
1814
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001815SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001817 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001818 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001819 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001820 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 struct file *swap_file, *victim;
1822 struct address_space *mapping;
1823 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001824 struct filename *pathname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 int i, type, prev;
1826 int err;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001827 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001828
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 if (!capable(CAP_SYS_ADMIN))
1830 return -EPERM;
1831
Al Viro191c5422012-02-13 03:58:52 +00001832 BUG_ON(!current->mm);
1833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001836 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Jeff Layton669abf42012-10-10 16:43:10 -04001838 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 err = PTR_ERR(victim);
1840 if (IS_ERR(victim))
1841 goto out;
1842
1843 mapping = victim->f_mapping;
1844 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001845 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001846 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1847 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001848 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (p->swap_file->f_mapping == mapping)
1850 break;
1851 }
1852 prev = type;
1853 }
1854 if (type < 0) {
1855 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001856 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 goto out_dput;
1858 }
Al Viro191c5422012-02-13 03:58:52 +00001859 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 vm_unacct_memory(p->pages);
1861 else {
1862 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001863 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 goto out_dput;
1865 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001866 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001868 else
1869 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 if (type == swap_list.next) {
1871 /* just pick something that's safe... */
1872 swap_list.next = swap_list.head;
1873 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001874 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001875 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001876 for (i = p->next; i >= 0; i = swap_info[i]->next)
1877 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001878 least_priority++;
1879 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001880 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 total_swap_pages -= p->pages;
1882 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001883 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001884 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001885
David Rientjese1e12d22012-12-11 16:02:56 -08001886 set_current_oom_origin();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001887 err = try_to_unuse(type, false, 0); /* force all pages to be unused */
David Rientjese1e12d22012-12-11 16:02:56 -08001888 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 if (err) {
1891 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001892 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 goto out_dput;
1894 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001895
Shaohua Li815c2c52013-09-11 14:20:30 -07001896 flush_work(&p->discard_work);
1897
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001898 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001899 if (p->flags & SWP_CONTINUED)
1900 free_swap_count_continuations(p);
1901
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001902 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001903 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001904 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001906
1907 /* wait for anyone still in scan_swap_map */
1908 p->highest_bit = 0; /* cuts scans short */
1909 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001910 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001911 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001912 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001913 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001914 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001915 }
1916
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001918 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 p->swap_file = NULL;
1920 p->max = 0;
1921 swap_map = p->swap_map;
1922 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001923 cluster_info = p->cluster_info;
1924 p->cluster_info = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 p->flags = 0;
Minchan Kim4f898492013-04-30 15:26:54 -07001926 frontswap_map = frontswap_map_get(p);
1927 frontswap_map_set(p, NULL);
Shaohua Liec8acf22013-02-22 16:34:38 -08001928 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001929 spin_unlock(&swap_lock);
Minchan Kim4f898492013-04-30 15:26:54 -07001930 frontswap_invalidate_area(type);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001931 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001932 free_percpu(p->percpu_cluster);
1933 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001935 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001936 vfree(frontswap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001937 /* Destroy swap account informatin */
1938 swap_cgroup_swapoff(type);
1939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 inode = mapping->host;
1941 if (S_ISBLK(inode->i_mode)) {
1942 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001943 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001944 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001946 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001948 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 }
1950 filp_close(swap_file, NULL);
1951 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001952 atomic_inc(&proc_poll_event);
1953 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
1955out_dput:
1956 filp_close(victim, NULL);
1957out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001958 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 return err;
