blob: 07a458d72fa880f5adc366b8acf03e610841880e [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>
17#include <linux/shm.h>
18#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>
24#include <linux/module.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080025#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/rmap.h>
27#include <linux/security.h>
28#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080029#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080030#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080032#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070033#include <linux/poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35#include <asm/pgtable.h>
36#include <asm/tlbflush.h>
37#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080038#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Hugh Dickins570a335b2009-12-14 17:58:46 -080040static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
41 unsigned char);
42static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080043static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080044
Adrian Bunk7c363b82008-07-25 19:46:24 -070045static DEFINE_SPINLOCK(swap_lock);
46static unsigned int nr_swapfiles;
Hugh Dickinsb9627162009-01-06 14:39:41 -080047long nr_swap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070049static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Linus Torvalds1da177e2005-04-16 15:20:36 -070051static const char Bad_file[] = "Bad swap file entry ";
52static const char Unused_file[] = "Unused swap file entry ";
53static const char Bad_offset[] = "Bad swap offset entry ";
54static const char Unused_offset[] = "Unused swap offset entry ";
55
Adrian Bunk7c363b82008-07-25 19:46:24 -070056static struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Hugh Dickinsefa90a92009-12-14 17:58:41 -080058static struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Ingo Molnarfc0abb12006-01-18 17:42:33 -080060static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Kay Sievers66d7dd52010-10-26 14:22:06 -070062static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
63/* Activity counter to indicate that a swapon or swapoff has occurred */
64static atomic_t proc_poll_event = ATOMIC_INIT(0);
65
Hugh Dickins8d69aae2009-12-14 17:58:45 -080066static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070067{
Hugh Dickins570a335b2009-12-14 17:58:46 -080068 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070069}
70
Hugh Dickinsefa90a92009-12-14 17:58:41 -080071/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070072static int
73__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
74{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080075 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070076 struct page *page;
77 int ret = 0;
78
79 page = find_get_page(&swapper_space, entry.val);
80 if (!page)
81 return 0;
82 /*
83 * This function is called from scan_swap_map() and it's called
84 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
85 * We have to use trylock for avoiding deadlock. This is a special
86 * case and you should use try_to_free_swap() with explicit lock_page()
87 * in usual operations.
88 */
89 if (trylock_page(page)) {
90 ret = try_to_free_swap(page);
91 unlock_page(page);
92 }
93 page_cache_release(page);
94 return ret;
95}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070096
Linus Torvalds1da177e2005-04-16 15:20:36 -070097/*
98 * We need this because the bdev->unplug_fn can sleep and we cannot
Hugh Dickins5d337b92005-09-03 15:54:41 -070099 * hold swap_lock while calling the unplug_fn. And swap_lock
Ingo Molnarfc0abb12006-01-18 17:42:33 -0800100 * cannot be turned into a mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
102static DECLARE_RWSEM(swap_unplug_sem);
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
105{
106 swp_entry_t entry;
107
108 down_read(&swap_unplug_sem);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700109 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 if (PageSwapCache(page)) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800111 struct block_device *bdev = swap_info[swp_type(entry)]->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 struct backing_dev_info *bdi;
113
114 /*
115 * If the page is removed from swapcache from under us (with a
116 * racy try_to_unuse/swapoff) we need an additional reference
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700117 * count to avoid reading garbage from page_private(page) above.
118 * If the WARN_ON triggers during a swapoff it maybe the race
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 * condition and it's harmless. However if it triggers without
120 * swapoff it signals a problem.
121 */
122 WARN_ON(page_count(page) <= 1);
123
124 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
McMullan, Jasonba323112005-05-16 21:53:40 -0700125 blk_run_backing_dev(bdi, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 }
127 up_read(&swap_unplug_sem);
128}
129
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800130/*
131 * swapon tell device that all the old swap contents can be discarded,
132 * to allow the swap device to optimize its wear-levelling.
133 */
134static int discard_swap(struct swap_info_struct *si)
135{
136 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800137 sector_t start_block;
138 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800139 int err = 0;
140
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800141 /* Do not discard the swap header page! */
142 se = &si->first_swap_extent;
143 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
144 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
145 if (nr_blocks) {
146 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200147 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800148 if (err)
149 return err;
150 cond_resched();
151 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800152
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800153 list_for_each_entry(se, &si->first_swap_extent.list, list) {
154 start_block = se->start_block << (PAGE_SHIFT - 9);
155 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800156
157 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200158 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800159 if (err)
160 break;
161
162 cond_resched();
163 }
164 return err; /* That will often be -EOPNOTSUPP */
165}
166
Hugh Dickins7992fde2009-01-06 14:39:53 -0800167/*
168 * swap allocation tell device that a cluster of swap can now be discarded,
169 * to allow the swap device to optimize its wear-levelling.
170 */
171static void discard_swap_cluster(struct swap_info_struct *si,
172 pgoff_t start_page, pgoff_t nr_pages)
173{
174 struct swap_extent *se = si->curr_swap_extent;
175 int found_extent = 0;
176
177 while (nr_pages) {
178 struct list_head *lh;
179
180 if (se->start_page <= start_page &&
181 start_page < se->start_page + se->nr_pages) {
182 pgoff_t offset = start_page - se->start_page;
183 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800184 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800185
186 if (nr_blocks > nr_pages)
187 nr_blocks = nr_pages;
188 start_page += nr_blocks;
189 nr_pages -= nr_blocks;
190
191 if (!found_extent++)
192 si->curr_swap_extent = se;
193
194 start_block <<= PAGE_SHIFT - 9;
195 nr_blocks <<= PAGE_SHIFT - 9;
196 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200197 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800198 break;
199 }
200
201 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800202 se = list_entry(lh, struct swap_extent, list);
203 }
204}
205
206static int wait_for_discard(void *word)
207{
208 schedule();
209 return 0;
210}
211
Hugh Dickins048c27f2005-09-03 15:54:40 -0700212#define SWAPFILE_CLUSTER 256
213#define LATENCY_LIMIT 256
214
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700215static inline unsigned long scan_swap_map(struct swap_info_struct *si,
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800216 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800218 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800219 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800220 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700221 int latency_ration = LATENCY_LIMIT;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800222 int found_free_cluster = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800224 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700225 * We try to cluster swap pages by allocating them sequentially
226 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
227 * way, however, we resort to first-free allocation, starting
228 * a new cluster. This prevents us from scattering swap pages
229 * all over the entire swap partition, so that we reduce
230 * overall disk seek times between swap pages. -- sct
231 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800232 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700233 */
234
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700235 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800236 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800237
238 if (unlikely(!si->cluster_nr--)) {
239 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
240 si->cluster_nr = SWAPFILE_CLUSTER - 1;
241 goto checks;
242 }
Hugh Dickins7992fde2009-01-06 14:39:53 -0800243 if (si->flags & SWP_DISCARDABLE) {
244 /*
245 * Start range check on racing allocations, in case
246 * they overlap the cluster we eventually decide on
247 * (we scan without swap_lock to allow preemption).
248 * It's hardly conceivable that cluster_nr could be
249 * wrapped during our scan, but don't depend on it.
250 */
251 if (si->lowest_alloc)
252 goto checks;
253 si->lowest_alloc = si->max;
254 si->highest_alloc = 0;
255 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700256 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700257
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800258 /*
259 * If seek is expensive, start searching for new cluster from
260 * start of partition, to minimize the span of allocated swap.
261 * But if seek is cheap, search from our current position, so
262 * that swap is allocated from all over the partition: if the
263 * Flash Translation Layer only remaps within limited zones,
264 * we don't want to wear out the first zone too quickly.
