blob: 6cd0a8f90dc74119d9db32fae7fb9f7ea01de13c [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34#include <asm/pgtable.h>
35#include <asm/tlbflush.h>
36#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080037#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Hugh Dickins570a3352009-12-14 17:58:46 -080039static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
40 unsigned char);
41static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080042static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a3352009-12-14 17:58:46 -080043
Adrian Bunk7c363b82008-07-25 19:46:24 -070044static DEFINE_SPINLOCK(swap_lock);
45static unsigned int nr_swapfiles;
Hugh Dickinsb9627162009-01-06 14:39:41 -080046long nr_swap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070048static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050static const char Bad_file[] = "Bad swap file entry ";
51static const char Unused_file[] = "Unused swap file entry ";
52static const char Bad_offset[] = "Bad swap offset entry ";
53static const char Unused_offset[] = "Unused swap offset entry ";
54
Adrian Bunk7c363b82008-07-25 19:46:24 -070055static struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Hugh Dickinsefa90a92009-12-14 17:58:41 -080057static struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Ingo Molnarfc0abb12006-01-18 17:42:33 -080059static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Hugh Dickins8d69aae2009-12-14 17:58:45 -080061static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070062{
Hugh Dickins570a3352009-12-14 17:58:46 -080063 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070064}
65
Hugh Dickinsefa90a92009-12-14 17:58:41 -080066/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070067static int
68__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
69{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080070 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070071 struct page *page;
72 int ret = 0;
73
74 page = find_get_page(&swapper_space, entry.val);
75 if (!page)
76 return 0;
77 /*
78 * This function is called from scan_swap_map() and it's called
79 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
80 * We have to use trylock for avoiding deadlock. This is a special
81 * case and you should use try_to_free_swap() with explicit lock_page()
82 * in usual operations.
83 */
84 if (trylock_page(page)) {
85 ret = try_to_free_swap(page);
86 unlock_page(page);
87 }
88 page_cache_release(page);
89 return ret;
90}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070091
Linus Torvalds1da177e2005-04-16 15:20:36 -070092/*
93 * We need this because the bdev->unplug_fn can sleep and we cannot
Hugh Dickins5d337b92005-09-03 15:54:41 -070094 * hold swap_lock while calling the unplug_fn. And swap_lock
Ingo Molnarfc0abb12006-01-18 17:42:33 -080095 * cannot be turned into a mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 */
97static DECLARE_RWSEM(swap_unplug_sem);
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
100{
101 swp_entry_t entry;
102
103 down_read(&swap_unplug_sem);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700104 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 if (PageSwapCache(page)) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800106 struct block_device *bdev = swap_info[swp_type(entry)]->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 struct backing_dev_info *bdi;
108
109 /*
110 * If the page is removed from swapcache from under us (with a
111 * racy try_to_unuse/swapoff) we need an additional reference
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700112 * count to avoid reading garbage from page_private(page) above.
113 * If the WARN_ON triggers during a swapoff it maybe the race
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * condition and it's harmless. However if it triggers without
115 * swapoff it signals a problem.
116 */
117 WARN_ON(page_count(page) <= 1);
118
119 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
McMullan, Jasonba323112005-05-16 21:53:40 -0700120 blk_run_backing_dev(bdi, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 }
122 up_read(&swap_unplug_sem);
123}
124
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800125/*
126 * swapon tell device that all the old swap contents can be discarded,
127 * to allow the swap device to optimize its wear-levelling.
128 */
129static int discard_swap(struct swap_info_struct *si)
130{
131 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800132 sector_t start_block;
133 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800134 int err = 0;
135
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800136 /* Do not discard the swap header page! */
137 se = &si->first_swap_extent;
138 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
139 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
140 if (nr_blocks) {
141 err = blkdev_issue_discard(si->bdev, start_block,
142 nr_blocks, GFP_KERNEL, DISCARD_FL_BARRIER);
143 if (err)
144 return err;
145 cond_resched();
146 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800147
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800148 list_for_each_entry(se, &si->first_swap_extent.list, list) {
149 start_block = se->start_block << (PAGE_SHIFT - 9);
150 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800151
152 err = blkdev_issue_discard(si->bdev, start_block,
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800153 nr_blocks, GFP_KERNEL, DISCARD_FL_BARRIER);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800154 if (err)
155 break;
156
157 cond_resched();
158 }
159 return err; /* That will often be -EOPNOTSUPP */
160}
161
Hugh Dickins7992fde2009-01-06 14:39:53 -0800162/*
163 * swap allocation tell device that a cluster of swap can now be discarded,
164 * to allow the swap device to optimize its wear-levelling.
165 */
166static void discard_swap_cluster(struct swap_info_struct *si,
167 pgoff_t start_page, pgoff_t nr_pages)
168{
169 struct swap_extent *se = si->curr_swap_extent;
170 int found_extent = 0;
171
172 while (nr_pages) {
173 struct list_head *lh;
174
175 if (se->start_page <= start_page &&
176 start_page < se->start_page + se->nr_pages) {
177 pgoff_t offset = start_page - se->start_page;
178 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800179 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800180
181 if (nr_blocks > nr_pages)
182 nr_blocks = nr_pages;
183 start_page += nr_blocks;
184 nr_pages -= nr_blocks;
185
186 if (!found_extent++)
187 si->curr_swap_extent = se;
188
189 start_block <<= PAGE_SHIFT - 9;
190 nr_blocks <<= PAGE_SHIFT - 9;
191 if (blkdev_issue_discard(si->bdev, start_block,
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800192 nr_blocks, GFP_NOIO, DISCARD_FL_BARRIER))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800193 break;
194 }
195
196 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800197 se = list_entry(lh, struct swap_extent, list);
198 }
199}
200
201static int wait_for_discard(void *word)
202{
203 schedule();
204 return 0;
205}
206
Hugh Dickins048c27f2005-09-03 15:54:40 -0700207#define SWAPFILE_CLUSTER 256
208#define LATENCY_LIMIT 256
209
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700210static inline unsigned long scan_swap_map(struct swap_info_struct *si,
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800211 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800213 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800214 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800215 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700216 int latency_ration = LATENCY_LIMIT;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800217 int found_free_cluster = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800219 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700220 * We try to cluster swap pages by allocating them sequentially
221 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
222 * way, however, we resort to first-free allocation, starting
223 * a new cluster. This prevents us from scattering swap pages
224 * all over the entire swap partition, so that we reduce
225 * overall disk seek times between swap pages. -- sct
226 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800227 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700228 */
229
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700230 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800231 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800232
233 if (unlikely(!si->cluster_nr--)) {
234 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
235 si->cluster_nr = SWAPFILE_CLUSTER - 1;
236 goto checks;
237 }
Hugh Dickins7992fde2009-01-06 14:39:53 -0800238 if (si->flags & SWP_DISCARDABLE) {
239 /*
240 * Start range check on racing allocations, in case
241 * they overlap the cluster we eventually decide on
242 * (we scan without swap_lock to allow preemption).
