blob: 71373d03fcee99d57c29d24a25db888228bf3417 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37#include <asm/pgtable.h>
38#include <asm/tlbflush.h>
39#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080040#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Hugh Dickins570a335b2009-12-14 17:58:46 -080042static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
43 unsigned char);
44static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080045static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080046
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060047DEFINE_SPINLOCK(swap_lock);
Adrian Bunk7c363b82008-07-25 19:46:24 -070048static unsigned int nr_swapfiles;
Hugh Dickinsb9627162009-01-06 14:39:41 -080049long nr_swap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070051static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053static const char Bad_file[] = "Bad swap file entry ";
54static const char Unused_file[] = "Unused swap file entry ";
55static const char Bad_offset[] = "Bad swap offset entry ";
56static const char Unused_offset[] = "Unused swap offset entry ";
57
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060058struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Dan Magenheimer38b5faf2012-04-09 17:08:06 -060060struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Ingo Molnarfc0abb12006-01-18 17:42:33 -080062static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Kay Sievers66d7dd52010-10-26 14:22:06 -070064static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
65/* Activity counter to indicate that a swapon or swapoff has occurred */
66static atomic_t proc_poll_event = ATOMIC_INIT(0);
67
Hugh Dickins8d69aae2009-12-14 17:58:45 -080068static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070069{
Hugh Dickins570a335b2009-12-14 17:58:46 -080070 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070071}
72
Hugh Dickinsefa90a92009-12-14 17:58:41 -080073/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070074static int
75__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
76{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080077 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070078 struct page *page;
79 int ret = 0;
80
81 page = find_get_page(&swapper_space, entry.val);
82 if (!page)
83 return 0;
84 /*
85 * This function is called from scan_swap_map() and it's called
86 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
87 * We have to use trylock for avoiding deadlock. This is a special
88 * case and you should use try_to_free_swap() with explicit lock_page()
89 * in usual operations.
90 */
91 if (trylock_page(page)) {
92 ret = try_to_free_swap(page);
93 unlock_page(page);
94 }
95 page_cache_release(page);
96 return ret;
97}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070098
Linus Torvalds1da177e2005-04-16 15:20:36 -070099/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800100 * swapon tell device that all the old swap contents can be discarded,
101 * to allow the swap device to optimize its wear-levelling.
102 */
103static int discard_swap(struct swap_info_struct *si)
104{
105 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800106 sector_t start_block;
107 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800108 int err = 0;
109
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800110 /* Do not discard the swap header page! */
111 se = &si->first_swap_extent;
112 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
113 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
114 if (nr_blocks) {
115 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200116 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800117 if (err)
118 return err;
119 cond_resched();
120 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800121
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800122 list_for_each_entry(se, &si->first_swap_extent.list, list) {
123 start_block = se->start_block << (PAGE_SHIFT - 9);
124 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800125
126 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200127 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800128 if (err)
129 break;
130
131 cond_resched();
132 }
133 return err; /* That will often be -EOPNOTSUPP */
134}
135
Hugh Dickins7992fde2009-01-06 14:39:53 -0800136/*
137 * swap allocation tell device that a cluster of swap can now be discarded,
138 * to allow the swap device to optimize its wear-levelling.
139 */
140static void discard_swap_cluster(struct swap_info_struct *si,
141 pgoff_t start_page, pgoff_t nr_pages)
142{
143 struct swap_extent *se = si->curr_swap_extent;
144 int found_extent = 0;
145
146 while (nr_pages) {
147 struct list_head *lh;
148
149 if (se->start_page <= start_page &&
150 start_page < se->start_page + se->nr_pages) {
151 pgoff_t offset = start_page - se->start_page;
152 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800153 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800154
155 if (nr_blocks > nr_pages)
156 nr_blocks = nr_pages;
157 start_page += nr_blocks;
158 nr_pages -= nr_blocks;
159
160 if (!found_extent++)
161 si->curr_swap_extent = se;
162
163 start_block <<= PAGE_SHIFT - 9;
164 nr_blocks <<= PAGE_SHIFT - 9;
165 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200166 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800167 break;
168 }
169
170 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800171 se = list_entry(lh, struct swap_extent, list);
172 }
173}
174
175static int wait_for_discard(void *word)
176{
177 schedule();
178 return 0;
179}
180
Hugh Dickins048c27f2005-09-03 15:54:40 -0700181#define SWAPFILE_CLUSTER 256
182#define LATENCY_LIMIT 256
183
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700184static unsigned long scan_swap_map(struct swap_info_struct *si,
185 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800187 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800188 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800189 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700190 int latency_ration = LATENCY_LIMIT;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800191 int found_free_cluster = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800193 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700194 * We try to cluster swap pages by allocating them sequentially
195 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
196 * way, however, we resort to first-free allocation, starting
197 * a new cluster. This prevents us from scattering swap pages
198 * all over the entire swap partition, so that we reduce
199 * overall disk seek times between swap pages. -- sct
200 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800201 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700202 */
203
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700204 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800205 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800206
207 if (unlikely(!si->cluster_nr--)) {
208 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
209 si->cluster_nr = SWAPFILE_CLUSTER - 1;
210 goto checks;
211 }
Hugh Dickins7992fde2009-01-06 14:39:53 -0800212 if (si->flags & SWP_DISCARDABLE) {
213 /*
214 * Start range check on racing allocations, in case
215 * they overlap the cluster we eventually decide on
216 * (we scan without swap_lock to allow preemption).
217 * It's hardly conceivable that cluster_nr could be
218 * wrapped during our scan, but don't depend on it.
219 */
220 if (si->lowest_alloc)
221 goto checks;
222 si->lowest_alloc = si->max;
223 si->highest_alloc = 0;
224 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700225 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700226
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800227 /*
228 * If seek is expensive, start searching for new cluster from
229 * start of partition, to minimize the span of allocated swap.
230 * But if seek is cheap, search from our current position, so
231 * that swap is allocated from all over the partition: if the
232 * Flash Translation Layer only remaps within limited zones,
233 * we don't want to wear out the first zone too quickly.
234 */
235 if (!(si->flags & SWP_SOLIDSTATE))
236 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700237 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
238
239 /* Locate the first empty (unaligned) cluster */
240 for (; last_in_cluster <= si->highest_bit; offset++) {
241 if (si->swap_map[offset])
242 last_in_cluster = offset + SWAPFILE_CLUSTER;
243 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700244 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800245 offset -= SWAPFILE_CLUSTER - 1;
246 si->cluster_next = offset;
247 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800248 found_free_cluster = 1;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800249 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700251 if (unlikely(--latency_ration < 0)) {
252 cond_resched();
253 latency_ration = LATENCY_LIMIT;
254 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700255 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800256
257 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800258 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
259
260 /* Locate the first empty (unaligned) cluster */
261 for (; last_in_cluster < scan_base; offset++) {
262 if (si->swap_map[offset])
263 last_in_cluster = offset + SWAPFILE_CLUSTER;
264 else if (offset == last_in_cluster) {
265 spin_lock(&swap_lock);
266 offset -= SWAPFILE_CLUSTER - 1;
267 si->cluster_next = offset;
268 si->cluster_nr = SWAPFILE_CLUSTER - 1;
269 found_free_cluster = 1;
270 goto checks;
271 }
272 if (unlikely(--latency_ration < 0)) {
273 cond_resched();
274 latency_ration = LATENCY_LIMIT;
275 }
276 }
277
278 offset = scan_base;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700279 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800280 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800281 si->lowest_alloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700283
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800284checks:
285 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700286 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700287 if (!si->highest_bit)
288 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800289 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800290 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700291
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700292 /* reuse swap entry of cache-only swap if not busy. */
293 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700294 int swap_was_freed;
295 spin_unlock(&swap_lock);
296 swap_was_freed = __try_to_reclaim_swap(si, offset);
297 spin_lock(&swap_lock);
298 /* entry was freed successfully, try to use this again */
299 if (swap_was_freed)
300 goto checks;
301 goto scan; /* check next one */
302 }
303
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800304 if (si->swap_map[offset])
305 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700306
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800307 if (offset == si->lowest_bit)
308 si->lowest_bit++;
309 if (offset == si->highest_bit)
310 si->highest_bit--;
311 si->inuse_pages++;
312 if (si->inuse_pages == si->pages) {
313 si->lowest_bit = si->max;
314 si->highest_bit = 0;
315 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800316 si->swap_map[offset] = usage;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800317 si->cluster_next = offset + 1;
318 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800319
320 if (si->lowest_alloc) {
321 /*
322 * Only set when SWP_DISCARDABLE, and there's a scan
323 * for a free cluster in progress or just completed.
