blob: d2c37365e2d6e429cf0f7cdbb62308f5e76aea93 [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) {
Hugh Dickins7992fde2009-01-06 14:39:53 -0800168 if (se->start_page <= start_page &&
169 start_page < se->start_page + se->nr_pages) {
170 pgoff_t offset = start_page - se->start_page;
171 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800172 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800173
174 if (nr_blocks > nr_pages)
175 nr_blocks = nr_pages;
176 start_page += nr_blocks;
177 nr_pages -= nr_blocks;
178
179 if (!found_extent++)
180 si->curr_swap_extent = se;
181
182 start_block <<= PAGE_SHIFT - 9;
183 nr_blocks <<= PAGE_SHIFT - 9;
184 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200185 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800186 break;
187 }
188
Geliang Tanga8ae4992016-01-14 15:20:45 -0800189 se = list_next_entry(se, list);
Hugh Dickins7992fde2009-01-06 14:39:53 -0800190 }
191}
192
Hugh Dickins048c27f2005-09-03 15:54:40 -0700193#define SWAPFILE_CLUSTER 256
194#define LATENCY_LIMIT 256
195
Shaohua Li2a8f9442013-09-11 14:20:28 -0700196static inline void cluster_set_flag(struct swap_cluster_info *info,
197 unsigned int flag)
198{
199 info->flags = flag;
200}
201
202static inline unsigned int cluster_count(struct swap_cluster_info *info)
203{
204 return info->data;
205}
206
207static inline void cluster_set_count(struct swap_cluster_info *info,
208 unsigned int c)
209{
210 info->data = c;
211}
212
213static inline void cluster_set_count_flag(struct swap_cluster_info *info,
214 unsigned int c, unsigned int f)
215{
216 info->flags = f;
217 info->data = c;
218}
219
220static inline unsigned int cluster_next(struct swap_cluster_info *info)
221{
222 return info->data;
223}
224
225static inline void cluster_set_next(struct swap_cluster_info *info,
226 unsigned int n)
227{
228 info->data = n;
229}
230
231static inline void cluster_set_next_flag(struct swap_cluster_info *info,
232 unsigned int n, unsigned int f)
233{
234 info->flags = f;
235 info->data = n;
236}
237
238static inline bool cluster_is_free(struct swap_cluster_info *info)
239{
240 return info->flags & CLUSTER_FLAG_FREE;
241}
242
243static inline bool cluster_is_null(struct swap_cluster_info *info)
244{
245 return info->flags & CLUSTER_FLAG_NEXT_NULL;
246}
247
248static inline void cluster_set_null(struct swap_cluster_info *info)
249{
250 info->flags = CLUSTER_FLAG_NEXT_NULL;
251 info->data = 0;
252}
253
Shaohua Li815c2c52013-09-11 14:20:30 -0700254/* Add a cluster to discard list and schedule it to do discard */
255static void swap_cluster_schedule_discard(struct swap_info_struct *si,
256 unsigned int idx)
257{
258 /*
259 * If scan_swap_map() can't find a free cluster, it will check
260 * si->swap_map directly. To make sure the discarding cluster isn't
261 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
262 * will be cleared after discard
263 */
264 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
265 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
266
267 if (cluster_is_null(&si->discard_cluster_head)) {
268 cluster_set_next_flag(&si->discard_cluster_head,
269 idx, 0);
270 cluster_set_next_flag(&si->discard_cluster_tail,
271 idx, 0);
272 } else {
273 unsigned int tail = cluster_next(&si->discard_cluster_tail);
274 cluster_set_next(&si->cluster_info[tail], idx);
275 cluster_set_next_flag(&si->discard_cluster_tail,
276 idx, 0);
277 }
278
279 schedule_work(&si->discard_work);
280}
281
282/*
283 * Doing discard actually. After a cluster discard is finished, the cluster
284 * will be added to free cluster list. caller should hold si->lock.
285*/
286static void swap_do_scheduled_discard(struct swap_info_struct *si)
287{
288 struct swap_cluster_info *info;
289 unsigned int idx;
290
291 info = si->cluster_info;
292
293 while (!cluster_is_null(&si->discard_cluster_head)) {
294 idx = cluster_next(&si->discard_cluster_head);
295
296 cluster_set_next_flag(&si->discard_cluster_head,
297 cluster_next(&info[idx]), 0);
298 if (cluster_next(&si->discard_cluster_tail) == idx) {
299 cluster_set_null(&si->discard_cluster_head);
300 cluster_set_null(&si->discard_cluster_tail);
301 }
302 spin_unlock(&si->lock);
303
304 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
305 SWAPFILE_CLUSTER);
306
307 spin_lock(&si->lock);
308 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
309 if (cluster_is_null(&si->free_cluster_head)) {
310 cluster_set_next_flag(&si->free_cluster_head,
311 idx, 0);
312 cluster_set_next_flag(&si->free_cluster_tail,
313 idx, 0);
314 } else {
315 unsigned int tail;
316
317 tail = cluster_next(&si->free_cluster_tail);
318 cluster_set_next(&info[tail], idx);
319 cluster_set_next_flag(&si->free_cluster_tail,
320 idx, 0);
321 }
322 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
323 0, SWAPFILE_CLUSTER);
324 }
325}
326
327static void swap_discard_work(struct work_struct *work)
328{
329 struct swap_info_struct *si;
330
331 si = container_of(work, struct swap_info_struct, discard_work);
332
333 spin_lock(&si->lock);
334 swap_do_scheduled_discard(si);
335 spin_unlock(&si->lock);
336}
337
Shaohua Li2a8f9442013-09-11 14:20:28 -0700338/*
339 * The cluster corresponding to page_nr will be used. The cluster will be
340 * removed from free cluster list and its usage counter will be increased.
341 */
342static void inc_cluster_info_page(struct swap_info_struct *p,
343 struct swap_cluster_info *cluster_info, unsigned long page_nr)
344{
345 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
346
347 if (!cluster_info)
348 return;
349 if (cluster_is_free(&cluster_info[idx])) {
350 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
351 cluster_set_next_flag(&p->free_cluster_head,
352 cluster_next(&cluster_info[idx]), 0);
353 if (cluster_next(&p->free_cluster_tail) == idx) {
354 cluster_set_null(&p->free_cluster_tail);
355 cluster_set_null(&p->free_cluster_head);
356 }
357 cluster_set_count_flag(&cluster_info[idx], 0, 0);
358 }
359
360 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
361 cluster_set_count(&cluster_info[idx],
362 cluster_count(&cluster_info[idx]) + 1);
363}
364
365/*
366 * The cluster corresponding to page_nr decreases one usage. If the usage
367 * counter becomes 0, which means no page in the cluster is in using, we can
368 * optionally discard the cluster and add it to free cluster list.
369 */
370static void dec_cluster_info_page(struct swap_info_struct *p,
371 struct swap_cluster_info *cluster_info, unsigned long page_nr)
372{
373 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
374
375 if (!cluster_info)
376 return;
377
378 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
379 cluster_set_count(&cluster_info[idx],
380 cluster_count(&cluster_info[idx]) - 1);
381
382 if (cluster_count(&cluster_info[idx]) == 0) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700383 /*
384 * If the swap is discardable, prepare discard the cluster
385 * instead of free it immediately. The cluster will be freed
386 * after discard.
387 */
Shaohua Liedfe23d2013-09-11 14:20:31 -0700388 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
389 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
Shaohua Li815c2c52013-09-11 14:20:30 -0700390 swap_cluster_schedule_discard(p, idx);
391 return;
392 }
393
Shaohua Li2a8f9442013-09-11 14:20:28 -0700394 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
395 if (cluster_is_null(&p->free_cluster_head)) {
396 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
397 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
398 } else {
399 unsigned int tail = cluster_next(&p->free_cluster_tail);
400 cluster_set_next(&cluster_info[tail], idx);
401 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
402 }
403 }
404}
405
406/*
407 * It's possible scan_swap_map() uses a free cluster in the middle of free
408 * cluster list. Avoiding such abuse to avoid list corruption.
409 */
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700410static bool
411scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
Shaohua Li2a8f9442013-09-11 14:20:28 -0700412 unsigned long offset)
413{
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700414 struct percpu_cluster *percpu_cluster;
415 bool conflict;
416
Shaohua Li2a8f9442013-09-11 14:20:28 -0700417 offset /= SWAPFILE_CLUSTER;
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700418 conflict = !cluster_is_null(&si->free_cluster_head) &&
Shaohua Li2a8f9442013-09-11 14:20:28 -0700419 offset != cluster_next(&si->free_cluster_head) &&
420 cluster_is_free(&si->cluster_info[offset]);
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700421
422 if (!conflict)
423 return false;
424
425 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
426 cluster_set_null(&percpu_cluster->index);
427 return true;
428}
429
430/*
431 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
432 * might involve allocating a new cluster for current CPU too.
433 */
434static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
435 unsigned long *offset, unsigned long *scan_base)
436{
437 struct percpu_cluster *cluster;
438 bool found_free;
439 unsigned long tmp;
440
441new_cluster:
442 cluster = this_cpu_ptr(si->percpu_cluster);
443 if (cluster_is_null(&cluster->index)) {
444 if (!cluster_is_null(&si->free_cluster_head)) {
445 cluster->index = si->free_cluster_head;
446 cluster->next = cluster_next(&cluster->index) *
447 SWAPFILE_CLUSTER;
448 } else if (!cluster_is_null(&si->discard_cluster_head)) {
449 /*
450 * we don't have free cluster but have some clusters in
451 * discarding, do discard now and reclaim them
452 */
453 swap_do_scheduled_discard(si);
454 *scan_base = *offset = si->cluster_next;
455 goto new_cluster;
456 } else
457 return;
458 }
459
460 found_free = false;
461
462 /*
463 * Other CPUs can use our cluster if they can't find a free cluster,
464 * check if there is still free entry in the cluster
465 */
466 tmp = cluster->next;
467 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
468 SWAPFILE_CLUSTER) {
469 if (!si->swap_map[tmp]) {
470 found_free = true;
471 break;
472 }
473 tmp++;
474 }
475 if (!found_free) {
476 cluster_set_null(&cluster->index);
477 goto new_cluster;
478 }
479 cluster->next = tmp + 1;
480 *offset = tmp;
481 *scan_base = tmp;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700482}
483
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700484static unsigned long scan_swap_map(struct swap_info_struct *si,
485 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800487 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800488 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800489 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700490 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800492 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700493 * We try to cluster swap pages by allocating them sequentially
494 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
495 * way, however, we resort to first-free allocation, starting
496 * a new cluster. This prevents us from scattering swap pages
497 * all over the entire swap partition, so that we reduce
498 * overall disk seek times between swap pages. -- sct
499 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800500 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700501 */
502
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700503 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800504 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800505
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700506 /* SSD algorithm */
507 if (si->cluster_info) {
508 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
509 goto checks;
510 }
511
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800512 if (unlikely(!si->cluster_nr--)) {
513 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
514 si->cluster_nr = SWAPFILE_CLUSTER - 1;
515 goto checks;
516 }
Shaohua Li2a8f9442013-09-11 14:20:28 -0700517
Shaohua Liec8acf22013-02-22 16:34:38 -0800518 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700519
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800520 /*
521 * If seek is expensive, start searching for new cluster from
522 * start of partition, to minimize the span of allocated swap.