243 * It's hardly conceivable that cluster_nr could be
244 * wrapped during our scan, but don't depend on it.
245 */
246 if (si->lowest_alloc)
247 goto checks;
248 si->lowest_alloc = si->max;
249 si->highest_alloc = 0;
250 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700251 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700252
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800253 /*
254 * If seek is expensive, start searching for new cluster from
255 * start of partition, to minimize the span of allocated swap.
256 * But if seek is cheap, search from our current position, so
257 * that swap is allocated from all over the partition: if the
258 * Flash Translation Layer only remaps within limited zones,
259 * we don't want to wear out the first zone too quickly.
260 */
261 if (!(si->flags & SWP_SOLIDSTATE))
262 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700263 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
264
265 /* Locate the first empty (unaligned) cluster */
266 for (; last_in_cluster <= si->highest_bit; offset++) {
267 if (si->swap_map[offset])
268 last_in_cluster = offset + SWAPFILE_CLUSTER;
269 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700270 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800271 offset -= SWAPFILE_CLUSTER - 1;
272 si->cluster_next = offset;
273 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800274 found_free_cluster = 1;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800275 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700277 if (unlikely(--latency_ration < 0)) {
278 cond_resched();
279 latency_ration = LATENCY_LIMIT;
280 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700281 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800282
283 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800284 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
285
286 /* Locate the first empty (unaligned) cluster */
287 for (; last_in_cluster < scan_base; offset++) {
288 if (si->swap_map[offset])
289 last_in_cluster = offset + SWAPFILE_CLUSTER;
290 else if (offset == last_in_cluster) {
291 spin_lock(&swap_lock);
292 offset -= SWAPFILE_CLUSTER - 1;
293 si->cluster_next = offset;
294 si->cluster_nr = SWAPFILE_CLUSTER - 1;
295 found_free_cluster = 1;
296 goto checks;
297 }
298 if (unlikely(--latency_ration < 0)) {
299 cond_resched();
300 latency_ration = LATENCY_LIMIT;
301 }
302 }
303
304 offset = scan_base;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700305 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800306 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800307 si->lowest_alloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700309
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800310checks:
311 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700312 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700313 if (!si->highest_bit)
314 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800315 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800316 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700317
318 /* reuse swap entry of cache-only swap if not busy. */
319 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
320 int swap_was_freed;
321 spin_unlock(&swap_lock);
322 swap_was_freed = __try_to_reclaim_swap(si, offset);
323 spin_lock(&swap_lock);
324 /* entry was freed successfully, try to use this again */
325 if (swap_was_freed)
326 goto checks;
327 goto scan; /* check next one */
328 }
329
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800330 if (si->swap_map[offset])
331 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700332
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800333 if (offset == si->lowest_bit)
334 si->lowest_bit++;
335 if (offset == si->highest_bit)
336 si->highest_bit--;
337 si->inuse_pages++;
338 if (si->inuse_pages == si->pages) {
339 si->lowest_bit = si->max;
340 si->highest_bit = 0;
341 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800342 si->swap_map[offset] = usage;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800343 si->cluster_next = offset + 1;
344 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800345
346 if (si->lowest_alloc) {
347 /*
348 * Only set when SWP_DISCARDABLE, and there's a scan
349 * for a free cluster in progress or just completed.
350 */
351 if (found_free_cluster) {
352 /*
353 * To optimize wear-levelling, discard the
354 * old data of the cluster, taking care not to
355 * discard any of its pages that have already
356 * been allocated by racing tasks (offset has
357 * already stepped over any at the beginning).
358 */
359 if (offset < si->highest_alloc &&
360 si->lowest_alloc <= last_in_cluster)
361 last_in_cluster = si->lowest_alloc - 1;
362 si->flags |= SWP_DISCARDING;
363 spin_unlock(&swap_lock);
364
365 if (offset < last_in_cluster)
366 discard_swap_cluster(si, offset,
367 last_in_cluster - offset + 1);
368
369 spin_lock(&swap_lock);
370 si->lowest_alloc = 0;
371 si->flags &= ~SWP_DISCARDING;
372
373 smp_mb(); /* wake_up_bit advises this */
374 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
375
376 } else if (si->flags & SWP_DISCARDING) {
377 /*
378 * Delay using pages allocated by racing tasks
379 * until the whole discard has been issued. We
380 * could defer that delay until swap_writepage,
381 * but it's easier to keep this self-contained.
382 */
383 spin_unlock(&swap_lock);
384 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
385 wait_for_discard, TASK_UNINTERRUPTIBLE);
386 spin_lock(&swap_lock);
387 } else {
388 /*
389 * Note pages allocated by racing tasks while
390 * scan for a free cluster is in progress, so
391 * that its final discard can exclude them.