324 */
325 if (found_free_cluster) {
326 /*
327 * To optimize wear-levelling, discard the
328 * old data of the cluster, taking care not to
329 * discard any of its pages that have already
330 * been allocated by racing tasks (offset has
331 * already stepped over any at the beginning).
332 */
333 if (offset < si->highest_alloc &&
334 si->lowest_alloc <= last_in_cluster)
335 last_in_cluster = si->lowest_alloc - 1;
336 si->flags |= SWP_DISCARDING;
337 spin_unlock(&swap_lock);
338
339 if (offset < last_in_cluster)
340 discard_swap_cluster(si, offset,
341 last_in_cluster - offset + 1);
342
343 spin_lock(&swap_lock);
344 si->lowest_alloc = 0;
345 si->flags &= ~SWP_DISCARDING;
346
347 smp_mb(); /* wake_up_bit advises this */
348 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
349
350 } else if (si->flags & SWP_DISCARDING) {
351 /*
352 * Delay using pages allocated by racing tasks
353 * until the whole discard has been issued. We
354 * could defer that delay until swap_writepage,
355 * but it's easier to keep this self-contained.
356 */
357 spin_unlock(&swap_lock);
358 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
359 wait_for_discard, TASK_UNINTERRUPTIBLE);
360 spin_lock(&swap_lock);
361 } else {
362 /*
363 * Note pages allocated by racing tasks while
364 * scan for a free cluster is in progress, so
365 * that its final discard can exclude them.
366 */
367 if (offset < si->lowest_alloc)
368 si->lowest_alloc = offset;
369 if (offset > si->highest_alloc)
370 si->highest_alloc = offset;
371 }
372 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800373 return offset;
374
375scan:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700376 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700377 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700378 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700379 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700380 goto checks;
381 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700382 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
383 spin_lock(&swap_lock);
384 goto checks;
385 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700386 if (unlikely(--latency_ration < 0)) {
387 cond_resched();
388 latency_ration = LATENCY_LIMIT;
389 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700390 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800391 offset = si->lowest_bit;
392 while (++offset < scan_base) {
393 if (!si->swap_map[offset]) {
394 spin_lock(&swap_lock);
395 goto checks;
396 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700397 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
398 spin_lock(&swap_lock);
399 goto checks;
400 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800401 if (unlikely(--latency_ration < 0)) {
402 cond_resched();
403 latency_ration = LATENCY_LIMIT;
404 }
405 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700406 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700407
408no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700409 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 return 0;
411}
412
413swp_entry_t get_swap_page(void)
414{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700415 struct swap_info_struct *si;
416 pgoff_t offset;
417 int type, next;
418 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Hugh Dickins5d337b92005-09-03 15:54:41 -0700420 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700422 goto noswap;
423 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700425 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800426 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700427 next = si->next;
428 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800429 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700430 next = swap_list.head;
431 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700433
434 if (!si->highest_bit)
435 continue;
436 if (!(si->flags & SWP_WRITEOK))
437 continue;
438
439 swap_list.next = next;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700440 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800441 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700442 if (offset) {
443 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700444 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700445 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700446 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700448
449 nr_swap_pages++;
450noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700451 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700452 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
Hugh Dickins910321e2010-09-09 16:38:07 -0700455/* The only caller of this function is now susupend routine */
456swp_entry_t get_swap_page_of_type(int type)
457{
458 struct swap_info_struct *si;
459 pgoff_t offset;
460
461 spin_lock(&swap_lock);
462 si = swap_info[type];
463 if (si && (si->flags & SWP_WRITEOK)) {
464 nr_swap_pages--;
465 /* This is called for allocating swap entry, not cache */
466 offset = scan_swap_map(si, 1);
467 if (offset) {
468 spin_unlock(&swap_lock);
469 return swp_entry(type, offset);
470 }
471 nr_swap_pages++;
472 }
473 spin_unlock(&swap_lock);
474 return (swp_entry_t) {0};
475}
476
Hugh Dickins73c34b62009-12-14 17:58:43 -0800477static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800479 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 unsigned long offset, type;
481
482 if (!entry.val)
483 goto out;
484 type = swp_type(entry);
485 if (type >= nr_swapfiles)
486 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800487 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 if (!(p->flags & SWP_USED))
489 goto bad_device;
490 offset = swp_offset(entry);
491 if (offset >= p->max)
492 goto bad_offset;
493 if (!p->swap_map[offset])
494 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700495 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return p;
497
498bad_free:
499 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
500 goto out;
501bad_offset:
502 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
503 goto out;
504bad_device:
505 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
506 goto out;
507bad_nofile:
508 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
509out:
510 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800511}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800513static unsigned char swap_entry_free(struct swap_info_struct *p,
514 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515{
Hugh Dickins253d5532009-12-14 17:58:44 -0800516 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800517 unsigned char count;
518 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700520 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800521 has_cache = count & SWAP_HAS_CACHE;
522 count &= ~SWAP_HAS_CACHE;
523
524 if (usage == SWAP_HAS_CACHE) {
525 VM_BUG_ON(!has_cache);
526 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800527 } else if (count == SWAP_MAP_SHMEM) {
528 /*
529 * Or we could insist on shmem.c using a special
530 * swap_shmem_free() and free_shmem_swap_and_cache()...
531 */
532 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800533 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
534 if (count == COUNT_CONTINUED) {
535 if (swap_count_continued(p, offset, count))
536 count = SWAP_MAP_MAX | COUNT_CONTINUED;
537 else
538 count = SWAP_MAP_MAX;
539 } else
540 count--;
541 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800542
543 if (!count)
544 mem_cgroup_uncharge_swap(entry);
545
546 usage = count | has_cache;
547 p->swap_map[offset] = usage;
548
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700549 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800550 if (!usage) {
Nitin Guptab3a27d02010-05-17 11:02:43 +0530551 struct gendisk *disk = p->bdev->bd_disk;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700552 if (offset < p->lowest_bit)
553 p->lowest_bit = offset;
554 if (offset > p->highest_bit)
555 p->highest_bit = offset;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800556 if (swap_list.next >= 0 &&
557 p->prio > swap_info[swap_list.next]->prio)
558 swap_list.next = p->type;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700559 nr_swap_pages++;
560 p->inuse_pages--;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600561 frontswap_invalidate_page(p->type, offset);
Nitin Guptab3a27d02010-05-17 11:02:43 +0530562 if ((p->flags & SWP_BLKDEV) &&
563 disk->fops->swap_slot_free_notify)
564 disk->fops->swap_slot_free_notify(p->bdev, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800566
567 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
570/*
571 * Caller has made sure that the swapdevice corresponding to entry
572 * is still around or has not been recycled.
573 */
574void swap_free(swp_entry_t entry)
575{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800576 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 p = swap_info_get(entry);
579 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800580 swap_entry_free(p, entry, 1);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700581 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
583}
584
585/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700586 * Called after dropping swapcache to decrease refcnt to swap entries.