265 */
266 if (!(si->flags & SWP_SOLIDSTATE))
267 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700268 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
269
270 /* Locate the first empty (unaligned) cluster */
271 for (; last_in_cluster <= si->highest_bit; offset++) {
272 if (si->swap_map[offset])
273 last_in_cluster = offset + SWAPFILE_CLUSTER;
274 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700275 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800276 offset -= SWAPFILE_CLUSTER - 1;
277 si->cluster_next = offset;
278 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800279 found_free_cluster = 1;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800280 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700282 if (unlikely(--latency_ration < 0)) {
283 cond_resched();
284 latency_ration = LATENCY_LIMIT;
285 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700286 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800287
288 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800289 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
290
291 /* Locate the first empty (unaligned) cluster */
292 for (; last_in_cluster < scan_base; offset++) {
293 if (si->swap_map[offset])
294 last_in_cluster = offset + SWAPFILE_CLUSTER;
295 else if (offset == last_in_cluster) {
296 spin_lock(&swap_lock);
297 offset -= SWAPFILE_CLUSTER - 1;
298 si->cluster_next = offset;
299 si->cluster_nr = SWAPFILE_CLUSTER - 1;
300 found_free_cluster = 1;
301 goto checks;
302 }
303 if (unlikely(--latency_ration < 0)) {
304 cond_resched();
305 latency_ration = LATENCY_LIMIT;
306 }
307 }
308
309 offset = scan_base;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700310 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800311 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800312 si->lowest_alloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700314
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800315checks:
316 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700317 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700318 if (!si->highest_bit)
319 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800320 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800321 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700322
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700323 /* reuse swap entry of cache-only swap if not busy. */
324 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700325 int swap_was_freed;
326 spin_unlock(&swap_lock);
327 swap_was_freed = __try_to_reclaim_swap(si, offset);
328 spin_lock(&swap_lock);
329 /* entry was freed successfully, try to use this again */
330 if (swap_was_freed)
331 goto checks;
332 goto scan; /* check next one */
333 }
334
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800335 if (si->swap_map[offset])
336 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700337
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800338 if (offset == si->lowest_bit)
339 si->lowest_bit++;
340 if (offset == si->highest_bit)
341 si->highest_bit--;
342 si->inuse_pages++;
343 if (si->inuse_pages == si->pages) {
344 si->lowest_bit = si->max;
345 si->highest_bit = 0;
346 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800347 si->swap_map[offset] = usage;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800348 si->cluster_next = offset + 1;
349 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800350
351 if (si->lowest_alloc) {
352 /*
353 * Only set when SWP_DISCARDABLE, and there's a scan
354 * for a free cluster in progress or just completed.
355 */
356 if (found_free_cluster) {
357 /*
358 * To optimize wear-levelling, discard the
359 * old data of the cluster, taking care not to
360 * discard any of its pages that have already
361 * been allocated by racing tasks (offset has
362 * already stepped over any at the beginning).
363 */
364 if (offset < si->highest_alloc &&
365 si->lowest_alloc <= last_in_cluster)
366 last_in_cluster = si->lowest_alloc - 1;
367 si->flags |= SWP_DISCARDING;
368 spin_unlock(&swap_lock);
369
370 if (offset < last_in_cluster)
371 discard_swap_cluster(si, offset,
372 last_in_cluster - offset + 1);
373
374 spin_lock(&swap_lock);
375 si->lowest_alloc = 0;
376 si->flags &= ~SWP_DISCARDING;
377
378 smp_mb(); /* wake_up_bit advises this */
379 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
380
381 } else if (si->flags & SWP_DISCARDING) {
382 /*
383 * Delay using pages allocated by racing tasks
384 * until the whole discard has been issued. We
385 * could defer that delay until swap_writepage,
386 * but it's easier to keep this self-contained.
387 */
388 spin_unlock(&swap_lock);
389 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
390 wait_for_discard, TASK_UNINTERRUPTIBLE);
391 spin_lock(&swap_lock);
392 } else {
393 /*
394 * Note pages allocated by racing tasks while
395 * scan for a free cluster is in progress, so
396 * that its final discard can exclude them.
397 */
398 if (offset < si->lowest_alloc)
399 si->lowest_alloc = offset;
400 if (offset > si->highest_alloc)
401 si->highest_alloc = offset;
402 }
403 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800404 return offset;
405
406scan:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700407 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700408 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700409 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700410 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700411 goto checks;
412 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700413 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
414 spin_lock(&swap_lock);
415 goto checks;
416 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700417 if (unlikely(--latency_ration < 0)) {
418 cond_resched();
419 latency_ration = LATENCY_LIMIT;
420 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700421 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800422 offset = si->lowest_bit;
423 while (++offset < scan_base) {
424 if (!si->swap_map[offset]) {
425 spin_lock(&swap_lock);
426 goto checks;
427 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700428 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
429 spin_lock(&swap_lock);
430 goto checks;
431 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800432 if (unlikely(--latency_ration < 0)) {
433 cond_resched();
434 latency_ration = LATENCY_LIMIT;
435 }
436 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700437 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700438
439no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700440 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 return 0;
442}
443
444swp_entry_t get_swap_page(void)
445{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700446 struct swap_info_struct *si;
447 pgoff_t offset;
448 int type, next;
449 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Hugh Dickins5d337b92005-09-03 15:54:41 -0700451 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700453 goto noswap;
454 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700456 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800457 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700458 next = si->next;
459 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800460 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700461 next = swap_list.head;
462 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700464
465 if (!si->highest_bit)
466 continue;
467 if (!(si->flags & SWP_WRITEOK))
468 continue;
469
470 swap_list.next = next;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700471 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800472 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700473 if (offset) {
474 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700475 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700476 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700477 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700479
480 nr_swap_pages++;
481noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700482 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700483 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
485
Hugh Dickins910321e2010-09-09 16:38:07 -0700486/* The only caller of this function is now susupend routine */
487swp_entry_t get_swap_page_of_type(int type)
488{
489 struct swap_info_struct *si;
490 pgoff_t offset;
491
492 spin_lock(&swap_lock);
493 si = swap_info[type];
494 if (si && (si->flags & SWP_WRITEOK)) {
495 nr_swap_pages--;
496 /* This is called for allocating swap entry, not cache */
497 offset = scan_swap_map(si, 1);
498 if (offset) {
499 spin_unlock(&swap_lock);
500 return swp_entry(type, offset);
501 }
502 nr_swap_pages++;
503 }
504 spin_unlock(&swap_lock);
505 return (swp_entry_t) {0};
506}
507
Hugh Dickins73c34b62009-12-14 17:58:43 -0800508static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800510 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 unsigned long offset, type;
512
513 if (!entry.val)
514 goto out;
515 type = swp_type(entry);
516 if (type >= nr_swapfiles)
517 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800518 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 if (!(p->flags & SWP_USED))
520 goto bad_device;
521 offset = swp_offset(entry);
522 if (offset >= p->max)
523 goto bad_offset;
524 if (!p->swap_map[offset])
525 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700526 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 return p;
528
529bad_free:
530 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
531 goto out;
532bad_offset:
533 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
534 goto out;
535bad_device:
536 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
537 goto out;
538bad_nofile:
539 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
540out:
541 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800542}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800544static unsigned char swap_entry_free(struct swap_info_struct *p,
545 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
Hugh Dickins253d5532009-12-14 17:58:44 -0800547 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800548 unsigned char count;
549 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700551 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800552 has_cache = count & SWAP_HAS_CACHE;
553 count &= ~SWAP_HAS_CACHE;
554
555 if (usage == SWAP_HAS_CACHE) {
556 VM_BUG_ON(!has_cache);
557 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800558 } else if (count == SWAP_MAP_SHMEM) {
559 /*
560 * Or we could insist on shmem.c using a special
561 * swap_shmem_free() and free_shmem_swap_and_cache()...
562 */
563 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800564 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
565 if (count == COUNT_CONTINUED) {
566 if (swap_count_continued(p, offset, count))
567 count = SWAP_MAP_MAX | COUNT_CONTINUED;
568 else
569 count = SWAP_MAP_MAX;
570 } else
571 count--;
572 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800573
574 if (!count)
575 mem_cgroup_uncharge_swap(entry);
576
577 usage = count | has_cache;
578 p->swap_map[offset] = usage;
579
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700580 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800581 if (!usage) {
Nitin Guptab3a27d02010-05-17 11:02:43 +0530582 struct gendisk *disk = p->bdev->bd_disk;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700583 if (offset < p->lowest_bit)
584 p->lowest_bit = offset;
585 if (offset > p->highest_bit)
586 p->highest_bit = offset;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800587 if (swap_list.next >= 0 &&
588 p->prio > swap_info[swap_list.next]->prio)
589 swap_list.next = p->type;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700590 nr_swap_pages++;
591 p->inuse_pages--;
Nitin Guptab3a27d02010-05-17 11:02:43 +0530592 if ((p->flags & SWP_BLKDEV) &&
593 disk->fops->swap_slot_free_notify)
594 disk->fops->swap_slot_free_notify(p->bdev, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800596
597 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598}
599
600/*
601 * Caller has made sure that the swapdevice corresponding to entry
602 * is still around or has not been recycled.
603 */
604void swap_free(swp_entry_t entry)
605{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800606 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 p = swap_info_get(entry);
609 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800610 swap_entry_free(p, entry, 1);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700611 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 }
613}
614
615/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700616 * Called after dropping swapcache to decrease refcnt to swap entries.
