blob: 58877312cf6b94b74da00b8913bf8db64d56ab26 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/mm.h>
9#include <linux/hugetlb.h>
10#include <linux/mman.h>
11#include <linux/slab.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/vmalloc.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
Hugh Dickins072441e2011-06-27 16:18:02 -070017#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080019#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080024#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070032#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070033#include <linux/oom.h>
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060034#include <linux/frontswap.h>
35#include <linux/swapfile.h>
Mel Gormanf981c592012-07-31 16:44:47 -070036#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/pgtable.h>
39#include <asm/tlbflush.h>
40#include <linux/swapops.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080041#include <linux/swap_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Hugh Dickins570a335b2009-12-14 17:58:46 -080043static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
44 unsigned char);
45static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080046static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080047
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060048DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070049static unsigned int nr_swapfiles;
Shaohua Liec8acf22013-02-22 16:34:38 -080050atomic_long_t nr_swap_pages;
51/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -070052long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070053static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055static const char Bad_file[] = "Bad swap file entry ";
56static const char Unused_file[] = "Unused swap file entry ";
57static const char Bad_offset[] = "Bad swap offset entry ";
58static const char Unused_offset[] = "Unused swap offset entry ";
59
Dan Streetmanadfab832014-06-04 16:09:53 -070060/*
61 * all active swap_info_structs
62 * protected with swap_lock, and ordered by priority.
63 */
Dan Streetman18ab4d42014-06-04 16:09:59 -070064PLIST_HEAD(swap_active_head);
65
66/*
67 * all available (active, not full) swap_info_structs
68 * protected with swap_avail_lock, ordered by priority.
69 * This is used by get_swap_page() instead of swap_active_head
70 * because swap_active_head includes all swap_info_structs,
71 * but get_swap_page() doesn't need to look at full ones.
72 * This uses its own lock instead of swap_lock because when a
73 * swap_info_struct changes between not-full/full, it needs to
74 * add/remove itself to/from this list, but the swap_info_struct->lock
75 * is held and the locking order requires swap_lock to be taken
76 * before any swap_info_struct->lock.
77 */
78static PLIST_HEAD(swap_avail_head);
79static DEFINE_SPINLOCK(swap_avail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060081struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Ingo Molnarfc0abb12006-01-18 17:42:33 -080083static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Kay Sievers66d7dd52010-10-26 14:22:06 -070085static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
86/* Activity counter to indicate that a swapon or swapoff has occurred */
87static atomic_t proc_poll_event = ATOMIC_INIT(0);
88
Hugh Dickins8d69aae2009-12-14 17:58:45 -080089static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070090{
Hugh Dickins570a335b2009-12-14 17:58:46 -080091 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070092}
93
Hugh Dickinsefa90a92009-12-14 17:58:41 -080094/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070095static int
96__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
97{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080098 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070099 struct page *page;
100 int ret = 0;
101
Shaohua Li33806f02013-02-22 16:34:37 -0800102 page = find_get_page(swap_address_space(entry), entry.val);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700103 if (!page)
104 return 0;
105 /*
106 * This function is called from scan_swap_map() and it's called
107 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
108 * We have to use trylock for avoiding deadlock. This is a special
109 * case and you should use try_to_free_swap() with explicit lock_page()
110 * in usual operations.
111 */
112 if (trylock_page(page)) {
113 ret = try_to_free_swap(page);
114 unlock_page(page);
115 }
116 page_cache_release(page);
117 return ret;
118}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800121 * swapon tell device that all the old swap contents can be discarded,
122 * to allow the swap device to optimize its wear-levelling.
123 */
124static int discard_swap(struct swap_info_struct *si)
125{
126 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800127 sector_t start_block;
128 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800129 int err = 0;
130
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800131 /* Do not discard the swap header page! */
132 se = &si->first_swap_extent;
133 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
134 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
135 if (nr_blocks) {
136 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200137 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800138 if (err)
139 return err;
140 cond_resched();
141 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800142
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800143 list_for_each_entry(se, &si->first_swap_extent.list, list) {
144 start_block = se->start_block << (PAGE_SHIFT - 9);
145 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800146
147 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200148 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800149 if (err)
150 break;
151
152 cond_resched();
153 }
154 return err; /* That will often be -EOPNOTSUPP */
155}
156
Hugh Dickins7992fde2009-01-06 14:39:53 -0800157/*
158 * swap allocation tell device that a cluster of swap can now be discarded,
159 * to allow the swap device to optimize its wear-levelling.
160 */
161static void discard_swap_cluster(struct swap_info_struct *si,
162 pgoff_t start_page, pgoff_t nr_pages)
163{
164 struct swap_extent *se = si->curr_swap_extent;
165 int found_extent = 0;
166
167 while (nr_pages) {
168 struct list_head *lh;
169
170 if (se->start_page <= start_page &&
171 start_page < se->start_page + se->nr_pages) {
172 pgoff_t offset = start_page - se->start_page;
173 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800174 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800175
176 if (nr_blocks > nr_pages)
177 nr_blocks = nr_pages;
178 start_page += nr_blocks;
179 nr_pages -= nr_blocks;
180
181 if (!found_extent++)
182 si->curr_swap_extent = se;
183
184 start_block <<= PAGE_SHIFT - 9;
185 nr_blocks <<= PAGE_SHIFT - 9;
186 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200187 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800188 break;
189 }
190
191 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800192 se = list_entry(lh, struct swap_extent, list);
193 }
194}
195
Hugh Dickins048c27f2005-09-03 15:54:40 -0700196#define SWAPFILE_CLUSTER 256
197#define LATENCY_LIMIT 256
198
Shaohua Li2a8f9442013-09-11 14:20:28 -0700199static inline void cluster_set_flag(struct swap_cluster_info *info,
200 unsigned int flag)
201{
202 info->flags = flag;
203}
204
205static inline unsigned int cluster_count(struct swap_cluster_info *info)
206{
207 return info->data;
208}
209
210static inline void cluster_set_count(struct swap_cluster_info *info,
211 unsigned int c)
212{
213 info->data = c;
214}
215
216static inline void cluster_set_count_flag(struct swap_cluster_info *info,
217 unsigned int c, unsigned int f)
218{
219 info->flags = f;
220 info->data = c;
221}
222
223static inline unsigned int cluster_next(struct swap_cluster_info *info)
224{
225 return info->data;
226}
227
228static inline void cluster_set_next(struct swap_cluster_info *info,
229 unsigned int n)
230{
231 info->data = n;
232}
233
234static inline void cluster_set_next_flag(struct swap_cluster_info *info,
235 unsigned int n, unsigned int f)
236{
237 info->flags = f;
238 info->data = n;
239}
240
241static inline bool cluster_is_free(struct swap_cluster_info *info)
242{
243 return info->flags & CLUSTER_FLAG_FREE;
244}
245
246static inline bool cluster_is_null(struct swap_cluster_info *info)
247{
248 return info->flags & CLUSTER_FLAG_NEXT_NULL;
249}
250
251static inline void cluster_set_null(struct swap_cluster_info *info)
252{
253 info->flags = CLUSTER_FLAG_NEXT_NULL;
254 info->data = 0;
255}
256
Shaohua Li815c2c52013-09-11 14:20:30 -0700257/* Add a cluster to discard list and schedule it to do discard */
258static void swap_cluster_schedule_discard(struct swap_info_struct *si,
259 unsigned int idx)
260{
261 /*
262 * If scan_swap_map() can't find a free cluster, it will check
263 * si->swap_map directly. To make sure the discarding cluster isn't
264 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
265 * will be cleared after discard
266 */
267 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
268 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
269
270 if (cluster_is_null(&si->discard_cluster_head)) {
271 cluster_set_next_flag(&si->discard_cluster_head,
272 idx, 0);
273 cluster_set_next_flag(&si->discard_cluster_tail,
274 idx, 0);
275 } else {
276 unsigned int tail = cluster_next(&si->discard_cluster_tail);
277 cluster_set_next(&si->cluster_info[tail], idx);
278 cluster_set_next_flag(&si->discard_cluster_tail,
279 idx, 0);
280 }
281
282 schedule_work(&si->discard_work);
283}
284
285/*
286 * Doing discard actually. After a cluster discard is finished, the cluster
287 * will be added to free cluster list. caller should hold si->lock.
288*/
289static void swap_do_scheduled_discard(struct swap_info_struct *si)
290{
291 struct swap_cluster_info *info;
292 unsigned int idx;
293
294 info = si->cluster_info;
295
296 while (!cluster_is_null(&si->discard_cluster_head)) {
297 idx = cluster_next(&si->discard_cluster_head);
298
299 cluster_set_next_flag(&si->discard_cluster_head,
300 cluster_next(&info[idx]), 0);
301 if (cluster_next(&si->discard_cluster_tail) == idx) {
302 cluster_set_null(&si->discard_cluster_head);
303 cluster_set_null(&si->discard_cluster_tail);
304 }
305 spin_unlock(&si->lock);
306
307 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
308 SWAPFILE_CLUSTER);
309
310 spin_lock(&si->lock);
311 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
312 if (cluster_is_null(&si->free_cluster_head)) {
313 cluster_set_next_flag(&si->free_cluster_head,
314 idx, 0);
315 cluster_set_next_flag(&si->free_cluster_tail,
316 idx, 0);
317 } else {
318 unsigned int tail;
319
320 tail = cluster_next(&si->free_cluster_tail);
321 cluster_set_next(&info[tail], idx);
322 cluster_set_next_flag(&si->free_cluster_tail,
323 idx, 0);
324 }
325 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
326 0, SWAPFILE_CLUSTER);
327 }
328}
329
330static void swap_discard_work(struct work_struct *work)
331{
332 struct swap_info_struct *si;
333
334 si = container_of(work, struct swap_info_struct, discard_work);
335
336 spin_lock(&si->lock);
337 swap_do_scheduled_discard(si);
338 spin_unlock(&si->lock);
339}
340
Shaohua Li2a8f9442013-09-11 14:20:28 -0700341/*
342 * The cluster corresponding to page_nr will be used. The cluster will be
343 * removed from free cluster list and its usage counter will be increased.