1960}
1961
1962#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001963static unsigned swaps_poll(struct file *file, poll_table *wait)
1964{
Kay Sieversf1514632011-07-12 20:48:39 +02001965 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001966
1967 poll_wait(file, &proc_poll_wait, wait);
1968
Kay Sieversf1514632011-07-12 20:48:39 +02001969 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1970 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001971 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1972 }
1973
1974 return POLLIN | POLLRDNORM;
1975}
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977/* iterator */
1978static void *swap_start(struct seq_file *swap, loff_t *pos)
1979{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001980 struct swap_info_struct *si;
1981 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 loff_t l = *pos;
1983
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001984 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001986 if (!l)
1987 return SEQ_START_TOKEN;
1988
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001989 for (type = 0; type < nr_swapfiles; type++) {
1990 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1991 si = swap_info[type];
1992 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001994 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001995 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 }
1997
1998 return NULL;
1999}
2000
2001static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2002{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002003 struct swap_info_struct *si = v;
2004 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002006 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002007 type = 0;
2008 else
2009 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002010
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002011 for (; type < nr_swapfiles; type++) {
2012 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2013 si = swap_info[type];
2014 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 continue;
2016 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002017 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 }
2019
2020 return NULL;
2021}
2022
2023static void swap_stop(struct seq_file *swap, void *v)
2024{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002025 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026}
2027
2028static int swap_show(struct seq_file *swap, void *v)
2029{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002030 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 struct file *file;
2032 int len;
2033
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002034 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002035 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2036 return 0;
2037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002039 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08002040 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002041 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002042 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002043 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002045 si->pages << (PAGE_SHIFT - 10),
2046 si->inuse_pages << (PAGE_SHIFT - 10),
2047 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return 0;
2049}
2050
Helge Deller15ad7cd2006-12-06 20:40:36 -08002051static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 .start = swap_start,
2053 .next = swap_next,
2054 .stop = swap_stop,
2055 .show = swap_show
2056};
2057
2058static int swaps_open(struct inode *inode, struct file *file)
2059{
Kay Sieversf1514632011-07-12 20:48:39 +02002060 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002061 int ret;
2062
Kay Sievers66d7dd52010-10-26 14:22:06 -07002063 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002064 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002065 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002066
Kay Sieversf1514632011-07-12 20:48:39 +02002067 seq = file->private_data;
2068 seq->poll_event = atomic_read(&proc_poll_event);
2069 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
2071
Helge Deller15ad7cd2006-12-06 20:40:36 -08002072static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 .open = swaps_open,
2074 .read = seq_read,
2075 .llseek = seq_lseek,
2076 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002077 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078};
2079
2080static int __init procswaps_init(void)
2081{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002082 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 return 0;
2084}
2085__initcall(procswaps_init);
2086#endif /* CONFIG_PROC_FS */
2087
Jan Beulich17963162008-12-16 11:35:24 +00002088#ifdef MAX_SWAPFILES_CHECK
2089static int __init max_swapfiles_check(void)
2090{
2091 MAX_SWAPFILES_CHECK();
2092 return 0;
2093}
2094late_initcall(max_swapfiles_check);
2095#endif
2096
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002097static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002099 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002101
2102 p = kzalloc(sizeof(*p), GFP_KERNEL);
2103 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002104 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002105
Hugh Dickins5d337b92005-09-03 15:54:41 -07002106 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002107 for (type = 0; type < nr_swapfiles; type++) {
2108 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002110 }
Christoph Lameter06972122006-06-23 02:03:35 -07002111 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002112 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002113 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002114 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002116 if (type >= nr_swapfiles) {
2117 p->type = type;
2118 swap_info[type] = p;
2119 /*
2120 * Write swap_info[type] before nr_swapfiles, in case a
2121 * racing procfs swap_start() or swap_next() is reading them.
2122 * (We never shrink nr_swapfiles, we never free this entry.)
2123 */
2124 smp_wmb();
2125 nr_swapfiles++;
2126 } else {
2127 kfree(p);
2128 p = swap_info[type];
2129 /*
2130 * Do not memset this entry: a racing procfs swap_next()
2131 * would be relying on p->type to remain valid.