617 */
618void swapcache_free(swp_entry_t entry, struct page *page)
619{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700620 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800621 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700622
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700623 p = swap_info_get(entry);
624 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800625 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
626 if (page)
627 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700628 spin_unlock(&swap_lock);
629 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700630}
631
632/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700633 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800634 * This does not give an exact answer when swap count is continued,
635 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700637static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700639 int count = 0;
640 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 swp_entry_t entry;
642
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700643 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 p = swap_info_get(entry);
645 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700646 count = swap_count(p->swap_map[swp_offset(entry)]);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700647 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700649 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
651
652/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800653 * We can write to an anon page without COW if there are no other references
654 * to it. And as a side-effect, free up its swap: because the old content
655 * on disk will never be read, and seeking back there to write new content
656 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800658int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700660 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Hugh Dickins51726b12009-01-06 14:39:25 -0800662 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800663 if (unlikely(PageKsm(page)))
664 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700665 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800666 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700667 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800668 if (count == 1 && !PageWriteback(page)) {
669 delete_from_swap_cache(page);
670 SetPageDirty(page);
671 }
672 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800673 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800677 * If swap is getting full, or if there are no more mappings of this page,
678 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800680int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Hugh Dickins51726b12009-01-06 14:39:25 -0800682 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
684 if (!PageSwapCache(page))
685 return 0;
686 if (PageWriteback(page))
687 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800688 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 return 0;
690
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700691 /*
692 * Once hibernation has begun to create its image of memory,
693 * there's a danger that one of the calls to try_to_free_swap()
694 * - most probably a call from __try_to_reclaim_swap() while
695 * hibernation is allocating its own swap pages for the image,
696 * but conceivably even a call from memory reclaim - will free
697 * the swap from a page which has already been recorded in the
698 * image as a clean swapcache page, and then reuse its swap for
699 * another page of the image. On waking from hibernation, the
700 * original page might be freed under memory pressure, then
701 * later read back in from swap, now with the wrong data.
702 *
703 * Hibernation clears bits from gfp_allowed_mask to prevent
704 * memory reclaim from writing to disk, so check that here.
705 */
706 if (!(gfp_allowed_mask & __GFP_IO))
707 return 0;
708
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800709 delete_from_swap_cache(page);
710 SetPageDirty(page);
711 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700712}
713
714/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 * Free the swap entry like above, but also try to
716 * free the page cache entry if it is the last user.
717 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800718int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800720 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 struct page *page = NULL;
722
Andi Kleena7420aa2009-09-16 11:50:05 +0200723 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800724 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 p = swap_info_get(entry);
727 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800728 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Nick Piggin93fac702006-03-31 02:29:56 -0800729 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700730 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800731 page_cache_release(page);
732 page = NULL;
733 }
734 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700735 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
737 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800738 /*
739 * Not mapped elsewhere, or swap space full? Free it!
740 * Also recheck PageSwapCache now page is locked (above).
741 */
Nick Piggin93fac702006-03-31 02:29:56 -0800742 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800743 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 delete_from_swap_cache(page);
745 SetPageDirty(page);
746 }
747 unlock_page(page);
748 page_cache_release(page);
749 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800750 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
Daisuke Nishimura02491442010-03-10 15:22:17 -0800753#ifdef CONFIG_CGROUP_MEM_RES_CTLR
754/**
755 * mem_cgroup_count_swap_user - count the user of a swap entry
756 * @ent: the swap entry to be checked
757 * @pagep: the pointer for the swap cache page of the entry to be stored
758 *
759 * Returns the number of the user of the swap entry. The number is valid only
760 * for swaps of anonymous pages.
761 * If the entry is found on swap cache, the page is stored to pagep with
762 * refcount of it being incremented.
763 */
764int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep)
765{
766 struct page *page;
767 struct swap_info_struct *p;
768 int count = 0;
769
770 page = find_get_page(&swapper_space, ent.val);
771 if (page)
772 count += page_mapcount(page);
773 p = swap_info_get(ent);
774 if (p) {
775 count += swap_count(p->swap_map[swp_offset(ent)]);
776 spin_unlock(&swap_lock);
777 }
778
779 *pagep = page;
780 return count;
781}
782#endif
783
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200784#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800785/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800786 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800787 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800788 * @offset - number of the PAGE_SIZE-sized block of the device, starting
789 * from 0, in which the swap header is expected to be located.
790 *
791 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800792 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800793int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800794{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800795 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800796 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800797
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800798 if (device)
799 bdev = bdget(device);
800
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800801 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800802 for (type = 0; type < nr_swapfiles; type++) {
803 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800804
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800805 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800806 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700807
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800808 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800809 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200810 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800811
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800812 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800813 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800814 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800815 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800816 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800817
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800818 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800819 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200820 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800821
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800822 spin_unlock(&swap_lock);
823 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800824 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800825 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800826 }
827 }
828 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800829 if (bdev)
830 bdput(bdev);
831
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800832 return -ENODEV;
833}
834
835/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800836 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
837 * corresponding to given index in swap_info (swap type).
838 */
839sector_t swapdev_block(int type, pgoff_t offset)
840{
841 struct block_device *bdev;
842
843 if ((unsigned int)type >= nr_swapfiles)
844 return 0;
845 if (!(swap_info[type]->flags & SWP_WRITEOK))
846 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800847 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800848}
849
850/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800851 * Return either the total number of swap pages of given type, or the number
852 * of free pages of that type (depending on @free)
853 *
854 * This is needed for software suspend
855 */
856unsigned int count_swap_pages(int type, int free)
857{
858 unsigned int n = 0;
859
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800860 spin_lock(&swap_lock);
861 if ((unsigned int)type < nr_swapfiles) {
862 struct swap_info_struct *sis = swap_info[type];
863
864 if (sis->flags & SWP_WRITEOK) {
865 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800866 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800867 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800868 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800869 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800870 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800871 return n;
872}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800873#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700876 * No need to decide whether this PTE shares the swap entry with others,
877 * just let do_wp_page work it out if a write is requested later - to
878 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800880static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 unsigned long addr, swp_entry_t entry, struct page *page)
882{
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800883 struct mem_cgroup *ptr = NULL;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800884 spinlock_t *ptl;
885 pte_t *pte;
886 int ret = 1;
887
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800888 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800889 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800890 goto out_nolock;
891 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800892
893 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
894 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
895 if (ret > 0)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800896 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800897 ret = 0;
898 goto out;
899 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800900
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800901 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800902 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 get_page(page);
904 set_pte_at(vma->vm_mm, addr, pte,
905 pte_mkold(mk_pte(page, vma->vm_page_prot)));
906 page_add_anon_rmap(page, vma, addr);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800907 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 swap_free(entry);
909 /*
910 * Move the page to the active list so it is not
911 * immediately swapped out again after swapon.
912 */
913 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800914out:
915 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800916out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800917 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918}
919
920static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
921 unsigned long addr, unsigned long end,
922 swp_entry_t entry, struct page *page)
923{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700925 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800926 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Hugh Dickins044d66c2008-02-07 00:14:04 -0800928 /*
929 * We don't actually need pte lock while scanning for swp_pte: since
930 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
931 * page table while we're scanning; though it could get zapped, and on
932 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
933 * of unmatched parts which look like swp_pte, so unuse_pte must
934 * recheck under pte lock. Scanning without pte lock lets it be
935 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
936 */
937 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 do {
939 /*
940 * swapoff spends a _lot_ of time in this loop!
941 * Test inline before going to call unuse_pte.
942 */
943 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800944 pte_unmap(pte);
945 ret = unuse_pte(vma, pmd, addr, entry, page);
946 if (ret)
947 goto out;
948 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 }
950 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800951 pte_unmap(pte - 1);
952out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800953 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954}
955
956static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
957 unsigned long addr, unsigned long end,
958 swp_entry_t entry, struct page *page)
959{
960 pmd_t *pmd;
961 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800962 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
964 pmd = pmd_offset(pud, addr);
965 do {
966 next = pmd_addr_end(addr, end);
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800967 if (unlikely(pmd_trans_huge(*pmd)))
968 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 if (pmd_none_or_clear_bad(pmd))
970 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800971 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
972 if (ret)
973 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 } while (pmd++, addr = next, addr != end);
975 return 0;
976}
977
978static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
979 unsigned long addr, unsigned long end,
980 swp_entry_t entry, struct page *page)
981{
982 pud_t *pud;
983 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800984 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 pud = pud_offset(pgd, addr);
987 do {
988 next = pud_addr_end(addr, end);
989 if (pud_none_or_clear_bad(pud))
990 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800991 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
992 if (ret)
993 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 } while (pud++, addr = next, addr != end);
995 return 0;
996}
997
998static int unuse_vma(struct vm_area_struct *vma,
999 swp_entry_t entry, struct page *page)
1000{
1001 pgd_t *pgd;
1002 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001003 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -08001005 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 addr = page_address_in_vma(page, vma);
1007 if (addr == -EFAULT)
1008 return 0;
1009 else
1010 end = addr + PAGE_SIZE;
1011 } else {
1012 addr = vma->vm_start;
1013 end = vma->vm_end;
1014 }
1015
1016 pgd = pgd_offset(vma->vm_mm, addr);
1017 do {
1018 next = pgd_addr_end(addr, end);
1019 if (pgd_none_or_clear_bad(pgd))
1020 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001021 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1022 if (ret)
1023 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 } while (pgd++, addr = next, addr != end);
1025 return 0;
1026}
1027
1028static int unuse_mm(struct mm_struct *mm,
1029 swp_entry_t entry, struct page *page)
1030{
1031 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001032 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 if (!down_read_trylock(&mm->mmap_sem)) {
1035 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001036 * Activate page so shrink_inactive_list is unlikely to unmap
1037 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001039 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 unlock_page(page);
1041 down_read(&mm->mmap_sem);
1042 lock_page(page);
1043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001045 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 break;
1047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001049 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
1051
1052/*
1053 * Scan swap_map from current position to next entry still in use.