392 */
393 if (offset < si->lowest_alloc)
394 si->lowest_alloc = offset;
395 if (offset > si->highest_alloc)
396 si->highest_alloc = offset;
397 }
398 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800399 return offset;
400
401scan:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700402 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700403 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700404 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700405 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700406 goto checks;
407 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700408 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
409 spin_lock(&swap_lock);
410 goto checks;
411 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700412 if (unlikely(--latency_ration < 0)) {
413 cond_resched();
414 latency_ration = LATENCY_LIMIT;
415 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700416 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800417 offset = si->lowest_bit;
418 while (++offset < scan_base) {
419 if (!si->swap_map[offset]) {
420 spin_lock(&swap_lock);
421 goto checks;
422 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700423 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
424 spin_lock(&swap_lock);
425 goto checks;
426 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800427 if (unlikely(--latency_ration < 0)) {
428 cond_resched();
429 latency_ration = LATENCY_LIMIT;
430 }
431 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700432 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700433
434no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700435 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return 0;
437}
438
439swp_entry_t get_swap_page(void)
440{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700441 struct swap_info_struct *si;
442 pgoff_t offset;
443 int type, next;
444 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Hugh Dickins5d337b92005-09-03 15:54:41 -0700446 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700448 goto noswap;
449 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700451 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800452 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700453 next = si->next;
454 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800455 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700456 next = swap_list.head;
457 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700459
460 if (!si->highest_bit)
461 continue;
462 if (!(si->flags & SWP_WRITEOK))
463 continue;
464
465 swap_list.next = next;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700466 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800467 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700468 if (offset) {
469 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700470 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700471 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700472 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700474
475 nr_swap_pages++;
476noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700477 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700478 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700481/* The only caller of this function is now susupend routine */
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800482swp_entry_t get_swap_page_of_type(int type)
483{
484 struct swap_info_struct *si;
485 pgoff_t offset;
486
487 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800488 si = swap_info[type];
489 if (si && (si->flags & SWP_WRITEOK)) {
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800490 nr_swap_pages--;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700491 /* This is called for allocating swap entry, not cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800492 offset = scan_swap_map(si, 1);
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800493 if (offset) {
494 spin_unlock(&swap_lock);
495 return swp_entry(type, offset);
496 }
497 nr_swap_pages++;
498 }
499 spin_unlock(&swap_lock);
500 return (swp_entry_t) {0};
501}
502
Hugh Dickins73c34b62009-12-14 17:58:43 -0800503static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800505 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 unsigned long offset, type;
507
508 if (!entry.val)
509 goto out;
510 type = swp_type(entry);
511 if (type >= nr_swapfiles)
512 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800513 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (!(p->flags & SWP_USED))
515 goto bad_device;
516 offset = swp_offset(entry);
517 if (offset >= p->max)
518 goto bad_offset;
519 if (!p->swap_map[offset])
520 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700521 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 return p;
523
524bad_free:
525 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
526 goto out;
527bad_offset:
528 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
529 goto out;
530bad_device:
531 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
532 goto out;
533bad_nofile:
534 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
535out:
536 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800537}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800539static unsigned char swap_entry_free(struct swap_info_struct *p,
540 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
Hugh Dickins253d5532009-12-14 17:58:44 -0800542 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800543 unsigned char count;
544 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700546 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800547 has_cache = count & SWAP_HAS_CACHE;
548 count &= ~SWAP_HAS_CACHE;
549
550 if (usage == SWAP_HAS_CACHE) {
551 VM_BUG_ON(!has_cache);
552 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800553 } else if (count == SWAP_MAP_SHMEM) {
554 /*
555 * Or we could insist on shmem.c using a special
556 * swap_shmem_free() and free_shmem_swap_and_cache()...
557 */
558 count = 0;
Hugh Dickins570a3352009-12-14 17:58:46 -0800559 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
560 if (count == COUNT_CONTINUED) {
561 if (swap_count_continued(p, offset, count))
562 count = SWAP_MAP_MAX | COUNT_CONTINUED;
563 else
564 count = SWAP_MAP_MAX;
565 } else
566 count--;
567 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800568
569 if (!count)
570 mem_cgroup_uncharge_swap(entry);
571
572 usage = count | has_cache;
573 p->swap_map[offset] = usage;
574
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700575 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800576 if (!usage) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700577 if (offset < p->lowest_bit)
578 p->lowest_bit = offset;
579 if (offset > p->highest_bit)
580 p->highest_bit = offset;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800581 if (swap_list.next >= 0 &&
582 p->prio > swap_info[swap_list.next]->prio)
583 swap_list.next = p->type;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700584 nr_swap_pages++;
585 p->inuse_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800587
588 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589}
590
591/*
592 * Caller has made sure that the swapdevice corresponding to entry
593 * is still around or has not been recycled.
594 */
595void swap_free(swp_entry_t entry)
596{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800597 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599 p = swap_info_get(entry);
600 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800601 swap_entry_free(p, entry, 1);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700602 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 }
604}
605
606/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700607 * Called after dropping swapcache to decrease refcnt to swap entries.
608 */
609void swapcache_free(swp_entry_t entry, struct page *page)
610{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700611 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800612 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700613
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700614 p = swap_info_get(entry);
615 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800616 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
617 if (page)
618 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700619 spin_unlock(&swap_lock);
620 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700621}
622
623/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700624 * How many references to page are currently swapped out?
Hugh Dickins570a3352009-12-14 17:58:46 -0800625 * This does not give an exact answer when swap count is continued,
626 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700628static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700630 int count = 0;
631 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 swp_entry_t entry;
633
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700634 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 p = swap_info_get(entry);
636 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700637 count = swap_count(p->swap_map[swp_offset(entry)]);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700638 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700640 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
642
643/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800644 * We can write to an anon page without COW if there are no other references
645 * to it. And as a side-effect, free up its swap: because the old content
646 * on disk will never be read, and seeking back there to write new content
647 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800649int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700651 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Hugh Dickins51726b12009-01-06 14:39:25 -0800653 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800654 if (unlikely(PageKsm(page)))
655 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700656 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800657 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700658 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800659 if (count == 1 && !PageWriteback(page)) {
660 delete_from_swap_cache(page);
661 SetPageDirty(page);
662 }
663 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800664 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
667/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800668 * If swap is getting full, or if there are no more mappings of this page,
669 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800671int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
Hugh Dickins51726b12009-01-06 14:39:25 -0800673 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 if (!PageSwapCache(page))
676 return 0;
677 if (PageWriteback(page))
678 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800679 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 return 0;
681
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800682 delete_from_swap_cache(page);
683 SetPageDirty(page);
684 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700685}
686
687/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 * Free the swap entry like above, but also try to
689 * free the page cache entry if it is the last user.
690 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800691int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800693 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 struct page *page = NULL;
695
Andi Kleena7420aa2009-09-16 11:50:05 +0200696 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800697 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 p = swap_info_get(entry);
700 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800701 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Nick Piggin93fac702006-03-31 02:29:56 -0800702 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700703 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800704 page_cache_release(page);
705 page = NULL;
706 }
707 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700708 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
710 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800711 /*
712 * Not mapped elsewhere, or swap space full? Free it!
713 * Also recheck PageSwapCache now page is locked (above).
714 */
Nick Piggin93fac702006-03-31 02:29:56 -0800715 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800716 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 delete_from_swap_cache(page);
718 SetPageDirty(page);
719 }
720 unlock_page(page);
721 page_cache_release(page);
722 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800723 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
725
Daisuke Nishimura02491442010-03-10 15:22:17 -0800726#ifdef CONFIG_CGROUP_MEM_RES_CTLR
727/**
728 * mem_cgroup_count_swap_user - count the user of a swap entry
729 * @ent: the swap entry to be checked
730 * @pagep: the pointer for the swap cache page of the entry to be stored
731 *
732 * Returns the number of the user of the swap entry. The number is valid only
733 * for swaps of anonymous pages.
734 * If the entry is found on swap cache, the page is stored to pagep with
735 * refcount of it being incremented.