2132 */
2133 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002134 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002137 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002138 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002139
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002140 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002141}
2142
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002143static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2144{
2145 int error;
2146
2147 if (S_ISBLK(inode->i_mode)) {
2148 p->bdev = bdgrab(I_BDEV(inode));
2149 error = blkdev_get(p->bdev,
2150 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
2151 sys_swapon);
2152 if (error < 0) {
2153 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002154 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002155 }
2156 p->old_block_size = block_size(p->bdev);
2157 error = set_blocksize(p->bdev, PAGE_SIZE);
2158 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002159 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002160 p->flags |= SWP_BLKDEV;
2161 } else if (S_ISREG(inode->i_mode)) {
2162 p->bdev = inode->i_sb->s_bdev;
2163 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002164 if (IS_SWAPFILE(inode))
2165 return -EBUSY;
2166 } else
2167 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002168
2169 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002170}
2171
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002172static unsigned long read_swap_header(struct swap_info_struct *p,
2173 union swap_header *swap_header,
2174 struct inode *inode)
2175{
2176 int i;
2177 unsigned long maxpages;
2178 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002179 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002180
2181 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002182 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002183 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002184 }
2185
2186 /* swap partition endianess hack... */
2187 if (swab32(swap_header->info.version) == 1) {
2188 swab32s(&swap_header->info.version);
2189 swab32s(&swap_header->info.last_page);
2190 swab32s(&swap_header->info.nr_badpages);
2191 for (i = 0; i < swap_header->info.nr_badpages; i++)
2192 swab32s(&swap_header->info.badpages[i]);
2193 }
2194 /* Check the swap header's sub-version */
2195 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002196 pr_warn("Unable to handle swap header version %d\n",
2197 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002198 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002199 }
2200
2201 p->lowest_bit = 1;
2202 p->cluster_next = 1;
2203 p->cluster_nr = 0;
2204
2205 /*
2206 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002207 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002208 * of bits for the swap offset in the swp_entry_t type, and
2209 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002210 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002211 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002212 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002213 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002214 * offset is extracted. This will mask all the bits from
2215 * the initial ~0UL mask that can't be encoded in either
2216 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002217 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002218 */
2219 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002220 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002221 last_page = swap_header->info.last_page;
2222 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002223 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002224 maxpages << (PAGE_SHIFT - 10),
2225 last_page << (PAGE_SHIFT - 10));
2226 }
2227 if (maxpages > last_page) {
2228 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002229 /* p->max is an unsigned int: don't overflow it */
2230 if ((unsigned int)maxpages == 0)
2231 maxpages = UINT_MAX;
2232 }
2233 p->highest_bit = maxpages - 1;
2234
2235 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002236 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002237 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2238 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002239 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002240 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002241 }
2242 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002243 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002244 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002245 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002246
2247 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002248}
2249
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002250static int setup_swap_map_and_extents(struct swap_info_struct *p,
2251 union swap_header *swap_header,
2252 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002253 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002254 unsigned long maxpages,
2255 sector_t *span)
2256{
2257 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002258 unsigned int nr_good_pages;
2259 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002260 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2261 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002262
2263 nr_good_pages = maxpages - 1; /* omit header page */
2264
Shaohua Li2a8f9442013-09-11 14:20:28 -07002265 cluster_set_null(&p->free_cluster_head);
2266 cluster_set_null(&p->free_cluster_tail);
Shaohua Li815c2c52013-09-11 14:20:30 -07002267 cluster_set_null(&p->discard_cluster_head);
2268 cluster_set_null(&p->discard_cluster_tail);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002269
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002270 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2271 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002272 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2273 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002274 if (page_nr < maxpages) {
2275 swap_map[page_nr] = SWAP_MAP_BAD;
2276 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002277 /*
2278 * Haven't marked the cluster free yet, no list
2279 * operation involved
2280 */
2281 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002282 }
2283 }
2284
Shaohua Li2a8f9442013-09-11 14:20:28 -07002285 /* Haven't marked the cluster free yet, no list operation involved */
2286 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2287 inc_cluster_info_page(p, cluster_info, i);
2288
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002289 if (nr_good_pages) {
2290 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002291 /*
2292 * Not mark the cluster free yet, no list
2293 * operation involved
2294 */
2295 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002296 p->max = maxpages;
2297 p->pages = nr_good_pages;
2298 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002299 if (nr_extents < 0)
2300 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002301 nr_good_pages = p->pages;
2302 }
2303 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002304 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002305 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002306 }
2307
Shaohua Li2a8f9442013-09-11 14:20:28 -07002308 if (!cluster_info)
2309 return nr_extents;
2310
2311 for (i = 0; i < nr_clusters; i++) {
2312 if (!cluster_count(&cluster_info[idx])) {
2313 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2314 if (cluster_is_null(&p->free_cluster_head)) {
2315 cluster_set_next_flag(&p->free_cluster_head,
2316 idx, 0);
2317 cluster_set_next_flag(&p->free_cluster_tail,
2318 idx, 0);
2319 } else {
2320 unsigned int tail;
2321
2322 tail = cluster_next(&p->free_cluster_tail);
2323 cluster_set_next(&cluster_info[tail], idx);
2324 cluster_set_next_flag(&p->free_cluster_tail,
2325 idx, 0);
2326 }
2327 }
2328 idx++;
2329 if (idx == nr_clusters)
2330 idx = 0;
2331 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002332 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002333}
2334
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002335/*
2336 * Helper to sys_swapon determining if a given swap
2337 * backing device queue supports DISCARD operations.