587 */
588void swapcache_free(swp_entry_t entry, struct page *page)
589{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700590 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800591 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700592
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700593 p = swap_info_get(entry);
594 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800595 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
596 if (page)
597 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700598 spin_unlock(&swap_lock);
599 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700600}
601
602/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700603 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800604 * This does not give an exact answer when swap count is continued,
605 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700607int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700609 int count = 0;
610 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 swp_entry_t entry;
612
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700613 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 p = swap_info_get(entry);
615 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700616 count = swap_count(p->swap_map[swp_offset(entry)]);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700617 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700619 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
622/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800623 * We can write to an anon page without COW if there are no other references
624 * to it. And as a side-effect, free up its swap: because the old content
625 * on disk will never be read, and seeking back there to write new content
626 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800628int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700630 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Hugh Dickins51726b12009-01-06 14:39:25 -0800632 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800633 if (unlikely(PageKsm(page)))
634 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700635 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800636 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700637 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800638 if (count == 1 && !PageWriteback(page)) {
639 delete_from_swap_cache(page);
640 SetPageDirty(page);
641 }
642 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800643 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
646/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800647 * If swap is getting full, or if there are no more mappings of this page,
648 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800650int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651{
Hugh Dickins51726b12009-01-06 14:39:25 -0800652 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 if (!PageSwapCache(page))
655 return 0;
656 if (PageWriteback(page))
657 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800658 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 return 0;
660
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700661 /*
662 * Once hibernation has begun to create its image of memory,
663 * there's a danger that one of the calls to try_to_free_swap()
664 * - most probably a call from __try_to_reclaim_swap() while
665 * hibernation is allocating its own swap pages for the image,
666 * but conceivably even a call from memory reclaim - will free
667 * the swap from a page which has already been recorded in the
668 * image as a clean swapcache page, and then reuse its swap for
669 * another page of the image. On waking from hibernation, the
670 * original page might be freed under memory pressure, then
671 * later read back in from swap, now with the wrong data.
672 *
Mel Gormanf90ac392012-01-10 15:07:15 -0800673 * Hibration suspends storage while it is writing the image
674 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700675 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800676 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700677 return 0;
678
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800679 delete_from_swap_cache(page);
680 SetPageDirty(page);
681 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700682}
683
684/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 * Free the swap entry like above, but also try to
686 * free the page cache entry if it is the last user.
687 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800688int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800690 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 struct page *page = NULL;
692
Andi Kleena7420aa2009-09-16 11:50:05 +0200693 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800694 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 p = swap_info_get(entry);
697 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800698 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Nick Piggin93fac702006-03-31 02:29:56 -0800699 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700700 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800701 page_cache_release(page);
702 page = NULL;
703 }
704 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700705 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
707 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800708 /*
709 * Not mapped elsewhere, or swap space full? Free it!
710 * Also recheck PageSwapCache now page is locked (above).
711 */
Nick Piggin93fac702006-03-31 02:29:56 -0800712 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800713 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 delete_from_swap_cache(page);
715 SetPageDirty(page);
716 }
717 unlock_page(page);
718 page_cache_release(page);
719 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800720 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721}
722
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200723#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800724/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800725 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800726 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800727 * @offset - number of the PAGE_SIZE-sized block of the device, starting
728 * from 0, in which the swap header is expected to be located.
729 *
730 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800731 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800732int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800733{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800734 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800735 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800736
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800737 if (device)
738 bdev = bdget(device);
739
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800740 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800741 for (type = 0; type < nr_swapfiles; type++) {
742 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800743
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800744 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800745 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700746
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800747 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800748 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200749 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800750
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800751 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800752 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800753 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800754 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800755 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800756
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800757 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800758 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200759 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800760
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800761 spin_unlock(&swap_lock);
762 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800763 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800764 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800765 }
766 }
767 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800768 if (bdev)
769 bdput(bdev);
770
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800771 return -ENODEV;
772}
773
774/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800775 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
776 * corresponding to given index in swap_info (swap type).
777 */
778sector_t swapdev_block(int type, pgoff_t offset)
779{
780 struct block_device *bdev;
781
782 if ((unsigned int)type >= nr_swapfiles)
783 return 0;
784 if (!(swap_info[type]->flags & SWP_WRITEOK))
785 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800786 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800787}
788
789/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800790 * Return either the total number of swap pages of given type, or the number
791 * of free pages of that type (depending on @free)
792 *
793 * This is needed for software suspend
794 */
795unsigned int count_swap_pages(int type, int free)
796{
797 unsigned int n = 0;
798
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800799 spin_lock(&swap_lock);
800 if ((unsigned int)type < nr_swapfiles) {
801 struct swap_info_struct *sis = swap_info[type];
802
803 if (sis->flags & SWP_WRITEOK) {
804 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800805 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800806 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800807 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800808 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800809 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800810 return n;
811}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800812#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700815 * No need to decide whether this PTE shares the swap entry with others,
816 * just let do_wp_page work it out if a write is requested later - to
817 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800819static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 unsigned long addr, swp_entry_t entry, struct page *page)
821{
Johannes Weiner72835c82012-01-12 17:18:32 -0800822 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800823 spinlock_t *ptl;
824 pte_t *pte;
825 int ret = 1;
826
Johannes Weiner72835c82012-01-12 17:18:32 -0800827 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
828 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800829 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800830 goto out_nolock;
831 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800832
833 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
834 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
835 if (ret > 0)
Johannes Weiner72835c82012-01-12 17:18:32 -0800836 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800837 ret = 0;
838 goto out;
839 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800840
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800841 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800842 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 get_page(page);
844 set_pte_at(vma->vm_mm, addr, pte,
845 pte_mkold(mk_pte(page, vma->vm_page_prot)));
846 page_add_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -0800847 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 swap_free(entry);
849 /*
850 * Move the page to the active list so it is not
851 * immediately swapped out again after swapon.
852 */
853 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800854out:
855 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800856out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800857 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858}
859
860static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
861 unsigned long addr, unsigned long end,
862 swp_entry_t entry, struct page *page)
863{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700865 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800866 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Hugh Dickins044d66c2008-02-07 00:14:04 -0800868 /*
869 * We don't actually need pte lock while scanning for swp_pte: since
870 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
871 * page table while we're scanning; though it could get zapped, and on
872 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
873 * of unmatched parts which look like swp_pte, so unuse_pte must
874 * recheck under pte lock. Scanning without pte lock lets it be
875 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
876 */
877 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 do {
879 /*
880 * swapoff spends a _lot_ of time in this loop!
881 * Test inline before going to call unuse_pte.
882 */
883 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800884 pte_unmap(pte);
885 ret = unuse_pte(vma, pmd, addr, entry, page);
886 if (ret)
887 goto out;
888 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800891 pte_unmap(pte - 1);
892out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800893 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
895
896static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
897 unsigned long addr, unsigned long end,
898 swp_entry_t entry, struct page *page)
899{
900 pmd_t *pmd;
901 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800902 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 pmd = pmd_offset(pud, addr);
905 do {
906 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700907 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800909 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
910 if (ret)
911 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 } while (pmd++, addr = next, addr != end);
913 return 0;
914}
915
916static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
917 unsigned long addr, unsigned long end,
918 swp_entry_t entry, struct page *page)
919{
920 pud_t *pud;
921 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800922 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
924 pud = pud_offset(pgd, addr);
925 do {
926 next = pud_addr_end(addr, end);
927 if (pud_none_or_clear_bad(pud))
928 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800929 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
930 if (ret)
931 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 } while (pud++, addr = next, addr != end);
933 return 0;
934}
935
936static int unuse_vma(struct vm_area_struct *vma,
937 swp_entry_t entry, struct page *page)
938{
939 pgd_t *pgd;
940 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800941 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800943 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 addr = page_address_in_vma(page, vma);
945 if (addr == -EFAULT)
946 return 0;
947 else
948 end = addr + PAGE_SIZE;
949 } else {
950 addr = vma->vm_start;
951 end = vma->vm_end;
952 }
953
954 pgd = pgd_offset(vma->vm_mm, addr);
955 do {
956 next = pgd_addr_end(addr, end);
957 if (pgd_none_or_clear_bad(pgd))
958 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800959 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
960 if (ret)
961 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 } while (pgd++, addr = next, addr != end);
963 return 0;
964}
965
966static int unuse_mm(struct mm_struct *mm,
967 swp_entry_t entry, struct page *page)
968{
969 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800970 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
972 if (!down_read_trylock(&mm->mmap_sem)) {
973 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -0700974 * Activate page so shrink_inactive_list is unlikely to unmap
975 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700977 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 unlock_page(page);
979 down_read(&mm->mmap_sem);
980 lock_page(page);
981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800983 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 break;
985 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800987 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988}
989
990/*
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600991 * Scan swap_map (or frontswap_map if frontswap parameter is true)
992 * from current position to next entry still in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 * Recycle to start on reaching the end, returning 0 when empty.