Chen Yucong50088c42014-06-04 16:10:57 -0700523 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
524 * case, just handled by scan_swap_map_try_ssd_cluster() above.
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800525 */
Chen Yucong50088c42014-06-04 16:10:57 -0700526 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700527 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
528
529 /* Locate the first empty (unaligned) cluster */
530 for (; last_in_cluster <= si->highest_bit; offset++) {
531 if (si->swap_map[offset])
532 last_in_cluster = offset + SWAPFILE_CLUSTER;
533 else if (offset == last_in_cluster) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800534 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800535 offset -= SWAPFILE_CLUSTER - 1;
536 si->cluster_next = offset;
537 si->cluster_nr = SWAPFILE_CLUSTER - 1;
538 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700540 if (unlikely(--latency_ration < 0)) {
541 cond_resched();
542 latency_ration = LATENCY_LIMIT;
543 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700544 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800545
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800546 offset = scan_base;
Shaohua Liec8acf22013-02-22 16:34:38 -0800547 spin_lock(&si->lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800548 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700550
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800551checks:
Shaohua Liebc2a1a2013-09-11 14:20:32 -0700552 if (si->cluster_info) {
553 while (scan_swap_map_ssd_cluster_conflict(si, offset))
554 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
555 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800556 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700557 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700558 if (!si->highest_bit)
559 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800560 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800561 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700562
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700563 /* reuse swap entry of cache-only swap if not busy. */
564 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700565 int swap_was_freed;
Shaohua Liec8acf22013-02-22 16:34:38 -0800566 spin_unlock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700567 swap_was_freed = __try_to_reclaim_swap(si, offset);
Shaohua Liec8acf22013-02-22 16:34:38 -0800568 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700569 /* entry was freed successfully, try to use this again */
570 if (swap_was_freed)
571 goto checks;
572 goto scan; /* check next one */
573 }
574
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800575 if (si->swap_map[offset])
576 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700577
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800578 if (offset == si->lowest_bit)
579 si->lowest_bit++;
580 if (offset == si->highest_bit)
581 si->highest_bit--;
582 si->inuse_pages++;
583 if (si->inuse_pages == si->pages) {
584 si->lowest_bit = si->max;
585 si->highest_bit = 0;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700586 spin_lock(&swap_avail_lock);
587 plist_del(&si->avail_list, &swap_avail_head);
588 spin_unlock(&swap_avail_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800589 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800590 si->swap_map[offset] = usage;
Shaohua Li2a8f9442013-09-11 14:20:28 -0700591 inc_cluster_info_page(si, si->cluster_info, offset);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800592 si->cluster_next = offset + 1;
593 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800594
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800595 return offset;
596
597scan:
Shaohua Liec8acf22013-02-22 16:34:38 -0800598 spin_unlock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700599 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700600 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800601 spin_lock(&si->lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700602 goto checks;
603 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700604 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800605 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700606 goto checks;
607 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700608 if (unlikely(--latency_ration < 0)) {
609 cond_resched();
610 latency_ration = LATENCY_LIMIT;
611 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700612 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800613 offset = si->lowest_bit;
Jamie Liua5998062014-01-23 15:53:40 -0800614 while (offset < scan_base) {
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800615 if (!si->swap_map[offset]) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800616 spin_lock(&si->lock);
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800617 goto checks;
618 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700619 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800620 spin_lock(&si->lock);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700621 goto checks;
622 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800623 if (unlikely(--latency_ration < 0)) {
624 cond_resched();
625 latency_ration = LATENCY_LIMIT;
626 }
Jamie Liua5998062014-01-23 15:53:40 -0800627 offset++;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800628 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800629 spin_lock(&si->lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700630
631no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700632 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return 0;
634}
635
636swp_entry_t get_swap_page(void)
637{
Dan Streetmanadfab832014-06-04 16:09:53 -0700638 struct swap_info_struct *si, *next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700639 pgoff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
Shaohua Liec8acf22013-02-22 16:34:38 -0800641 if (atomic_long_read(&nr_swap_pages) <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700642 goto noswap;
Shaohua Liec8acf22013-02-22 16:34:38 -0800643 atomic_long_dec(&nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Dan Streetman18ab4d42014-06-04 16:09:59 -0700645 spin_lock(&swap_avail_lock);
646
647start_over:
648 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
649 /* requeue si to after same-priority siblings */
650 plist_requeue(&si->avail_list, &swap_avail_head);
651 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -0800652 spin_lock(&si->lock);
Dan Streetmanadfab832014-06-04 16:09:53 -0700653 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
Dan Streetman18ab4d42014-06-04 16:09:59 -0700654 spin_lock(&swap_avail_lock);
655 if (plist_node_empty(&si->avail_list)) {
656 spin_unlock(&si->lock);
657 goto nextsi;
658 }
659 WARN(!si->highest_bit,
660 "swap_info %d in list but !highest_bit\n",
661 si->type);
662 WARN(!(si->flags & SWP_WRITEOK),
663 "swap_info %d in list but !SWP_WRITEOK\n",
664 si->type);
665 plist_del(&si->avail_list, &swap_avail_head);
Shaohua Liec8acf22013-02-22 16:34:38 -0800666 spin_unlock(&si->lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700667 goto nextsi;
Shaohua Liec8acf22013-02-22 16:34:38 -0800668 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700669
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700670 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800671 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800672 spin_unlock(&si->lock);
673 if (offset)
Dan Streetmanadfab832014-06-04 16:09:53 -0700674 return swp_entry(si->type, offset);
Dan Streetman18ab4d42014-06-04 16:09:59 -0700675 pr_debug("scan_swap_map of si %d failed to find offset\n",
676 si->type);
677 spin_lock(&swap_avail_lock);
678nextsi:
Dan Streetmanadfab832014-06-04 16:09:53 -0700679 /*
680 * if we got here, it's likely that si was almost full before,
681 * and since scan_swap_map() can drop the si->lock, multiple
682 * callers probably all tried to get a page from the same si
Dan Streetman18ab4d42014-06-04 16:09:59 -0700683 * and it filled up before we could get one; or, the si filled
684 * up between us dropping swap_avail_lock and taking si->lock.
685 * Since we dropped the swap_avail_lock, the swap_avail_head
686 * list may have been modified; so if next is still in the
687 * swap_avail_head list then try it, otherwise start over.
Dan Streetmanadfab832014-06-04 16:09:53 -0700688 */
Dan Streetman18ab4d42014-06-04 16:09:59 -0700689 if (plist_node_empty(&next->avail_list))
690 goto start_over;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700692
Dan Streetman18ab4d42014-06-04 16:09:59 -0700693 spin_unlock(&swap_avail_lock);
694
Shaohua Liec8acf22013-02-22 16:34:38 -0800695 atomic_long_inc(&nr_swap_pages);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700696noswap:
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700697 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698}
699
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800700/* The only caller of this function is now suspend routine */
Hugh Dickins910321e2010-09-09 16:38:07 -0700701swp_entry_t get_swap_page_of_type(int type)
702{
703 struct swap_info_struct *si;
704 pgoff_t offset;
705
Hugh Dickins910321e2010-09-09 16:38:07 -0700706 si = swap_info[type];
Shaohua Liec8acf22013-02-22 16:34:38 -0800707 spin_lock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700708 if (si && (si->flags & SWP_WRITEOK)) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800709 atomic_long_dec(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700710 /* This is called for allocating swap entry, not cache */
711 offset = scan_swap_map(si, 1);
712 if (offset) {
Shaohua Liec8acf22013-02-22 16:34:38 -0800713 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700714 return swp_entry(type, offset);
715 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800716 atomic_long_inc(&nr_swap_pages);
Hugh Dickins910321e2010-09-09 16:38:07 -0700717 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800718 spin_unlock(&si->lock);
Hugh Dickins910321e2010-09-09 16:38:07 -0700719 return (swp_entry_t) {0};
720}
721
Hugh Dickins73c34b62009-12-14 17:58:43 -0800722static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800724 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 unsigned long offset, type;
726
727 if (!entry.val)
728 goto out;
729 type = swp_type(entry);
730 if (type >= nr_swapfiles)
731 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800732 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 if (!(p->flags & SWP_USED))
734 goto bad_device;
735 offset = swp_offset(entry);
736 if (offset >= p->max)
737 goto bad_offset;
738 if (!p->swap_map[offset])
739 goto bad_free;
Shaohua Liec8acf22013-02-22 16:34:38 -0800740 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 return p;
742
743bad_free:
Andrew Morton465c47f2013-09-11 14:20:17 -0700744 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 goto out;
746bad_offset:
Andrew Morton465c47f2013-09-11 14:20:17 -0700747 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 goto out;
749bad_device:
Andrew Morton465c47f2013-09-11 14:20:17 -0700750 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 goto out;
752bad_nofile:
Andrew Morton465c47f2013-09-11 14:20:17 -0700753 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754out:
755 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800756}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800758static unsigned char swap_entry_free(struct swap_info_struct *p,
759 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
Hugh Dickins253d5532009-12-14 17:58:44 -0800761 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800762 unsigned char count;
763 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700765 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800766 has_cache = count & SWAP_HAS_CACHE;
767 count &= ~SWAP_HAS_CACHE;
768
769 if (usage == SWAP_HAS_CACHE) {
770 VM_BUG_ON(!has_cache);
771 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800772 } else if (count == SWAP_MAP_SHMEM) {
773 /*
774 * Or we could insist on shmem.c using a special
775 * swap_shmem_free() and free_shmem_swap_and_cache()...