2338 */
2339static bool swap_discardable(struct swap_info_struct *si)
2340{
2341 struct request_queue *q = bdev_get_queue(si->bdev);
2342
2343 if (!q || !blk_queue_discard(q))
2344 return false;
2345
2346 return true;
2347}
2348
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002349SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2350{
2351 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002352 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002353 struct file *swap_file = NULL;
2354 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002355 int i;
2356 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002357 int error;
2358 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002359 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002360 sector_t span;
2361 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002362 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002363 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002364 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002365 struct page *page = NULL;
2366 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002367
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002368 if (swap_flags & ~SWAP_FLAGS_VALID)
2369 return -EINVAL;
2370
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002371 if (!capable(CAP_SYS_ADMIN))
2372 return -EPERM;
2373
2374 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002375 if (IS_ERR(p))
2376 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002377
Shaohua Li815c2c52013-09-11 14:20:30 -07002378 INIT_WORK(&p->discard_work, swap_discard_work);
2379
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002382 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002384 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 }
Jeff Layton669abf42012-10-10 16:43:10 -04002386 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002388 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002390 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 }
2392
2393 p->swap_file = swap_file;
2394 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002397 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002399 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002401 if (mapping == q->swap_file->f_mapping) {
2402 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002404 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 }
2406
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002407 inode = mapping->host;
2408 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002409 error = claim_swapfile(p, inode);
2410 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 /*
2414 * Read the swap header.
2415 */
2416 if (!mapping->a_ops->readpage) {
2417 error = -EINVAL;
2418 goto bad_swap;
2419 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002420 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 if (IS_ERR(page)) {
2422 error = PTR_ERR(page);
2423 goto bad_swap;
2424 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002425 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002427 maxpages = read_swap_header(p, swap_header, inode);
2428 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 error = -EINVAL;
2430 goto bad_swap;
2431 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002432
Hugh Dickins81e33972009-01-06 14:39:49 -08002433 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002434 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002435 if (!swap_map) {
2436 error = -ENOMEM;
2437 goto bad_swap;
2438 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002439 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2440 p->flags |= SWP_SOLIDSTATE;
2441 /*
2442 * select a random position to start with to help wear leveling
2443 * SSD
2444 */
2445 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2446
2447 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2448 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2449 if (!cluster_info) {
2450 error = -ENOMEM;
2451 goto bad_swap;
2452 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002453 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2454 if (!p->percpu_cluster) {
2455 error = -ENOMEM;
2456 goto bad_swap;
2457 }
2458 for_each_possible_cpu(i) {
2459 struct percpu_cluster *cluster;
2460 cluster = per_cpu_ptr(p->percpu_cluster, i);
2461 cluster_set_null(&cluster->index);
2462 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002463 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002464
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002465 error = swap_cgroup_swapon(p->type, maxpages);
2466 if (error)
2467 goto bad_swap;
2468
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002469 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002470 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002471 if (unlikely(nr_extents < 0)) {
2472 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 goto bad_swap;
2474 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002475 /* frontswap enabled? set up bit-per-page map for frontswap */
2476 if (frontswap_enabled)
Akinobu Mita7b579762013-06-12 14:05:08 -07002477 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478
Shaohua Li2a8f9442013-09-11 14:20:28 -07002479 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2480 /*
2481 * When discard is enabled for swap with no particular
2482 * policy flagged, we set all swap discard flags here in
2483 * order to sustain backward compatibility with older
2484 * swapon(8) releases.