1054 * Recycle to start on reaching the end, returning 0 when empty.
1055 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001056static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1057 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001059 unsigned int max = si->max;
1060 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001061 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
1063 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001064 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 * for whether an entry is in use, not modifying it; false
1066 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001067 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 */
1069 for (;;) {
1070 if (++i >= max) {
1071 if (!prev) {
1072 i = 0;
1073 break;
1074 }
1075 /*
1076 * No entries in use at top of swap_map,
1077 * loop back to start and recheck there.
1078 */
1079 max = prev + 1;
1080 prev = 0;
1081 i = 1;
1082 }
1083 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001084 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 break;
1086 }
1087 return i;
1088}
1089
1090/*
1091 * We completely avoid races by reading each swap page in advance,
1092 * and then search for the process using it. All the necessary
1093 * page table adjustments can then be made atomically.
1094 */
1095static int try_to_unuse(unsigned int type)
1096{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001097 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001099 unsigned char *swap_map;
1100 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 struct page *page;
1102 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001103 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106 /*
1107 * When searching mms for an entry, a good strategy is to
1108 * start at the first mm we freed the previous entry from
1109 * (though actually we don't notice whether we or coincidence
1110 * freed the entry). Initialize this start_mm with a hold.
1111 *
1112 * A simpler strategy would be to start at the last mm we
1113 * freed the previous entry from; but that would take less
1114 * advantage of mmlist ordering, which clusters forked mms
1115 * together, child after parent. If we race with dup_mmap(), we
1116 * prefer to resolve parent before child, lest we miss entries
1117 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001118 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 */
1120 start_mm = &init_mm;
1121 atomic_inc(&init_mm.mm_users);
1122
1123 /*
1124 * Keep on scanning until all entries have gone. Usually,
1125 * one pass through swap_map is enough, but not necessarily:
1126 * there are races when an instance of an entry might be missed.
1127 */
1128 while ((i = find_next_to_unuse(si, i)) != 0) {
1129 if (signal_pending(current)) {
1130 retval = -EINTR;
1131 break;
1132 }
1133
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001134 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 * Get a page for the entry, using the existing swap
1136 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001137 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 */
1139 swap_map = &si->swap_map[i];
1140 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001141 page = read_swap_cache_async(entry,
1142 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (!page) {
1144 /*
1145 * Either swap_duplicate() failed because entry
1146 * has been freed independently, and will not be
1147 * reused since sys_swapoff() already disabled
1148 * allocation from here, or alloc_page() failed.
1149 */
1150 if (!*swap_map)
1151 continue;
1152 retval = -ENOMEM;
1153 break;
1154 }
1155
1156 /*
1157 * Don't hold on to start_mm if it looks like exiting.
1158 */
1159 if (atomic_read(&start_mm->mm_users) == 1) {
1160 mmput(start_mm);
1161 start_mm = &init_mm;
1162 atomic_inc(&init_mm.mm_users);
1163 }
1164
1165 /*
1166 * Wait for and lock page. When do_swap_page races with
1167 * try_to_unuse, do_swap_page can handle the fault much
1168 * faster than try_to_unuse can locate the entry. This
1169 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1170 * defer to do_swap_page in such a case - in some tests,
1171 * do_swap_page and try_to_unuse repeatedly compete.
1172 */
1173 wait_on_page_locked(page);
1174 wait_on_page_writeback(page);
1175 lock_page(page);
1176 wait_on_page_writeback(page);
1177
1178 /*
1179 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001182 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1183 retval = shmem_unuse(entry, page);
1184 /* page has already been unlocked and released */
1185 if (retval < 0)
1186 break;
1187 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001189 if (swap_count(swcount) && start_mm != &init_mm)
1190 retval = unuse_mm(start_mm, entry, page);
1191
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001192 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 int set_start_mm = (*swap_map >= swcount);
1194 struct list_head *p = &start_mm->mmlist;
1195 struct mm_struct *new_start_mm = start_mm;
1196 struct mm_struct *prev_mm = start_mm;
1197 struct mm_struct *mm;
1198
1199 atomic_inc(&new_start_mm->mm_users);
1200 atomic_inc(&prev_mm->mm_users);
1201 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001202 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 (p = p->next) != &start_mm->mmlist) {
1204 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001205 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 spin_unlock(&mmlist_lock);
1208 mmput(prev_mm);
1209 prev_mm = mm;
1210
1211 cond_resched();
1212
1213 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001214 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001216 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001218 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001220
Bo Liu32c5fc12009-11-02 16:50:33 +00001221 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 mmput(new_start_mm);
1223 atomic_inc(&mm->mm_users);
1224 new_start_mm = mm;
1225 set_start_mm = 0;
1226 }
1227 spin_lock(&mmlist_lock);
1228 }
1229 spin_unlock(&mmlist_lock);
1230 mmput(prev_mm);
1231 mmput(start_mm);
1232 start_mm = new_start_mm;
1233 }
1234 if (retval) {
1235 unlock_page(page);
1236 page_cache_release(page);
1237 break;
1238 }
1239
1240 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 * If a reference remains (rare), we would like to leave
1242 * the page in the swap cache; but try_to_unmap could
1243 * then re-duplicate the entry once we drop page lock,
1244 * so we might loop indefinitely; also, that page could
1245 * not be swapped out to other storage meanwhile. So:
1246 * delete from cache even if there's another reference,
1247 * after ensuring that the data has been saved to disk -
1248 * since if the reference remains (rarer), it will be
1249 * read from disk into another page. Splitting into two
1250 * pages would be incorrect if swap supported "shared
1251 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001252 *
1253 * Given how unuse_vma() targets one particular offset
1254 * in an anon_vma, once the anon_vma has been determined,
1255 * this splitting happens to be just what is needed to
1256 * handle where KSM pages have been swapped out: re-reading
1257 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001259 if (swap_count(*swap_map) &&
1260 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 struct writeback_control wbc = {
1262 .sync_mode = WB_SYNC_NONE,
1263 };
1264
1265 swap_writepage(page, &wbc);
1266 lock_page(page);
1267 wait_on_page_writeback(page);
1268 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001269
1270 /*
1271 * It is conceivable that a racing task removed this page from
1272 * swap cache just before we acquired the page lock at the top,
1273 * or while we dropped it in unuse_mm(). The page might even
1274 * be back in swap cache on another swap area: that we must not
1275 * delete, since it may not have been written out to swap yet.
1276 */
1277 if (PageSwapCache(page) &&
1278 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001279 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
1281 /*
1282 * So we could skip searching mms once swap count went
1283 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001284 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 */
1286 SetPageDirty(page);
1287 unlock_page(page);
1288 page_cache_release(page);
1289
1290 /*
1291 * Make sure that we aren't completely killing
1292 * interactive performance.
1293 */
1294 cond_resched();
1295 }
1296
1297 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 return retval;
1299}
1300
1301/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001302 * After a successful try_to_unuse, if no swap is now in use, we know
1303 * we can empty the mmlist. swap_lock must be held on entry and exit.
1304 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 * added to the mmlist just after page_duplicate - before would be racy.
1306 */
1307static void drain_mmlist(void)
1308{
1309 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001310 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001312 for (type = 0; type < nr_swapfiles; type++)
1313 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return;
1315 spin_lock(&mmlist_lock);
1316 list_for_each_safe(p, next, &init_mm.mmlist)
1317 list_del_init(p);
1318 spin_unlock(&mmlist_lock);
1319}
1320
1321/*
1322 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001323 * corresponds to page offset for the specified swap entry.
1324 * Note that the type of this function is sector_t, but it returns page offset
1325 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001327static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001329 struct swap_info_struct *sis;
1330 struct swap_extent *start_se;
1331 struct swap_extent *se;
1332 pgoff_t offset;
1333
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001334 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001335 *bdev = sis->bdev;
1336
1337 offset = swp_offset(entry);
1338 start_se = sis->curr_swap_extent;
1339 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
1341 for ( ; ; ) {
1342 struct list_head *lh;
1343
1344 if (se->start_page <= offset &&
1345 offset < (se->start_page + se->nr_pages)) {
1346 return se->start_block + (offset - se->start_page);
1347 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001348 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 se = list_entry(lh, struct swap_extent, list);
1350 sis->curr_swap_extent = se;
1351 BUG_ON(se == start_se); /* It *must* be present */
1352 }
1353}
1354
1355/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001356 * Returns the page offset into bdev for the specified page's swap entry.