776 */
777 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800778 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
779 if (count == COUNT_CONTINUED) {
780 if (swap_count_continued(p, offset, count))
781 count = SWAP_MAP_MAX | COUNT_CONTINUED;
782 else
783 count = SWAP_MAP_MAX;
784 } else
785 count--;
786 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800787
Hugh Dickins253d5532009-12-14 17:58:44 -0800788 usage = count | has_cache;
789 p->swap_map[offset] = usage;
790
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700791 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800792 if (!usage) {
Vladimir Davydov37e84352016-01-20 15:02:56 -0800793 mem_cgroup_uncharge_swap(entry);
Shaohua Li2a8f9442013-09-11 14:20:28 -0700794 dec_cluster_info_page(p, p->cluster_info, offset);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700795 if (offset < p->lowest_bit)
796 p->lowest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700797 if (offset > p->highest_bit) {
798 bool was_full = !p->highest_bit;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700799 p->highest_bit = offset;
Dan Streetman18ab4d42014-06-04 16:09:59 -0700800 if (was_full && (p->flags & SWP_WRITEOK)) {
801 spin_lock(&swap_avail_lock);
802 WARN_ON(!plist_node_empty(&p->avail_list));
803 if (plist_node_empty(&p->avail_list))
804 plist_add(&p->avail_list,
805 &swap_avail_head);
806 spin_unlock(&swap_avail_lock);
807 }
808 }
Shaohua Liec8acf22013-02-22 16:34:38 -0800809 atomic_long_inc(&nr_swap_pages);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700810 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600811 frontswap_invalidate_page(p->type, offset);
Mel Gorman73744922012-07-31 16:45:20 -0700812 if (p->flags & SWP_BLKDEV) {
813 struct gendisk *disk = p->bdev->bd_disk;
814 if (disk->fops->swap_slot_free_notify)
815 disk->fops->swap_slot_free_notify(p->bdev,
816 offset);
817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800819
820 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821}
822
823/*
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800824 * Caller has made sure that the swap device corresponding to entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 * is still around or has not been recycled.
826 */
827void swap_free(swp_entry_t entry)
828{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800829 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831 p = swap_info_get(entry);
832 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800833 swap_entry_free(p, entry, 1);
Shaohua Liec8acf22013-02-22 16:34:38 -0800834 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 }
836}
837
838/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700839 * Called after dropping swapcache to decrease refcnt to swap entries.
840 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700841void swapcache_free(swp_entry_t entry)
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700842{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700843 struct swap_info_struct *p;
844
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700845 p = swap_info_get(entry);
846 if (p) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700847 swap_entry_free(p, entry, SWAP_HAS_CACHE);
Shaohua Liec8acf22013-02-22 16:34:38 -0800848 spin_unlock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700849 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700850}
851
852/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700853 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800854 * This does not give an exact answer when swap count is continued,
855 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700857int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700859 int count = 0;
860 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 swp_entry_t entry;
862
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700863 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 p = swap_info_get(entry);
865 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700866 count = swap_count(p->swap_map[swp_offset(entry)]);
Shaohua Liec8acf22013-02-22 16:34:38 -0800867 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700869 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870}
871
872/*
Minchan Kim8334b962015-09-08 15:00:24 -0700873 * How many references to @entry are currently swapped out?
874 * This considers COUNT_CONTINUED so it returns exact answer.
875 */
876int swp_swapcount(swp_entry_t entry)
877{
878 int count, tmp_count, n;
879 struct swap_info_struct *p;
880 struct page *page;
881 pgoff_t offset;
882 unsigned char *map;
883
884 p = swap_info_get(entry);
885 if (!p)
886 return 0;
887
888 count = swap_count(p->swap_map[swp_offset(entry)]);
889 if (!(count & COUNT_CONTINUED))
890 goto out;
891
892 count &= ~COUNT_CONTINUED;
893 n = SWAP_MAP_MAX + 1;
894
895 offset = swp_offset(entry);
896 page = vmalloc_to_page(p->swap_map + offset);
897 offset &= ~PAGE_MASK;
898 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
899
900 do {
Geliang Tanga8ae4992016-01-14 15:20:45 -0800901 page = list_next_entry(page, lru);
Minchan Kim8334b962015-09-08 15:00:24 -0700902 map = kmap_atomic(page);
903 tmp_count = map[offset];
904 kunmap_atomic(map);
905
906 count += (tmp_count & ~COUNT_CONTINUED) * n;
907 n *= (SWAP_CONT_MAX + 1);
908 } while (tmp_count & COUNT_CONTINUED);
909out:
910 spin_unlock(&p->lock);
911 return count;
912}
913
914/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800915 * We can write to an anon page without COW if there are no other references
916 * to it. And as a side-effect, free up its swap: because the old content
917 * on disk will never be read, and seeking back there to write new content
918 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800920int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700922 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
Sasha Levin309381fea2014-01-23 15:52:54 -0800924 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800925 if (unlikely(PageKsm(page)))
926 return 0;
Kirill A. Shutemov1f25fe22016-01-15 16:52:24 -0800927 /* The page is part of THP and cannot be reused */
928 if (PageTransCompound(page))
929 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700930 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800931 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700932 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800933 if (count == 1 && !PageWriteback(page)) {
934 delete_from_swap_cache(page);
935 SetPageDirty(page);
936 }
937 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800938 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939}
940
941/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800942 * If swap is getting full, or if there are no more mappings of this page,
943 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800945int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946{
Sasha Levin309381fea2014-01-23 15:52:54 -0800947 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
949 if (!PageSwapCache(page))
950 return 0;
951 if (PageWriteback(page))
952 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800953 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 return 0;
955
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700956 /*
957 * Once hibernation has begun to create its image of memory,
958 * there's a danger that one of the calls to try_to_free_swap()
959 * - most probably a call from __try_to_reclaim_swap() while
960 * hibernation is allocating its own swap pages for the image,
961 * but conceivably even a call from memory reclaim - will free
962 * the swap from a page which has already been recorded in the
963 * image as a clean swapcache page, and then reuse its swap for
964 * another page of the image. On waking from hibernation, the
965 * original page might be freed under memory pressure, then
966 * later read back in from swap, now with the wrong data.
967 *
Seth Jennings2de1a7e2013-11-12 15:07:46 -0800968 * Hibernation suspends storage while it is writing the image
Mel Gormanf90ac392012-01-10 15:07:15 -0800969 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700970 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800971 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700972 return 0;
973
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800974 delete_from_swap_cache(page);
975 SetPageDirty(page);
976 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700977}
978
979/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 * Free the swap entry like above, but also try to
981 * free the page cache entry if it is the last user.
982 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800983int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800985 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 struct page *page = NULL;
987
Andi Kleena7420aa2009-09-16 11:50:05 +0200988 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800989 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 p = swap_info_get(entry);
992 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800993 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Shaohua Li33806f02013-02-22 16:34:37 -0800994 page = find_get_page(swap_address_space(entry),
995 entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700996 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800997 page_cache_release(page);
998 page = NULL;
999 }
1000 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001001 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 }
1003 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08001004 /*
1005 * Not mapped elsewhere, or swap space full? Free it!
1006 * Also recheck PageSwapCache now page is locked (above).
1007 */
Nick Piggin93fac702006-03-31 02:29:56 -08001008 if (PageSwapCache(page) && !PageWriteback(page) &&
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08001009 (!page_mapped(page) || mem_cgroup_swap_full(page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 delete_from_swap_cache(page);
1011 SetPageDirty(page);
1012 }
1013 unlock_page(page);
1014 page_cache_release(page);
1015 }
Hugh Dickins2509ef22009-01-06 14:40:10 -08001016 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001019#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001020/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001021 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001022 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001023 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1024 * from 0, in which the swap header is expected to be located.
1025 *
1026 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001027 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001028int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001029{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001030 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001031 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001032
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001033 if (device)
1034 bdev = bdget(device);
1035
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001036 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001037 for (type = 0; type < nr_swapfiles; type++) {
1038 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001039
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001040 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001041 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -07001042
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001043 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001044 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001045 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001046
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001047 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001048 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -08001049 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001050 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001051 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001052
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001053 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001054 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +02001055 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -08001056
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001057 spin_unlock(&swap_lock);
1058 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001059 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001060 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001061 }
1062 }
1063 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -08001064 if (bdev)
1065 bdput(bdev);
1066
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001067 return -ENODEV;
1068}
1069
1070/*
Hugh Dickins73c34b62009-12-14 17:58:43 -08001071 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1072 * corresponding to given index in swap_info (swap type).
1073 */
1074sector_t swapdev_block(int type, pgoff_t offset)
1075{
1076 struct block_device *bdev;
1077
1078 if ((unsigned int)type >= nr_swapfiles)
1079 return 0;
1080 if (!(swap_info[type]->flags & SWP_WRITEOK))
1081 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001082 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -08001083}
1084
1085/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001086 * Return either the total number of swap pages of given type, or the number
1087 * of free pages of that type (depending on @free)
1088 *
1089 * This is needed for software suspend
1090 */
1091unsigned int count_swap_pages(int type, int free)
1092{
1093 unsigned int n = 0;
1094
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001095 spin_lock(&swap_lock);
1096 if ((unsigned int)type < nr_swapfiles) {
1097 struct swap_info_struct *sis = swap_info[type];
1098
Shaohua Liec8acf22013-02-22 16:34:38 -08001099 spin_lock(&sis->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001100 if (sis->flags & SWP_WRITEOK) {
1101 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001102 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001103 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001104 }
Shaohua Liec8acf22013-02-22 16:34:38 -08001105 spin_unlock(&sis->lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001106 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001107 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001108 return n;
1109}
Hugh Dickins73c34b62009-12-14 17:58:43 -08001110#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -08001111
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001112static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001113{
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001114 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001115}
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117/*
Hugh Dickins72866f62005-10-29 18:15:55 -07001118 * No need to decide whether this PTE shares the swap entry with others,
1119 * just let do_wp_page work it out if a write is requested later - to
1120 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 */
Hugh Dickins044d66c2008-02-07 00:14:04 -08001122static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 unsigned long addr, swp_entry_t entry, struct page *page)
1124{
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001125 struct page *swapcache;
Johannes Weiner72835c82012-01-12 17:18:32 -08001126 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -08001127 spinlock_t *ptl;
1128 pte_t *pte;
1129 int ret = 1;
1130
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001131 swapcache = page;
1132 page = ksm_might_need_to_copy(page, vma, addr);
1133 if (unlikely(!page))
1134 return -ENOMEM;
1135
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001136 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1137 &memcg, false)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001138 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001139 goto out_nolock;
1140 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001141
1142 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001143 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001144 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001145 ret = 0;
1146 goto out;
1147 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001148
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001149 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001150 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 get_page(page);
1152 set_pte_at(vma->vm_mm, addr, pte,
1153 pte_mkold(mk_pte(page, vma->vm_page_prot)));
Johannes Weiner00501b52014-08-08 14:19:20 -07001154 if (page == swapcache) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001155 page_add_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001156 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001157 } else { /* ksm created a completely new copy */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001158 page_add_new_anon_rmap(page, vma, addr, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001159 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001160 lru_cache_add_active_or_unevictable(page, vma);
1161 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 swap_free(entry);
1163 /*
1164 * Move the page to the active list so it is not
1165 * immediately swapped out again after swapon.