344 */
345static void inc_cluster_info_page(struct swap_info_struct *p,
346 struct swap_cluster_info *cluster_info, unsigned long page_nr)
347{
348 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
349
350 if (!cluster_info)
351 return;
352 if (cluster_is_free(&cluster_info[idx])) {
353 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
354 cluster_set_next_flag(&p->free_cluster_head,
355 cluster_next(&cluster_info[idx]), 0);
356 if (cluster_next(&p->free_cluster_tail) == idx) {
357 cluster_set_null(&p->free_cluster_tail);
358 cluster_set_null(&p->free_cluster_head);
359 }
360 cluster_set_count_flag(&cluster_info[idx], 0, 0);
361 }
362
363 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
364 cluster_set_count(&cluster_info[idx],
365 cluster_count(&cluster_info[idx]) + 1);
366}
367
368/*
369 * The cluster corresponding to page_nr decreases one usage. If the usage
370 * counter becomes 0, which means no page in the cluster is in using, we can
371 * optionally discard the cluster and add it to free cluster list.
372 */
373static void dec_cluster_info_page(struct swap_info_struct *p,
374 struct swap_cluster_info *cluster_info, unsigned long page_nr)
375{
376 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
377
378 if (!cluster_info)
379 return;
380
381 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
382 cluster_set_count(&cluster_info[idx],
383 cluster_count(&cluster_info[idx]) - 1);
384
385 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700386 /*
387 * If the swap is discardable, prepare discard the cluster
388 * instead of free it immediately. The cluster will be freed
389 * after discard.
390 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700391 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
392 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700393 swap_cluster_schedule_discard(p, idx);
394 return;
395 }
396
Shaohua Li2a8f9442013-09-11 14:20:28 -0700397 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
398 if (cluster_is_null(&p->free_cluster_head)) {
399 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
400 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
401 } else {
402 unsigned int tail = cluster_next(&p->free_cluster_tail);
403 cluster_set_next(&cluster_info[tail], idx);
404 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
405 }
406 }
407}
408
409/*
410 * It's possible scan_swap_map() uses a free cluster in the middle of free
411 * cluster list. Avoiding such abuse to avoid list corruption.
412 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700413static bool
414scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700415 unsigned long offset)
416{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700417 struct percpu_cluster *percpu_cluster;
418 bool conflict;
419
Shaohua Li2a8f9442013-09-11 14:20:28 -0700420 offset /= SWAPFILE_CLUSTER;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700421 conflict = !cluster_is_null(&si->free_cluster_head) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700422 offset != cluster_next(&si->free_cluster_head) &&
423 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700424
425 if (!conflict)
426 return false;
427
428 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
429 cluster_set_null(&percpu_cluster->index);
430 return true;
431}
432
433/*
434 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
435 * might involve allocating a new cluster for current CPU too.
436 */
437static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
438 unsigned long *offset, unsigned long *scan_base)
439{
440 struct percpu_cluster *cluster;
441 bool found_free;
442 unsigned long tmp;
443
444new_cluster:
445 cluster = this_cpu_ptr(si->percpu_cluster);
446 if (cluster_is_null(&cluster->index)) {
447 if (!cluster_is_null(&si->free_cluster_head)) {
448 cluster->index = si->free_cluster_head;
449 cluster->next = cluster_next(&cluster->index) *
450 SWAPFILE_CLUSTER;
451 } else if (!cluster_is_null(&si->discard_cluster_head)) {
452 /*
453 * we don't have free cluster but have some clusters in
454 * discarding, do discard now and reclaim them
455 */
456 swap_do_scheduled_discard(si);
457 *scan_base = *offset = si->cluster_next;
458 goto new_cluster;
459 } else
460 return;
461 }
462
463 found_free = false;
464
465 /*
466 * Other CPUs can use our cluster if they can't find a free cluster,
467 * check if there is still free entry in the cluster
468 */
469 tmp = cluster->next;
470 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
471 SWAPFILE_CLUSTER) {
472 if (!si->swap_map[tmp]) {
473 found_free = true;
474 break;
475 }
476 tmp++;
477 }
478 if (!found_free) {
479 cluster_set_null(&cluster->index);
480 goto new_cluster;
481 }
482 cluster->next = tmp + 1;
483 *offset = tmp;
484 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700485}
486
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700487static unsigned long scan_swap_map(struct swap_info_struct *si,
488 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800490 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800491 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800492 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700493 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800495 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700496 * We try to cluster swap pages by allocating them sequentially
497 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
498 * way, however, we resort to first-free allocation, starting
499 * a new cluster. This prevents us from scattering swap pages
500 * all over the entire swap partition, so that we reduce
501 * overall disk seek times between swap pages. -- sct
502 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800503 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700504 */
505
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700506 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800507 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800508
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700509 /* SSD algorithm */
510 if (si->cluster_info) {
511 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
512 goto checks;
513 }
514
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800515 if (unlikely(!si->cluster_nr--)) {
516 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
517 si->cluster_nr = SWAPFILE_CLUSTER - 1;
518 goto checks;
519 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700520
Shaohua Liec8acf22013-02-22 16:34:38 -0800521 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700522
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800523 /*
524 * If seek is expensive, start searching for new cluster from
525 * start of partition, to minimize the span of allocated swap.
Chen Yucong50088c42014-06-04 16:10:57 -0700526 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
527 * case, just handled by scan_swap_map_try_ssd_cluster() above.
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800528 */
Chen Yucong50088c42014-06-04 16:10:57 -0700529 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700530 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
531
532 /* Locate the first empty (unaligned) cluster */
533 for (; last_in_cluster <= si->highest_bit; offset++) {
534 if (si->swap_map[offset])
535 last_in_cluster = offset + SWAPFILE_CLUSTER;
536 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800537 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800538 offset -= SWAPFILE_CLUSTER - 1;
539 si->cluster_next = offset;
540 si->cluster_nr = SWAPFILE_CLUSTER - 1;
541 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700543 if (unlikely(--latency_ration < 0)) {
544 cond_resched();
545 latency_ration = LATENCY_LIMIT;
546 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700547 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800548
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800549 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800550 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800551 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700553
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800554checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700555 if (si->cluster_info) {
556 while (scan_swap_map_ssd_cluster_conflict(si, offset))
557 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
558 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800559 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700560 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700561 if (!si->highest_bit)
562 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800563 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800564 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700565
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700566 /* reuse swap entry of cache-only swap if not busy. */
567 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700568 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800569 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700570 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800571 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700572 /* entry was freed successfully, try to use this again */
573 if (swap_was_freed)
574 goto checks;
575 goto scan; /* check next one */
576 }
577
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800578 if (si->swap_map[offset])
579 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700580
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800581 if (offset == si->lowest_bit)
582 si->lowest_bit++;
583 if (offset == si->highest_bit)
584 si->highest_bit--;
585 si->inuse_pages++;
586 if (si->inuse_pages == si->pages) {
587 si->lowest_bit = si->max;
588 si->highest_bit = 0;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700589 spin_lock(&swap_avail_lock);
590 plist_del(&si->avail_list, &swap_avail_head);
591 spin_unlock(&swap_avail_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800592 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800593 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700594 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800595 si->cluster_next = offset + 1;
596 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800597
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800598 return offset;
599
600scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800601 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700602 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700603 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800604 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700605 goto checks;
606 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700607 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800608 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700609 goto checks;
610 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700611 if (unlikely(--latency_ration < 0)) {
612 cond_resched();
613 latency_ration = LATENCY_LIMIT;
614 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700615 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800616 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800617 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800618 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800619 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800620 goto checks;
621 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700622 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800623 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700624 goto checks;
625 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800626 if (unlikely(--latency_ration < 0)) {
627 cond_resched();
628 latency_ration = LATENCY_LIMIT;
629 }
Jamie Liua5998062014-01-23 15:53:40 -0800630 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800631 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800632 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700633
634no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700635 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 return 0;
637}
638
639swp_entry_t get_swap_page(void)
640{
Dan Streetmanadfab832014-06-04 16:09:53 -0700641 struct swap_info_struct *si, *next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700642 pgoff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Shaohua Liec8acf22013-02-22 16:34:38 -0800644 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700645 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800646 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Dan Streetman18ab4d42014-06-04 16:09:59 -0700648 spin_lock(&swap_avail_lock);
649
650start_over:
651 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
652 /* requeue si to after same-priority siblings */
653 plist_requeue(&si->avail_list, &swap_avail_head);
654 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800655 spin_lock(&si->lock);
Dan Streetmanadfab832014-06-04 16:09:53 -0700656 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
Dan Streetman18ab4d42014-06-04 16:09:59 -0700657 spin_lock(&swap_avail_lock);
658 if (plist_node_empty(&si->avail_list)) {
659 spin_unlock(&si->lock);
660 goto nextsi;
661 }
662 WARN(!si->highest_bit,
663 "swap_info %d in list but !highest_bit\n",
664 si->type);
665 WARN(!(si->flags & SWP_WRITEOK),
666 "swap_info %d in list but !SWP_WRITEOK\n",
667 si->type);
668 plist_del(&si->avail_list, &swap_avail_head);
Shaohua Liec8acf22013-02-22 16:34:38 -0800669 spin_unlock(&si->lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700670 goto nextsi;
Shaohua Liec8acf22013-02-22 16:34:38 -0800671 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700672
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700673 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800674 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800675 spin_unlock(&si->lock);
676 if (offset)
Dan Streetmanadfab832014-06-04 16:09:53 -0700677 return swp_entry(si->type, offset);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700678 pr_debug("scan_swap_map of si %d failed to find offset\n",
679 si->type);
680 spin_lock(&swap_avail_lock);
681nextsi:
Dan Streetmanadfab832014-06-04 16:09:53 -0700682 /*
683 * if we got here, it's likely that si was almost full before,
684 * and since scan_swap_map() can drop the si->lock, multiple
685 * callers probably all tried to get a page from the same si
Dan Streetman18ab4d42014-06-04 16:09:59 -0700686 * and it filled up before we could get one; or, the si filled
687 * up between us dropping swap_avail_lock and taking si->lock.