994 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700995static unsigned int find_next_to_unuse(struct swap_info_struct *si,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -0600996 unsigned int prev, bool frontswap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700998 unsigned int max = si->max;
999 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001000 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
1002 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001003 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 * for whether an entry is in use, not modifying it; false
1005 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001006 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 */
1008 for (;;) {
1009 if (++i >= max) {
1010 if (!prev) {
1011 i = 0;
1012 break;
1013 }
1014 /*
1015 * No entries in use at top of swap_map,
1016 * loop back to start and recheck there.
1017 */
1018 max = prev + 1;
1019 prev = 0;
1020 i = 1;
1021 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001022 if (frontswap) {
1023 if (frontswap_test(si, i))
1024 break;
1025 else
1026 continue;
1027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001029 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 break;
1031 }
1032 return i;
1033}
1034
1035/*
1036 * We completely avoid races by reading each swap page in advance,
1037 * and then search for the process using it. All the necessary
1038 * page table adjustments can then be made atomically.
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001039 *
1040 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1041 * pages_to_unuse==0 means all pages; ignored if frontswap is false
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001043int try_to_unuse(unsigned int type, bool frontswap,
1044 unsigned long pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001046 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001048 unsigned char *swap_map;
1049 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 struct page *page;
1051 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001052 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055 /*
1056 * When searching mms for an entry, a good strategy is to
1057 * start at the first mm we freed the previous entry from
1058 * (though actually we don't notice whether we or coincidence
1059 * freed the entry). Initialize this start_mm with a hold.
1060 *
1061 * A simpler strategy would be to start at the last mm we
1062 * freed the previous entry from; but that would take less
1063 * advantage of mmlist ordering, which clusters forked mms
1064 * together, child after parent. If we race with dup_mmap(), we
1065 * prefer to resolve parent before child, lest we miss entries
1066 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001067 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 */
1069 start_mm = &init_mm;
1070 atomic_inc(&init_mm.mm_users);
1071
1072 /*
1073 * Keep on scanning until all entries have gone. Usually,
1074 * one pass through swap_map is enough, but not necessarily:
1075 * there are races when an instance of an entry might be missed.
1076 */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001077 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 if (signal_pending(current)) {
1079 retval = -EINTR;
1080 break;
1081 }
1082
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001083 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 * Get a page for the entry, using the existing swap
1085 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001086 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 */
1088 swap_map = &si->swap_map[i];
1089 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001090 page = read_swap_cache_async(entry,
1091 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 if (!page) {
1093 /*
1094 * Either swap_duplicate() failed because entry
1095 * has been freed independently, and will not be
1096 * reused since sys_swapoff() already disabled
1097 * allocation from here, or alloc_page() failed.
1098 */
1099 if (!*swap_map)
1100 continue;
1101 retval = -ENOMEM;
1102 break;
1103 }
1104
1105 /*
1106 * Don't hold on to start_mm if it looks like exiting.
1107 */
1108 if (atomic_read(&start_mm->mm_users) == 1) {
1109 mmput(start_mm);
1110 start_mm = &init_mm;
1111 atomic_inc(&init_mm.mm_users);
1112 }
1113
1114 /*
1115 * Wait for and lock page. When do_swap_page races with
1116 * try_to_unuse, do_swap_page can handle the fault much
1117 * faster than try_to_unuse can locate the entry. This
1118 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1119 * defer to do_swap_page in such a case - in some tests,
1120 * do_swap_page and try_to_unuse repeatedly compete.
1121 */
1122 wait_on_page_locked(page);
1123 wait_on_page_writeback(page);
1124 lock_page(page);
1125 wait_on_page_writeback(page);
1126
1127 /*
1128 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001131 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1132 retval = shmem_unuse(entry, page);
1133 /* page has already been unlocked and released */
1134 if (retval < 0)
1135 break;
1136 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001138 if (swap_count(swcount) && start_mm != &init_mm)
1139 retval = unuse_mm(start_mm, entry, page);
1140
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001141 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 int set_start_mm = (*swap_map >= swcount);
1143 struct list_head *p = &start_mm->mmlist;
1144 struct mm_struct *new_start_mm = start_mm;
1145 struct mm_struct *prev_mm = start_mm;
1146 struct mm_struct *mm;
1147
1148 atomic_inc(&new_start_mm->mm_users);
1149 atomic_inc(&prev_mm->mm_users);
1150 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001151 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 (p = p->next) != &start_mm->mmlist) {
1153 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001154 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 spin_unlock(&mmlist_lock);
1157 mmput(prev_mm);
1158 prev_mm = mm;
1159
1160 cond_resched();
1161
1162 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001163 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001165 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001167 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001169
Bo Liu32c5fc12009-11-02 16:50:33 +00001170 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 mmput(new_start_mm);
1172 atomic_inc(&mm->mm_users);
1173 new_start_mm = mm;
1174 set_start_mm = 0;
1175 }
1176 spin_lock(&mmlist_lock);
1177 }
1178 spin_unlock(&mmlist_lock);
1179 mmput(prev_mm);
1180 mmput(start_mm);
1181 start_mm = new_start_mm;
1182 }
1183 if (retval) {
1184 unlock_page(page);
1185 page_cache_release(page);
1186 break;
1187 }
1188
1189 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 * If a reference remains (rare), we would like to leave
1191 * the page in the swap cache; but try_to_unmap could
1192 * then re-duplicate the entry once we drop page lock,
1193 * so we might loop indefinitely; also, that page could
1194 * not be swapped out to other storage meanwhile. So:
1195 * delete from cache even if there's another reference,
1196 * after ensuring that the data has been saved to disk -
1197 * since if the reference remains (rarer), it will be
1198 * read from disk into another page. Splitting into two
1199 * pages would be incorrect if swap supported "shared
1200 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001201 *
1202 * Given how unuse_vma() targets one particular offset
1203 * in an anon_vma, once the anon_vma has been determined,
1204 * this splitting happens to be just what is needed to
1205 * handle where KSM pages have been swapped out: re-reading
1206 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001208 if (swap_count(*swap_map) &&
1209 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 struct writeback_control wbc = {
1211 .sync_mode = WB_SYNC_NONE,
1212 };
1213
1214 swap_writepage(page, &wbc);
1215 lock_page(page);
1216 wait_on_page_writeback(page);
1217 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001218
1219 /*
1220 * It is conceivable that a racing task removed this page from
1221 * swap cache just before we acquired the page lock at the top,
1222 * or while we dropped it in unuse_mm(). The page might even
1223 * be back in swap cache on another swap area: that we must not
1224 * delete, since it may not have been written out to swap yet.
1225 */
1226 if (PageSwapCache(page) &&
1227 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001228 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 /*
1231 * So we could skip searching mms once swap count went
1232 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001233 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 */
1235 SetPageDirty(page);
1236 unlock_page(page);
1237 page_cache_release(page);
1238
1239 /*
1240 * Make sure that we aren't completely killing
1241 * interactive performance.
1242 */
1243 cond_resched();
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001244 if (frontswap && pages_to_unuse > 0) {
1245 if (!--pages_to_unuse)
1246 break;
1247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 }
1249
1250 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 return retval;
1252}
1253
1254/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001255 * After a successful try_to_unuse, if no swap is now in use, we know
1256 * we can empty the mmlist. swap_lock must be held on entry and exit.
1257 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 * added to the mmlist just after page_duplicate - before would be racy.
1259 */
1260static void drain_mmlist(void)
1261{
1262 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001263 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001265 for (type = 0; type < nr_swapfiles; type++)
1266 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 return;
1268 spin_lock(&mmlist_lock);
1269 list_for_each_safe(p, next, &init_mm.mmlist)
1270 list_del_init(p);
1271 spin_unlock(&mmlist_lock);
1272}
1273
1274/*
1275 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001276 * corresponds to page offset for the specified swap entry.