2485 */
2486 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2487 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002488
Shaohua Li2a8f9442013-09-11 14:20:28 -07002489 /*
2490 * By flagging sys_swapon, a sysadmin can tell us to
2491 * either do single-time area discards only, or to just
2492 * perform discards for released swap page-clusters.
2493 * Now it's time to adjust the p->flags accordingly.
2494 */
2495 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2496 p->flags &= ~SWP_PAGE_DISCARD;
2497 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2498 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002499
Shaohua Li2a8f9442013-09-11 14:20:28 -07002500 /* issue a swapon-time discard if it's still required */
2501 if (p->flags & SWP_AREA_DISCARD) {
2502 int err = discard_swap(p);
2503 if (unlikely(err))
2504 pr_err("swapon: discard_swap(%p): %d\n",
2505 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002506 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002507 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002508
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002509 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002510 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002511 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002512 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002513 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002514 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002515
Andrew Morton465c47f2013-09-11 14:20:17 -07002516 pr_info("Adding %uk swap on %s. "
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002517 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002518 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002519 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2520 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002521 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002522 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2523 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002524 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002525
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002526 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002527 atomic_inc(&proc_poll_event);
2528 wake_up_interruptible(&proc_poll_wait);
2529
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002530 if (S_ISREG(inode->i_mode))
2531 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 error = 0;
2533 goto out;
2534bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002535 free_percpu(p->percpu_cluster);
2536 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002537 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002538 set_blocksize(p->bdev, p->old_block_size);
2539 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002541 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002542 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002543 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002546 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002548 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002549 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002550 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002551 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002552 inode = NULL;
2553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556out:
2557 if (page && !IS_ERR(page)) {
2558 kunmap(page);
2559 page_cache_release(page);
2560 }
2561 if (name)
2562 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002563 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002564 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 return error;
2566}
2567
2568void si_swapinfo(struct sysinfo *val)
2569{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002570 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 unsigned long nr_to_be_unused = 0;
2572
Hugh Dickins5d337b92005-09-03 15:54:41 -07002573 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002574 for (type = 0; type < nr_swapfiles; type++) {
2575 struct swap_info_struct *si = swap_info[type];
2576
2577 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2578 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002580 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002582 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583}
2584
2585/*
2586 * Verify that a swap entry is valid and increment its swap map count.
2587 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002588 * Returns error code in following case.
2589 * - success -> 0
2590 * - swp_entry is invalid -> EINVAL
2591 * - swp_entry is migration entry -> EINVAL
2592 * - swap-cache reference is requested but there is already one. -> EEXIST
2593 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002594 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002596static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002598 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002600 unsigned char count;
2601 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002602 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
Andi Kleena7420aa2009-09-16 11:50:05 +02002604 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002605 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 type = swp_type(entry);
2608 if (type >= nr_swapfiles)
2609 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002610 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 offset = swp_offset(entry);
2612
Shaohua Liec8acf22013-02-22 16:34:38 -08002613 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002614 if (unlikely(offset >= p->max))
2615 goto unlock_out;
2616
Hugh Dickins253d5532009-12-14 17:58:44 -08002617 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002618
2619 /*
2620 * swapin_readahead() doesn't check if a swap entry is valid, so the
2621 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2622 */
2623 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2624 err = -ENOENT;
2625 goto unlock_out;
2626 }
2627
Hugh Dickins253d5532009-12-14 17:58:44 -08002628 has_cache = count & SWAP_HAS_CACHE;
2629 count &= ~SWAP_HAS_CACHE;
2630 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002631
Hugh Dickins253d5532009-12-14 17:58:44 -08002632 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002633
2634 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002635 if (!has_cache && count)
2636 has_cache = SWAP_HAS_CACHE;
2637 else if (has_cache) /* someone else added cache */
2638 err = -EEXIST;
2639 else /* no users remaining */
2640 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002641
2642 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002643
Hugh Dickins570a335b2009-12-14 17:58:46 -08002644 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2645 count += usage;
2646 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002647 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002648 else if (swap_count_continued(p, offset, count))
2649 count = COUNT_CONTINUED;
2650 else
2651 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002652 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002653 err = -ENOENT; /* unused swap entry */
2654
2655 p->swap_map[offset] = count | has_cache;
2656
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002657unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002658 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002660 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
2662bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002663 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 goto out;
2665}
Hugh Dickins253d5532009-12-14 17:58:44 -08002666
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002667/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002668 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2669 * (in which case its reference count is never incremented).