736 */
737int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep)
738{
739 struct page *page;
740 struct swap_info_struct *p;
741 int count = 0;
742
743 page = find_get_page(&swapper_space, ent.val);
744 if (page)
745 count += page_mapcount(page);
746 p = swap_info_get(ent);
747 if (p) {
748 count += swap_count(p->swap_map[swp_offset(ent)]);
749 spin_unlock(&swap_lock);
750 }
751
752 *pagep = page;
753 return count;
754}
755#endif
756
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200757#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800758/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800759 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800760 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800761 * @offset - number of the PAGE_SIZE-sized block of the device, starting
762 * from 0, in which the swap header is expected to be located.
763 *
764 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800765 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800766int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800767{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800768 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800769 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800770
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800771 if (device)
772 bdev = bdget(device);
773
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800774 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800775 for (type = 0; type < nr_swapfiles; type++) {
776 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800777
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800778 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800779 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700780
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800781 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800782 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200783 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800784
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800785 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800786 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800787 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800788 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800789 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800790
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800791 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800792 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200793 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800794
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800795 spin_unlock(&swap_lock);
796 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800797 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800798 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800799 }
800 }
801 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800802 if (bdev)
803 bdput(bdev);
804
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800805 return -ENODEV;
806}
807
808/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800809 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
810 * corresponding to given index in swap_info (swap type).
811 */
812sector_t swapdev_block(int type, pgoff_t offset)
813{
814 struct block_device *bdev;
815
816 if ((unsigned int)type >= nr_swapfiles)
817 return 0;
818 if (!(swap_info[type]->flags & SWP_WRITEOK))
819 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800820 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800821}
822
823/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800824 * Return either the total number of swap pages of given type, or the number
825 * of free pages of that type (depending on @free)
826 *
827 * This is needed for software suspend
828 */
829unsigned int count_swap_pages(int type, int free)
830{
831 unsigned int n = 0;
832
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800833 spin_lock(&swap_lock);
834 if ((unsigned int)type < nr_swapfiles) {
835 struct swap_info_struct *sis = swap_info[type];
836
837 if (sis->flags & SWP_WRITEOK) {
838 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800839 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800840 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800841 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800842 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800843 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800844 return n;
845}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800846#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700849 * No need to decide whether this PTE shares the swap entry with others,
850 * just let do_wp_page work it out if a write is requested later - to
851 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800853static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 unsigned long addr, swp_entry_t entry, struct page *page)
855{
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800856 struct mem_cgroup *ptr = NULL;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800857 spinlock_t *ptl;
858 pte_t *pte;
859 int ret = 1;
860
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800861 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800862 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800863 goto out_nolock;
864 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800865
866 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
867 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
868 if (ret > 0)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800869 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800870 ret = 0;
871 goto out;
872 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800873
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800874 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800875 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 get_page(page);
877 set_pte_at(vma->vm_mm, addr, pte,
878 pte_mkold(mk_pte(page, vma->vm_page_prot)));
879 page_add_anon_rmap(page, vma, addr);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800880 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 swap_free(entry);
882 /*
883 * Move the page to the active list so it is not
884 * immediately swapped out again after swapon.
885 */
886 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800887out:
888 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800889out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800890 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891}
892
893static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
894 unsigned long addr, unsigned long end,
895 swp_entry_t entry, struct page *page)
896{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700898 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800899 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
Hugh Dickins044d66c2008-02-07 00:14:04 -0800901 /*
902 * We don't actually need pte lock while scanning for swp_pte: since
903 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
904 * page table while we're scanning; though it could get zapped, and on
905 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
906 * of unmatched parts which look like swp_pte, so unuse_pte must
907 * recheck under pte lock. Scanning without pte lock lets it be
908 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
909 */
910 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 do {
912 /*
913 * swapoff spends a _lot_ of time in this loop!
914 * Test inline before going to call unuse_pte.
915 */
916 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800917 pte_unmap(pte);
918 ret = unuse_pte(vma, pmd, addr, entry, page);
919 if (ret)
920 goto out;
921 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
923 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800924 pte_unmap(pte - 1);
925out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800926 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927}
928
929static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
930 unsigned long addr, unsigned long end,
931 swp_entry_t entry, struct page *page)
932{
933 pmd_t *pmd;
934 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800935 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 pmd = pmd_offset(pud, addr);
938 do {
939 next = pmd_addr_end(addr, end);
940 if (pmd_none_or_clear_bad(pmd))
941 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800942 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
943 if (ret)
944 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 } while (pmd++, addr = next, addr != end);
946 return 0;
947}
948
949static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
950 unsigned long addr, unsigned long end,
951 swp_entry_t entry, struct page *page)
952{
953 pud_t *pud;
954 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800955 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957 pud = pud_offset(pgd, addr);
958 do {
959 next = pud_addr_end(addr, end);
960 if (pud_none_or_clear_bad(pud))
961 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800962 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
963 if (ret)
964 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 } while (pud++, addr = next, addr != end);
966 return 0;
967}
968
969static int unuse_vma(struct vm_area_struct *vma,
970 swp_entry_t entry, struct page *page)
971{
972 pgd_t *pgd;
973 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800974 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800976 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 addr = page_address_in_vma(page, vma);
978 if (addr == -EFAULT)
979 return 0;
980 else
981 end = addr + PAGE_SIZE;
982 } else {
983 addr = vma->vm_start;
984 end = vma->vm_end;
985 }
986
987 pgd = pgd_offset(vma->vm_mm, addr);
988 do {
989 next = pgd_addr_end(addr, end);
990 if (pgd_none_or_clear_bad(pgd))
991 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800992 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
993 if (ret)
994 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 } while (pgd++, addr = next, addr != end);
996 return 0;
997}
998
999static int unuse_mm(struct mm_struct *mm,
1000 swp_entry_t entry, struct page *page)
1001{
1002 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001003 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
1005 if (!down_read_trylock(&mm->mmap_sem)) {
1006 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001007 * Activate page so shrink_inactive_list is unlikely to unmap
1008 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001010 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 unlock_page(page);
1012 down_read(&mm->mmap_sem);
1013 lock_page(page);
1014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001016 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 break;
1018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001020 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021}
1022
1023/*
1024 * Scan swap_map from current position to next entry still in use.
1025 * Recycle to start on reaching the end, returning 0 when empty.
1026 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001027static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1028 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001030 unsigned int max = si->max;
1031 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001032 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001035 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 * for whether an entry is in use, not modifying it; false
1037 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001038 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 */
1040 for (;;) {
1041 if (++i >= max) {
1042 if (!prev) {
1043 i = 0;
1044 break;
1045 }
1046 /*
1047 * No entries in use at top of swap_map,
1048 * loop back to start and recheck there.