688 * Since we dropped the swap_avail_lock, the swap_avail_head
689 * list may have been modified; so if next is still in the
690 * swap_avail_head list then try it, otherwise start over.
Dan Streetmanadfab832014-06-04 16:09:53 -0700691 */
Dan Streetman18ab4d42014-06-04 16:09:59 -0700692 if (plist_node_empty(&next->avail_list))
693 goto start_over;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700695
Dan Streetman18ab4d42014-06-04 16:09:59 -0700696 spin_unlock(&swap_avail_lock);
697
Shaohua Liec8acf22013-02-22 16:34:38 -0800698 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700699noswap:
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700700 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800703/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700704swp_entry_t get_swap_page_of_type(int type)
705{
706 struct swap_info_struct *si;
707 pgoff_t offset;
708
Hugh Dickins910321e2010-09-09 16:38:07 -0700709 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800710 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700711 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800712 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700713 /* This is called for allocating swap entry, not cache */
714 offset = scan_swap_map(si, 1);
715 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800716 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700717 return swp_entry(type, offset);
718 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800719 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700720 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800721 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700722 return (swp_entry_t) {0};
723}
724
Hugh Dickins73c34b62009-12-14 17:58:43 -0800725static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800727 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 unsigned long offset, type;
729
730 if (!entry.val)
731 goto out;
732 type = swp_type(entry);
733 if (type >= nr_swapfiles)
734 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800735 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 if (!(p->flags & SWP_USED))
737 goto bad_device;
738 offset = swp_offset(entry);
739 if (offset >= p->max)
740 goto bad_offset;
741 if (!p->swap_map[offset])
742 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800743 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 return p;
745
746bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700747 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 goto out;
749bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700750 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 goto out;
752bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700753 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 goto out;
755bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700756 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757out:
758 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800759}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800761static unsigned char swap_entry_free(struct swap_info_struct *p,
762 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763{
Hugh Dickins253d5532009-12-14 17:58:44 -0800764 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800765 unsigned char count;
766 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700768 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800769 has_cache = count & SWAP_HAS_CACHE;
770 count &= ~SWAP_HAS_CACHE;
771
772 if (usage == SWAP_HAS_CACHE) {
773 VM_BUG_ON(!has_cache);
774 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800775 } else if (count == SWAP_MAP_SHMEM) {
776 /*
777 * Or we could insist on shmem.c using a special
778 * swap_shmem_free() and free_shmem_swap_and_cache()...
779 */
780 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800781 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
782 if (count == COUNT_CONTINUED) {
783 if (swap_count_continued(p, offset, count))
784 count = SWAP_MAP_MAX | COUNT_CONTINUED;
785 else
786 count = SWAP_MAP_MAX;
787 } else
788 count--;
789 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800790
791 if (!count)
792 mem_cgroup_uncharge_swap(entry);
793
794 usage = count | has_cache;
795 p->swap_map[offset] = usage;
796
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700797 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800798 if (!usage) {
Shaohua Li2a8f9442013-09-11 14:20:28 -0700799 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700800 if (offset < p->lowest_bit)
801 p->lowest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700802 if (offset > p->highest_bit) {
803 bool was_full = !p->highest_bit;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700804 p->highest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700805 if (was_full && (p->flags & SWP_WRITEOK)) {
806 spin_lock(&swap_avail_lock);
807 WARN_ON(!plist_node_empty(&p->avail_list));
808 if (plist_node_empty(&p->avail_list))
809 plist_add(&p->avail_list,
810 &swap_avail_head);
811 spin_unlock(&swap_avail_lock);
812 }
813 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800814 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700815 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600816 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700817 if (p->flags & SWP_BLKDEV) {
818 struct gendisk *disk = p->bdev->bd_disk;
819 if (disk->fops->swap_slot_free_notify)
820 disk->fops->swap_slot_free_notify(p->bdev,
821 offset);
822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800824
825 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826}
827
828/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800829 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 * is still around or has not been recycled.
831 */
832void swap_free(swp_entry_t entry)
833{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800834 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836 p = swap_info_get(entry);
837 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800838 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800839 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
841}
842
843/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700844 * Called after dropping swapcache to decrease refcnt to swap entries.
845 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700846void swapcache_free(swp_entry_t entry)
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700847{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700848 struct swap_info_struct *p;
849
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700850 p = swap_info_get(entry);
851 if (p) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700852 swap_entry_free(p, entry, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800853 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700854 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700855}
856
857/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700858 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800859 * This does not give an exact answer when swap count is continued,
860 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700862int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700864 int count = 0;
865 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 swp_entry_t entry;
867
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700868 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 p = swap_info_get(entry);
870 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700871 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800872 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700874 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877/*
Minchan Kim8334b962015-09-08 15:00:24 -0700878 * How many references to @entry are currently swapped out?
879 * This considers COUNT_CONTINUED so it returns exact answer.
880 */
881int swp_swapcount(swp_entry_t entry)
882{
883 int count, tmp_count, n;
884 struct swap_info_struct *p;
885 struct page *page;
886 pgoff_t offset;
887 unsigned char *map;
888
889 p = swap_info_get(entry);
890 if (!p)
891 return 0;
892
893 count = swap_count(p->swap_map[swp_offset(entry)]);
894 if (!(count & COUNT_CONTINUED))
895 goto out;
896
897 count &= ~COUNT_CONTINUED;
898 n = SWAP_MAP_MAX + 1;
899
900 offset = swp_offset(entry);
901 page = vmalloc_to_page(p->swap_map + offset);
902 offset &= ~PAGE_MASK;
903 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
904
905 do {
906 page = list_entry(page->lru.next, struct page, lru);
907 map = kmap_atomic(page);
908 tmp_count = map[offset];
909 kunmap_atomic(map);
910
911 count += (tmp_count & ~COUNT_CONTINUED) * n;
912 n *= (SWAP_CONT_MAX + 1);
913 } while (tmp_count & COUNT_CONTINUED);
914out:
915 spin_unlock(&p->lock);
916 return count;
917}
918
919/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800920 * We can write to an anon page without COW if there are no other references
921 * to it. And as a side-effect, free up its swap: because the old content
922 * on disk will never be read, and seeking back there to write new content
923 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800925int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700927 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
Sasha Levin309381fea2014-01-23 15:52:54 -0800929 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800930 if (unlikely(PageKsm(page)))
931 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700932 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800933 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700934 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800935 if (count == 1 && !PageWriteback(page)) {
936 delete_from_swap_cache(page);
937 SetPageDirty(page);
938 }
939 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800940 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941}
942
943/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800944 * If swap is getting full, or if there are no more mappings of this page,
945 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800947int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948{
Sasha Levin309381fea2014-01-23 15:52:54 -0800949 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
951 if (!PageSwapCache(page))
952 return 0;
953 if (PageWriteback(page))
954 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800955 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 return 0;
957
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700958 /*
959 * Once hibernation has begun to create its image of memory,
960 * there's a danger that one of the calls to try_to_free_swap()
961 * - most probably a call from __try_to_reclaim_swap() while
962 * hibernation is allocating its own swap pages for the image,
963 * but conceivably even a call from memory reclaim - will free
964 * the swap from a page which has already been recorded in the
965 * image as a clean swapcache page, and then reuse its swap for
966 * another page of the image. On waking from hibernation, the
967 * original page might be freed under memory pressure, then
968 * later read back in from swap, now with the wrong data.
969 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800970 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -0800971 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700972 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800973 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700974 return 0;
975
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800976 delete_from_swap_cache(page);
977 SetPageDirty(page);
978 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700979}
980
981/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 * Free the swap entry like above, but also try to
983 * free the page cache entry if it is the last user.
984 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800985int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800987 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 struct page *page = NULL;
989
Andi Kleena7420aa2009-09-16 11:50:05 +0200990 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800991 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 p = swap_info_get(entry);
994 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800995 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -0800996 page = find_get_page(swap_address_space(entry),
997 entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700998 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800999 page_cache_release(page);
1000 page = NULL;
1001 }
1002 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001003 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001006 /*
1007 * Not mapped elsewhere, or swap space full? Free it!
1008 * Also recheck PageSwapCache now page is locked (above).
1009 */
Nick Piggin93fac702006-03-31 02:29:56 -08001010 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001011 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 delete_from_swap_cache(page);
1013 SetPageDirty(page);
1014 }
1015 unlock_page(page);
1016 page_cache_release(page);
1017 }
Hugh Dickins2509ef22009-01-06 14:40:10 -08001018 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019}
1020
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001021#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001022/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001023 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001024 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001025 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1026 * from 0, in which the swap header is expected to be located.