1357 */
1358sector_t map_swap_page(struct page *page, struct block_device **bdev)
1359{
1360 swp_entry_t entry;
1361 entry.val = page_private(page);
1362 return map_swap_entry(entry, bdev);
1363}
1364
1365/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 * Free all of a swapdev's extent information
1367 */
1368static void destroy_swap_extents(struct swap_info_struct *sis)
1369{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001370 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 struct swap_extent *se;
1372
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001373 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 struct swap_extent, list);
1375 list_del(&se->list);
1376 kfree(se);
1377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378}
1379
1380/*
1381 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001382 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001384 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 */
1386static int
1387add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1388 unsigned long nr_pages, sector_t start_block)
1389{
1390 struct swap_extent *se;
1391 struct swap_extent *new_se;
1392 struct list_head *lh;
1393
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001394 if (start_page == 0) {
1395 se = &sis->first_swap_extent;
1396 sis->curr_swap_extent = se;
1397 se->start_page = 0;
1398 se->nr_pages = nr_pages;
1399 se->start_block = start_block;
1400 return 1;
1401 } else {
1402 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001404 BUG_ON(se->start_page + se->nr_pages != start_page);
1405 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 /* Merge it */
1407 se->nr_pages += nr_pages;
1408 return 0;
1409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 }
1411
1412 /*
1413 * No merge. Insert a new extent, preserving ordering.
1414 */
1415 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1416 if (new_se == NULL)
1417 return -ENOMEM;
1418 new_se->start_page = start_page;
1419 new_se->nr_pages = nr_pages;
1420 new_se->start_block = start_block;
1421
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001422 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001423 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
1425
1426/*
1427 * A `swap extent' is a simple thing which maps a contiguous range of pages
1428 * onto a contiguous range of disk blocks. An ordered list of swap extents
1429 * is built at swapon time and is then used at swap_writepage/swap_readpage
1430 * time for locating where on disk a page belongs.
1431 *
1432 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1433 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1434 * swap files identically.
1435 *
1436 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1437 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1438 * swapfiles are handled *identically* after swapon time.
1439 *
1440 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1441 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1442 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1443 * requirements, they are simply tossed out - we will never use those blocks
1444 * for swapping.
1445 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001446 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1448 * which will scribble on the fs.
1449 *
1450 * The amount of disk space which a single swap extent represents varies.
1451 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1452 * extents in the list. To avoid much list walking, we cache the previous
1453 * search location in `curr_swap_extent', and start new searches from there.
1454 * This is extremely effective. The average number of iterations in
1455 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1456 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001457static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
1459 struct inode *inode;
1460 unsigned blocks_per_page;
1461 unsigned long page_no;
1462 unsigned blkbits;
1463 sector_t probe_block;
1464 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001465 sector_t lowest_block = -1;
1466 sector_t highest_block = 0;
1467 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 int ret;
1469
1470 inode = sis->swap_file->f_mapping->host;
1471 if (S_ISBLK(inode->i_mode)) {
1472 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001473 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001474 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 }
1476
1477 blkbits = inode->i_blkbits;
1478 blocks_per_page = PAGE_SIZE >> blkbits;
1479
1480 /*
1481 * Map all the blocks into the extent list. This code doesn't try
1482 * to be very smart.
1483 */
1484 probe_block = 0;
1485 page_no = 0;
1486 last_block = i_size_read(inode) >> blkbits;
1487 while ((probe_block + blocks_per_page) <= last_block &&
1488 page_no < sis->max) {
1489 unsigned block_in_page;
1490 sector_t first_block;
1491
1492 first_block = bmap(inode, probe_block);
1493 if (first_block == 0)
1494 goto bad_bmap;
1495
1496 /*
1497 * It must be PAGE_SIZE aligned on-disk
1498 */
1499 if (first_block & (blocks_per_page - 1)) {
1500 probe_block++;
1501 goto reprobe;
1502 }
1503
1504 for (block_in_page = 1; block_in_page < blocks_per_page;
1505 block_in_page++) {
1506 sector_t block;
1507
1508 block = bmap(inode, probe_block + block_in_page);
1509 if (block == 0)
1510 goto bad_bmap;
1511 if (block != first_block + block_in_page) {
1512 /* Discontiguity */
1513 probe_block++;
1514 goto reprobe;
1515 }
1516 }
1517
Hugh Dickins53092a72005-09-03 15:54:34 -07001518 first_block >>= (PAGE_SHIFT - blkbits);
1519 if (page_no) { /* exclude the header page */
1520 if (first_block < lowest_block)
1521 lowest_block = first_block;
1522 if (first_block > highest_block)
1523 highest_block = first_block;
1524 }
1525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 /*
1527 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1528 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001529 ret = add_swap_extent(sis, page_no, 1, first_block);
1530 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001532 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 page_no++;
1534 probe_block += blocks_per_page;
1535reprobe:
1536 continue;
1537 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001538 ret = nr_extents;
1539 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001541 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001543 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001545out:
1546 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547bad_bmap:
1548 printk(KERN_ERR "swapon: swapfile has holes\n");
1549 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001550 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551}
1552
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001553SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001555 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001556 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 struct file *swap_file, *victim;
1558 struct address_space *mapping;
1559 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001560 char *pathname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 int i, type, prev;
1562 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 if (!capable(CAP_SYS_ADMIN))
1565 return -EPERM;
1566
1567 pathname = getname(specialfile);
1568 err = PTR_ERR(pathname);
1569 if (IS_ERR(pathname))
1570 goto out;
1571
1572 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1573 putname(pathname);
1574 err = PTR_ERR(victim);
1575 if (IS_ERR(victim))
1576 goto out;
1577
1578 mapping = victim->f_mapping;
1579 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001580 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001581 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1582 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001583 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 if (p->swap_file->f_mapping == mapping)
1585 break;
1586 }
1587 prev = type;
1588 }
1589 if (type < 0) {
1590 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001591 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 goto out_dput;
1593 }
1594 if (!security_vm_enough_memory(p->pages))
1595 vm_unacct_memory(p->pages);
1596 else {
1597 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001598 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 goto out_dput;
1600 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001601 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001603 else
1604 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 if (type == swap_list.next) {
1606 /* just pick something that's safe... */
1607 swap_list.next = swap_list.head;
1608 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001609 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001610 for (i = p->next; i >= 0; i = swap_info[i]->next)
1611 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001612 least_priority++;
1613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 nr_swap_pages -= p->pages;
1615 total_swap_pages -= p->pages;
1616 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001617 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001618
Hugh Dickins35451be2009-09-21 17:02:27 -07001619 current->flags |= PF_OOM_ORIGIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 err = try_to_unuse(type);
Hugh Dickins35451be2009-09-21 17:02:27 -07001621 current->flags &= ~PF_OOM_ORIGIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 if (err) {
1624 /* re-insert swap space back into swap_list */
Hugh Dickins5d337b92005-09-03 15:54:41 -07001625 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001626 if (p->prio < 0)
1627 p->prio = --least_priority;
1628 prev = -1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001629 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1630 if (p->prio >= swap_info[i]->prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 break;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001632 prev = i;
1633 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 p->next = i;
1635 if (prev < 0)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001636 swap_list.head = swap_list.next = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 else
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001638 swap_info[prev]->next = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 nr_swap_pages += p->pages;
1640 total_swap_pages += p->pages;
1641 p->flags |= SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001642 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 goto out_dput;
1644 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001645
1646 /* wait for any unplug function to finish */
1647 down_write(&swap_unplug_sem);
1648 up_write(&swap_unplug_sem);
1649
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001650 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001651 if (p->flags & SWP_CONTINUED)
1652 free_swap_count_continuations(p);
1653
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001654 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001655 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001657
1658 /* wait for anyone still in scan_swap_map */
1659 p->highest_bit = 0; /* cuts scans short */
1660 while (p->flags >= SWP_SCANNING) {
1661 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001662 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001663 spin_lock(&swap_lock);
1664 }
1665
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 swap_file = p->swap_file;
1667 p->swap_file = NULL;
1668 p->max = 0;
1669 swap_map = p->swap_map;
1670 p->swap_map = NULL;
1671 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001672 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001673 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 vfree(swap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001675 /* Destroy swap account informatin */
1676 swap_cgroup_swapoff(type);
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 inode = mapping->host;
1679 if (S_ISBLK(inode->i_mode)) {
1680 struct block_device *bdev = I_BDEV(inode);