1277 * Note that the type of this function is sector_t, but it returns page offset
1278 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001280static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001282 struct swap_info_struct *sis;
1283 struct swap_extent *start_se;
1284 struct swap_extent *se;
1285 pgoff_t offset;
1286
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001287 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001288 *bdev = sis->bdev;
1289
1290 offset = swp_offset(entry);
1291 start_se = sis->curr_swap_extent;
1292 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
1294 for ( ; ; ) {
1295 struct list_head *lh;
1296
1297 if (se->start_page <= offset &&
1298 offset < (se->start_page + se->nr_pages)) {
1299 return se->start_block + (offset - se->start_page);
1300 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001301 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 se = list_entry(lh, struct swap_extent, list);
1303 sis->curr_swap_extent = se;
1304 BUG_ON(se == start_se); /* It *must* be present */
1305 }
1306}
1307
1308/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001309 * Returns the page offset into bdev for the specified page's swap entry.
1310 */
1311sector_t map_swap_page(struct page *page, struct block_device **bdev)
1312{
1313 swp_entry_t entry;
1314 entry.val = page_private(page);
1315 return map_swap_entry(entry, bdev);
1316}
1317
1318/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 * Free all of a swapdev's extent information
1320 */
1321static void destroy_swap_extents(struct swap_info_struct *sis)
1322{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001323 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 struct swap_extent *se;
1325
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001326 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 struct swap_extent, list);
1328 list_del(&se->list);
1329 kfree(se);
1330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331}
1332
1333/*
1334 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001335 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001337 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 */
1339static int
1340add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1341 unsigned long nr_pages, sector_t start_block)
1342{
1343 struct swap_extent *se;
1344 struct swap_extent *new_se;
1345 struct list_head *lh;
1346
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001347 if (start_page == 0) {
1348 se = &sis->first_swap_extent;
1349 sis->curr_swap_extent = se;
1350 se->start_page = 0;
1351 se->nr_pages = nr_pages;
1352 se->start_block = start_block;
1353 return 1;
1354 } else {
1355 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001357 BUG_ON(se->start_page + se->nr_pages != start_page);
1358 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 /* Merge it */
1360 se->nr_pages += nr_pages;
1361 return 0;
1362 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364
1365 /*
1366 * No merge. Insert a new extent, preserving ordering.
1367 */
1368 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1369 if (new_se == NULL)
1370 return -ENOMEM;
1371 new_se->start_page = start_page;
1372 new_se->nr_pages = nr_pages;
1373 new_se->start_block = start_block;
1374
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001375 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001376 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377}
1378
1379/*
1380 * A `swap extent' is a simple thing which maps a contiguous range of pages
1381 * onto a contiguous range of disk blocks. An ordered list of swap extents
1382 * is built at swapon time and is then used at swap_writepage/swap_readpage
1383 * time for locating where on disk a page belongs.
1384 *
1385 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1386 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1387 * swap files identically.
1388 *
1389 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1390 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1391 * swapfiles are handled *identically* after swapon time.
1392 *
1393 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1394 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1395 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1396 * requirements, they are simply tossed out - we will never use those blocks
1397 * for swapping.
1398 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001399 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1401 * which will scribble on the fs.
1402 *
1403 * The amount of disk space which a single swap extent represents varies.
1404 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1405 * extents in the list. To avoid much list walking, we cache the previous
1406 * search location in `curr_swap_extent', and start new searches from there.
1407 * This is extremely effective. The average number of iterations in
1408 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1409 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001410static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
1412 struct inode *inode;
1413 unsigned blocks_per_page;
1414 unsigned long page_no;
1415 unsigned blkbits;
1416 sector_t probe_block;
1417 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001418 sector_t lowest_block = -1;
1419 sector_t highest_block = 0;
1420 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 int ret;
1422
1423 inode = sis->swap_file->f_mapping->host;
1424 if (S_ISBLK(inode->i_mode)) {
1425 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001426 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001427 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 }
1429
1430 blkbits = inode->i_blkbits;
1431 blocks_per_page = PAGE_SIZE >> blkbits;
1432
1433 /*
1434 * Map all the blocks into the extent list. This code doesn't try
1435 * to be very smart.
1436 */
1437 probe_block = 0;
1438 page_no = 0;
1439 last_block = i_size_read(inode) >> blkbits;
1440 while ((probe_block + blocks_per_page) <= last_block &&
1441 page_no < sis->max) {
1442 unsigned block_in_page;
1443 sector_t first_block;
1444
1445 first_block = bmap(inode, probe_block);
1446 if (first_block == 0)
1447 goto bad_bmap;
1448
1449 /*
1450 * It must be PAGE_SIZE aligned on-disk
1451 */
1452 if (first_block & (blocks_per_page - 1)) {
1453 probe_block++;
1454 goto reprobe;
1455 }
1456
1457 for (block_in_page = 1; block_in_page < blocks_per_page;
1458 block_in_page++) {
1459 sector_t block;
1460
1461 block = bmap(inode, probe_block + block_in_page);
1462 if (block == 0)
1463 goto bad_bmap;
1464 if (block != first_block + block_in_page) {
1465 /* Discontiguity */
1466 probe_block++;
1467 goto reprobe;
1468 }
1469 }
1470
Hugh Dickins53092a72005-09-03 15:54:34 -07001471 first_block >>= (PAGE_SHIFT - blkbits);
1472 if (page_no) { /* exclude the header page */
1473 if (first_block < lowest_block)
1474 lowest_block = first_block;
1475 if (first_block > highest_block)
1476 highest_block = first_block;
1477 }
1478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 /*
1480 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1481 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001482 ret = add_swap_extent(sis, page_no, 1, first_block);
1483 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001485 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 page_no++;
1487 probe_block += blocks_per_page;
1488reprobe:
1489 continue;
1490 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001491 ret = nr_extents;
1492 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001494 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001496 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001498out:
1499 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500bad_bmap:
1501 printk(KERN_ERR "swapon: swapfile has holes\n");
1502 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001503 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
1505
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001506static void enable_swap_info(struct swap_info_struct *p, int prio,
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001507 unsigned char *swap_map,
1508 unsigned long *frontswap_map)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001509{
1510 int i, prev;
1511
1512 spin_lock(&swap_lock);
1513 if (prio >= 0)
1514 p->prio = prio;
1515 else
1516 p->prio = --least_priority;
1517 p->swap_map = swap_map;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001518 frontswap_map_set(p, frontswap_map);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001519 p->flags |= SWP_WRITEOK;
1520 nr_swap_pages += p->pages;
1521 total_swap_pages += p->pages;
1522
1523 /* insert swap space into swap_list: */
1524 prev = -1;
1525 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1526 if (p->prio >= swap_info[i]->prio)
1527 break;
1528 prev = i;
1529 }
1530 p->next = i;
1531 if (prev < 0)
1532 swap_list.head = swap_list.next = p->type;
1533 else
1534 swap_info[prev]->next = p->type;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001535 frontswap_init(p->type);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001536 spin_unlock(&swap_lock);
1537}
1538
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001539SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001541 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001542 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 struct file *swap_file, *victim;
1544 struct address_space *mapping;
1545 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001546 char *pathname;
David Rientjes72788c32011-05-24 17:11:40 -07001547 int oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 int i, type, prev;
1549 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 if (!capable(CAP_SYS_ADMIN))
1552 return -EPERM;
1553
Al Viro191c5422012-02-13 03:58:52 +00001554 BUG_ON(!current->mm);
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 pathname = getname(specialfile);
1557 err = PTR_ERR(pathname);
1558 if (IS_ERR(pathname))
1559 goto out;
1560
1561 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1562 putname(pathname);
1563 err = PTR_ERR(victim);
1564 if (IS_ERR(victim))
1565 goto out;
1566
1567 mapping = victim->f_mapping;
1568 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001569 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001570 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1571 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001572 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 if (p->swap_file->f_mapping == mapping)
1574 break;
1575 }
1576 prev = type;
1577 }
1578 if (type < 0) {
1579 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001580 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 goto out_dput;
1582 }
Al Viro191c5422012-02-13 03:58:52 +00001583 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 vm_unacct_memory(p->pages);
1585 else {
1586 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001587 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 goto out_dput;
1589 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001590 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001592 else
1593 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (type == swap_list.next) {
1595 /* just pick something that's safe... */
1596 swap_list.next = swap_list.head;
1597 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001598 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001599 for (i = p->next; i >= 0; i = swap_info[i]->next)
1600 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001601 least_priority++;
1602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 nr_swap_pages -= p->pages;
1604 total_swap_pages -= p->pages;
1605 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001606 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001607
David Rientjes72788c32011-05-24 17:11:40 -07001608 oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001609 err = try_to_unuse(type, false, 0); /* force all pages to be unused */
David Rientjes43362a42011-10-31 17:07:18 -07001610 compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX, oom_score_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (err) {
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001613 /*
1614 * reading p->prio and p->swap_map outside the lock is
1615 * safe here because only sys_swapon and sys_swapoff
1616 * change them, and there can be no other sys_swapon or
1617 * sys_swapoff for this swap_info_struct at this point.