1166 */
1167 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001168out:
1169 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -08001170out_nolock:
Hugh Dickins9e16b7f2013-02-22 16:36:09 -08001171 if (page != swapcache) {
1172 unlock_page(page);
1173 put_page(page);
1174 }
Hugh Dickins044d66c2008-02-07 00:14:04 -08001175 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176}
1177
1178static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1179 unsigned long addr, unsigned long end,
1180 swp_entry_t entry, struct page *page)
1181{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -07001183 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001184 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Hugh Dickins044d66c2008-02-07 00:14:04 -08001186 /*
1187 * We don't actually need pte lock while scanning for swp_pte: since
1188 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1189 * page table while we're scanning; though it could get zapped, and on
1190 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1191 * of unmatched parts which look like swp_pte, so unuse_pte must
1192 * recheck under pte lock. Scanning without pte lock lets it be
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001193 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
Hugh Dickins044d66c2008-02-07 00:14:04 -08001194 */
1195 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 do {
1197 /*
1198 * swapoff spends a _lot_ of time in this loop!
1199 * Test inline before going to call unuse_pte.
1200 */
Hugh Dickins9f8bdb32016-01-15 16:57:34 -08001201 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -08001202 pte_unmap(pte);
1203 ret = unuse_pte(vma, pmd, addr, entry, page);
1204 if (ret)
1205 goto out;
1206 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 }
1208 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -08001209 pte_unmap(pte - 1);
1210out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001211 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212}
1213
1214static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1215 unsigned long addr, unsigned long end,
1216 swp_entry_t entry, struct page *page)
1217{
1218 pmd_t *pmd;
1219 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001220 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
1222 pmd = pmd_offset(pud, addr);
1223 do {
1224 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001225 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001227 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1228 if (ret)
1229 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 } while (pmd++, addr = next, addr != end);
1231 return 0;
1232}
1233
1234static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1235 unsigned long addr, unsigned long end,
1236 swp_entry_t entry, struct page *page)
1237{
1238 pud_t *pud;
1239 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001240 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
1242 pud = pud_offset(pgd, addr);
1243 do {
1244 next = pud_addr_end(addr, end);
1245 if (pud_none_or_clear_bad(pud))
1246 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001247 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1248 if (ret)
1249 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 } while (pud++, addr = next, addr != end);
1251 return 0;
1252}
1253
1254static int unuse_vma(struct vm_area_struct *vma,
1255 swp_entry_t entry, struct page *page)
1256{
1257 pgd_t *pgd;
1258 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001259 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001261 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 addr = page_address_in_vma(page, vma);
1263 if (addr == -EFAULT)
1264 return 0;
1265 else
1266 end = addr + PAGE_SIZE;
1267 } else {
1268 addr = vma->vm_start;
1269 end = vma->vm_end;
1270 }
1271
1272 pgd = pgd_offset(vma->vm_mm, addr);
1273 do {
1274 next = pgd_addr_end(addr, end);
1275 if (pgd_none_or_clear_bad(pgd))
1276 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001277 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1278 if (ret)
1279 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 } while (pgd++, addr = next, addr != end);
1281 return 0;
1282}
1283
1284static int unuse_mm(struct mm_struct *mm,
1285 swp_entry_t entry, struct page *page)
1286{
1287 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001288 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
1290 if (!down_read_trylock(&mm->mmap_sem)) {
1291 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001292 * Activate page so shrink_inactive_list is unlikely to unmap
1293 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001295 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 unlock_page(page);
1297 down_read(&mm->mmap_sem);
1298 lock_page(page);
1299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001301 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 break;
1303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001305 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306}
1307
1308/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001309 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1310 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 * Recycle to start on reaching the end, returning 0 when empty.
1312 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001313static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001314 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001316 unsigned int max = si->max;
1317 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001318 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
1320 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001321 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 * for whether an entry is in use, not modifying it; false
1323 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001324 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 */
1326 for (;;) {
1327 if (++i >= max) {
1328 if (!prev) {
1329 i = 0;
1330 break;
1331 }
1332 /*
1333 * No entries in use at top of swap_map,
1334 * loop back to start and recheck there.
1335 */
1336 max = prev + 1;
1337 prev = 0;
1338 i = 1;
1339 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001340 if (frontswap) {
1341 if (frontswap_test(si, i))
1342 break;
1343 else
1344 continue;
1345 }
Jason Low4db0c3c2015-04-15 16:14:08 -07001346 count = READ_ONCE(si->swap_map[i]);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001347 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 break;
1349 }
1350 return i;
1351}
1352
1353/*
1354 * We completely avoid races by reading each swap page in advance,
1355 * and then search for the process using it. All the necessary
1356 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001357 *
1358 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1359 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001361int try_to_unuse(unsigned int type, bool frontswap,
1362 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001364 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 struct mm_struct *start_mm;
Shaohua Liedfe23d2013-09-11 14:20:31 -07001366 volatile unsigned char *swap_map; /* swap_map is accessed without
1367 * locking. Mark it as volatile
1368 * to prevent compiler doing
1369 * something odd.
1370 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001371 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 struct page *page;
1373 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001374 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
1377 /*
1378 * When searching mms for an entry, a good strategy is to
1379 * start at the first mm we freed the previous entry from
1380 * (though actually we don't notice whether we or coincidence
1381 * freed the entry). Initialize this start_mm with a hold.
1382 *
1383 * A simpler strategy would be to start at the last mm we
1384 * freed the previous entry from; but that would take less
1385 * advantage of mmlist ordering, which clusters forked mms
1386 * together, child after parent. If we race with dup_mmap(), we
1387 * prefer to resolve parent before child, lest we miss entries
1388 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001389 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 */
1391 start_mm = &init_mm;
1392 atomic_inc(&init_mm.mm_users);
1393
1394 /*
1395 * Keep on scanning until all entries have gone. Usually,
1396 * one pass through swap_map is enough, but not necessarily:
1397 * there are races when an instance of an entry might be missed.
1398 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001399 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 if (signal_pending(current)) {
1401 retval = -EINTR;
1402 break;
1403 }
1404
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001405 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 * Get a page for the entry, using the existing swap
1407 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001408 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 */
1410 swap_map = &si->swap_map[i];
1411 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001412 page = read_swap_cache_async(entry,
1413 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 if (!page) {
1415 /*
1416 * Either swap_duplicate() failed because entry
1417 * has been freed independently, and will not be
1418 * reused since sys_swapoff() already disabled
1419 * allocation from here, or alloc_page() failed.
1420 */
Shaohua Liedfe23d2013-09-11 14:20:31 -07001421 swcount = *swap_map;
1422 /*
1423 * We don't hold lock here, so the swap entry could be
1424 * SWAP_MAP_BAD (when the cluster is discarding).
1425 * Instead of fail out, We can just skip the swap
1426 * entry because swapoff will wait for discarding
1427 * finish anyway.
1428 */
1429 if (!swcount || swcount == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 continue;
1431 retval = -ENOMEM;
1432 break;
1433 }
1434
1435 /*
1436 * Don't hold on to start_mm if it looks like exiting.
1437 */
1438 if (atomic_read(&start_mm->mm_users) == 1) {
1439 mmput(start_mm);
1440 start_mm = &init_mm;
1441 atomic_inc(&init_mm.mm_users);
1442 }
1443
1444 /*
1445 * Wait for and lock page. When do_swap_page races with
1446 * try_to_unuse, do_swap_page can handle the fault much
1447 * faster than try_to_unuse can locate the entry. This
1448 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1449 * defer to do_swap_page in such a case - in some tests,
1450 * do_swap_page and try_to_unuse repeatedly compete.
1451 */
1452 wait_on_page_locked(page);
1453 wait_on_page_writeback(page);
1454 lock_page(page);
1455 wait_on_page_writeback(page);
1456
1457 /*
1458 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001461 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1462 retval = shmem_unuse(entry, page);
1463 /* page has already been unlocked and released */
1464 if (retval < 0)
1465 break;
1466 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001468 if (swap_count(swcount) && start_mm != &init_mm)
1469 retval = unuse_mm(start_mm, entry, page);
1470
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001471 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 int set_start_mm = (*swap_map >= swcount);
1473 struct list_head *p = &start_mm->mmlist;
1474 struct mm_struct *new_start_mm = start_mm;
1475 struct mm_struct *prev_mm = start_mm;
1476 struct mm_struct *mm;
1477
1478 atomic_inc(&new_start_mm->mm_users);
1479 atomic_inc(&prev_mm->mm_users);
1480 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001481 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 (p = p->next) != &start_mm->mmlist) {
1483 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001484 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 spin_unlock(&mmlist_lock);
1487 mmput(prev_mm);
1488 prev_mm = mm;
1489
1490 cond_resched();
1491
1492 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001493 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001495 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001497 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001499
Bo Liu32c5fc12009-11-02 16:50:33 +00001500 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 mmput(new_start_mm);
1502 atomic_inc(&mm->mm_users);
1503 new_start_mm = mm;
1504 set_start_mm = 0;
1505 }
1506 spin_lock(&mmlist_lock);
1507 }
1508 spin_unlock(&mmlist_lock);
1509 mmput(prev_mm);
1510 mmput(start_mm);
1511 start_mm = new_start_mm;
1512 }
1513 if (retval) {
1514 unlock_page(page);
1515 page_cache_release(page);
1516 break;
1517 }
1518
1519 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 * If a reference remains (rare), we would like to leave
1521 * the page in the swap cache; but try_to_unmap could
1522 * then re-duplicate the entry once we drop page lock,
1523 * so we might loop indefinitely; also, that page could
1524 * not be swapped out to other storage meanwhile. So:
1525 * delete from cache even if there's another reference,
1526 * after ensuring that the data has been saved to disk -
1527 * since if the reference remains (rarer), it will be
1528 * read from disk into another page. Splitting into two
1529 * pages would be incorrect if swap supported "shared
1530 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001531 *
1532 * Given how unuse_vma() targets one particular offset
1533 * in an anon_vma, once the anon_vma has been determined,
1534 * this splitting happens to be just what is needed to
1535 * handle where KSM pages have been swapped out: re-reading
1536 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001538 if (swap_count(*swap_map) &&
1539 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 struct writeback_control wbc = {
1541 .sync_mode = WB_SYNC_NONE,
1542 };
1543
1544 swap_writepage(page, &wbc);
1545 lock_page(page);
1546 wait_on_page_writeback(page);
1547 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001548
1549 /*
1550 * It is conceivable that a racing task removed this page from
1551 * swap cache just before we acquired the page lock at the top,
1552 * or while we dropped it in unuse_mm(). The page might even
1553 * be back in swap cache on another swap area: that we must not
1554 * delete, since it may not have been written out to swap yet.
1555 */
1556 if (PageSwapCache(page) &&
1557 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001558 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
1560 /*
1561 * So we could skip searching mms once swap count went
1562 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001563 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 */
1565 SetPageDirty(page);
1566 unlock_page(page);
1567 page_cache_release(page);
1568
1569 /*
1570 * Make sure that we aren't completely killing
1571 * interactive performance.