1681 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001682 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001684 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001686 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 }
1688 filp_close(swap_file, NULL);
1689 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001690 atomic_inc(&proc_poll_event);
1691 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693out_dput:
1694 filp_close(victim, NULL);
1695out:
1696 return err;
1697}
1698
1699#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001700struct proc_swaps {
1701 struct seq_file seq;
1702 int event;
1703};
1704
1705static unsigned swaps_poll(struct file *file, poll_table *wait)
1706{
1707 struct proc_swaps *s = file->private_data;
1708
1709 poll_wait(file, &proc_poll_wait, wait);
1710
1711 if (s->event != atomic_read(&proc_poll_event)) {
1712 s->event = atomic_read(&proc_poll_event);
1713 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1714 }
1715
1716 return POLLIN | POLLRDNORM;
1717}
1718
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719/* iterator */
1720static void *swap_start(struct seq_file *swap, loff_t *pos)
1721{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001722 struct swap_info_struct *si;
1723 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 loff_t l = *pos;
1725
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001726 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001728 if (!l)
1729 return SEQ_START_TOKEN;
1730
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001731 for (type = 0; type < nr_swapfiles; type++) {
1732 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1733 si = swap_info[type];
1734 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001736 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001737 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 }
1739
1740 return NULL;
1741}
1742
1743static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1744{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001745 struct swap_info_struct *si = v;
1746 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001748 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001749 type = 0;
1750 else
1751 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001752
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001753 for (; type < nr_swapfiles; type++) {
1754 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1755 si = swap_info[type];
1756 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 continue;
1758 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001759 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 }
1761
1762 return NULL;
1763}
1764
1765static void swap_stop(struct seq_file *swap, void *v)
1766{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001767 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768}
1769
1770static int swap_show(struct seq_file *swap, void *v)
1771{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001772 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 struct file *file;
1774 int len;
1775
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001776 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001777 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1778 return 0;
1779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001781 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001782 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001783 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001784 len < 40 ? 40 - len : 1, " ",
1785 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001787 si->pages << (PAGE_SHIFT - 10),
1788 si->inuse_pages << (PAGE_SHIFT - 10),
1789 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 return 0;
1791}
1792
Helge Deller15ad7cd2006-12-06 20:40:36 -08001793static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 .start = swap_start,
1795 .next = swap_next,
1796 .stop = swap_stop,
1797 .show = swap_show
1798};
1799
1800static int swaps_open(struct inode *inode, struct file *file)
1801{
Kay Sievers66d7dd52010-10-26 14:22:06 -07001802 struct proc_swaps *s;
1803 int ret;
1804
1805 s = kmalloc(sizeof(struct proc_swaps), GFP_KERNEL);
1806 if (!s)
1807 return -ENOMEM;
1808
1809 file->private_data = s;
1810
1811 ret = seq_open(file, &swaps_op);
1812 if (ret) {
1813 kfree(s);
1814 return ret;
1815 }
1816
1817 s->seq.private = s;
1818 s->event = atomic_read(&proc_poll_event);
1819 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820}
1821
Helge Deller15ad7cd2006-12-06 20:40:36 -08001822static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 .open = swaps_open,
1824 .read = seq_read,
1825 .llseek = seq_lseek,
1826 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001827 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828};
1829
1830static int __init procswaps_init(void)
1831{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001832 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 return 0;
1834}
1835__initcall(procswaps_init);
1836#endif /* CONFIG_PROC_FS */
1837
Jan Beulich17963162008-12-16 11:35:24 +00001838#ifdef MAX_SWAPFILES_CHECK
1839static int __init max_swapfiles_check(void)
1840{
1841 MAX_SWAPFILES_CHECK();
1842 return 0;
1843}
1844late_initcall(max_swapfiles_check);
1845#endif
1846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847/*
1848 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1849 *
1850 * The swapon system call
1851 */
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001852SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001854 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 char *name = NULL;
1856 struct block_device *bdev = NULL;
1857 struct file *swap_file = NULL;
1858 struct address_space *mapping;
1859 unsigned int type;
1860 int i, prev;
1861 int error;
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001862 union swap_header *swap_header;
1863 unsigned int nr_good_pages;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001864 int nr_extents = 0;
Hugh Dickins53092a72005-09-03 15:54:34 -07001865 sector_t span;
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001866 unsigned long maxpages;
Hugh Dickins73fd8742009-01-06 14:39:47 -08001867 unsigned long swapfilepages;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001868 unsigned char *swap_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 struct page *page = NULL;
1870 struct inode *inode = NULL;
1871 int did_down = 0;
1872
1873 if (!capable(CAP_SYS_ADMIN))
1874 return -EPERM;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001875
1876 p = kzalloc(sizeof(*p), GFP_KERNEL);
1877 if (!p)
1878 return -ENOMEM;
1879
Hugh Dickins5d337b92005-09-03 15:54:41 -07001880 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001881 for (type = 0; type < nr_swapfiles; type++) {
1882 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 error = -EPERM;
Christoph Lameter06972122006-06-23 02:03:35 -07001886 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001887 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001888 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 goto out;
1890 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001891 if (type >= nr_swapfiles) {
1892 p->type = type;
1893 swap_info[type] = p;
1894 /*
1895 * Write swap_info[type] before nr_swapfiles, in case a
1896 * racing procfs swap_start() or swap_next() is reading them.
1897 * (We never shrink nr_swapfiles, we never free this entry.)
1898 */
1899 smp_wmb();
1900 nr_swapfiles++;
1901 } else {
1902 kfree(p);
1903 p = swap_info[type];
1904 /*
1905 * Do not memset this entry: a racing procfs swap_next()
1906 * would be relying on p->type to remain valid.
1907 */
1908 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001909 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001912 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 name = getname(specialfile);
1915 error = PTR_ERR(name);
1916 if (IS_ERR(name)) {
1917 name = NULL;
1918 goto bad_swap_2;
1919 }
1920 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1921 error = PTR_ERR(swap_file);
1922 if (IS_ERR(swap_file)) {
1923 swap_file = NULL;
1924 goto bad_swap_2;
1925 }
1926
1927 p->swap_file = swap_file;
1928 mapping = swap_file->f_mapping;
1929 inode = mapping->host;
1930
1931 error = -EBUSY;
1932 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001933 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
1935 if (i == type || !q->swap_file)
1936 continue;
1937 if (mapping == q->swap_file->f_mapping)
1938 goto bad_swap;
1939 }
1940
1941 error = -EINVAL;
1942 if (S_ISBLK(inode->i_mode)) {
1943 bdev = I_BDEV(inode);
Tejun Heoe525fd82010-11-13 11:55:17 +01001944 error = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1945 sys_swapon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 if (error < 0) {
1947 bdev = NULL;
Rob Landleyf7b3a432005-09-22 21:44:27 -07001948 error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 goto bad_swap;
1950 }
1951 p->old_block_size = block_size(bdev);
1952 error = set_blocksize(bdev, PAGE_SIZE);
1953 if (error < 0)
1954 goto bad_swap;
1955 p->bdev = bdev;
Nitin Guptab2725642010-05-17 11:02:42 +05301956 p->flags |= SWP_BLKDEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 } else if (S_ISREG(inode->i_mode)) {
1958 p->bdev = inode->i_sb->s_bdev;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001959 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 did_down = 1;
1961 if (IS_SWAPFILE(inode)) {
1962 error = -EBUSY;
1963 goto bad_swap;
1964 }
1965 } else {
1966 goto bad_swap;
1967 }
1968
Hugh Dickins73fd8742009-01-06 14:39:47 -08001969 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
1971 /*
1972 * Read the swap header.
1973 */
1974 if (!mapping->a_ops->readpage) {
1975 error = -EINVAL;
1976 goto bad_swap;
1977 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07001978 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 if (IS_ERR(page)) {
1980 error = PTR_ERR(page);
1981 goto bad_swap;
1982 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001983 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Hugh Dickins81e33972009-01-06 14:39:49 -08001985 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Jesper Juhle97a3112006-01-11 01:50:28 +01001986 printk(KERN_ERR "Unable to find swap-space signature\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 error = -EINVAL;
1988 goto bad_swap;
1989 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001990
Hugh Dickins81e33972009-01-06 14:39:49 -08001991 /* swap partition endianess hack... */
1992 if (swab32(swap_header->info.version) == 1) {
1993 swab32s(&swap_header->info.version);
1994 swab32s(&swap_header->info.last_page);
1995 swab32s(&swap_header->info.nr_badpages);
1996 for (i = 0; i < swap_header->info.nr_badpages; i++)
1997 swab32s(&swap_header->info.badpages[i]);
1998 }
1999 /* Check the swap header's sub-version */
2000 if (swap_header->info.version != 1) {
2001 printk(KERN_WARNING
2002 "Unable to handle swap header version %d\n",
2003 swap_header->info.version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 error = -EINVAL;
2005 goto bad_swap;
Hugh Dickins81e33972009-01-06 14:39:49 -08002006 }
2007
2008 p->lowest_bit = 1;
2009 p->cluster_next = 1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002010 p->cluster_nr = 0;
Hugh Dickins81e33972009-01-06 14:39:49 -08002011
2012 /*
2013 * Find out how many pages are allowed for a single swap
2014 * device. There are two limiting factors: 1) the number of
2015 * bits for the swap offset in the swp_entry_t type and
2016 * 2) the number of bits in the a swap pte as defined by
2017 * the different architectures. In order to find the
2018 * largest possible bit mask a swap entry with swap type 0
2019 * and swap offset ~0UL is created, encoded to a swap pte,
2020 * decoded to a swp_entry_t again and finally the swap
2021 * offset is extracted. This will mask all the bits from
2022 * the initial ~0UL mask that can't be encoded in either
2023 * the swp_entry_t or the architecture definition of a
2024 * swap pte.