1618 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 /* re-insert swap space back into swap_list */
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001620 enable_swap_info(p, p->prio, p->swap_map, frontswap_map_get(p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 goto out_dput;
1622 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001623
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001624 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001625 if (p->flags & SWP_CONTINUED)
1626 free_swap_count_continuations(p);
1627
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001628 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001629 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001631
1632 /* wait for anyone still in scan_swap_map */
1633 p->highest_bit = 0; /* cuts scans short */
1634 while (p->flags >= SWP_SCANNING) {
1635 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001636 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001637 spin_lock(&swap_lock);
1638 }
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 swap_file = p->swap_file;
1641 p->swap_file = NULL;
1642 p->max = 0;
1643 swap_map = p->swap_map;
1644 p->swap_map = NULL;
1645 p->flags = 0;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001646 frontswap_invalidate_area(type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001647 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001648 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 vfree(swap_map);
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06001650 vfree(frontswap_map_get(p));
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001651 /* Destroy swap account informatin */
1652 swap_cgroup_swapoff(type);
1653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 inode = mapping->host;
1655 if (S_ISBLK(inode->i_mode)) {
1656 struct block_device *bdev = I_BDEV(inode);
1657 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001658 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001660 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001662 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
1664 filp_close(swap_file, NULL);
1665 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001666 atomic_inc(&proc_poll_event);
1667 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669out_dput:
1670 filp_close(victim, NULL);
1671out:
1672 return err;
1673}
1674
1675#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001676static unsigned swaps_poll(struct file *file, poll_table *wait)
1677{
Kay Sieversf1514632011-07-12 20:48:39 +02001678 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001679
1680 poll_wait(file, &proc_poll_wait, wait);
1681
Kay Sieversf1514632011-07-12 20:48:39 +02001682 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1683 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001684 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1685 }
1686
1687 return POLLIN | POLLRDNORM;
1688}
1689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690/* iterator */
1691static void *swap_start(struct seq_file *swap, loff_t *pos)
1692{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001693 struct swap_info_struct *si;
1694 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 loff_t l = *pos;
1696
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001697 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001699 if (!l)
1700 return SEQ_START_TOKEN;
1701
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001702 for (type = 0; type < nr_swapfiles; type++) {
1703 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1704 si = swap_info[type];
1705 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001707 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001708 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 }
1710
1711 return NULL;
1712}
1713
1714static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1715{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001716 struct swap_info_struct *si = v;
1717 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001719 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001720 type = 0;
1721 else
1722 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001723
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001724 for (; type < nr_swapfiles; type++) {
1725 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1726 si = swap_info[type];
1727 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 continue;
1729 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001730 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 }
1732
1733 return NULL;
1734}
1735
1736static void swap_stop(struct seq_file *swap, void *v)
1737{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001738 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739}
1740
1741static int swap_show(struct seq_file *swap, void *v)
1742{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001743 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 struct file *file;
1745 int len;
1746
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001747 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001748 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1749 return 0;
1750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001752 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001753 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001754 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001755 len < 40 ? 40 - len : 1, " ",
1756 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001758 si->pages << (PAGE_SHIFT - 10),
1759 si->inuse_pages << (PAGE_SHIFT - 10),
1760 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 return 0;
1762}
1763
Helge Deller15ad7cd2006-12-06 20:40:36 -08001764static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 .start = swap_start,
1766 .next = swap_next,
1767 .stop = swap_stop,
1768 .show = swap_show
1769};
1770
1771static int swaps_open(struct inode *inode, struct file *file)
1772{
Kay Sieversf1514632011-07-12 20:48:39 +02001773 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001774 int ret;
1775
Kay Sievers66d7dd52010-10-26 14:22:06 -07001776 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02001777 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07001778 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001779
Kay Sieversf1514632011-07-12 20:48:39 +02001780 seq = file->private_data;
1781 seq->poll_event = atomic_read(&proc_poll_event);
1782 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783}
1784
Helge Deller15ad7cd2006-12-06 20:40:36 -08001785static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 .open = swaps_open,
1787 .read = seq_read,
1788 .llseek = seq_lseek,
1789 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001790 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791};
1792
1793static int __init procswaps_init(void)
1794{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001795 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 return 0;
1797}
1798__initcall(procswaps_init);
1799#endif /* CONFIG_PROC_FS */
1800
Jan Beulich17963162008-12-16 11:35:24 +00001801#ifdef MAX_SWAPFILES_CHECK
1802static int __init max_swapfiles_check(void)
1803{
1804 MAX_SWAPFILES_CHECK();
1805 return 0;
1806}
1807late_initcall(max_swapfiles_check);
1808#endif
1809
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001810static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001812 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001814
1815 p = kzalloc(sizeof(*p), GFP_KERNEL);
1816 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001817 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001818
Hugh Dickins5d337b92005-09-03 15:54:41 -07001819 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001820 for (type = 0; type < nr_swapfiles; type++) {
1821 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001823 }
Christoph Lameter06972122006-06-23 02:03:35 -07001824 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001825 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001826 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07001827 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001829 if (type >= nr_swapfiles) {
1830 p->type = type;
1831 swap_info[type] = p;
1832 /*
1833 * Write swap_info[type] before nr_swapfiles, in case a
1834 * racing procfs swap_start() or swap_next() is reading them.
1835 * (We never shrink nr_swapfiles, we never free this entry.)
1836 */
1837 smp_wmb();
1838 nr_swapfiles++;
1839 } else {
1840 kfree(p);
1841 p = swap_info[type];
1842 /*
1843 * Do not memset this entry: a racing procfs swap_next()
1844 * would be relying on p->type to remain valid.