1572 */
1573 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001574 if (frontswap && pages_to_unuse > 0) {
1575 if (!--pages_to_unuse)
1576 break;
1577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 }
1579
1580 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 return retval;
1582}
1583
1584/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001585 * After a successful try_to_unuse, if no swap is now in use, we know
1586 * we can empty the mmlist. swap_lock must be held on entry and exit.
1587 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 * added to the mmlist just after page_duplicate - before would be racy.
1589 */
1590static void drain_mmlist(void)
1591{
1592 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001593 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001595 for (type = 0; type < nr_swapfiles; type++)
1596 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 return;
1598 spin_lock(&mmlist_lock);
1599 list_for_each_safe(p, next, &init_mm.mmlist)
1600 list_del_init(p);
1601 spin_unlock(&mmlist_lock);
1602}
1603
1604/*
1605 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001606 * corresponds to page offset for the specified swap entry.
1607 * Note that the type of this function is sector_t, but it returns page offset
1608 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001610static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001612 struct swap_info_struct *sis;
1613 struct swap_extent *start_se;
1614 struct swap_extent *se;
1615 pgoff_t offset;
1616
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001617 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001618 *bdev = sis->bdev;
1619
1620 offset = swp_offset(entry);
1621 start_se = sis->curr_swap_extent;
1622 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 if (se->start_page <= offset &&
1626 offset < (se->start_page + se->nr_pages)) {
1627 return se->start_block + (offset - se->start_page);
1628 }
Geliang Tanga8ae4992016-01-14 15:20:45 -08001629 se = list_next_entry(se, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 sis->curr_swap_extent = se;
1631 BUG_ON(se == start_se); /* It *must* be present */
1632 }
1633}
1634
1635/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001636 * Returns the page offset into bdev for the specified page's swap entry.
1637 */
1638sector_t map_swap_page(struct page *page, struct block_device **bdev)
1639{
1640 swp_entry_t entry;
1641 entry.val = page_private(page);
1642 return map_swap_entry(entry, bdev);
1643}
1644
1645/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 * Free all of a swapdev's extent information
1647 */
1648static void destroy_swap_extents(struct swap_info_struct *sis)
1649{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001650 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 struct swap_extent *se;
1652
Geliang Tanga8ae4992016-01-14 15:20:45 -08001653 se = list_first_entry(&sis->first_swap_extent.list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 struct swap_extent, list);
1655 list_del(&se->list);
1656 kfree(se);
1657 }
Mel Gorman62c230b2012-07-31 16:44:55 -07001658
1659 if (sis->flags & SWP_FILE) {
1660 struct file *swap_file = sis->swap_file;
1661 struct address_space *mapping = swap_file->f_mapping;
1662
1663 sis->flags &= ~SWP_FILE;
1664 mapping->a_ops->swap_deactivate(swap_file);
1665 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666}
1667
1668/*
1669 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001670 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001672 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 */
Mel Gormana509bc12012-07-31 16:44:57 -07001674int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1676 unsigned long nr_pages, sector_t start_block)
1677{
1678 struct swap_extent *se;
1679 struct swap_extent *new_se;
1680 struct list_head *lh;
1681
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001682 if (start_page == 0) {
1683 se = &sis->first_swap_extent;
1684 sis->curr_swap_extent = se;
1685 se->start_page = 0;
1686 se->nr_pages = nr_pages;
1687 se->start_block = start_block;
1688 return 1;
1689 } else {
1690 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001692 BUG_ON(se->start_page + se->nr_pages != start_page);
1693 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 /* Merge it */
1695 se->nr_pages += nr_pages;
1696 return 0;
1697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 }
1699
1700 /*
1701 * No merge. Insert a new extent, preserving ordering.
1702 */
1703 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1704 if (new_se == NULL)
1705 return -ENOMEM;
1706 new_se->start_page = start_page;
1707 new_se->nr_pages = nr_pages;
1708 new_se->start_block = start_block;
1709
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001710 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001711 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712}
1713
1714/*
1715 * A `swap extent' is a simple thing which maps a contiguous range of pages
1716 * onto a contiguous range of disk blocks. An ordered list of swap extents
1717 * is built at swapon time and is then used at swap_writepage/swap_readpage
1718 * time for locating where on disk a page belongs.
1719 *
1720 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1721 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1722 * swap files identically.
1723 *
1724 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1725 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1726 * swapfiles are handled *identically* after swapon time.
1727 *
1728 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1729 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1730 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1731 * requirements, they are simply tossed out - we will never use those blocks
1732 * for swapping.
1733 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001734 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1736 * which will scribble on the fs.
1737 *
1738 * The amount of disk space which a single swap extent represents varies.
1739 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1740 * extents in the list. To avoid much list walking, we cache the previous
1741 * search location in `curr_swap_extent', and start new searches from there.
1742 * This is extremely effective. The average number of iterations in
1743 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1744 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001745static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746{
Mel Gorman62c230b2012-07-31 16:44:55 -07001747 struct file *swap_file = sis->swap_file;
1748 struct address_space *mapping = swap_file->f_mapping;
1749 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 int ret;
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 if (S_ISBLK(inode->i_mode)) {
1753 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001754 *span = sis->pages;
Mel Gormana509bc12012-07-31 16:44:57 -07001755 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 }
1757
Mel Gorman62c230b2012-07-31 16:44:55 -07001758 if (mapping->a_ops->swap_activate) {
Mel Gormana509bc12012-07-31 16:44:57 -07001759 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
Mel Gorman62c230b2012-07-31 16:44:55 -07001760 if (!ret) {
1761 sis->flags |= SWP_FILE;
1762 ret = add_swap_extent(sis, 0, sis->max, 0);
1763 *span = sis->pages;
1764 }
Mel Gormana509bc12012-07-31 16:44:57 -07001765 return ret;
Mel Gorman62c230b2012-07-31 16:44:55 -07001766 }
1767
Mel Gormana509bc12012-07-31 16:44:57 -07001768 return generic_swapfile_activate(sis, swap_file, span);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769}
1770
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001771static void _enable_swap_info(struct swap_info_struct *p, int prio,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001772 unsigned char *swap_map,
1773 struct swap_cluster_info *cluster_info)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001774{
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001775 if (prio >= 0)
1776 p->prio = prio;
1777 else
1778 p->prio = --least_priority;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001779 /*
1780 * the plist prio is negated because plist ordering is
1781 * low-to-high, while swap ordering is high-to-low
1782 */
1783 p->list.prio = -p->prio;
1784 p->avail_list.prio = -p->prio;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001785 p->swap_map = swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001786 p->cluster_info = cluster_info;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001787 p->flags |= SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001788 atomic_long_add(p->pages, &nr_swap_pages);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001789 total_swap_pages += p->pages;
1790
Dan Streetmanadfab832014-06-04 16:09:53 -07001791 assert_spin_locked(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001792 /*
Dan Streetman18ab4d42014-06-04 16:09:59 -07001793 * both lists are plists, and thus priority ordered.
1794 * swap_active_head needs to be priority ordered for swapoff(),
1795 * which on removal of any swap_info_struct with an auto-assigned
1796 * (i.e. negative) priority increments the auto-assigned priority
1797 * of any lower-priority swap_info_structs.
1798 * swap_avail_head needs to be priority ordered for get_swap_page(),
1799 * which allocates swap pages from the highest available priority
1800 * swap_info_struct.
Dan Streetmanadfab832014-06-04 16:09:53 -07001801 */
Dan Streetman18ab4d42014-06-04 16:09:59 -07001802 plist_add(&p->list, &swap_active_head);
1803 spin_lock(&swap_avail_lock);
1804 plist_add(&p->avail_list, &swap_avail_head);
1805 spin_unlock(&swap_avail_lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001806}
1807
1808static void enable_swap_info(struct swap_info_struct *p, int prio,
1809 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07001810 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001811 unsigned long *frontswap_map)
1812{
Minchan Kim4f898492013-04-30 15:26:54 -07001813 frontswap_init(p->type, frontswap_map);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001814 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001815 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001816 _enable_swap_info(p, prio, swap_map, cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001817 spin_unlock(&p->lock);
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001818 spin_unlock(&swap_lock);
1819}
1820
1821static void reinsert_swap_info(struct swap_info_struct *p)
1822{
1823 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001824 spin_lock(&p->lock);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001825 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
Shaohua Liec8acf22013-02-22 16:34:38 -08001826 spin_unlock(&p->lock);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001827 spin_unlock(&swap_lock);
1828}
1829
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001830SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001832 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001833 unsigned char *swap_map;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001834 struct swap_cluster_info *cluster_info;
Minchan Kim4f898492013-04-30 15:26:54 -07001835 unsigned long *frontswap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 struct file *swap_file, *victim;
1837 struct address_space *mapping;
1838 struct inode *inode;
Jeff Layton91a27b22012-10-10 15:25:28 -04001839 struct filename *pathname;
Dan Streetmanadfab832014-06-04 16:09:53 -07001840 int err, found = 0;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001841 unsigned int old_block_size;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 if (!capable(CAP_SYS_ADMIN))
1844 return -EPERM;
1845
Al Viro191c5422012-02-13 03:58:52 +00001846 BUG_ON(!current->mm);
1847
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 pathname = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (IS_ERR(pathname))
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001850 return PTR_ERR(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
Jeff Layton669abf42012-10-10 16:43:10 -04001852 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 err = PTR_ERR(victim);
1854 if (IS_ERR(victim))
1855 goto out;
1856
1857 mapping = victim->f_mapping;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001858 spin_lock(&swap_lock);
Dan Streetman18ab4d42014-06-04 16:09:59 -07001859 plist_for_each_entry(p, &swap_active_head, list) {
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001860 if (p->flags & SWP_WRITEOK) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001861 if (p->swap_file->f_mapping == mapping) {