1049 */
1050 max = prev + 1;
1051 prev = 0;
1052 i = 1;
1053 }
1054 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001055 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 break;
1057 }
1058 return i;
1059}
1060
1061/*
1062 * We completely avoid races by reading each swap page in advance,
1063 * and then search for the process using it. All the necessary
1064 * page table adjustments can then be made atomically.
1065 */
1066static int try_to_unuse(unsigned int type)
1067{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001068 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001070 unsigned char *swap_map;
1071 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 struct page *page;
1073 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001074 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077 /*
1078 * When searching mms for an entry, a good strategy is to
1079 * start at the first mm we freed the previous entry from
1080 * (though actually we don't notice whether we or coincidence
1081 * freed the entry). Initialize this start_mm with a hold.
1082 *
1083 * A simpler strategy would be to start at the last mm we
1084 * freed the previous entry from; but that would take less
1085 * advantage of mmlist ordering, which clusters forked mms
1086 * together, child after parent. If we race with dup_mmap(), we
1087 * prefer to resolve parent before child, lest we miss entries
1088 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a3352009-12-14 17:58:46 -08001089 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 */
1091 start_mm = &init_mm;
1092 atomic_inc(&init_mm.mm_users);
1093
1094 /*
1095 * Keep on scanning until all entries have gone. Usually,
1096 * one pass through swap_map is enough, but not necessarily:
1097 * there are races when an instance of an entry might be missed.
1098 */
1099 while ((i = find_next_to_unuse(si, i)) != 0) {
1100 if (signal_pending(current)) {
1101 retval = -EINTR;
1102 break;
1103 }
1104
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001105 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 * Get a page for the entry, using the existing swap
1107 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001108 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 */
1110 swap_map = &si->swap_map[i];
1111 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001112 page = read_swap_cache_async(entry,
1113 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 if (!page) {
1115 /*
1116 * Either swap_duplicate() failed because entry
1117 * has been freed independently, and will not be
1118 * reused since sys_swapoff() already disabled
1119 * allocation from here, or alloc_page() failed.
1120 */
1121 if (!*swap_map)
1122 continue;
1123 retval = -ENOMEM;
1124 break;
1125 }
1126
1127 /*
1128 * Don't hold on to start_mm if it looks like exiting.
1129 */
1130 if (atomic_read(&start_mm->mm_users) == 1) {
1131 mmput(start_mm);
1132 start_mm = &init_mm;
1133 atomic_inc(&init_mm.mm_users);
1134 }
1135
1136 /*
1137 * Wait for and lock page. When do_swap_page races with
1138 * try_to_unuse, do_swap_page can handle the fault much
1139 * faster than try_to_unuse can locate the entry. This
1140 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1141 * defer to do_swap_page in such a case - in some tests,
1142 * do_swap_page and try_to_unuse repeatedly compete.
1143 */
1144 wait_on_page_locked(page);
1145 wait_on_page_writeback(page);
1146 lock_page(page);
1147 wait_on_page_writeback(page);
1148
1149 /*
1150 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001153 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1154 retval = shmem_unuse(entry, page);
1155 /* page has already been unlocked and released */
1156 if (retval < 0)
1157 break;
1158 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001160 if (swap_count(swcount) && start_mm != &init_mm)
1161 retval = unuse_mm(start_mm, entry, page);
1162
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001163 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 int set_start_mm = (*swap_map >= swcount);
1165 struct list_head *p = &start_mm->mmlist;
1166 struct mm_struct *new_start_mm = start_mm;
1167 struct mm_struct *prev_mm = start_mm;
1168 struct mm_struct *mm;
1169
1170 atomic_inc(&new_start_mm->mm_users);
1171 atomic_inc(&prev_mm->mm_users);
1172 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001173 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 (p = p->next) != &start_mm->mmlist) {
1175 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001176 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 spin_unlock(&mmlist_lock);
1179 mmput(prev_mm);
1180 prev_mm = mm;
1181
1182 cond_resched();
1183
1184 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001185 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001187 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001189 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001191
Bo Liu32c5fc12009-11-02 16:50:33 +00001192 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 mmput(new_start_mm);
1194 atomic_inc(&mm->mm_users);
1195 new_start_mm = mm;
1196 set_start_mm = 0;
1197 }
1198 spin_lock(&mmlist_lock);
1199 }
1200 spin_unlock(&mmlist_lock);
1201 mmput(prev_mm);
1202 mmput(start_mm);
1203 start_mm = new_start_mm;
1204 }
1205 if (retval) {
1206 unlock_page(page);
1207 page_cache_release(page);
1208 break;
1209 }
1210
1211 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 * If a reference remains (rare), we would like to leave
1213 * the page in the swap cache; but try_to_unmap could
1214 * then re-duplicate the entry once we drop page lock,
1215 * so we might loop indefinitely; also, that page could
1216 * not be swapped out to other storage meanwhile. So:
1217 * delete from cache even if there's another reference,
1218 * after ensuring that the data has been saved to disk -
1219 * since if the reference remains (rarer), it will be
1220 * read from disk into another page. Splitting into two
1221 * pages would be incorrect if swap supported "shared
1222 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001223 *
1224 * Given how unuse_vma() targets one particular offset
1225 * in an anon_vma, once the anon_vma has been determined,
1226 * this splitting happens to be just what is needed to
1227 * handle where KSM pages have been swapped out: re-reading
1228 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001230 if (swap_count(*swap_map) &&
1231 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 struct writeback_control wbc = {
1233 .sync_mode = WB_SYNC_NONE,
1234 };
1235
1236 swap_writepage(page, &wbc);
1237 lock_page(page);
1238 wait_on_page_writeback(page);
1239 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001240
1241 /*
1242 * It is conceivable that a racing task removed this page from
1243 * swap cache just before we acquired the page lock at the top,
1244 * or while we dropped it in unuse_mm(). The page might even
1245 * be back in swap cache on another swap area: that we must not
1246 * delete, since it may not have been written out to swap yet.
1247 */
1248 if (PageSwapCache(page) &&
1249 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001250 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
1252 /*
1253 * So we could skip searching mms once swap count went
1254 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001255 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 */
1257 SetPageDirty(page);
1258 unlock_page(page);
1259 page_cache_release(page);
1260
1261 /*
1262 * Make sure that we aren't completely killing
1263 * interactive performance.
1264 */
1265 cond_resched();
1266 }
1267
1268 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 return retval;
1270}
1271
1272/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001273 * After a successful try_to_unuse, if no swap is now in use, we know
1274 * we can empty the mmlist. swap_lock must be held on entry and exit.