1027 *
1028 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001029 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001030int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001031{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001032 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001033 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001034
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001035 if (device)
1036 bdev = bdget(device);
1037
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001038 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001039 for (type = 0; type < nr_swapfiles; type++) {
1040 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001041
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001042 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001043 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001044
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001045 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001046 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001047 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001048
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001049 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001050 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001051 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001052 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001053 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001054
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001055 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001056 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001057 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001058
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001059 spin_unlock(&swap_lock);
1060 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001061 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001062 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001063 }
1064 }
1065 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001066 if (bdev)
1067 bdput(bdev);
1068
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001069 return -ENODEV;
1070}
1071
1072/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001073 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1074 * corresponding to given index in swap_info (swap type).
1075 */
1076sector_t swapdev_block(int type, pgoff_t offset)
1077{
1078 struct block_device *bdev;
1079
1080 if ((unsigned int)type >= nr_swapfiles)
1081 return 0;
1082 if (!(swap_info[type]->flags & SWP_WRITEOK))
1083 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001084 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001085}
1086
1087/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001088 * Return either the total number of swap pages of given type, or the number
1089 * of free pages of that type (depending on @free)
1090 *
1091 * This is needed for software suspend
1092 */
1093unsigned int count_swap_pages(int type, int free)
1094{
1095 unsigned int n = 0;
1096
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001097 spin_lock(&swap_lock);
1098 if ((unsigned int)type < nr_swapfiles) {
1099 struct swap_info_struct *sis = swap_info[type];
1100
Shaohua Liec8acf22013-02-22 16:34:38 -08001101 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001102 if (sis->flags & SWP_WRITEOK) {
1103 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001104 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001105 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001106 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001107 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001108 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001109 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001110 return n;
1111}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001112#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001113
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001114static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1115{
1116#ifdef CONFIG_MEM_SOFT_DIRTY
1117 /*
1118 * When pte keeps soft dirty bit the pte generated
1119 * from swap entry does not has it, still it's same
1120 * pte from logical point of view.
1121 */
1122 pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1123 return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1124#else
1125 return pte_same(pte, swp_pte);
1126#endif
1127}
1128
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001130 * No need to decide whether this PTE shares the swap entry with others,
1131 * just let do_wp_page work it out if a write is requested later - to
1132 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001134static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 unsigned long addr, swp_entry_t entry, struct page *page)
1136{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001137 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001138 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001139 spinlock_t *ptl;
1140 pte_t *pte;
1141 int ret = 1;
1142
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001143 swapcache = page;
1144 page = ksm_might_need_to_copy(page, vma, addr);
1145 if (unlikely(!page))
1146 return -ENOMEM;
1147
Johannes Weiner00501b52014-08-08 14:19:20 -07001148 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001149 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001150 goto out_nolock;
1151 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001152
1153 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001154 if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
Johannes Weiner00501b52014-08-08 14:19:20 -07001155 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001156 ret = 0;
1157 goto out;
1158 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001159
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001160 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001161 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 get_page(page);
1163 set_pte_at(vma->vm_mm, addr, pte,
1164 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001165 if (page == swapcache) {
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001166 page_add_anon_rmap(page, vma, addr);
Johannes Weiner00501b52014-08-08 14:19:20 -07001167 mem_cgroup_commit_charge(page, memcg, true);
1168 } else { /* ksm created a completely new copy */
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001169 page_add_new_anon_rmap(page, vma, addr);
Johannes Weiner00501b52014-08-08 14:19:20 -07001170 mem_cgroup_commit_charge(page, memcg, false);
1171 lru_cache_add_active_or_unevictable(page, vma);
1172 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 swap_free(entry);
1174 /*
1175 * Move the page to the active list so it is not
1176 * immediately swapped out again after swapon.
1177 */
1178 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001179out:
1180 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001181out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001182 if (page != swapcache) {
1183 unlock_page(page);
1184 put_page(page);
1185 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001186 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
1189static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1190 unsigned long addr, unsigned long end,
1191 swp_entry_t entry, struct page *page)
1192{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001194 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001195 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Hugh Dickins044d66c2008-02-07 00:14:04 -08001197 /*
1198 * We don't actually need pte lock while scanning for swp_pte: since
1199 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1200 * page table while we're scanning; though it could get zapped, and on
1201 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1202 * of unmatched parts which look like swp_pte, so unuse_pte must
1203 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001204 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001205 */
1206 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 do {
1208 /*
1209 * swapoff spends a _lot_ of time in this loop!
1210 * Test inline before going to call unuse_pte.
1211 */
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001212 if (unlikely(maybe_same_pte(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001213 pte_unmap(pte);
1214 ret = unuse_pte(vma, pmd, addr, entry, page);
1215 if (ret)
1216 goto out;
1217 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 }
1219 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001220 pte_unmap(pte - 1);
1221out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001222 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223}
1224
1225static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1226 unsigned long addr, unsigned long end,
1227 swp_entry_t entry, struct page *page)
1228{
1229 pmd_t *pmd;
1230 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001231 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 pmd = pmd_offset(pud, addr);
1234 do {
1235 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001236 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001238 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1239 if (ret)
1240 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 } while (pmd++, addr = next, addr != end);
1242 return 0;
1243}
1244
1245static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1246 unsigned long addr, unsigned long end,
1247 swp_entry_t entry, struct page *page)
1248{
1249 pud_t *pud;
1250 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001251 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
1253 pud = pud_offset(pgd, addr);
1254 do {
1255 next = pud_addr_end(addr, end);
1256 if (pud_none_or_clear_bad(pud))
1257 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001258 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1259 if (ret)
1260 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 } while (pud++, addr = next, addr != end);
1262 return 0;
1263}
1264
1265static int unuse_vma(struct vm_area_struct *vma,
1266 swp_entry_t entry, struct page *page)
1267{
1268 pgd_t *pgd;
1269 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001270 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001272 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 addr = page_address_in_vma(page, vma);
1274 if (addr == -EFAULT)
1275 return 0;
1276 else
1277 end = addr + PAGE_SIZE;
1278 } else {
1279 addr = vma->vm_start;
1280 end = vma->vm_end;
1281 }
1282
1283 pgd = pgd_offset(vma->vm_mm, addr);
1284 do {
1285 next = pgd_addr_end(addr, end);
1286 if (pgd_none_or_clear_bad(pgd))
1287 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001288 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1289 if (ret)
1290 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 } while (pgd++, addr = next, addr != end);
1292 return 0;
1293}
1294
1295static int unuse_mm(struct mm_struct *mm,
1296 swp_entry_t entry, struct page *page)
1297{
1298 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001299 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
1301 if (!down_read_trylock(&mm->mmap_sem)) {
1302 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001303 * Activate page so shrink_inactive_list is unlikely to unmap
1304 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001306 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 unlock_page(page);
1308 down_read(&mm->mmap_sem);
1309 lock_page(page);
1310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001312 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 break;
1314 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001316 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
1319/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001320 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1321 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 * Recycle to start on reaching the end, returning 0 when empty.
1323 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001324static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001325 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001327 unsigned int max = si->max;
1328 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001329 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
1331 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001332 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 * for whether an entry is in use, not modifying it; false
1334 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001335 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 */
1337 for (;;) {
1338 if (++i >= max) {
1339 if (!prev) {
1340 i = 0;
1341 break;
1342 }
1343 /*
1344 * No entries in use at top of swap_map,
1345 * loop back to start and recheck there.
1346 */
1347 max = prev + 1;
1348 prev = 0;
1349 i = 1;
1350 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001351 if (frontswap) {
1352 if (frontswap_test(si, i))
1353 break;
1354 else
1355 continue;
1356 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001357 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001358 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 break;
1360 }
1361 return i;
1362}
1363
1364/*
1365 * We completely avoid races by reading each swap page in advance,
1366 * and then search for the process using it. All the necessary
1367 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001368 *
1369 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1370 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001372int try_to_unuse(unsigned int type, bool frontswap,
1373 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001375 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001377 volatile unsigned char *swap_map; /* swap_map is accessed without
1378 * locking. Mark it as volatile
1379 * to prevent compiler doing
1380 * something odd.
1381 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001382 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 struct page *page;
1384 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001385 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
1388 /*
1389 * When searching mms for an entry, a good strategy is to
1390 * start at the first mm we freed the previous entry from
1391 * (though actually we don't notice whether we or coincidence
1392 * freed the entry). Initialize this start_mm with a hold.
1393 *
1394 * A simpler strategy would be to start at the last mm we
1395 * freed the previous entry from; but that would take less
1396 * advantage of mmlist ordering, which clusters forked mms
1397 * together, child after parent. If we race with dup_mmap(), we
1398 * prefer to resolve parent before child, lest we miss entries
1399 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001400 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 */
1402 start_mm = &init_mm;
1403 atomic_inc(&init_mm.mm_users);
1404
1405 /*
1406 * Keep on scanning until all entries have gone. Usually,
1407 * one pass through swap_map is enough, but not necessarily:
1408 * there are races when an instance of an entry might be missed.
1409 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001410 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 if (signal_pending(current)) {
1412 retval = -EINTR;
1413 break;
1414 }
1415
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001416 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 * Get a page for the entry, using the existing swap
1418 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001419 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 */
1421 swap_map = &si->swap_map[i];
1422 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001423 page = read_swap_cache_async(entry,
1424 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 if (!page) {
1426 /*
1427 * Either swap_duplicate() failed because entry
1428 * has been freed independently, and will not be
1429 * reused since sys_swapoff() already disabled
1430 * allocation from here, or alloc_page() failed.
1431 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001432 swcount = *swap_map;
1433 /*
1434 * We don't hold lock here, so the swap entry could be
1435 * SWAP_MAP_BAD (when the cluster is discarding).
1436 * Instead of fail out, We can just skip the swap
1437 * entry because swapoff will wait for discarding
1438 * finish anyway.
1439 */
1440 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 continue;
1442 retval = -ENOMEM;
1443 break;
1444 }
1445
1446 /*
1447 * Don't hold on to start_mm if it looks like exiting.