1862 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 break;
Dan Streetmanadfab832014-06-04 16:09:53 -07001864 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 }
Dan Streetmanadfab832014-06-04 16:09:53 -07001867 if (!found) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001869 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 goto out_dput;
1871 }
Al Viro191c5422012-02-13 03:58:52 +00001872 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 vm_unacct_memory(p->pages);
1874 else {
1875 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001876 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 goto out_dput;
1878 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001879 spin_lock(&swap_avail_lock);
1880 plist_del(&p->avail_list, &swap_avail_head);
1881 spin_unlock(&swap_avail_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001882 spin_lock(&p->lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001883 if (p->prio < 0) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001884 struct swap_info_struct *si = p;
1885
Dan Streetman18ab4d42014-06-04 16:09:59 -07001886 plist_for_each_entry_continue(si, &swap_active_head, list) {
Dan Streetmanadfab832014-06-04 16:09:53 -07001887 si->prio++;
Dan Streetman18ab4d42014-06-04 16:09:59 -07001888 si->list.prio--;
1889 si->avail_list.prio--;
Dan Streetmanadfab832014-06-04 16:09:53 -07001890 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001891 least_priority++;
1892 }
Dan Streetman18ab4d42014-06-04 16:09:59 -07001893 plist_del(&p->list, &swap_active_head);
Shaohua Liec8acf22013-02-22 16:34:38 -08001894 atomic_long_sub(p->pages, &nr_swap_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 total_swap_pages -= p->pages;
1896 p->flags &= ~SWP_WRITEOK;
Shaohua Liec8acf22013-02-22 16:34:38 -08001897 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001898 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001899
David Rientjese1e12d22012-12-11 16:02:56 -08001900 set_current_oom_origin();
Dan Streetmanadfab832014-06-04 16:09:53 -07001901 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
David Rientjese1e12d22012-12-11 16:02:56 -08001902 clear_current_oom_origin();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 if (err) {
1905 /* re-insert swap space back into swap_list */
Cesar Eduardo Barroscf0cac02012-12-11 16:01:13 -08001906 reinsert_swap_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 goto out_dput;
1908 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001909
Shaohua Li815c2c52013-09-11 14:20:30 -07001910 flush_work(&p->discard_work);
1911
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001912 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001913 if (p->flags & SWP_CONTINUED)
1914 free_swap_count_continuations(p);
1915
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001916 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001917 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001918 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001920
1921 /* wait for anyone still in scan_swap_map */
1922 p->highest_bit = 0; /* cuts scans short */
1923 while (p->flags >= SWP_SCANNING) {
Shaohua Liec8acf22013-02-22 16:34:38 -08001924 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001925 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001926 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001927 spin_lock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08001928 spin_lock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001929 }
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 swap_file = p->swap_file;
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001932 old_block_size = p->old_block_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 p->swap_file = NULL;
1934 p->max = 0;
1935 swap_map = p->swap_map;
1936 p->swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07001937 cluster_info = p->cluster_info;
1938 p->cluster_info = NULL;
Minchan Kim4f898492013-04-30 15:26:54 -07001939 frontswap_map = frontswap_map_get(p);
Shaohua Liec8acf22013-02-22 16:34:38 -08001940 spin_unlock(&p->lock);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001941 spin_unlock(&swap_lock);
Dan Streetmanadfab832014-06-04 16:09:53 -07001942 frontswap_invalidate_area(p->type);
Krzysztof Kozlowski58e97ba2013-11-12 15:07:47 -08001943 frontswap_map_set(p, NULL);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001944 mutex_unlock(&swapon_mutex);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07001945 free_percpu(p->percpu_cluster);
1946 p->percpu_cluster = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07001948 vfree(cluster_info);
Minchan Kim4f898492013-04-30 15:26:54 -07001949 vfree(frontswap_map);
Seth Jennings2de1a7e2013-11-12 15:07:46 -08001950 /* Destroy swap account information */
Dan Streetmanadfab832014-06-04 16:09:53 -07001951 swap_cgroup_swapoff(p->type);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 inode = mapping->host;
1954 if (S_ISBLK(inode->i_mode)) {
1955 struct block_device *bdev = I_BDEV(inode);
Krzysztof Kozlowski5b808a22013-10-16 13:47:06 -07001956 set_blocksize(bdev, old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001957 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 } else {
Al Viro59551022016-01-22 15:40:57 -05001959 inode_lock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 inode->i_flags &= ~S_SWAPFILE;
Al Viro59551022016-01-22 15:40:57 -05001961 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 }
1963 filp_close(swap_file, NULL);
Weijie Yangf893ab42014-02-06 12:04:23 -08001964
1965 /*
1966 * Clear the SWP_USED flag after all resources are freed so that swapon
1967 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1968 * not hold p->lock after we cleared its SWP_WRITEOK.
1969 */
1970 spin_lock(&swap_lock);
1971 p->flags = 0;
1972 spin_unlock(&swap_lock);
1973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001975 atomic_inc(&proc_poll_event);
1976 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
1978out_dput:
1979 filp_close(victim, NULL);
1980out:
Xiaotian Fengf58b59c2012-11-16 14:14:55 -08001981 putname(pathname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 return err;
1983}
1984
1985#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001986static unsigned swaps_poll(struct file *file, poll_table *wait)
1987{
Kay Sieversf1514632011-07-12 20:48:39 +02001988 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001989
1990 poll_wait(file, &proc_poll_wait, wait);
1991
Kay Sieversf1514632011-07-12 20:48:39 +02001992 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1993 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001994 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1995 }
1996
1997 return POLLIN | POLLRDNORM;
1998}
1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000/* iterator */
2001static void *swap_start(struct seq_file *swap, loff_t *pos)
2002{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002003 struct swap_info_struct *si;
2004 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 loff_t l = *pos;
2006
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002007 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002009 if (!l)
2010 return SEQ_START_TOKEN;
2011
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002012 for (type = 0; type < nr_swapfiles; type++) {
2013 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2014 si = swap_info[type];
2015 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002017 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002018 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
2020
2021 return NULL;
2022}
2023
2024static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2025{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002026 struct swap_info_struct *si = v;
2027 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002029 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002030 type = 0;
2031 else
2032 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002033
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002034 for (; type < nr_swapfiles; type++) {
2035 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2036 si = swap_info[type];
2037 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 continue;
2039 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002040 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
2042
2043 return NULL;
2044}
2045
2046static void swap_stop(struct seq_file *swap, void *v)
2047{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002048 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049}
2050
2051static int swap_show(struct seq_file *swap, void *v)
2052{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002053 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 struct file *file;
2055 int len;
2056
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002057 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08002058 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2059 return 0;
2060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002062 file = si->swap_file;
Miklos Szeredi2726d562015-06-19 10:30:28 +02002063 len = seq_file_path(swap, file, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002064 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002065 len < 40 ? 40 - len : 1, " ",
Al Viro496ad9a2013-01-23 17:07:38 -05002066 S_ISBLK(file_inode(file)->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002068 si->pages << (PAGE_SHIFT - 10),
2069 si->inuse_pages << (PAGE_SHIFT - 10),
2070 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 return 0;
2072}
2073
Helge Deller15ad7cd2006-12-06 20:40:36 -08002074static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 .start = swap_start,
2076 .next = swap_next,
2077 .stop = swap_stop,
2078 .show = swap_show
2079};
2080
2081static int swaps_open(struct inode *inode, struct file *file)
2082{
Kay Sieversf1514632011-07-12 20:48:39 +02002083 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002084 int ret;
2085
Kay Sievers66d7dd52010-10-26 14:22:06 -07002086 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02002087 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07002088 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07002089
Kay Sieversf1514632011-07-12 20:48:39 +02002090 seq = file->private_data;
2091 seq->poll_event = atomic_read(&proc_poll_event);
2092 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093}
2094
Helge Deller15ad7cd2006-12-06 20:40:36 -08002095static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 .open = swaps_open,
2097 .read = seq_read,
2098 .llseek = seq_lseek,
2099 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07002100 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101};
2102
2103static int __init procswaps_init(void)
2104{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07002105 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return 0;
2107}
2108__initcall(procswaps_init);
2109#endif /* CONFIG_PROC_FS */
2110
Jan Beulich17963162008-12-16 11:35:24 +00002111#ifdef MAX_SWAPFILES_CHECK
2112static int __init max_swapfiles_check(void)
2113{
2114 MAX_SWAPFILES_CHECK();
2115 return 0;
2116}
2117late_initcall(max_swapfiles_check);
2118#endif
2119
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002120static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002122 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002124
2125 p = kzalloc(sizeof(*p), GFP_KERNEL);
2126 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002127 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002128
Hugh Dickins5d337b92005-09-03 15:54:41 -07002129 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002130 for (type = 0; type < nr_swapfiles; type++) {
2131 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002133 }
Christoph Lameter06972122006-06-23 02:03:35 -07002134 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07002135 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002136 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07002137 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002139 if (type >= nr_swapfiles) {
2140 p->type = type;
2141 swap_info[type] = p;
2142 /*
2143 * Write swap_info[type] before nr_swapfiles, in case a
2144 * racing procfs swap_start() or swap_next() is reading them.
2145 * (We never shrink nr_swapfiles, we never free this entry.)
2146 */
2147 smp_wmb();
2148 nr_swapfiles++;
2149 } else {
2150 kfree(p);
2151 p = swap_info[type];
2152 /*
2153 * Do not memset this entry: a racing procfs swap_next()
2154 * would be relying on p->type to remain valid.