2670 */
2671void swap_shmem_alloc(swp_entry_t entry)
2672{
2673 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2674}
2675
2676/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002677 * Increase reference count of swap entry by 1.
2678 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2679 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2680 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2681 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002682 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002683int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002684{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002685 int err = 0;
2686
2687 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2688 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2689 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002690}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002692/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002693 * @entry: swap entry for which we allocate swap cache.
2694 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002695 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002696 * This can return error codes. Returns 0 at success.
2697 * -EBUSY means there is a swap cache.
2698 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002699 */
2700int swapcache_prepare(swp_entry_t entry)
2701{
Hugh Dickins253d5532009-12-14 17:58:44 -08002702 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002703}
2704
Mel Gormanf981c592012-07-31 16:44:47 -07002705struct swap_info_struct *page_swap_info(struct page *page)
2706{
2707 swp_entry_t swap = { .val = page_private(page) };
2708 BUG_ON(!PageSwapCache(page));
2709 return swap_info[swp_type(swap)];
2710}
2711
2712/*
2713 * out-of-line __page_file_ methods to avoid include hell.
2714 */
2715struct address_space *__page_file_mapping(struct page *page)
2716{
2717 VM_BUG_ON(!PageSwapCache(page));
2718 return page_swap_info(page)->swap_file->f_mapping;
2719}
2720EXPORT_SYMBOL_GPL(__page_file_mapping);
2721
2722pgoff_t __page_file_index(struct page *page)
2723{
2724 swp_entry_t swap = { .val = page_private(page) };
2725 VM_BUG_ON(!PageSwapCache(page));
2726 return swp_offset(swap);
2727}
2728EXPORT_SYMBOL_GPL(__page_file_index);
2729
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002731 * add_swap_count_continuation - called when a swap count is duplicated
2732 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2733 * page of the original vmalloc'ed swap_map, to hold the continuation count
2734 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2735 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2736 *
2737 * These continuation pages are seldom referenced: the common paths all work
2738 * on the original swap_map, only referring to a continuation page when the
2739 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2740 *
2741 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2742 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2743 * can be called after dropping locks.
2744 */
2745int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2746{
2747 struct swap_info_struct *si;
2748 struct page *head;
2749 struct page *page;
2750 struct page *list_page;
2751 pgoff_t offset;
2752 unsigned char count;
2753
2754 /*
2755 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2756 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2757 */
2758 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2759
2760 si = swap_info_get(entry);
2761 if (!si) {
2762 /*
2763 * An acceptable race has occurred since the failing
2764 * __swap_duplicate(): the swap entry has been freed,
2765 * perhaps even the whole swap_map cleared for swapoff.
2766 */
2767 goto outer;
2768 }
2769
2770 offset = swp_offset(entry);
2771 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2772
2773 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2774 /*
2775 * The higher the swap count, the more likely it is that tasks
2776 * will race to add swap count continuation: we need to avoid
2777 * over-provisioning.
2778 */
2779 goto out;
2780 }
2781
2782 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002783 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002784 return -ENOMEM;
2785 }
2786
2787 /*
2788 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2789 * no architecture is using highmem pages for kernel pagetables: so it
2790 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2791 */
2792 head = vmalloc_to_page(si->swap_map + offset);
2793 offset &= ~PAGE_MASK;
2794
2795 /*
2796 * Page allocation does not initialize the page's lru field,
2797 * but it does always reset its private field.