1845 */
1846 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001847 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001850 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001851
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001852 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001853}
1854
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001855static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1856{
1857 int error;
1858
1859 if (S_ISBLK(inode->i_mode)) {
1860 p->bdev = bdgrab(I_BDEV(inode));
1861 error = blkdev_get(p->bdev,
1862 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1863 sys_swapon);
1864 if (error < 0) {
1865 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001866 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001867 }
1868 p->old_block_size = block_size(p->bdev);
1869 error = set_blocksize(p->bdev, PAGE_SIZE);
1870 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001871 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001872 p->flags |= SWP_BLKDEV;
1873 } else if (S_ISREG(inode->i_mode)) {
1874 p->bdev = inode->i_sb->s_bdev;
1875 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001876 if (IS_SWAPFILE(inode))
1877 return -EBUSY;
1878 } else
1879 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001880
1881 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001882}
1883
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001884static unsigned long read_swap_header(struct swap_info_struct *p,
1885 union swap_header *swap_header,
1886 struct inode *inode)
1887{
1888 int i;
1889 unsigned long maxpages;
1890 unsigned long swapfilepages;
1891
1892 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1893 printk(KERN_ERR "Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001894 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001895 }
1896
1897 /* swap partition endianess hack... */
1898 if (swab32(swap_header->info.version) == 1) {
1899 swab32s(&swap_header->info.version);
1900 swab32s(&swap_header->info.last_page);
1901 swab32s(&swap_header->info.nr_badpages);
1902 for (i = 0; i < swap_header->info.nr_badpages; i++)
1903 swab32s(&swap_header->info.badpages[i]);
1904 }
1905 /* Check the swap header's sub-version */
1906 if (swap_header->info.version != 1) {
1907 printk(KERN_WARNING
1908 "Unable to handle swap header version %d\n",
1909 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001910 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001911 }
1912
1913 p->lowest_bit = 1;
1914 p->cluster_next = 1;
1915 p->cluster_nr = 0;
1916
1917 /*
1918 * Find out how many pages are allowed for a single swap
Hugh Dickins9b15b812012-06-15 17:55:50 -07001919 * device. There are two limiting factors: 1) the number
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001920 * of bits for the swap offset in the swp_entry_t type, and
1921 * 2) the number of bits in the swap pte as defined by the
Hugh Dickins9b15b812012-06-15 17:55:50 -07001922 * different architectures. In order to find the
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001923 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001924 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001925 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001926 * offset is extracted. This will mask all the bits from
1927 * the initial ~0UL mask that can't be encoded in either
1928 * the swp_entry_t or the architecture definition of a
Hugh Dickins9b15b812012-06-15 17:55:50 -07001929 * swap pte.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001930 */
1931 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickins9b15b812012-06-15 17:55:50 -07001932 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001933 if (maxpages > swap_header->info.last_page) {
1934 maxpages = swap_header->info.last_page + 1;
1935 /* p->max is an unsigned int: don't overflow it */
1936 if ((unsigned int)maxpages == 0)
1937 maxpages = UINT_MAX;
1938 }
1939 p->highest_bit = maxpages - 1;
1940
1941 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001942 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001943 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1944 if (swapfilepages && maxpages > swapfilepages) {
1945 printk(KERN_WARNING
1946 "Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001947 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001948 }
1949 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001950 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001951 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001952 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001953
1954 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001955}
1956
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001957static int setup_swap_map_and_extents(struct swap_info_struct *p,
1958 union swap_header *swap_header,
1959 unsigned char *swap_map,
1960 unsigned long maxpages,
1961 sector_t *span)
1962{
1963 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001964 unsigned int nr_good_pages;
1965 int nr_extents;
1966
1967 nr_good_pages = maxpages - 1; /* omit header page */
1968
1969 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1970 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001971 if (page_nr == 0 || page_nr > swap_header->info.last_page)
1972 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001973 if (page_nr < maxpages) {
1974 swap_map[page_nr] = SWAP_MAP_BAD;
1975 nr_good_pages--;
1976 }
1977 }
1978
1979 if (nr_good_pages) {
1980 swap_map[0] = SWAP_MAP_BAD;
1981 p->max = maxpages;
1982 p->pages = nr_good_pages;
1983 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001984 if (nr_extents < 0)
1985 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001986 nr_good_pages = p->pages;
1987 }
1988 if (!nr_good_pages) {
1989 printk(KERN_WARNING "Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001990 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001991 }
1992
1993 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001994}
1995
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001996SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
1997{
1998 struct swap_info_struct *p;
Cesar Eduardo Barros28b36bd2011-03-22 16:33:20 -07001999 char *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002000 struct file *swap_file = NULL;
2001 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002002 int i;
2003 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002004 int error;
2005 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002006 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002007 sector_t span;
2008 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002009 unsigned char *swap_map = NULL;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002010 unsigned long *frontswap_map = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002011 struct page *page = NULL;
2012 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002013
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002014 if (swap_flags & ~SWAP_FLAGS_VALID)
2015 return -EINVAL;
2016
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002017 if (!capable(CAP_SYS_ADMIN))
2018 return -EPERM;
2019
2020 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002021 if (IS_ERR(p))
2022 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002023
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002026 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002028 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 }
2030 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002032 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002034 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 }
2036
2037 p->swap_file = swap_file;
2038 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002041 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002043 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002045 if (mapping == q->swap_file->f_mapping) {
2046 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
2050
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002051 inode = mapping->host;
2052 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002053 error = claim_swapfile(p, inode);
2054 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 /*
2058 * Read the swap header.
2059 */
2060 if (!mapping->a_ops->readpage) {
2061 error = -EINVAL;
2062 goto bad_swap;
2063 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002064 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 if (IS_ERR(page)) {
2066 error = PTR_ERR(page);
2067 goto bad_swap;
2068 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002069 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002071 maxpages = read_swap_header(p, swap_header, inode);
2072 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 error = -EINVAL;
2074 goto bad_swap;
2075 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002076
Hugh Dickins81e33972009-01-06 14:39:49 -08002077 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002078 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002079 if (!swap_map) {
2080 error = -ENOMEM;
2081 goto bad_swap;
2082 }
2083
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002084 error = swap_cgroup_swapon(p->type, maxpages);
2085 if (error)
2086 goto bad_swap;
2087
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002088 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2089 maxpages, &span);
2090 if (unlikely(nr_extents < 0)) {
2091 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 goto bad_swap;
2093 }
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002094 /* frontswap enabled? set up bit-per-page map for frontswap */
2095 if (frontswap_enabled)
2096 frontswap_map = vzalloc(maxpages / sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002098 if (p->bdev) {
2099 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2100 p->flags |= SWP_SOLIDSTATE;
2101 p->cluster_next = 1 + (random32() % p->highest_bit);
2102 }
Shaohua Li052b1982012-03-21 16:34:17 -07002103 if ((swap_flags & SWAP_FLAG_DISCARD) && discard_swap(p) == 0)
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002104 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002105 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002106
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002107 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002108 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002109 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002110 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002111 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002112 enable_swap_info(p, prio, swap_map, frontswap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002113
2114 printk(KERN_INFO "Adding %uk swap on %s. "
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002115 "Priority:%d extents:%d across:%lluk %s%s%s\n",
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002116 p->pages<<(PAGE_SHIFT-10), name, p->prio,
2117 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2118 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
Dan Magenheimer38b5faf2012-04-09 17:08:06 -06002119 (p->flags & SWP_DISCARDABLE) ? "D" : "",
2120 (frontswap_map) ? "FS" : "");
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002121
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002122 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002123 atomic_inc(&proc_poll_event);
2124 wake_up_interruptible(&proc_poll_wait);
2125
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002126 if (S_ISREG(inode->i_mode))
2127 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 error = 0;
2129 goto out;
2130bad_swap:
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002131 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002132 set_blocksize(p->bdev, p->old_block_size);
2133 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002135 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002136 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002137 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002140 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 vfree(swap_map);
Mel Gorman52c50562011-03-22 16:30:08 -07002142 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002143 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002144 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002145 inode = NULL;
2146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149out:
2150 if (page && !IS_ERR(page)) {
2151 kunmap(page);
2152 page_cache_release(page);
2153 }
2154 if (name)
2155 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002156 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002157 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 return error;
2159}
2160
2161void si_swapinfo(struct sysinfo *val)
2162{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002163 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 unsigned long nr_to_be_unused = 0;
2165
Hugh Dickins5d337b92005-09-03 15:54:41 -07002166 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002167 for (type = 0; type < nr_swapfiles; type++) {
2168 struct swap_info_struct *si = swap_info[type];
2169
2170 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2171 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 }
2173 val->freeswap = nr_swap_pages + nr_to_be_unused;
2174 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002175 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176}
2177
2178/*
2179 * Verify that a swap entry is valid and increment its swap map count.
2180 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002181 * Returns error code in following case.