1275 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 * added to the mmlist just after page_duplicate - before would be racy.
1277 */
1278static void drain_mmlist(void)
1279{
1280 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001281 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001283 for (type = 0; type < nr_swapfiles; type++)
1284 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 return;
1286 spin_lock(&mmlist_lock);
1287 list_for_each_safe(p, next, &init_mm.mmlist)
1288 list_del_init(p);
1289 spin_unlock(&mmlist_lock);
1290}
1291
1292/*
1293 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001294 * corresponds to page offset for the specified swap entry.
1295 * Note that the type of this function is sector_t, but it returns page offset
1296 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001298static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001300 struct swap_info_struct *sis;
1301 struct swap_extent *start_se;
1302 struct swap_extent *se;
1303 pgoff_t offset;
1304
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001305 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001306 *bdev = sis->bdev;
1307
1308 offset = swp_offset(entry);
1309 start_se = sis->curr_swap_extent;
1310 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 for ( ; ; ) {
1313 struct list_head *lh;
1314
1315 if (se->start_page <= offset &&
1316 offset < (se->start_page + se->nr_pages)) {
1317 return se->start_block + (offset - se->start_page);
1318 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001319 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 se = list_entry(lh, struct swap_extent, list);
1321 sis->curr_swap_extent = se;
1322 BUG_ON(se == start_se); /* It *must* be present */
1323 }
1324}
1325
1326/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001327 * Returns the page offset into bdev for the specified page's swap entry.
1328 */
1329sector_t map_swap_page(struct page *page, struct block_device **bdev)
1330{
1331 swp_entry_t entry;
1332 entry.val = page_private(page);
1333 return map_swap_entry(entry, bdev);
1334}
1335
1336/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * Free all of a swapdev's extent information
1338 */
1339static void destroy_swap_extents(struct swap_info_struct *sis)
1340{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001341 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 struct swap_extent *se;
1343
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001344 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 struct swap_extent, list);
1346 list_del(&se->list);
1347 kfree(se);
1348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349}
1350
1351/*
1352 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001353 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001355 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 */
1357static int
1358add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1359 unsigned long nr_pages, sector_t start_block)
1360{
1361 struct swap_extent *se;
1362 struct swap_extent *new_se;
1363 struct list_head *lh;
1364
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001365 if (start_page == 0) {
1366 se = &sis->first_swap_extent;
1367 sis->curr_swap_extent = se;
1368 se->start_page = 0;
1369 se->nr_pages = nr_pages;
1370 se->start_block = start_block;
1371 return 1;
1372 } else {
1373 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001375 BUG_ON(se->start_page + se->nr_pages != start_page);
1376 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 /* Merge it */
1378 se->nr_pages += nr_pages;
1379 return 0;
1380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
1382
1383 /*
1384 * No merge. Insert a new extent, preserving ordering.
1385 */
1386 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1387 if (new_se == NULL)
1388 return -ENOMEM;
1389 new_se->start_page = start_page;
1390 new_se->nr_pages = nr_pages;
1391 new_se->start_block = start_block;
1392
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001393 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001394 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395}
1396
1397/*
1398 * A `swap extent' is a simple thing which maps a contiguous range of pages
1399 * onto a contiguous range of disk blocks. An ordered list of swap extents
1400 * is built at swapon time and is then used at swap_writepage/swap_readpage
1401 * time for locating where on disk a page belongs.
1402 *
1403 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1404 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1405 * swap files identically.
1406 *
1407 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1408 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1409 * swapfiles are handled *identically* after swapon time.
1410 *
1411 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1412 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1413 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1414 * requirements, they are simply tossed out - we will never use those blocks
1415 * for swapping.
1416 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001417 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1419 * which will scribble on the fs.
1420 *
1421 * The amount of disk space which a single swap extent represents varies.
1422 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1423 * extents in the list. To avoid much list walking, we cache the previous
1424 * search location in `curr_swap_extent', and start new searches from there.
1425 * This is extremely effective. The average number of iterations in
1426 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1427 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001428static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429{
1430 struct inode *inode;
1431 unsigned blocks_per_page;
1432 unsigned long page_no;
1433 unsigned blkbits;
1434 sector_t probe_block;
1435 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001436 sector_t lowest_block = -1;
1437 sector_t highest_block = 0;
1438 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 int ret;
1440
1441 inode = sis->swap_file->f_mapping->host;
1442 if (S_ISBLK(inode->i_mode)) {
1443 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001444 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001445 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 }
1447
1448 blkbits = inode->i_blkbits;
1449 blocks_per_page = PAGE_SIZE >> blkbits;
1450
1451 /*
1452 * Map all the blocks into the extent list. This code doesn't try
1453 * to be very smart.