1448 */
1449 if (atomic_read(&start_mm->mm_users) == 1) {
1450 mmput(start_mm);
1451 start_mm = &init_mm;
1452 atomic_inc(&init_mm.mm_users);
1453 }
1454
1455 /*
1456 * Wait for and lock page. When do_swap_page races with
1457 * try_to_unuse, do_swap_page can handle the fault much
1458 * faster than try_to_unuse can locate the entry. This
1459 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1460 * defer to do_swap_page in such a case - in some tests,
1461 * do_swap_page and try_to_unuse repeatedly compete.
1462 */
1463 wait_on_page_locked(page);
1464 wait_on_page_writeback(page);
1465 lock_page(page);
1466 wait_on_page_writeback(page);
1467
1468 /*
1469 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001472 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1473 retval = shmem_unuse(entry, page);
1474 /* page has already been unlocked and released */
1475 if (retval < 0)
1476 break;
1477 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001479 if (swap_count(swcount) && start_mm != &init_mm)
1480 retval = unuse_mm(start_mm, entry, page);
1481
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001482 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 int set_start_mm = (*swap_map >= swcount);
1484 struct list_head *p = &start_mm->mmlist;
1485 struct mm_struct *new_start_mm = start_mm;
1486 struct mm_struct *prev_mm = start_mm;
1487 struct mm_struct *mm;
1488
1489 atomic_inc(&new_start_mm->mm_users);
1490 atomic_inc(&prev_mm->mm_users);
1491 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001492 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 (p = p->next) != &start_mm->mmlist) {
1494 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001495 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 spin_unlock(&mmlist_lock);
1498 mmput(prev_mm);
1499 prev_mm = mm;
1500
1501 cond_resched();
1502
1503 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001504 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001506 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001508 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001510
Bo Liu32c5fc12009-11-02 16:50:33 +00001511 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 mmput(new_start_mm);
1513 atomic_inc(&mm->mm_users);
1514 new_start_mm = mm;
1515 set_start_mm = 0;
1516 }
1517 spin_lock(&mmlist_lock);
1518 }
1519 spin_unlock(&mmlist_lock);
1520 mmput(prev_mm);
1521 mmput(start_mm);
1522 start_mm = new_start_mm;
1523 }
1524 if (retval) {
1525 unlock_page(page);
1526 page_cache_release(page);
1527 break;
1528 }
1529
1530 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 * If a reference remains (rare), we would like to leave
1532 * the page in the swap cache; but try_to_unmap could
1533 * then re-duplicate the entry once we drop page lock,
1534 * so we might loop indefinitely; also, that page could
1535 * not be swapped out to other storage meanwhile. So:
1536 * delete from cache even if there's another reference,
1537 * after ensuring that the data has been saved to disk -
1538 * since if the reference remains (rarer), it will be
1539 * read from disk into another page. Splitting into two
1540 * pages would be incorrect if swap supported "shared
1541 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001542 *
1543 * Given how unuse_vma() targets one particular offset
1544 * in an anon_vma, once the anon_vma has been determined,
1545 * this splitting happens to be just what is needed to
1546 * handle where KSM pages have been swapped out: re-reading
1547 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001549 if (swap_count(*swap_map) &&
1550 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 struct writeback_control wbc = {
1552 .sync_mode = WB_SYNC_NONE,
1553 };
1554
1555 swap_writepage(page, &wbc);
1556 lock_page(page);
1557 wait_on_page_writeback(page);
1558 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001559
1560 /*
1561 * It is conceivable that a racing task removed this page from
1562 * swap cache just before we acquired the page lock at the top,
1563 * or while we dropped it in unuse_mm(). The page might even
1564 * be back in swap cache on another swap area: that we must not
1565 * delete, since it may not have been written out to swap yet.
1566 */
1567 if (PageSwapCache(page) &&
1568 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001569 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
1571 /*
1572 * So we could skip searching mms once swap count went
1573 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001574 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 */
1576 SetPageDirty(page);
1577 unlock_page(page);
1578 page_cache_release(page);
1579
1580 /*
1581 * Make sure that we aren't completely killing
1582 * interactive performance.
1583 */
1584 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001585 if (frontswap && pages_to_unuse > 0) {
1586 if (!--pages_to_unuse)
1587 break;
1588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 }
1590
1591 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 return retval;
1593}
1594
1595/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001596 * After a successful try_to_unuse, if no swap is now in use, we know
1597 * we can empty the mmlist. swap_lock must be held on entry and exit.
1598 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 * added to the mmlist just after page_duplicate - before would be racy.
1600 */
1601static void drain_mmlist(void)
1602{
1603 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001604 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001606 for (type = 0; type < nr_swapfiles; type++)
1607 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 return;
1609 spin_lock(&mmlist_lock);
1610 list_for_each_safe(p, next, &init_mm.mmlist)
1611 list_del_init(p);
1612 spin_unlock(&mmlist_lock);
1613}
1614
1615/*
1616 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001617 * corresponds to page offset for the specified swap entry.
1618 * Note that the type of this function is sector_t, but it returns page offset
1619 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001621static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001623 struct swap_info_struct *sis;
1624 struct swap_extent *start_se;
1625 struct swap_extent *se;
1626 pgoff_t offset;
1627
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001628 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001629 *bdev = sis->bdev;
1630
1631 offset = swp_offset(entry);
1632 start_se = sis->curr_swap_extent;
1633 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
1635 for ( ; ; ) {
1636 struct list_head *lh;
1637
1638 if (se->start_page <= offset &&
1639 offset < (se->start_page + se->nr_pages)) {
1640 return se->start_block + (offset - se->start_page);
1641 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001642 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 se = list_entry(lh, struct swap_extent, list);
1644 sis->curr_swap_extent = se;
1645 BUG_ON(se == start_se); /* It *must* be present */
1646 }
1647}
1648
1649/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001650 * Returns the page offset into bdev for the specified page's swap entry.
1651 */
1652sector_t map_swap_page(struct page *page, struct block_device **bdev)
1653{
1654 swp_entry_t entry;
1655 entry.val = page_private(page);
1656 return map_swap_entry(entry, bdev);
1657}
1658
1659/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 * Free all of a swapdev's extent information
1661 */
1662static void destroy_swap_extents(struct swap_info_struct *sis)
1663{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001664 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 struct swap_extent *se;
1666
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001667 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 struct swap_extent, list);
1669 list_del(&se->list);
1670 kfree(se);
1671 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001672
1673 if (sis->flags & SWP_FILE) {
1674 struct file *swap_file = sis->swap_file;
1675 struct address_space *mapping = swap_file->f_mapping;
1676
1677 sis->flags &= ~SWP_FILE;
1678 mapping->a_ops->swap_deactivate(swap_file);
1679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680}
1681
1682/*
1683 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001684 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001686 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 */
Mel Gormana509bc12012-07-31 16:44:57 -07001688int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1690 unsigned long nr_pages, sector_t start_block)
1691{
1692 struct swap_extent *se;
1693 struct swap_extent *new_se;
1694 struct list_head *lh;
1695
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001696 if (start_page == 0) {
1697 se = &sis->first_swap_extent;
1698 sis->curr_swap_extent = se;
1699 se->start_page = 0;
1700 se->nr_pages = nr_pages;
1701 se->start_block = start_block;
1702 return 1;
1703 } else {
1704 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001706 BUG_ON(se->start_page + se->nr_pages != start_page);
1707 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 /* Merge it */
1709 se->nr_pages += nr_pages;
1710 return 0;
1711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 }
1713
1714 /*
1715 * No merge. Insert a new extent, preserving ordering.
1716 */
1717 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1718 if (new_se == NULL)
1719 return -ENOMEM;
1720 new_se->start_page = start_page;
1721 new_se->nr_pages = nr_pages;
1722 new_se->start_block = start_block;
1723
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001724 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001725 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726}
1727
1728/*
1729 * A `swap extent' is a simple thing which maps a contiguous range of pages
1730 * onto a contiguous range of disk blocks. An ordered list of swap extents
1731 * is built at swapon time and is then used at swap_writepage/swap_readpage
1732 * time for locating where on disk a page belongs.
1733 *
1734 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1735 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1736 * swap files identically.
1737 *
1738 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1739 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1740 * swapfiles are handled *identically* after swapon time.
1741 *
1742 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1743 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1744 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1745 * requirements, they are simply tossed out - we will never use those blocks
1746 * for swapping.
1747 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001748 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1750 * which will scribble on the fs.
1751 *
1752 * The amount of disk space which a single swap extent represents varies.
1753 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1754 * extents in the list. To avoid much list walking, we cache the previous
1755 * search location in `curr_swap_extent', and start new searches from there.
1756 * This is extremely effective. The average number of iterations in
1757 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1758 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001759static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
Mel Gorman62c230b2012-07-31 16:44:55 -07001761 struct file *swap_file = sis->swap_file;
1762 struct address_space *mapping = swap_file->f_mapping;
1763 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 int ret;
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (S_ISBLK(inode->i_mode)) {
1767 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001768 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001769 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 }
1771
Mel Gorman62c230b2012-07-31 16:44:55 -07001772 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001773 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001774 if (!ret) {
1775 sis->flags |= SWP_FILE;
1776 ret = add_swap_extent(sis, 0, sis->max, 0);
1777 *span = sis->pages;
1778 }
Mel Gormana509bc12012-07-31 16:44:57 -07001779 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001780 }
1781
Mel Gormana509bc12012-07-31 16:44:57 -07001782 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783}
1784
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001785static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001786 unsigned char *swap_map,
1787 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001788{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001789 if (prio >= 0)
1790 p->prio = prio;
1791 else
1792 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001793 /*
1794 * the plist prio is negated because plist ordering is
1795 * low-to-high, while swap ordering is high-to-low
1796 */
1797 p->list.prio = -p->prio;
1798 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001799 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001800 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001801 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001802 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001803 total_swap_pages += p->pages;
1804
Dan Streetmanadfab832014-06-04 16:09:53 -07001805 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001806 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07001807 * both lists are plists, and thus priority ordered.