2155 */
2156 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08002157 INIT_LIST_HEAD(&p->first_swap_extent.list);
Dan Streetman18ab4d42014-06-04 16:09:59 -07002158 plist_node_init(&p->list, 0);
2159 plist_node_init(&p->avail_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 p->flags = SWP_USED;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002161 spin_unlock(&swap_lock);
Shaohua Liec8acf22013-02-22 16:34:38 -08002162 spin_lock_init(&p->lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002163
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002164 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002165}
2166
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002167static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2168{
2169 int error;
2170
2171 if (S_ISBLK(inode->i_mode)) {
2172 p->bdev = bdgrab(I_BDEV(inode));
2173 error = blkdev_get(p->bdev,
Hugh Dickins6f179af2015-08-17 17:34:27 -07002174 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002175 if (error < 0) {
2176 p->bdev = NULL;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002177 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002178 }
2179 p->old_block_size = block_size(p->bdev);
2180 error = set_blocksize(p->bdev, PAGE_SIZE);
2181 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002182 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002183 p->flags |= SWP_BLKDEV;
2184 } else if (S_ISREG(inode->i_mode)) {
2185 p->bdev = inode->i_sb->s_bdev;
Al Viro59551022016-01-22 15:40:57 -05002186 inode_lock(inode);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07002187 if (IS_SWAPFILE(inode))
2188 return -EBUSY;
2189 } else
2190 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002191
2192 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002193}
2194
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002195static unsigned long read_swap_header(struct swap_info_struct *p,
2196 union swap_header *swap_header,
2197 struct inode *inode)
2198{
2199 int i;
2200 unsigned long maxpages;
2201 unsigned long swapfilepages;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002202 unsigned long last_page;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002203
2204 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002205 pr_err("Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002206 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002207 }
2208
2209 /* swap partition endianess hack... */
2210 if (swab32(swap_header->info.version) == 1) {
2211 swab32s(&swap_header->info.version);
2212 swab32s(&swap_header->info.last_page);
2213 swab32s(&swap_header->info.nr_badpages);
2214 for (i = 0; i < swap_header->info.nr_badpages; i++)
2215 swab32s(&swap_header->info.badpages[i]);
2216 }
2217 /* Check the swap header's sub-version */
2218 if (swap_header->info.version != 1) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002219 pr_warn("Unable to handle swap header version %d\n",
2220 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002221 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002222 }
2223
2224 p->lowest_bit = 1;
2225 p->cluster_next = 1;
2226 p->cluster_nr = 0;
2227
2228 /*
2229 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07002230 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002231 * of bits for the swap offset in the swp_entry_t type, and
2232 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07002233 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002234 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002235 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07002236 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002237 * offset is extracted. This will mask all the bits from
2238 * the initial ~0UL mask that can't be encoded in either
2239 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07002240 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002241 */
2242 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07002243 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002244 last_page = swap_header->info.last_page;
2245 if (last_page > maxpages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002246 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
Raymond Jenningsd6bbbd22013-09-11 14:20:16 -07002247 maxpages << (PAGE_SHIFT - 10),
2248 last_page << (PAGE_SHIFT - 10));
2249 }
2250 if (maxpages > last_page) {
2251 maxpages = last_page + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002252 /* p->max is an unsigned int: don't overflow it */
2253 if ((unsigned int)maxpages == 0)
2254 maxpages = UINT_MAX;
2255 }
2256 p->highest_bit = maxpages - 1;
2257
2258 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002259 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002260 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2261 if (swapfilepages && maxpages > swapfilepages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002262 pr_warn("Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002263 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002264 }
2265 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002266 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002267 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07002268 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002269
2270 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002271}
2272
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002273static int setup_swap_map_and_extents(struct swap_info_struct *p,
2274 union swap_header *swap_header,
2275 unsigned char *swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002276 struct swap_cluster_info *cluster_info,
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002277 unsigned long maxpages,
2278 sector_t *span)
2279{
2280 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002281 unsigned int nr_good_pages;
2282 int nr_extents;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002283 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2284 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002285
2286 nr_good_pages = maxpages - 1; /* omit header page */
2287
Shaohua Li2a8f9442013-09-11 14:20:28 -07002288 cluster_set_null(&p->free_cluster_head);
2289 cluster_set_null(&p->free_cluster_tail);
Shaohua Li815c2c52013-09-11 14:20:30 -07002290 cluster_set_null(&p->discard_cluster_head);
2291 cluster_set_null(&p->discard_cluster_tail);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002292
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002293 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2294 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002295 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2296 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002297 if (page_nr < maxpages) {
2298 swap_map[page_nr] = SWAP_MAP_BAD;
2299 nr_good_pages--;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002300 /*
2301 * Haven't marked the cluster free yet, no list
2302 * operation involved
2303 */
2304 inc_cluster_info_page(p, cluster_info, page_nr);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002305 }
2306 }
2307
Shaohua Li2a8f9442013-09-11 14:20:28 -07002308 /* Haven't marked the cluster free yet, no list operation involved */
2309 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2310 inc_cluster_info_page(p, cluster_info, i);
2311
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002312 if (nr_good_pages) {
2313 swap_map[0] = SWAP_MAP_BAD;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002314 /*
2315 * Not mark the cluster free yet, no list
2316 * operation involved
2317 */
2318 inc_cluster_info_page(p, cluster_info, 0);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002319 p->max = maxpages;
2320 p->pages = nr_good_pages;
2321 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002322 if (nr_extents < 0)
2323 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002324 nr_good_pages = p->pages;
2325 }
2326 if (!nr_good_pages) {
Andrew Morton465c47f2013-09-11 14:20:17 -07002327 pr_warn("Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002328 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002329 }
2330
Shaohua Li2a8f9442013-09-11 14:20:28 -07002331 if (!cluster_info)
2332 return nr_extents;
2333
2334 for (i = 0; i < nr_clusters; i++) {
2335 if (!cluster_count(&cluster_info[idx])) {
2336 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2337 if (cluster_is_null(&p->free_cluster_head)) {
2338 cluster_set_next_flag(&p->free_cluster_head,
2339 idx, 0);
2340 cluster_set_next_flag(&p->free_cluster_tail,
2341 idx, 0);
2342 } else {
2343 unsigned int tail;
2344
2345 tail = cluster_next(&p->free_cluster_tail);
2346 cluster_set_next(&cluster_info[tail], idx);
2347 cluster_set_next_flag(&p->free_cluster_tail,
2348 idx, 0);
2349 }
2350 }
2351 idx++;
2352 if (idx == nr_clusters)
2353 idx = 0;
2354 }
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002355 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002356}
2357
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002358/*
2359 * Helper to sys_swapon determining if a given swap
2360 * backing device queue supports DISCARD operations.
2361 */
2362static bool swap_discardable(struct swap_info_struct *si)
2363{
2364 struct request_queue *q = bdev_get_queue(si->bdev);
2365
2366 if (!q || !blk_queue_discard(q))
2367 return false;
2368
2369 return true;
2370}
2371
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002372SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2373{
2374 struct swap_info_struct *p;
Jeff Layton91a27b22012-10-10 15:25:28 -04002375 struct filename *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002376 struct file *swap_file = NULL;
2377 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002378 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002379 int error;
2380 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002381 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002382 sector_t span;
2383 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002384 unsigned char *swap_map = NULL;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002385 struct swap_cluster_info *cluster_info = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002386 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002387 struct page *page = NULL;
2388 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002389
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002390 if (swap_flags & ~SWAP_FLAGS_VALID)
2391 return -EINVAL;
2392
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002393 if (!capable(CAP_SYS_ADMIN))
2394 return -EPERM;
2395
2396 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002397 if (IS_ERR(p))
2398 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002399
Shaohua Li815c2c52013-09-11 14:20:30 -07002400 INIT_WORK(&p->discard_work, swap_discard_work);
2401
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002404 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002406 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 }
Jeff Layton669abf42012-10-10 16:43:10 -04002408 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002410 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002412 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
2414
2415 p->swap_file = swap_file;
2416 mapping = swap_file->f_mapping;
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002417 inode = mapping->host;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002418
Al Viro59551022016-01-22 15:40:57 -05002419 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002420 error = claim_swapfile(p, inode);
2421 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 /*
2425 * Read the swap header.
2426 */
2427 if (!mapping->a_ops->readpage) {
2428 error = -EINVAL;
2429 goto bad_swap;
2430 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002431 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 if (IS_ERR(page)) {
2433 error = PTR_ERR(page);
2434 goto bad_swap;
2435 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002436 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002438 maxpages = read_swap_header(p, swap_header, inode);
2439 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 error = -EINVAL;
2441 goto bad_swap;
2442 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002443
Hugh Dickins81e33972009-01-06 14:39:49 -08002444 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002445 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002446 if (!swap_map) {
2447 error = -ENOMEM;
2448 goto bad_swap;
2449 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002450 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
Hugh Dickins6f179af2015-08-17 17:34:27 -07002451 int cpu;
2452
Shaohua Li2a8f9442013-09-11 14:20:28 -07002453 p->flags |= SWP_SOLIDSTATE;
2454 /*
2455 * select a random position to start with to help wear leveling
2456 * SSD
2457 */
2458 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2459
2460 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2461 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2462 if (!cluster_info) {
2463 error = -ENOMEM;
2464 goto bad_swap;
2465 }
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002466 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2467 if (!p->percpu_cluster) {
2468 error = -ENOMEM;
2469 goto bad_swap;
2470 }
Hugh Dickins6f179af2015-08-17 17:34:27 -07002471 for_each_possible_cpu(cpu) {
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002472 struct percpu_cluster *cluster;
Hugh Dickins6f179af2015-08-17 17:34:27 -07002473 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002474 cluster_set_null(&cluster->index);
2475 }
Shaohua Li2a8f9442013-09-11 14:20:28 -07002476 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002477
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002478 error = swap_cgroup_swapon(p->type, maxpages);
2479 if (error)
2480 goto bad_swap;
2481
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002482 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
Shaohua Li2a8f9442013-09-11 14:20:28 -07002483 cluster_info, maxpages, &span);
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002484 if (unlikely(nr_extents < 0)) {
2485 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 goto bad_swap;
2487 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002488 /* frontswap enabled? set up bit-per-page map for frontswap */
2489 if (frontswap_enabled)
Akinobu Mita7b579762013-06-12 14:05:08 -07002490 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Shaohua Li2a8f9442013-09-11 14:20:28 -07002492 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2493 /*
2494 * When discard is enabled for swap with no particular
2495 * policy flagged, we set all swap discard flags here in
2496 * order to sustain backward compatibility with older
2497 * swapon(8) releases.
2498 */
2499 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2500 SWP_PAGE_DISCARD);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002501
Shaohua Li2a8f9442013-09-11 14:20:28 -07002502 /*
2503 * By flagging sys_swapon, a sysadmin can tell us to
2504 * either do single-time area discards only, or to just
2505 * perform discards for released swap page-clusters.
2506 * Now it's time to adjust the p->flags accordingly.
2507 */
2508 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2509 p->flags &= ~SWP_PAGE_DISCARD;
2510 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2511 p->flags &= ~SWP_AREA_DISCARD;
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002512
Shaohua Li2a8f9442013-09-11 14:20:28 -07002513 /* issue a swapon-time discard if it's still required */
2514 if (p->flags & SWP_AREA_DISCARD) {
2515 int err = discard_swap(p);
2516 if (unlikely(err))
2517 pr_err("swapon: discard_swap(%p): %d\n",
2518 p, err);
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002519 }
Hugh Dickins20137a42009-01-06 14:39:54 -08002520 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002521
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002522 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002523 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002524 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002525 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002526 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Shaohua Li2a8f9442013-09-11 14:20:28 -07002527 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002528
Andrew Morton465c47f2013-09-11 14:20:17 -07002529 pr_info("Adding %uk swap on %s. "
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002530 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
Jeff Layton91a27b22012-10-10 15:25:28 -04002531 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002532 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2533 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002534 (p->flags & SWP_DISCARDABLE) ? "D" : "",
Rafael Aquinidcf6b7d2013-07-03 15:02:46 -07002535 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2536 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002537 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002538
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002539 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002540 atomic_inc(&proc_poll_event);
2541 wake_up_interruptible(&proc_poll_wait);
2542
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002543 if (S_ISREG(inode->i_mode))
2544 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 error = 0;
2546 goto out;
2547bad_swap:
Shaohua Liebc2a1a2013-09-11 14:20:32 -07002548 free_percpu(p->percpu_cluster);
2549 p->percpu_cluster = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002550 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002551 set_blocksize(p->bdev, p->old_block_size);
2552 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002554 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002555 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002556 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002559 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 vfree(swap_map);
Shaohua Li2a8f9442013-09-11 14:20:28 -07002561 vfree(cluster_info);
Mel Gorman52c50562011-03-22 16:30:08 -07002562 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002563 if (inode && S_ISREG(inode->i_mode)) {
Al Viro59551022016-01-22 15:40:57 -05002564 inode_unlock(inode);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002565 inode = NULL;
2566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569out:
2570 if (page && !IS_ERR(page)) {
2571 kunmap(page);
2572 page_cache_release(page);
2573 }
2574 if (name)
2575 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002576 if (inode && S_ISREG(inode->i_mode))
Al Viro59551022016-01-22 15:40:57 -05002577 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 return error;
2579}
2580
2581void si_swapinfo(struct sysinfo *val)
2582{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002583 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 unsigned long nr_to_be_unused = 0;
2585
Hugh Dickins5d337b92005-09-03 15:54:41 -07002586 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002587 for (type = 0; type < nr_swapfiles; type++) {
2588 struct swap_info_struct *si = swap_info[type];
2589
2590 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2591 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 }
Shaohua Liec8acf22013-02-22 16:34:38 -08002593 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002595 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596}
2597
2598/*
2599 * Verify that a swap entry is valid and increment its swap map count.