1454 */
1455 probe_block = 0;
1456 page_no = 0;
1457 last_block = i_size_read(inode) >> blkbits;
1458 while ((probe_block + blocks_per_page) <= last_block &&
1459 page_no < sis->max) {
1460 unsigned block_in_page;
1461 sector_t first_block;
1462
1463 first_block = bmap(inode, probe_block);
1464 if (first_block == 0)
1465 goto bad_bmap;
1466
1467 /*
1468 * It must be PAGE_SIZE aligned on-disk
1469 */
1470 if (first_block & (blocks_per_page - 1)) {
1471 probe_block++;
1472 goto reprobe;
1473 }
1474
1475 for (block_in_page = 1; block_in_page < blocks_per_page;
1476 block_in_page++) {
1477 sector_t block;
1478
1479 block = bmap(inode, probe_block + block_in_page);
1480 if (block == 0)
1481 goto bad_bmap;
1482 if (block != first_block + block_in_page) {
1483 /* Discontiguity */
1484 probe_block++;
1485 goto reprobe;
1486 }
1487 }
1488
Hugh Dickins53092a72005-09-03 15:54:34 -07001489 first_block >>= (PAGE_SHIFT - blkbits);
1490 if (page_no) { /* exclude the header page */
1491 if (first_block < lowest_block)
1492 lowest_block = first_block;
1493 if (first_block > highest_block)
1494 highest_block = first_block;
1495 }
1496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 /*
1498 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1499 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001500 ret = add_swap_extent(sis, page_no, 1, first_block);
1501 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001503 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 page_no++;
1505 probe_block += blocks_per_page;
1506reprobe:
1507 continue;
1508 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001509 ret = nr_extents;
1510 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001512 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001514 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001516out:
1517 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518bad_bmap:
1519 printk(KERN_ERR "swapon: swapfile has holes\n");
1520 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001521 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522}
1523
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001524SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001526 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001527 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 struct file *swap_file, *victim;
1529 struct address_space *mapping;
1530 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001531 char *pathname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 int i, type, prev;
1533 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001534
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 if (!capable(CAP_SYS_ADMIN))
1536 return -EPERM;
1537
1538 pathname = getname(specialfile);
1539 err = PTR_ERR(pathname);
1540 if (IS_ERR(pathname))
1541 goto out;
1542
1543 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1544 putname(pathname);
1545 err = PTR_ERR(victim);
1546 if (IS_ERR(victim))
1547 goto out;
1548
1549 mapping = victim->f_mapping;
1550 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001551 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001552 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1553 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001554 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (p->swap_file->f_mapping == mapping)
1556 break;
1557 }
1558 prev = type;
1559 }
1560 if (type < 0) {
1561 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001562 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 goto out_dput;
1564 }
1565 if (!security_vm_enough_memory(p->pages))
1566 vm_unacct_memory(p->pages);
1567 else {
1568 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001569 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 goto out_dput;
1571 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001572 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001574 else
1575 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 if (type == swap_list.next) {
1577 /* just pick something that's safe... */
1578 swap_list.next = swap_list.head;
1579 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001580 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001581 for (i = p->next; i >= 0; i = swap_info[i]->next)
1582 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001583 least_priority++;
1584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 nr_swap_pages -= p->pages;
1586 total_swap_pages -= p->pages;
1587 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001588 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001589
Hugh Dickins35451be2009-09-21 17:02:27 -07001590 current->flags |= PF_OOM_ORIGIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 err = try_to_unuse(type);
Hugh Dickins35451be2009-09-21 17:02:27 -07001592 current->flags &= ~PF_OOM_ORIGIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (err) {
1595 /* re-insert swap space back into swap_list */
Hugh Dickins5d337b92005-09-03 15:54:41 -07001596 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001597 if (p->prio < 0)
1598 p->prio = --least_priority;
1599 prev = -1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001600 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1601 if (p->prio >= swap_info[i]->prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 break;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001603 prev = i;
1604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 p->next = i;
1606 if (prev < 0)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001607 swap_list.head = swap_list.next = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 else
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001609 swap_info[prev]->next = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 nr_swap_pages += p->pages;
1611 total_swap_pages += p->pages;
1612 p->flags |= SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001613 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 goto out_dput;
1615 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001616
1617 /* wait for any unplug function to finish */
1618 down_write(&swap_unplug_sem);
1619 up_write(&swap_unplug_sem);
1620
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001621 destroy_swap_extents(p);
Hugh Dickins570a3352009-12-14 17:58:46 -08001622 if (p->flags & SWP_CONTINUED)
1623 free_swap_count_continuations(p);
1624
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001625 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001626 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001628
1629 /* wait for anyone still in scan_swap_map */
1630 p->highest_bit = 0; /* cuts scans short */
1631 while (p->flags >= SWP_SCANNING) {
1632 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001633 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001634 spin_lock(&swap_lock);
1635 }
1636
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 swap_file = p->swap_file;
1638 p->swap_file = NULL;
1639 p->max = 0;
1640 swap_map = p->swap_map;
1641 p->swap_map = NULL;
1642 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001643 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001644 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 vfree(swap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001646 /* Destroy swap account informatin */
1647 swap_cgroup_swapoff(type);
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 inode = mapping->host;
1650 if (S_ISBLK(inode->i_mode)) {
1651 struct block_device *bdev = I_BDEV(inode);
1652 set_blocksize(bdev, p->old_block_size);
1653 bd_release(bdev);
1654 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001655 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001657 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 }
1659 filp_close(swap_file, NULL);
1660 err = 0;
1661
1662out_dput:
1663 filp_close(victim, NULL);
1664out:
1665 return err;
1666}
1667
1668#ifdef CONFIG_PROC_FS
1669/* iterator */
1670static void *swap_start(struct seq_file *swap, loff_t *pos)
1671{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001672 struct swap_info_struct *si;
1673 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 loff_t l = *pos;
1675
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001676 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001678 if (!l)
1679 return SEQ_START_TOKEN;
1680
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001681 for (type = 0; type < nr_swapfiles; type++) {
1682 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1683 si = swap_info[type];
1684 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001686 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001687 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 }
1689
1690 return NULL;
1691}
1692
1693static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1694{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001695 struct swap_info_struct *si = v;
1696 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001698 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001699 type = 0;
1700 else
1701 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001702
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001703 for (; type < nr_swapfiles; type++) {
1704 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1705 si = swap_info[type];
1706 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 continue;
1708 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001709 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
1711
1712 return NULL;
1713}
1714
1715static void swap_stop(struct seq_file *swap, void *v)
1716{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001717 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718}
1719
1720static int swap_show(struct seq_file *swap, void *v)
1721{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001722 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 struct file *file;
1724 int len;
1725
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001726 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001727 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1728 return 0;
1729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001731 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001732 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001733 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001734 len < 40 ? 40 - len : 1, " ",
1735 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001737 si->pages << (PAGE_SHIFT - 10),
1738 si->inuse_pages << (PAGE_SHIFT - 10),
1739 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 return 0;
1741}
1742
Helge Deller15ad7cd2006-12-06 20:40:36 -08001743static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 .start = swap_start,
1745 .next = swap_next,
1746 .stop = swap_stop,
1747 .show = swap_show
1748};
1749
1750static int swaps_open(struct inode *inode, struct file *file)
1751{
1752 return seq_open(file, &swaps_op);
1753}
1754
Helge Deller15ad7cd2006-12-06 20:40:36 -08001755static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 .open = swaps_open,
1757 .read = seq_read,
1758 .llseek = seq_lseek,
1759 .release = seq_release,
1760};
1761
1762static int __init procswaps_init(void)
1763{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001764 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 return 0;
1766}
1767__initcall(procswaps_init);
1768#endif /* CONFIG_PROC_FS */
1769
Jan Beulich17963162008-12-16 11:35:24 +00001770#ifdef MAX_SWAPFILES_CHECK
1771static int __init max_swapfiles_check(void)
1772{
1773 MAX_SWAPFILES_CHECK();
1774 return 0;
1775}
1776late_initcall(max_swapfiles_check);
1777#endif
1778
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779/*
1780 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1781 *
1782 * The swapon system call
1783 */
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001784SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001786 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 char *name = NULL;
1788 struct block_device *bdev = NULL;
1789 struct file *swap_file = NULL;
1790 struct address_space *mapping;
1791 unsigned int type;
1792 int i, prev;
1793 int error;
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001794 union swap_header *swap_header;
1795 unsigned int nr_good_pages;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001796 int nr_extents = 0;
Hugh Dickins53092a72005-09-03 15:54:34 -07001797 sector_t span;
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001798 unsigned long maxpages;
Hugh Dickins73fd8742009-01-06 14:39:47 -08001799 unsigned long swapfilepages;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001800 unsigned char *swap_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 struct page *page = NULL;
1802 struct inode *inode = NULL;
1803 int did_down = 0;
1804
1805 if (!capable(CAP_SYS_ADMIN))
1806 return -EPERM;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001807
1808 p = kzalloc(sizeof(*p), GFP_KERNEL);
1809 if (!p)
1810 return -ENOMEM;
1811
Hugh Dickins5d337b92005-09-03 15:54:41 -07001812 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001813 for (type = 0; type < nr_swapfiles; type++) {
1814 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 error = -EPERM;
Christoph Lameter06972122006-06-23 02:03:35 -07001818 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001819 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001820 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 goto out;
1822 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001823 if (type >= nr_swapfiles) {
1824 p->type = type;
1825 swap_info[type] = p;
1826 /*
1827 * Write swap_info[type] before nr_swapfiles, in case a
1828 * racing procfs swap_start() or swap_next() is reading them.