1808 * swap_active_head needs to be priority ordered for swapoff(),
1809 * which on removal of any swap_info_struct with an auto-assigned
1810 * (i.e. negative) priority increments the auto-assigned priority
1811 * of any lower-priority swap_info_structs.
1812 * swap_avail_head needs to be priority ordered for get_swap_page(),
1813 * which allocates swap pages from the highest available priority
1814 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07001815 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07001816 plist_add(&p->list, &swap_active_head);
1817 spin_lock(&swap_avail_lock);
1818 plist_add(&p->avail_list, &swap_avail_head);
1819 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001820}
1821
1822static void enable_swap_info(struct swap_info_struct *p, int prio,
1823 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001824 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001825 unsigned long *frontswap_map)
1826{
Minchan Kim4f898492013-04-30 15:26:54 -07001827 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001828 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001829 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001830 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001831 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001832 spin_unlock(&swap_lock);
1833}
1834
1835static void reinsert_swap_info(struct swap_info_struct *p)
1836{
1837 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001838 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001839 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001840 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001841 spin_unlock(&swap_lock);
1842}
1843
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001844SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001846 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001847 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001848 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001849 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 struct file *swap_file, *victim;
1851 struct address_space *mapping;
1852 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001853 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07001854 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001855 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001856
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 if (!capable(CAP_SYS_ADMIN))
1858 return -EPERM;
1859
Al Viro191c5422012-02-13 03:58:52 +00001860 BUG_ON(!current->mm);
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001864 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
Jeff Layton669abf42012-10-10 16:43:10 -04001866 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 err = PTR_ERR(victim);
1868 if (IS_ERR(victim))
1869 goto out;
1870
1871 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001872 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07001873 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8fd2009-01-06 14:39:48 -08001874 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001875 if (p->swap_file->f_mapping == mapping) {
1876 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07001878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 }
Dan Streetmanadfab832014-06-04 16:09:53 -07001881 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001883 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 goto out_dput;
1885 }
Al Viro191c5422012-02-13 03:58:52 +00001886 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 vm_unacct_memory(p->pages);
1888 else {
1889 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001890 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 goto out_dput;
1892 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001893 spin_lock(&swap_avail_lock);
1894 plist_del(&p->avail_list, &swap_avail_head);
1895 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001896 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001897 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001898 struct swap_info_struct *si = p;
1899
Dan Streetman18ab4d42014-06-04 16:09:59 -07001900 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001901 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001902 si->list.prio--;
1903 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07001904 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001905 least_priority++;
1906 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001907 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08001908 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 total_swap_pages -= p->pages;
1910 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001911 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001912 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001913
David Rientjese1e12d22012-12-11 16:02:56 -08001914 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07001915 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08001916 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 if (err) {
1919 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001920 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 goto out_dput;
1922 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001923
Shaohua Li815c2c52013-09-11 14:20:30 -07001924 flush_work(&p->discard_work);
1925
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001926 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001927 if (p->flags & SWP_CONTINUED)
1928 free_swap_count_continuations(p);
1929
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001930 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001931 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001932 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001934
1935 /* wait for anyone still in scan_swap_map */
1936 p->highest_bit = 0; /* cuts scans short */
1937 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001938 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001939 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001940 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001941 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001942 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001943 }
1944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001946 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 p->swap_file = NULL;
1948 p->max = 0;
1949 swap_map = p->swap_map;
1950 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001951 cluster_info = p->cluster_info;
1952 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07001953 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08001954 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001955 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001956 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08001957 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001958 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001959 free_percpu(p->percpu_cluster);
1960 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001962 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001963 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001964 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07001965 swap_cgroup_swapoff(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 inode = mapping->host;
1968 if (S_ISBLK(inode->i_mode)) {
1969 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001970 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001971 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001973 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001975 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 }
1977 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08001978
1979 /*
1980 * Clear the SWP_USED flag after all resources are freed so that swapon
1981 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1982 * not hold p->lock after we cleared its SWP_WRITEOK.
1983 */
1984 spin_lock(&swap_lock);
1985 p->flags = 0;
1986 spin_unlock(&swap_lock);
1987
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001989 atomic_inc(&proc_poll_event);
1990 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
1992out_dput:
1993 filp_close(victim, NULL);
1994out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001995 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return err;
1997}
1998
1999#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07002000static unsigned swaps_poll(struct file *file, poll_table *wait)
2001{
Kay Sieversf1514632011-07-12 20:48:39 +02002002 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002003
2004 poll_wait(file, &proc_poll_wait, wait);
2005
Kay Sieversf1514632011-07-12 20:48:39 +02002006 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2007 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002008 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2009 }
2010
2011 return POLLIN | POLLRDNORM;
2012}
2013
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014/* iterator */
2015static void *swap_start(struct seq_file *swap, loff_t *pos)
2016{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002017 struct swap_info_struct *si;
2018 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 loff_t l = *pos;
2020
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002021 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002023 if (!l)
2024 return SEQ_START_TOKEN;
2025
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002026 for (type = 0; type < nr_swapfiles; type++) {
2027 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2028 si = swap_info[type];
2029 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002031 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002032 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 }
2034
2035 return NULL;
2036}
2037
2038static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2039{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002040 struct swap_info_struct *si = v;
2041 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002043 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002044 type = 0;
2045 else
2046 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002047
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002048 for (; type < nr_swapfiles; type++) {
2049 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2050 si = swap_info[type];
2051 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 continue;
2053 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002054 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 }
2056
2057 return NULL;
2058}
2059
2060static void swap_stop(struct seq_file *swap, void *v)
2061{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002062 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063}
2064
2065static int swap_show(struct seq_file *swap, void *v)
2066{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002067 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 struct file *file;
2069 int len;
2070
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002071 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002072 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2073 return 0;
2074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002076 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002077 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002078 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002079 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002080 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002082 si->pages << (PAGE_SHIFT - 10),
2083 si->inuse_pages << (PAGE_SHIFT - 10),
2084 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 return 0;
2086}
2087
Helge Deller15ad7cd2006-12-06 20:40:36 -08002088static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 .start = swap_start,
2090 .next = swap_next,
2091 .stop = swap_stop,
2092 .show = swap_show
2093};
2094
2095static int swaps_open(struct inode *inode, struct file *file)
2096{
Kay Sieversf1514632011-07-12 20:48:39 +02002097 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002098 int ret;
2099
Kay Sievers66d7dd52010-10-26 14:22:06 -07002100 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002101 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002102 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002103
Kay Sieversf1514632011-07-12 20:48:39 +02002104 seq = file->private_data;
2105 seq->poll_event = atomic_read(&proc_poll_event);
2106 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107}
2108
Helge Deller15ad7cd2006-12-06 20:40:36 -08002109static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 .open = swaps_open,
2111 .read = seq_read,
2112 .llseek = seq_lseek,
2113 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002114 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115};
2116
2117static int __init procswaps_init(void)
2118{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002119 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 return 0;
2121}
2122__initcall(procswaps_init);
2123#endif /* CONFIG_PROC_FS */
2124
Jan Beulich17963162008-12-16 11:35:24 +00002125#ifdef MAX_SWAPFILES_CHECK
2126static int __init max_swapfiles_check(void)
2127{
2128 MAX_SWAPFILES_CHECK();
2129 return 0;
2130}
2131late_initcall(max_swapfiles_check);
2132#endif
2133
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002134static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002136 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002138
2139 p = kzalloc(sizeof(*p), GFP_KERNEL);
2140 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002141 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002142
Hugh Dickins5d337b92005-09-03 15:54:41 -07002143 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002144 for (type = 0; type < nr_swapfiles; type++) {
2145 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002147 }
Christoph Lameter06972122006-06-23 02:03:35 -07002148 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002149 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002150 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002151 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002153 if (type >= nr_swapfiles) {
2154 p->type = type;
2155 swap_info[type] = p;
2156 /*
2157 * Write swap_info[type] before nr_swapfiles, in case a
2158 * racing procfs swap_start() or swap_next() is reading them.
2159 * (We never shrink nr_swapfiles, we never free this entry.)
2160 */
2161 smp_wmb();
2162 nr_swapfiles++;
2163 } else {
2164 kfree(p);
2165 p = swap_info[type];
2166 /*
2167 * Do not memset this entry: a racing procfs swap_next()
2168 * would be relying on p->type to remain valid.