2600 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002601 * Returns error code in following case.
2602 * - success -> 0
2603 * - swp_entry is invalid -> EINVAL
2604 * - swp_entry is migration entry -> EINVAL
2605 * - swap-cache reference is requested but there is already one. -> EEXIST
2606 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002607 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002609static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002611 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002613 unsigned char count;
2614 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002615 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Andi Kleena7420aa2009-09-16 11:50:05 +02002617 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002618 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002619
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 type = swp_type(entry);
2621 if (type >= nr_swapfiles)
2622 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002623 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 offset = swp_offset(entry);
2625
Shaohua Liec8acf22013-02-22 16:34:38 -08002626 spin_lock(&p->lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002627 if (unlikely(offset >= p->max))
2628 goto unlock_out;
2629
Hugh Dickins253d5532009-12-14 17:58:44 -08002630 count = p->swap_map[offset];
Shaohua Liedfe23d2013-09-11 14:20:31 -07002631
2632 /*
2633 * swapin_readahead() doesn't check if a swap entry is valid, so the
2634 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2635 */
2636 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2637 err = -ENOENT;
2638 goto unlock_out;
2639 }
2640
Hugh Dickins253d5532009-12-14 17:58:44 -08002641 has_cache = count & SWAP_HAS_CACHE;
2642 count &= ~SWAP_HAS_CACHE;
2643 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002644
Hugh Dickins253d5532009-12-14 17:58:44 -08002645 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002646
2647 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002648 if (!has_cache && count)
2649 has_cache = SWAP_HAS_CACHE;
2650 else if (has_cache) /* someone else added cache */
2651 err = -EEXIST;
2652 else /* no users remaining */
2653 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002654
2655 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002656
Hugh Dickins570a335b2009-12-14 17:58:46 -08002657 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2658 count += usage;
2659 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002660 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002661 else if (swap_count_continued(p, offset, count))
2662 count = COUNT_CONTINUED;
2663 else
2664 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002665 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002666 err = -ENOENT; /* unused swap entry */
2667
2668 p->swap_map[offset] = count | has_cache;
2669
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002670unlock_out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002671 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002673 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674
2675bad_file:
Andrew Morton465c47f2013-09-11 14:20:17 -07002676 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 goto out;
2678}
Hugh Dickins253d5532009-12-14 17:58:44 -08002679
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002680/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002681 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2682 * (in which case its reference count is never incremented).
2683 */
2684void swap_shmem_alloc(swp_entry_t entry)
2685{
2686 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2687}
2688
2689/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002690 * Increase reference count of swap entry by 1.
2691 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2692 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2693 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2694 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002695 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002696int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002697{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002698 int err = 0;
2699
2700 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2701 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2702 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002703}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002705/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002706 * @entry: swap entry for which we allocate swap cache.
2707 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002708 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002709 * This can return error codes. Returns 0 at success.
2710 * -EBUSY means there is a swap cache.
2711 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002712 */
2713int swapcache_prepare(swp_entry_t entry)
2714{
Hugh Dickins253d5532009-12-14 17:58:44 -08002715 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002716}
2717
Mel Gormanf981c592012-07-31 16:44:47 -07002718struct swap_info_struct *page_swap_info(struct page *page)
2719{
2720 swp_entry_t swap = { .val = page_private(page) };
2721 BUG_ON(!PageSwapCache(page));
2722 return swap_info[swp_type(swap)];
2723}
2724
2725/*
2726 * out-of-line __page_file_ methods to avoid include hell.
2727 */
2728struct address_space *__page_file_mapping(struct page *page)
2729{
Sasha Levin309381fea2014-01-23 15:52:54 -08002730 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002731 return page_swap_info(page)->swap_file->f_mapping;
2732}
2733EXPORT_SYMBOL_GPL(__page_file_mapping);
2734
2735pgoff_t __page_file_index(struct page *page)
2736{
2737 swp_entry_t swap = { .val = page_private(page) };
Sasha Levin309381fea2014-01-23 15:52:54 -08002738 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
Mel Gormanf981c592012-07-31 16:44:47 -07002739 return swp_offset(swap);
2740}
2741EXPORT_SYMBOL_GPL(__page_file_index);
2742
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002744 * add_swap_count_continuation - called when a swap count is duplicated
2745 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2746 * page of the original vmalloc'ed swap_map, to hold the continuation count
2747 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2748 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2749 *
2750 * These continuation pages are seldom referenced: the common paths all work
2751 * on the original swap_map, only referring to a continuation page when the
2752 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2753 *
2754 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2755 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2756 * can be called after dropping locks.
2757 */
2758int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2759{
2760 struct swap_info_struct *si;
2761 struct page *head;
2762 struct page *page;
2763 struct page *list_page;
2764 pgoff_t offset;
2765 unsigned char count;
2766
2767 /*
2768 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2769 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2770 */
2771 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2772
2773 si = swap_info_get(entry);
2774 if (!si) {
2775 /*
2776 * An acceptable race has occurred since the failing
2777 * __swap_duplicate(): the swap entry has been freed,
2778 * perhaps even the whole swap_map cleared for swapoff.
2779 */
2780 goto outer;
2781 }
2782
2783 offset = swp_offset(entry);
2784 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2785
2786 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2787 /*
2788 * The higher the swap count, the more likely it is that tasks
2789 * will race to add swap count continuation: we need to avoid
2790 * over-provisioning.
2791 */
2792 goto out;
2793 }
2794
2795 if (!page) {
Shaohua Liec8acf22013-02-22 16:34:38 -08002796 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002797 return -ENOMEM;
2798 }
2799
2800 /*
2801 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
Seth Jennings2de1a7e2013-11-12 15:07:46 -08002802 * no architecture is using highmem pages for kernel page tables: so it
2803 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
Hugh Dickins570a335b2009-12-14 17:58:46 -08002804 */
2805 head = vmalloc_to_page(si->swap_map + offset);
2806 offset &= ~PAGE_MASK;
2807
2808 /*
2809 * Page allocation does not initialize the page's lru field,
2810 * but it does always reset its private field.
2811 */
2812 if (!page_private(head)) {
2813 BUG_ON(count & COUNT_CONTINUED);
2814 INIT_LIST_HEAD(&head->lru);
2815 set_page_private(head, SWP_CONTINUED);
2816 si->flags |= SWP_CONTINUED;
2817 }
2818
2819 list_for_each_entry(list_page, &head->lru, lru) {
2820 unsigned char *map;
2821
2822 /*
2823 * If the previous map said no continuation, but we've found
2824 * a continuation page, free our allocation and use this one.
2825 */
2826 if (!(count & COUNT_CONTINUED))
2827 goto out;
2828
Cong Wang9b04c5f2011-11-25 23:14:39 +08002829 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002830 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002831 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002832
2833 /*
2834 * If this continuation count now has some space in it,
2835 * free our allocation and use this one.
2836 */
2837 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2838 goto out;
2839 }
2840
2841 list_add_tail(&page->lru, &head->lru);
2842 page = NULL; /* now it's attached, don't free it */
2843out:
Shaohua Liec8acf22013-02-22 16:34:38 -08002844 spin_unlock(&si->lock);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002845outer:
2846 if (page)
2847 __free_page(page);
2848 return 0;
2849}
2850
2851/*
2852 * swap_count_continued - when the original swap_map count is incremented
2853 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2854 * into, carry if so, or else fail until a new continuation page is allocated;
2855 * when the original swap_map count is decremented from 0 with continuation,
2856 * borrow from the continuation and report whether it still holds more.
2857 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2858 */
2859static bool swap_count_continued(struct swap_info_struct *si,
2860 pgoff_t offset, unsigned char count)
2861{
2862 struct page *head;
2863 struct page *page;
2864 unsigned char *map;
2865
2866 head = vmalloc_to_page(si->swap_map + offset);
2867 if (page_private(head) != SWP_CONTINUED) {
2868 BUG_ON(count & COUNT_CONTINUED);
2869 return false; /* need to add count continuation */
2870 }
2871
2872 offset &= ~PAGE_MASK;
2873 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002874 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002875
2876 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2877 goto init_map; /* jump over SWAP_CONT_MAX checks */
2878
2879 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2880 /*
2881 * Think of how you add 1 to 999
2882 */
2883 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002884 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002885 page = list_entry(page->lru.next, struct page, lru);
2886 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002887 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002888 }
2889 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002890 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002891 page = list_entry(page->lru.next, struct page, lru);
2892 if (page == head)
2893 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002894 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002895init_map: *map = 0; /* we didn't zero the page */
2896 }
2897 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002898 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002899 page = list_entry(page->lru.prev, struct page, lru);
2900 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002901 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002902 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002903 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002904 page = list_entry(page->lru.prev, struct page, lru);
2905 }
2906 return true; /* incremented */
2907
2908 } else { /* decrementing */
2909 /*
2910 * Think of how you subtract 1 from 1000
2911 */
2912 BUG_ON(count != COUNT_CONTINUED);
2913 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002914 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002915 page = list_entry(page->lru.next, struct page, lru);
2916 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002917 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002918 }
2919 BUG_ON(*map == 0);
2920 *map -= 1;
2921 if (*map == 0)
2922 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002923 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002924 page = list_entry(page->lru.prev, struct page, lru);
2925 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002926 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002927 *map = SWAP_CONT_MAX | count;
2928 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002929 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002930 page = list_entry(page->lru.prev, struct page, lru);
2931 }
2932 return count == COUNT_CONTINUED;
2933 }
2934}
2935
2936/*
2937 * free_swap_count_continuations - swapoff free all the continuation pages
2938 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2939 */
2940static void free_swap_count_continuations(struct swap_info_struct *si)
2941{
2942 pgoff_t offset;
2943
2944 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2945 struct page *head;
2946 head = vmalloc_to_page(si->swap_map + offset);
2947 if (page_private(head)) {
Geliang Tang0d576d22016-01-14 15:21:49 -08002948 struct page *page, *next;
2949
2950 list_for_each_entry_safe(page, next, &head->lru, lru) {
2951 list_del(&page->lru);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002952 __free_page(page);
2953 }
2954 }
2955 }
2956}