2025 */
2026 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08002027 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2028 if (maxpages > swap_header->info.last_page) {
2029 maxpages = swap_header->info.last_page + 1;
2030 /* p->max is an unsigned int: don't overflow it */
2031 if ((unsigned int)maxpages == 0)
2032 maxpages = UINT_MAX;
2033 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002034 p->highest_bit = maxpages - 1;
2035
2036 error = -EINVAL;
2037 if (!maxpages)
2038 goto bad_swap;
2039 if (swapfilepages && maxpages > swapfilepages) {
2040 printk(KERN_WARNING
2041 "Swap area shorter than signature indicates\n");
2042 goto bad_swap;
2043 }
2044 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2045 goto bad_swap;
2046 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2047 goto bad_swap;
2048
2049 /* OK, set up the swap map and apply the bad block list */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002050 swap_map = vmalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002051 if (!swap_map) {
2052 error = -ENOMEM;
2053 goto bad_swap;
2054 }
2055
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002056 memset(swap_map, 0, maxpages);
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08002057 nr_good_pages = maxpages - 1; /* omit header page */
2058
Hugh Dickins81e33972009-01-06 14:39:49 -08002059 for (i = 0; i < swap_header->info.nr_badpages; i++) {
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08002060 unsigned int page_nr = swap_header->info.badpages[i];
2061 if (page_nr == 0 || page_nr > swap_header->info.last_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 error = -EINVAL;
2063 goto bad_swap;
2064 }
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08002065 if (page_nr < maxpages) {
2066 swap_map[page_nr] = SWAP_MAP_BAD;
2067 nr_good_pages--;
2068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 }
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002070
2071 error = swap_cgroup_swapon(type, maxpages);
2072 if (error)
2073 goto bad_swap;
2074
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002075 if (nr_good_pages) {
Hugh Dickins78ecba02008-07-23 21:28:23 -07002076 swap_map[0] = SWAP_MAP_BAD;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002077 p->max = maxpages;
2078 p->pages = nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07002079 nr_extents = setup_swap_extents(p, &span);
2080 if (nr_extents < 0) {
2081 error = nr_extents;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002082 goto bad_swap;
Hugh Dickins53092a72005-09-03 15:54:34 -07002083 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002084 nr_good_pages = p->pages;
2085 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 if (!nr_good_pages) {
2087 printk(KERN_WARNING "Empty swap-file\n");
2088 error = -EINVAL;
2089 goto bad_swap;
2090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002092 if (p->bdev) {
2093 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2094 p->flags |= SWP_SOLIDSTATE;
2095 p->cluster_next = 1 + (random32() % p->highest_bit);
2096 }
Hugh Dickins33994462010-09-09 16:38:11 -07002097 if (discard_swap(p) == 0 && (swap_flags & SWAP_FLAG_DISCARD))
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002098 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002099 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002100
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002101 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002102 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07002103 if (swap_flags & SWAP_FLAG_PREFER)
2104 p->prio =
2105 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2106 else
2107 p->prio = --least_priority;
2108 p->swap_map = swap_map;
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08002109 p->flags |= SWP_WRITEOK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 nr_swap_pages += nr_good_pages;
2111 total_swap_pages += nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07002112
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002113 printk(KERN_INFO "Adding %uk swap on %s. "
Hugh Dickins20137a42009-01-06 14:39:54 -08002114 "Priority:%d extents:%d across:%lluk %s%s\n",
Hugh Dickins53092a72005-09-03 15:54:34 -07002115 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002116 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
Hugh Dickins20137a42009-01-06 14:39:54 -08002117 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2118 (p->flags & SWP_DISCARDABLE) ? "D" : "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
2120 /* insert swap space into swap_list: */
2121 prev = -1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002122 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
2123 if (p->prio >= swap_info[i]->prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 prev = i;
2126 }
2127 p->next = i;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002128 if (prev < 0)
2129 swap_list.head = swap_list.next = type;
2130 else
2131 swap_info[prev]->next = type;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002132 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002133 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002134 atomic_inc(&proc_poll_event);
2135 wake_up_interruptible(&proc_poll_wait);
2136
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 error = 0;
2138 goto out;
2139bad_swap:
2140 if (bdev) {
2141 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01002142 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002144 destroy_swap_extents(p);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002145 swap_cgroup_swapoff(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146bad_swap_2:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002147 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002150 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 vfree(swap_map);
2152 if (swap_file)
2153 filp_close(swap_file, NULL);
2154out:
2155 if (page && !IS_ERR(page)) {
2156 kunmap(page);
2157 page_cache_release(page);
2158 }
2159 if (name)
2160 putname(name);
2161 if (did_down) {
2162 if (!error)
2163 inode->i_flags |= S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002164 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 }
2166 return error;
2167}
2168
2169void si_swapinfo(struct sysinfo *val)
2170{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002171 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 unsigned long nr_to_be_unused = 0;
2173
Hugh Dickins5d337b92005-09-03 15:54:41 -07002174 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002175 for (type = 0; type < nr_swapfiles; type++) {
2176 struct swap_info_struct *si = swap_info[type];
2177
2178 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2179 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 }
2181 val->freeswap = nr_swap_pages + nr_to_be_unused;
2182 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002183 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184}
2185
2186/*
2187 * Verify that a swap entry is valid and increment its swap map count.
2188 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002189 * Returns error code in following case.
2190 * - success -> 0
2191 * - swp_entry is invalid -> EINVAL
2192 * - swp_entry is migration entry -> EINVAL
2193 * - swap-cache reference is requested but there is already one. -> EEXIST
2194 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002195 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002197static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002199 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002201 unsigned char count;
2202 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002203 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Andi Kleena7420aa2009-09-16 11:50:05 +02002205 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002206 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002207
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 type = swp_type(entry);
2209 if (type >= nr_swapfiles)
2210 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002211 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 offset = swp_offset(entry);
2213
Hugh Dickins5d337b92005-09-03 15:54:41 -07002214 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002215 if (unlikely(offset >= p->max))
2216 goto unlock_out;
2217
Hugh Dickins253d5532009-12-14 17:58:44 -08002218 count = p->swap_map[offset];
2219 has_cache = count & SWAP_HAS_CACHE;
2220 count &= ~SWAP_HAS_CACHE;
2221 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002222
Hugh Dickins253d5532009-12-14 17:58:44 -08002223 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002224
2225 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002226 if (!has_cache && count)
2227 has_cache = SWAP_HAS_CACHE;
2228 else if (has_cache) /* someone else added cache */
2229 err = -EEXIST;
2230 else /* no users remaining */
2231 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002232
2233 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002234
Hugh Dickins570a335b2009-12-14 17:58:46 -08002235 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2236 count += usage;
2237 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002238 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002239 else if (swap_count_continued(p, offset, count))
2240 count = COUNT_CONTINUED;
2241 else
2242 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002243 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002244 err = -ENOENT; /* unused swap entry */
2245
2246 p->swap_map[offset] = count | has_cache;
2247
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002248unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002249 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002251 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253bad_file:
2254 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2255 goto out;
2256}
Hugh Dickins253d5532009-12-14 17:58:44 -08002257
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002258/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002259 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2260 * (in which case its reference count is never incremented).
2261 */
2262void swap_shmem_alloc(swp_entry_t entry)
2263{
2264 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2265}
2266
2267/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002268 * Increase reference count of swap entry by 1.
2269 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2270 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2271 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2272 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002273 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002274int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002275{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002276 int err = 0;
2277
2278 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2279 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2280 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002281}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002283/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002284 * @entry: swap entry for which we allocate swap cache.
2285 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002286 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002287 * This can return error codes. Returns 0 at success.
2288 * -EBUSY means there is a swap cache.
2289 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002290 */
2291int swapcache_prepare(swp_entry_t entry)
2292{
Hugh Dickins253d5532009-12-14 17:58:44 -08002293 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002294}
2295
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07002297 * swap_lock prevents swap_map being freed. Don't grab an extra
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 * reference on the swaphandle, it doesn't matter if it becomes unused.