2169 */
2170 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002171 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002172 plist_node_init(&p->list, 0);
2173 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002175 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002176 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002177
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002178 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002179}
2180
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002181static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2182{
2183 int error;
2184
2185 if (S_ISBLK(inode->i_mode)) {
2186 p->bdev = bdgrab(I_BDEV(inode));
2187 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002188 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002189 if (error < 0) {
2190 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002191 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002192 }
2193 p->old_block_size = block_size(p->bdev);
2194 error = set_blocksize(p->bdev, PAGE_SIZE);
2195 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002196 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002197 p->flags |= SWP_BLKDEV;
2198 } else if (S_ISREG(inode->i_mode)) {
2199 p->bdev = inode->i_sb->s_bdev;
2200 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002201 if (IS_SWAPFILE(inode))
2202 return -EBUSY;
2203 } else
2204 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002205
2206 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002207}
2208
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002209static unsigned long read_swap_header(struct swap_info_struct *p,
2210 union swap_header *swap_header,
2211 struct inode *inode)
2212{
2213 int i;
2214 unsigned long maxpages;
2215 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002216 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002217
2218 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002219 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002220 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002221 }
2222
2223 /* swap partition endianess hack... */
2224 if (swab32(swap_header->info.version) == 1) {
2225 swab32s(&swap_header->info.version);
2226 swab32s(&swap_header->info.last_page);
2227 swab32s(&swap_header->info.nr_badpages);
2228 for (i = 0; i < swap_header->info.nr_badpages; i++)
2229 swab32s(&swap_header->info.badpages[i]);
2230 }
2231 /* Check the swap header's sub-version */
2232 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002233 pr_warn("Unable to handle swap header version %d\n",
2234 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002235 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002236 }
2237
2238 p->lowest_bit = 1;
2239 p->cluster_next = 1;
2240 p->cluster_nr = 0;
2241
2242 /*
2243 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002244 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002245 * of bits for the swap offset in the swp_entry_t type, and
2246 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002247 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002248 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002249 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002250 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002251 * offset is extracted. This will mask all the bits from
2252 * the initial ~0UL mask that can't be encoded in either
2253 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002254 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002255 */
2256 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002257 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002258 last_page = swap_header->info.last_page;
2259 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002260 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002261 maxpages << (PAGE_SHIFT - 10),
2262 last_page << (PAGE_SHIFT - 10));
2263 }
2264 if (maxpages > last_page) {
2265 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002266 /* p->max is an unsigned int: don't overflow it */
2267 if ((unsigned int)maxpages == 0)
2268 maxpages = UINT_MAX;
2269 }
2270 p->highest_bit = maxpages - 1;
2271
2272 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002273 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002274 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2275 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002276 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002277 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002278 }
2279 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002280 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002281 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002282 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002283
2284 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002285}
2286
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002287static int setup_swap_map_and_extents(struct swap_info_struct *p,
2288 union swap_header *swap_header,
2289 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002290 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002291 unsigned long maxpages,
2292 sector_t *span)
2293{
2294 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002295 unsigned int nr_good_pages;
2296 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002297 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2298 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002299
2300 nr_good_pages = maxpages - 1; /* omit header page */
2301
Shaohua Li2a8f9442013-09-11 14:20:28 -07002302 cluster_set_null(&p->free_cluster_head);
2303 cluster_set_null(&p->free_cluster_tail);
Shaohua Li815c2c52013-09-11 14:20:30 -07002304 cluster_set_null(&p->discard_cluster_head);
2305 cluster_set_null(&p->discard_cluster_tail);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002306
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002307 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2308 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002309 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2310 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002311 if (page_nr < maxpages) {
2312 swap_map[page_nr] = SWAP_MAP_BAD;
2313 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002314 /*
2315 * Haven't marked the cluster free yet, no list
2316 * operation involved
2317 */
2318 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002319 }
2320 }
2321
Shaohua Li2a8f9442013-09-11 14:20:28 -07002322 /* Haven't marked the cluster free yet, no list operation involved */
2323 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2324 inc_cluster_info_page(p, cluster_info, i);
2325
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002326 if (nr_good_pages) {
2327 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002328 /*
2329 * Not mark the cluster free yet, no list
2330 * operation involved
2331 */
2332 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002333 p->max = maxpages;
2334 p->pages = nr_good_pages;
2335 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002336 if (nr_extents < 0)
2337 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002338 nr_good_pages = p->pages;
2339 }
2340 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002341 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002342 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002343 }
2344
Shaohua Li2a8f9442013-09-11 14:20:28 -07002345 if (!cluster_info)
2346 return nr_extents;
2347
2348 for (i = 0; i < nr_clusters; i++) {
2349 if (!cluster_count(&cluster_info[idx])) {
2350 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2351 if (cluster_is_null(&p->free_cluster_head)) {
2352 cluster_set_next_flag(&p->free_cluster_head,
2353 idx, 0);
2354 cluster_set_next_flag(&p->free_cluster_tail,
2355 idx, 0);
2356 } else {
2357 unsigned int tail;
2358
2359 tail = cluster_next(&p->free_cluster_tail);
2360 cluster_set_next(&cluster_info[tail], idx);
2361 cluster_set_next_flag(&p->free_cluster_tail,
2362 idx, 0);
2363 }
2364 }
2365 idx++;
2366 if (idx == nr_clusters)
2367 idx = 0;
2368 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002369 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002370}
2371
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002372/*
2373 * Helper to sys_swapon determining if a given swap
2374 * backing device queue supports DISCARD operations.
2375 */
2376static bool swap_discardable(struct swap_info_struct *si)
2377{
2378 struct request_queue *q = bdev_get_queue(si->bdev);
2379
2380 if (!q || !blk_queue_discard(q))
2381 return false;
2382
2383 return true;
2384}
2385
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002386SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2387{
2388 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002389 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002390 struct file *swap_file = NULL;
2391 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002392 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002393 int error;
2394 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002395 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002396 sector_t span;
2397 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002398 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002399 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002400 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002401 struct page *page = NULL;
2402 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002403
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002404 if (swap_flags & ~SWAP_FLAGS_VALID)
2405 return -EINVAL;
2406
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002407 if (!capable(CAP_SYS_ADMIN))
2408 return -EPERM;
2409
2410 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002411 if (IS_ERR(p))
2412 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002413
Shaohua Li815c2c52013-09-11 14:20:30 -07002414 INIT_WORK(&p->discard_work, swap_discard_work);
2415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002418 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002420 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 }
Jeff Layton669abf42012-10-10 16:43:10 -04002422 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002424 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002426 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 }
2428
2429 p->swap_file = swap_file;
2430 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002431 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002432
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002433 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002434 error = claim_swapfile(p, inode);
2435 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 /*
2439 * Read the swap header.
2440 */
2441 if (!mapping->a_ops->readpage) {
2442 error = -EINVAL;
2443 goto bad_swap;
2444 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002445 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 if (IS_ERR(page)) {
2447 error = PTR_ERR(page);
2448 goto bad_swap;
2449 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002450 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002452 maxpages = read_swap_header(p, swap_header, inode);
2453 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 error = -EINVAL;
2455 goto bad_swap;
2456 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002457
Hugh Dickins81e33972009-01-06 14:39:49 -08002458 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002459 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002460 if (!swap_map) {
2461 error = -ENOMEM;
2462 goto bad_swap;
2463 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002464 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002465 int cpu;
2466
Shaohua Li2a8f9442013-09-11 14:20:28 -07002467 p->flags |= SWP_SOLIDSTATE;
2468 /*
2469 * select a random position to start with to help wear leveling
2470 * SSD
2471 */
2472 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2473
2474 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2475 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2476 if (!cluster_info) {
2477 error = -ENOMEM;
2478 goto bad_swap;
2479 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002480 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2481 if (!p->percpu_cluster) {
2482 error = -ENOMEM;
2483 goto bad_swap;
2484 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002485 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002486 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002487 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002488 cluster_set_null(&cluster->index);
2489 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002490 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002491
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002492 error = swap_cgroup_swapon(p->type, maxpages);
2493 if (error)
2494 goto bad_swap;
2495
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002496 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002497 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002498 if (unlikely(nr_extents < 0)) {
2499 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 goto bad_swap;
2501 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002502 /* frontswap enabled? set up bit-per-page map for frontswap */
2503 if (frontswap_enabled)
Akinobu Mita7b579762013-06-12 14:05:08 -07002504 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Shaohua Li2a8f9442013-09-11 14:20:28 -07002506 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2507 /*
2508 * When discard is enabled for swap with no particular
2509 * policy flagged, we set all swap discard flags here in
2510 * order to sustain backward compatibility with older
2511 * swapon(8) releases.
2512 */
2513 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2514 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002515
Shaohua Li2a8f9442013-09-11 14:20:28 -07002516 /*
2517 * By flagging sys_swapon, a sysadmin can tell us to
2518 * either do single-time area discards only, or to just
2519 * perform discards for released swap page-clusters.
2520 * Now it's time to adjust the p->flags accordingly.
2521 */
2522 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2523 p->flags &= ~SWP_PAGE_DISCARD;
2524 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2525 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002526
Shaohua Li2a8f9442013-09-11 14:20:28 -07002527 /* issue a swapon-time discard if it's still required */
2528 if (p->flags & SWP_AREA_DISCARD) {
2529 int err = discard_swap(p);
2530 if (unlikely(err))
2531 pr_err("swapon: discard_swap(%p): %d\n",
2532 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002533 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002534 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002535
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002536 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002537 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002538 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002539 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002540 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002541 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002542
Andrew Morton465c47f2013-09-11 14:20:17 -07002543 pr_info("Adding %uk swap on %s. "
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002544 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002545 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002546 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2547 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002548 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002549 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2550 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002551 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002552
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002553 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002554 atomic_inc(&proc_poll_event);
2555 wake_up_interruptible(&proc_poll_wait);
2556
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002557 if (S_ISREG(inode->i_mode))
2558 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 error = 0;
2560 goto out;
2561bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002562 free_percpu(p->percpu_cluster);
2563 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002564 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002565 set_blocksize(p->bdev, p->old_block_size);
2566 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002568 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002569 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002570 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002573 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002575 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002576 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002577 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002578 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002579 inode = NULL;
2580 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583out:
2584 if (page && !IS_ERR(page)) {
2585 kunmap(page);
2586 page_cache_release(page);
2587 }
2588 if (name)
2589 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002590 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002591 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 return error;
2593}
2594
2595void si_swapinfo(struct sysinfo *val)
2596{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002597 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 unsigned long nr_to_be_unused = 0;
2599
Hugh Dickins5d337b92005-09-03 15:54:41 -07002600 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002601 for (type = 0; type < nr_swapfiles; type++) {
2602 struct swap_info_struct *si = swap_info[type];
2603
2604 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2605 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002607 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002609 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610}
2611
2612/*
2613 * Verify that a swap entry is valid and increment its swap map count.