2182 * - success -> 0
2183 * - swp_entry is invalid -> EINVAL
2184 * - swp_entry is migration entry -> EINVAL
2185 * - swap-cache reference is requested but there is already one. -> EEXIST
2186 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002187 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002189static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002191 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002193 unsigned char count;
2194 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002195 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
Andi Kleena7420aa2009-09-16 11:50:05 +02002197 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002198 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002199
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 type = swp_type(entry);
2201 if (type >= nr_swapfiles)
2202 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002203 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 offset = swp_offset(entry);
2205
Hugh Dickins5d337b92005-09-03 15:54:41 -07002206 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002207 if (unlikely(offset >= p->max))
2208 goto unlock_out;
2209
Hugh Dickins253d5532009-12-14 17:58:44 -08002210 count = p->swap_map[offset];
2211 has_cache = count & SWAP_HAS_CACHE;
2212 count &= ~SWAP_HAS_CACHE;
2213 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002214
Hugh Dickins253d5532009-12-14 17:58:44 -08002215 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002216
2217 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002218 if (!has_cache && count)
2219 has_cache = SWAP_HAS_CACHE;
2220 else if (has_cache) /* someone else added cache */
2221 err = -EEXIST;
2222 else /* no users remaining */
2223 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002224
2225 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002226
Hugh Dickins570a335b2009-12-14 17:58:46 -08002227 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2228 count += usage;
2229 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002230 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002231 else if (swap_count_continued(p, offset, count))
2232 count = COUNT_CONTINUED;
2233 else
2234 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002235 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002236 err = -ENOENT; /* unused swap entry */
2237
2238 p->swap_map[offset] = count | has_cache;
2239
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002240unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002241 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002243 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245bad_file:
2246 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2247 goto out;
2248}
Hugh Dickins253d5532009-12-14 17:58:44 -08002249
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002250/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002251 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2252 * (in which case its reference count is never incremented).
2253 */
2254void swap_shmem_alloc(swp_entry_t entry)
2255{
2256 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2257}
2258
2259/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002260 * Increase reference count of swap entry by 1.
2261 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2262 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2263 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2264 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002265 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002266int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002267{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002268 int err = 0;
2269
2270 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2271 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2272 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002273}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002275/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002276 * @entry: swap entry for which we allocate swap cache.
2277 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002278 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002279 * This can return error codes. Returns 0 at success.
2280 * -EBUSY means there is a swap cache.
2281 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002282 */
2283int swapcache_prepare(swp_entry_t entry)
2284{
Hugh Dickins253d5532009-12-14 17:58:44 -08002285 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002286}
2287
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002289 * add_swap_count_continuation - called when a swap count is duplicated
2290 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2291 * page of the original vmalloc'ed swap_map, to hold the continuation count
2292 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2293 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2294 *
2295 * These continuation pages are seldom referenced: the common paths all work
2296 * on the original swap_map, only referring to a continuation page when the
2297 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2298 *
2299 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2300 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2301 * can be called after dropping locks.
2302 */
2303int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2304{
2305 struct swap_info_struct *si;
2306 struct page *head;
2307 struct page *page;
2308 struct page *list_page;
2309 pgoff_t offset;
2310 unsigned char count;
2311
2312 /*
2313 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2314 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2315 */
2316 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2317
2318 si = swap_info_get(entry);
2319 if (!si) {
2320 /*
2321 * An acceptable race has occurred since the failing
2322 * __swap_duplicate(): the swap entry has been freed,
2323 * perhaps even the whole swap_map cleared for swapoff.
2324 */
2325 goto outer;
2326 }
2327
2328 offset = swp_offset(entry);
2329 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2330
2331 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2332 /*
2333 * The higher the swap count, the more likely it is that tasks
2334 * will race to add swap count continuation: we need to avoid
2335 * over-provisioning.
2336 */
2337 goto out;
2338 }
2339
2340 if (!page) {
2341 spin_unlock(&swap_lock);
2342 return -ENOMEM;
2343 }
2344
2345 /*
2346 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2347 * no architecture is using highmem pages for kernel pagetables: so it
2348 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2349 */
2350 head = vmalloc_to_page(si->swap_map + offset);
2351 offset &= ~PAGE_MASK;
2352
2353 /*
2354 * Page allocation does not initialize the page's lru field,
2355 * but it does always reset its private field.
2356 */
2357 if (!page_private(head)) {
2358 BUG_ON(count & COUNT_CONTINUED);
2359 INIT_LIST_HEAD(&head->lru);
2360 set_page_private(head, SWP_CONTINUED);
2361 si->flags |= SWP_CONTINUED;
2362 }
2363
2364 list_for_each_entry(list_page, &head->lru, lru) {
2365 unsigned char *map;
2366
2367 /*
2368 * If the previous map said no continuation, but we've found
2369 * a continuation page, free our allocation and use this one.
2370 */
2371 if (!(count & COUNT_CONTINUED))
2372 goto out;
2373
Cong Wang9b04c5f2011-11-25 23:14:39 +08002374 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002375 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002376 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002377
2378 /*
2379 * If this continuation count now has some space in it,
2380 * free our allocation and use this one.
2381 */
2382 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2383 goto out;
2384 }
2385
2386 list_add_tail(&page->lru, &head->lru);
2387 page = NULL; /* now it's attached, don't free it */
2388out:
2389 spin_unlock(&swap_lock);
2390outer:
2391 if (page)
2392 __free_page(page);
2393 return 0;
2394}
2395
2396/*
2397 * swap_count_continued - when the original swap_map count is incremented
2398 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2399 * into, carry if so, or else fail until a new continuation page is allocated;
2400 * when the original swap_map count is decremented from 0 with continuation,
2401 * borrow from the continuation and report whether it still holds more.
2402 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2403 */
2404static bool swap_count_continued(struct swap_info_struct *si,
2405 pgoff_t offset, unsigned char count)
2406{
2407 struct page *head;
2408 struct page *page;
2409 unsigned char *map;
2410
2411 head = vmalloc_to_page(si->swap_map + offset);
2412 if (page_private(head) != SWP_CONTINUED) {
2413 BUG_ON(count & COUNT_CONTINUED);
2414 return false; /* need to add count continuation */
2415 }
2416
2417 offset &= ~PAGE_MASK;
2418 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002419 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002420
2421 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2422 goto init_map; /* jump over SWAP_CONT_MAX checks */
2423
2424 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2425 /*
2426 * Think of how you add 1 to 999
2427 */
2428 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002429 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002430 page = list_entry(page->lru.next, struct page, lru);
2431 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002432 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002433 }
2434 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002435 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002436 page = list_entry(page->lru.next, struct page, lru);
2437 if (page == head)
2438 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002439 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002440init_map: *map = 0; /* we didn't zero the page */
2441 }
2442 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002443 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002444 page = list_entry(page->lru.prev, struct page, lru);
2445 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002446 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002447 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002448 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002449 page = list_entry(page->lru.prev, struct page, lru);
2450 }
2451 return true; /* incremented */
2452
2453 } else { /* decrementing */
2454 /*
2455 * Think of how you subtract 1 from 1000
2456 */
2457 BUG_ON(count != COUNT_CONTINUED);
2458 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002459 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002460 page = list_entry(page->lru.next, struct page, lru);
2461 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002462 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002463 }
2464 BUG_ON(*map == 0);
2465 *map -= 1;
2466 if (*map == 0)
2467 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002468 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002469 page = list_entry(page->lru.prev, struct page, lru);
2470 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002471 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002472 *map = SWAP_CONT_MAX | count;
2473 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002474 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002475 page = list_entry(page->lru.prev, struct page, lru);
2476 }
2477 return count == COUNT_CONTINUED;
2478 }
2479}
2480
2481/*
2482 * free_swap_count_continuations - swapoff free all the continuation pages
2483 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2484 */
2485static void free_swap_count_continuations(struct swap_info_struct *si)
2486{
2487 pgoff_t offset;
2488
2489 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2490 struct page *head;
2491 head = vmalloc_to_page(si->swap_map + offset);
2492 if (page_private(head)) {
2493 struct list_head *this, *next;
2494 list_for_each_safe(this, next, &head->lru) {
2495 struct page *page;
2496 page = list_entry(this, struct page, lru);
2497 list_del(this);
2498 __free_page(page);
2499 }
2500 }
2501 }
2502}