2798 */
2799 if (!page_private(head)) {
2800 BUG_ON(count & COUNT_CONTINUED);
2801 INIT_LIST_HEAD(&head->lru);
2802 set_page_private(head, SWP_CONTINUED);
2803 si->flags |= SWP_CONTINUED;
2804 }
2805
2806 list_for_each_entry(list_page, &head->lru, lru) {
2807 unsigned char *map;
2808
2809 /*
2810 * If the previous map said no continuation, but we've found
2811 * a continuation page, free our allocation and use this one.
2812 */
2813 if (!(count & COUNT_CONTINUED))
2814 goto out;
2815
Cong Wang9b04c5f2011-11-25 23:14:39 +08002816 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002817 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002818 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002819
2820 /*
2821 * If this continuation count now has some space in it,
2822 * free our allocation and use this one.
2823 */
2824 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2825 goto out;
2826 }
2827
2828 list_add_tail(&page->lru, &head->lru);
2829 page = NULL; /* now it's attached, don't free it */
2830out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002831 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002832outer:
2833 if (page)
2834 __free_page(page);
2835 return 0;
2836}
2837
2838/*
2839 * swap_count_continued - when the original swap_map count is incremented
2840 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2841 * into, carry if so, or else fail until a new continuation page is allocated;
2842 * when the original swap_map count is decremented from 0 with continuation,
2843 * borrow from the continuation and report whether it still holds more.
2844 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2845 */
2846static bool swap_count_continued(struct swap_info_struct *si,
2847 pgoff_t offset, unsigned char count)
2848{
2849 struct page *head;
2850 struct page *page;
2851 unsigned char *map;
2852
2853 head = vmalloc_to_page(si->swap_map + offset);
2854 if (page_private(head) != SWP_CONTINUED) {
2855 BUG_ON(count & COUNT_CONTINUED);
2856 return false; /* need to add count continuation */
2857 }
2858
2859 offset &= ~PAGE_MASK;
2860 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002861 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002862
2863 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2864 goto init_map; /* jump over SWAP_CONT_MAX checks */
2865
2866 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2867 /*
2868 * Think of how you add 1 to 999
2869 */
2870 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002871 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002872 page = list_entry(page->lru.next, struct page, lru);
2873 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002874 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002875 }
2876 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002877 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002878 page = list_entry(page->lru.next, struct page, lru);
2879 if (page == head)
2880 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002881 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002882init_map: *map = 0; /* we didn't zero the page */
2883 }
2884 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002885 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002886 page = list_entry(page->lru.prev, struct page, lru);
2887 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002888 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002889 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002890 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002891 page = list_entry(page->lru.prev, struct page, lru);
2892 }
2893 return true; /* incremented */
2894
2895 } else { /* decrementing */
2896 /*
2897 * Think of how you subtract 1 from 1000
2898 */
2899 BUG_ON(count != COUNT_CONTINUED);
2900 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002901 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002902 page = list_entry(page->lru.next, struct page, lru);
2903 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002904 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002905 }
2906 BUG_ON(*map == 0);
2907 *map -= 1;
2908 if (*map == 0)
2909 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002910 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002911 page = list_entry(page->lru.prev, struct page, lru);
2912 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002913 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002914 *map = SWAP_CONT_MAX | count;
2915 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002916 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002917 page = list_entry(page->lru.prev, struct page, lru);
2918 }
2919 return count == COUNT_CONTINUED;
2920 }
2921}
2922
2923/*
2924 * free_swap_count_continuations - swapoff free all the continuation pages
2925 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2926 */
2927static void free_swap_count_continuations(struct swap_info_struct *si)
2928{
2929 pgoff_t offset;
2930
2931 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2932 struct page *head;
2933 head = vmalloc_to_page(si->swap_map + offset);
2934 if (page_private(head)) {
2935 struct list_head *this, *next;
2936 list_for_each_safe(this, next, &head->lru) {
2937 struct page *page;
2938 page = list_entry(this, struct page, lru);
2939 list_del(this);
2940 __free_page(page);
2941 }
2942 }
2943 }
2944}