2614 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002615 * Returns error code in following case.
2616 * - success -> 0
2617 * - swp_entry is invalid -> EINVAL
2618 * - swp_entry is migration entry -> EINVAL
2619 * - swap-cache reference is requested but there is already one. -> EEXIST
2620 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002621 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002623static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002625 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002627 unsigned char count;
2628 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002629 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
Andi Kleena7420aa2009-09-16 11:50:05 +02002631 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002632 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002633
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 type = swp_type(entry);
2635 if (type >= nr_swapfiles)
2636 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002637 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 offset = swp_offset(entry);
2639
Shaohua Liec8acf22013-02-22 16:34:38 -08002640 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002641 if (unlikely(offset >= p->max))
2642 goto unlock_out;
2643
Hugh Dickins253d5532009-12-14 17:58:44 -08002644 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002645
2646 /*
2647 * swapin_readahead() doesn't check if a swap entry is valid, so the
2648 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2649 */
2650 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2651 err = -ENOENT;
2652 goto unlock_out;
2653 }
2654
Hugh Dickins253d5532009-12-14 17:58:44 -08002655 has_cache = count & SWAP_HAS_CACHE;
2656 count &= ~SWAP_HAS_CACHE;
2657 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002658
Hugh Dickins253d5532009-12-14 17:58:44 -08002659 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002660
2661 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002662 if (!has_cache && count)
2663 has_cache = SWAP_HAS_CACHE;
2664 else if (has_cache) /* someone else added cache */
2665 err = -EEXIST;
2666 else /* no users remaining */
2667 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002668
2669 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002670
Hugh Dickins570a335b2009-12-14 17:58:46 -08002671 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2672 count += usage;
2673 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002674 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002675 else if (swap_count_continued(p, offset, count))
2676 count = COUNT_CONTINUED;
2677 else
2678 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002679 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002680 err = -ENOENT; /* unused swap entry */
2681
2682 p->swap_map[offset] = count | has_cache;
2683
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002684unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002685 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002687 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
2689bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002690 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 goto out;
2692}
Hugh Dickins253d5532009-12-14 17:58:44 -08002693
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002694/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002695 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2696 * (in which case its reference count is never incremented).
2697 */
2698void swap_shmem_alloc(swp_entry_t entry)
2699{
2700 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2701}
2702
2703/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002704 * Increase reference count of swap entry by 1.
2705 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2706 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2707 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2708 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002709 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002710int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002711{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002712 int err = 0;
2713
2714 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2715 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2716 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002717}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002719/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002720 * @entry: swap entry for which we allocate swap cache.
2721 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002722 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002723 * This can return error codes. Returns 0 at success.
2724 * -EBUSY means there is a swap cache.
2725 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002726 */
2727int swapcache_prepare(swp_entry_t entry)
2728{
Hugh Dickins253d5532009-12-14 17:58:44 -08002729 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002730}
2731
Mel Gormanf981c592012-07-31 16:44:47 -07002732struct swap_info_struct *page_swap_info(struct page *page)
2733{
2734 swp_entry_t swap = { .val = page_private(page) };
2735 BUG_ON(!PageSwapCache(page));
2736 return swap_info[swp_type(swap)];
2737}
2738
2739/*
2740 * out-of-line __page_file_ methods to avoid include hell.
2741 */
2742struct address_space *__page_file_mapping(struct page *page)
2743{
Sasha Levin309381fea2014-01-23 15:52:54 -08002744 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002745 return page_swap_info(page)->swap_file->f_mapping;
2746}
2747EXPORT_SYMBOL_GPL(__page_file_mapping);
2748
2749pgoff_t __page_file_index(struct page *page)
2750{
2751 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002752 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002753 return swp_offset(swap);
2754}
2755EXPORT_SYMBOL_GPL(__page_file_index);
2756
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002758 * add_swap_count_continuation - called when a swap count is duplicated
2759 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2760 * page of the original vmalloc'ed swap_map, to hold the continuation count
2761 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2762 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2763 *
2764 * These continuation pages are seldom referenced: the common paths all work
2765 * on the original swap_map, only referring to a continuation page when the
2766 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2767 *
2768 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2769 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2770 * can be called after dropping locks.
2771 */
2772int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2773{
2774 struct swap_info_struct *si;
2775 struct page *head;
2776 struct page *page;
2777 struct page *list_page;
2778 pgoff_t offset;
2779 unsigned char count;
2780
2781 /*
2782 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2783 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2784 */
2785 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2786
2787 si = swap_info_get(entry);
2788 if (!si) {
2789 /*
2790 * An acceptable race has occurred since the failing
2791 * __swap_duplicate(): the swap entry has been freed,
2792 * perhaps even the whole swap_map cleared for swapoff.
2793 */
2794 goto outer;
2795 }
2796
2797 offset = swp_offset(entry);
2798 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2799
2800 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2801 /*
2802 * The higher the swap count, the more likely it is that tasks
2803 * will race to add swap count continuation: we need to avoid
2804 * over-provisioning.
2805 */
2806 goto out;
2807 }
2808
2809 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002810 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002811 return -ENOMEM;
2812 }
2813
2814 /*
2815 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002816 * no architecture is using highmem pages for kernel page tables: so it
2817 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002818 */
2819 head = vmalloc_to_page(si->swap_map + offset);
2820 offset &= ~PAGE_MASK;
2821
2822 /*
2823 * Page allocation does not initialize the page's lru field,
2824 * but it does always reset its private field.
2825 */
2826 if (!page_private(head)) {
2827 BUG_ON(count & COUNT_CONTINUED);
2828 INIT_LIST_HEAD(&head->lru);
2829 set_page_private(head, SWP_CONTINUED);
2830 si->flags |= SWP_CONTINUED;
2831 }
2832
2833 list_for_each_entry(list_page, &head->lru, lru) {
2834 unsigned char *map;
2835
2836 /*
2837 * If the previous map said no continuation, but we've found
2838 * a continuation page, free our allocation and use this one.
2839 */
2840 if (!(count & COUNT_CONTINUED))
2841 goto out;
2842
Cong Wang9b04c5f2011-11-25 23:14:39 +08002843 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002844 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002845 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002846
2847 /*
2848 * If this continuation count now has some space in it,
2849 * free our allocation and use this one.
2850 */
2851 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2852 goto out;
2853 }
2854
2855 list_add_tail(&page->lru, &head->lru);
2856 page = NULL; /* now it's attached, don't free it */
2857out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002858 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002859outer:
2860 if (page)
2861 __free_page(page);
2862 return 0;
2863}
2864
2865/*
2866 * swap_count_continued - when the original swap_map count is incremented
2867 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2868 * into, carry if so, or else fail until a new continuation page is allocated;
2869 * when the original swap_map count is decremented from 0 with continuation,
2870 * borrow from the continuation and report whether it still holds more.
2871 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2872 */
2873static bool swap_count_continued(struct swap_info_struct *si,
2874 pgoff_t offset, unsigned char count)
2875{
2876 struct page *head;
2877 struct page *page;
2878 unsigned char *map;
2879
2880 head = vmalloc_to_page(si->swap_map + offset);
2881 if (page_private(head) != SWP_CONTINUED) {
2882 BUG_ON(count & COUNT_CONTINUED);
2883 return false; /* need to add count continuation */
2884 }
2885
2886 offset &= ~PAGE_MASK;
2887 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002888 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002889
2890 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2891 goto init_map; /* jump over SWAP_CONT_MAX checks */
2892
2893 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2894 /*
2895 * Think of how you add 1 to 999
2896 */
2897 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002898 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002899 page = list_entry(page->lru.next, struct page, lru);
2900 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002901 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002902 }
2903 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002904 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002905 page = list_entry(page->lru.next, struct page, lru);
2906 if (page == head)
2907 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002908 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002909init_map: *map = 0; /* we didn't zero the page */
2910 }
2911 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002912 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002913 page = list_entry(page->lru.prev, struct page, lru);
2914 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002915 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002916 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002917 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002918 page = list_entry(page->lru.prev, struct page, lru);
2919 }
2920 return true; /* incremented */
2921
2922 } else { /* decrementing */
2923 /*
2924 * Think of how you subtract 1 from 1000
2925 */
2926 BUG_ON(count != COUNT_CONTINUED);
2927 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002928 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002929 page = list_entry(page->lru.next, struct page, lru);
2930 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002931 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002932 }
2933 BUG_ON(*map == 0);
2934 *map -= 1;
2935 if (*map == 0)
2936 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002937 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002938 page = list_entry(page->lru.prev, struct page, lru);
2939 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002940 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002941 *map = SWAP_CONT_MAX | count;
2942 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002943 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002944 page = list_entry(page->lru.prev, struct page, lru);
2945 }
2946 return count == COUNT_CONTINUED;
2947 }
2948}
2949
2950/*
2951 * free_swap_count_continuations - swapoff free all the continuation pages
2952 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2953 */
2954static void free_swap_count_continuations(struct swap_info_struct *si)
2955{
2956 pgoff_t offset;
2957
2958 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2959 struct page *head;
2960 head = vmalloc_to_page(si->swap_map + offset);
2961 if (page_private(head)) {
2962 struct list_head *this, *next;
2963 list_for_each_safe(this, next, &head->lru) {
2964 struct page *page;
2965 page = list_entry(this, struct page, lru);
2966 list_del(this);
2967 __free_page(page);
2968 }
2969 }
2970 }
2971}