2299 */
2300int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2301{
Hugh Dickins89528982008-02-04 22:28:45 -08002302 struct swap_info_struct *si;
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002303 int our_page_cluster = page_cluster;
Hugh Dickins89528982008-02-04 22:28:45 -08002304 pgoff_t target, toff;
2305 pgoff_t base, end;
2306 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002308 if (!our_page_cluster) /* no readahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 return 0;
Hugh Dickins89528982008-02-04 22:28:45 -08002310
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002311 si = swap_info[swp_type(entry)];
Hugh Dickins89528982008-02-04 22:28:45 -08002312 target = swp_offset(entry);
2313 base = (target >> our_page_cluster) << our_page_cluster;
2314 end = base + (1 << our_page_cluster);
2315 if (!base) /* first page is swap header */
2316 base++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Hugh Dickins5d337b92005-09-03 15:54:41 -07002318 spin_lock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002319 if (end > si->max) /* don't go beyond end of map */
2320 end = si->max;
2321
2322 /* Count contiguous allocated slots above our target */
2323 for (toff = target; ++toff < end; nr_pages++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 /* Don't read in free or bad pages */
Hugh Dickins89528982008-02-04 22:28:45 -08002325 if (!si->swap_map[toff])
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002327 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 break;
Hugh Dickins89528982008-02-04 22:28:45 -08002329 }
2330 /* Count contiguous allocated slots below our target */
2331 for (toff = target; --toff >= base; nr_pages++) {
2332 /* Don't read in free or bad pages */
2333 if (!si->swap_map[toff])
2334 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002335 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Hugh Dickins89528982008-02-04 22:28:45 -08002336 break;
2337 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07002338 spin_unlock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002339
2340 /*
2341 * Indicate starting offset, and return number of pages to get:
2342 * if only 1, say 0, since there's then no readahead to be done.
2343 */
2344 *offset = ++toff;
2345 return nr_pages? ++nr_pages: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346}
Hugh Dickins570a335b2009-12-14 17:58:46 -08002347
2348/*
2349 * add_swap_count_continuation - called when a swap count is duplicated
2350 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2351 * page of the original vmalloc'ed swap_map, to hold the continuation count
2352 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2353 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2354 *
2355 * These continuation pages are seldom referenced: the common paths all work
2356 * on the original swap_map, only referring to a continuation page when the
2357 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2358 *
2359 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2360 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2361 * can be called after dropping locks.
2362 */
2363int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2364{
2365 struct swap_info_struct *si;
2366 struct page *head;
2367 struct page *page;
2368 struct page *list_page;
2369 pgoff_t offset;
2370 unsigned char count;
2371
2372 /*
2373 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2374 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2375 */
2376 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2377
2378 si = swap_info_get(entry);
2379 if (!si) {
2380 /*
2381 * An acceptable race has occurred since the failing
2382 * __swap_duplicate(): the swap entry has been freed,
2383 * perhaps even the whole swap_map cleared for swapoff.
2384 */
2385 goto outer;
2386 }
2387
2388 offset = swp_offset(entry);
2389 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2390
2391 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2392 /*
2393 * The higher the swap count, the more likely it is that tasks
2394 * will race to add swap count continuation: we need to avoid
2395 * over-provisioning.
2396 */
2397 goto out;
2398 }
2399
2400 if (!page) {
2401 spin_unlock(&swap_lock);
2402 return -ENOMEM;
2403 }
2404
2405 /*
2406 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2407 * no architecture is using highmem pages for kernel pagetables: so it
2408 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2409 */
2410 head = vmalloc_to_page(si->swap_map + offset);
2411 offset &= ~PAGE_MASK;
2412
2413 /*
2414 * Page allocation does not initialize the page's lru field,
2415 * but it does always reset its private field.
2416 */
2417 if (!page_private(head)) {
2418 BUG_ON(count & COUNT_CONTINUED);
2419 INIT_LIST_HEAD(&head->lru);
2420 set_page_private(head, SWP_CONTINUED);
2421 si->flags |= SWP_CONTINUED;
2422 }
2423
2424 list_for_each_entry(list_page, &head->lru, lru) {
2425 unsigned char *map;
2426
2427 /*
2428 * If the previous map said no continuation, but we've found
2429 * a continuation page, free our allocation and use this one.
2430 */
2431 if (!(count & COUNT_CONTINUED))
2432 goto out;
2433
2434 map = kmap_atomic(list_page, KM_USER0) + offset;
2435 count = *map;
2436 kunmap_atomic(map, KM_USER0);
2437
2438 /*
2439 * If this continuation count now has some space in it,
2440 * free our allocation and use this one.
2441 */
2442 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2443 goto out;
2444 }
2445
2446 list_add_tail(&page->lru, &head->lru);
2447 page = NULL; /* now it's attached, don't free it */
2448out:
2449 spin_unlock(&swap_lock);
2450outer:
2451 if (page)
2452 __free_page(page);
2453 return 0;
2454}
2455
2456/*
2457 * swap_count_continued - when the original swap_map count is incremented
2458 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2459 * into, carry if so, or else fail until a new continuation page is allocated;
2460 * when the original swap_map count is decremented from 0 with continuation,
2461 * borrow from the continuation and report whether it still holds more.
2462 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2463 */
2464static bool swap_count_continued(struct swap_info_struct *si,
2465 pgoff_t offset, unsigned char count)
2466{
2467 struct page *head;
2468 struct page *page;
2469 unsigned char *map;
2470
2471 head = vmalloc_to_page(si->swap_map + offset);
2472 if (page_private(head) != SWP_CONTINUED) {
2473 BUG_ON(count & COUNT_CONTINUED);
2474 return false; /* need to add count continuation */
2475 }
2476
2477 offset &= ~PAGE_MASK;
2478 page = list_entry(head->lru.next, struct page, lru);
2479 map = kmap_atomic(page, KM_USER0) + offset;
2480
2481 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2482 goto init_map; /* jump over SWAP_CONT_MAX checks */
2483
2484 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2485 /*
2486 * Think of how you add 1 to 999
2487 */
2488 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2489 kunmap_atomic(map, KM_USER0);
2490 page = list_entry(page->lru.next, struct page, lru);
2491 BUG_ON(page == head);
2492 map = kmap_atomic(page, KM_USER0) + offset;
2493 }
2494 if (*map == SWAP_CONT_MAX) {
2495 kunmap_atomic(map, KM_USER0);
2496 page = list_entry(page->lru.next, struct page, lru);
2497 if (page == head)
2498 return false; /* add count continuation */
2499 map = kmap_atomic(page, KM_USER0) + offset;
2500init_map: *map = 0; /* we didn't zero the page */
2501 }
2502 *map += 1;
2503 kunmap_atomic(map, KM_USER0);
2504 page = list_entry(page->lru.prev, struct page, lru);
2505 while (page != head) {
2506 map = kmap_atomic(page, KM_USER0) + offset;
2507 *map = COUNT_CONTINUED;
2508 kunmap_atomic(map, KM_USER0);
2509 page = list_entry(page->lru.prev, struct page, lru);
2510 }
2511 return true; /* incremented */
2512
2513 } else { /* decrementing */
2514 /*
2515 * Think of how you subtract 1 from 1000
2516 */
2517 BUG_ON(count != COUNT_CONTINUED);
2518 while (*map == COUNT_CONTINUED) {
2519 kunmap_atomic(map, KM_USER0);
2520 page = list_entry(page->lru.next, struct page, lru);
2521 BUG_ON(page == head);
2522 map = kmap_atomic(page, KM_USER0) + offset;
2523 }
2524 BUG_ON(*map == 0);
2525 *map -= 1;
2526 if (*map == 0)
2527 count = 0;
2528 kunmap_atomic(map, KM_USER0);
2529 page = list_entry(page->lru.prev, struct page, lru);
2530 while (page != head) {
2531 map = kmap_atomic(page, KM_USER0) + offset;
2532 *map = SWAP_CONT_MAX | count;
2533 count = COUNT_CONTINUED;
2534 kunmap_atomic(map, KM_USER0);
2535 page = list_entry(page->lru.prev, struct page, lru);
2536 }
2537 return count == COUNT_CONTINUED;
2538 }
2539}
2540
2541/*
2542 * free_swap_count_continuations - swapoff free all the continuation pages
2543 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2544 */
2545static void free_swap_count_continuations(struct swap_info_struct *si)
2546{
2547 pgoff_t offset;
2548
2549 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2550 struct page *head;
2551 head = vmalloc_to_page(si->swap_map + offset);
2552 if (page_private(head)) {
2553 struct list_head *this, *next;
2554 list_for_each_safe(this, next, &head->lru) {
2555 struct page *page;
2556 page = list_entry(this, struct page, lru);
2557 list_del(this);
2558 __free_page(page);
2559 }
2560 }
2561 }
2562}