1829 * (We never shrink nr_swapfiles, we never free this entry.)
1830 */
1831 smp_wmb();
1832 nr_swapfiles++;
1833 } else {
1834 kfree(p);
1835 p = swap_info[type];
1836 /*
1837 * Do not memset this entry: a racing procfs swap_next()
1838 * would be relying on p->type to remain valid.
1839 */
1840 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001841 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001844 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 name = getname(specialfile);
1847 error = PTR_ERR(name);
1848 if (IS_ERR(name)) {
1849 name = NULL;
1850 goto bad_swap_2;
1851 }
1852 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1853 error = PTR_ERR(swap_file);
1854 if (IS_ERR(swap_file)) {
1855 swap_file = NULL;
1856 goto bad_swap_2;
1857 }
1858
1859 p->swap_file = swap_file;
1860 mapping = swap_file->f_mapping;
1861 inode = mapping->host;
1862
1863 error = -EBUSY;
1864 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001865 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
1867 if (i == type || !q->swap_file)
1868 continue;
1869 if (mapping == q->swap_file->f_mapping)
1870 goto bad_swap;
1871 }
1872
1873 error = -EINVAL;
1874 if (S_ISBLK(inode->i_mode)) {
1875 bdev = I_BDEV(inode);
1876 error = bd_claim(bdev, sys_swapon);
1877 if (error < 0) {
1878 bdev = NULL;
Rob Landleyf7b3a432005-09-22 21:44:27 -07001879 error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 goto bad_swap;
1881 }
1882 p->old_block_size = block_size(bdev);
1883 error = set_blocksize(bdev, PAGE_SIZE);
1884 if (error < 0)
1885 goto bad_swap;
1886 p->bdev = bdev;
1887 } else if (S_ISREG(inode->i_mode)) {
1888 p->bdev = inode->i_sb->s_bdev;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001889 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 did_down = 1;
1891 if (IS_SWAPFILE(inode)) {
1892 error = -EBUSY;
1893 goto bad_swap;
1894 }
1895 } else {
1896 goto bad_swap;
1897 }
1898
Hugh Dickins73fd8742009-01-06 14:39:47 -08001899 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
1901 /*
1902 * Read the swap header.
1903 */
1904 if (!mapping->a_ops->readpage) {
1905 error = -EINVAL;
1906 goto bad_swap;
1907 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07001908 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (IS_ERR(page)) {
1910 error = PTR_ERR(page);
1911 goto bad_swap;
1912 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001913 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Hugh Dickins81e33972009-01-06 14:39:49 -08001915 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Jesper Juhle97a3112006-01-11 01:50:28 +01001916 printk(KERN_ERR "Unable to find swap-space signature\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 error = -EINVAL;
1918 goto bad_swap;
1919 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001920
Hugh Dickins81e33972009-01-06 14:39:49 -08001921 /* swap partition endianess hack... */
1922 if (swab32(swap_header->info.version) == 1) {
1923 swab32s(&swap_header->info.version);
1924 swab32s(&swap_header->info.last_page);
1925 swab32s(&swap_header->info.nr_badpages);
1926 for (i = 0; i < swap_header->info.nr_badpages; i++)
1927 swab32s(&swap_header->info.badpages[i]);
1928 }
1929 /* Check the swap header's sub-version */
1930 if (swap_header->info.version != 1) {
1931 printk(KERN_WARNING
1932 "Unable to handle swap header version %d\n",
1933 swap_header->info.version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 error = -EINVAL;
1935 goto bad_swap;
Hugh Dickins81e33972009-01-06 14:39:49 -08001936 }
1937
1938 p->lowest_bit = 1;
1939 p->cluster_next = 1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001940 p->cluster_nr = 0;
Hugh Dickins81e33972009-01-06 14:39:49 -08001941
1942 /*
1943 * Find out how many pages are allowed for a single swap
1944 * device. There are two limiting factors: 1) the number of
1945 * bits for the swap offset in the swp_entry_t type and
1946 * 2) the number of bits in the a swap pte as defined by
1947 * the different architectures. In order to find the
1948 * largest possible bit mask a swap entry with swap type 0
1949 * and swap offset ~0UL is created, encoded to a swap pte,
1950 * decoded to a swp_entry_t again and finally the swap
1951 * offset is extracted. This will mask all the bits from
1952 * the initial ~0UL mask that can't be encoded in either
1953 * the swp_entry_t or the architecture definition of a
1954 * swap pte.
1955 */
1956 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001957 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
1958 if (maxpages > swap_header->info.last_page) {
1959 maxpages = swap_header->info.last_page + 1;
1960 /* p->max is an unsigned int: don't overflow it */
1961 if ((unsigned int)maxpages == 0)
1962 maxpages = UINT_MAX;
1963 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001964 p->highest_bit = maxpages - 1;
1965
1966 error = -EINVAL;
1967 if (!maxpages)
1968 goto bad_swap;
1969 if (swapfilepages && maxpages > swapfilepages) {
1970 printk(KERN_WARNING
1971 "Swap area shorter than signature indicates\n");
1972 goto bad_swap;
1973 }
1974 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1975 goto bad_swap;
1976 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1977 goto bad_swap;
1978
1979 /* OK, set up the swap map and apply the bad block list */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001980 swap_map = vmalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08001981 if (!swap_map) {
1982 error = -ENOMEM;
1983 goto bad_swap;
1984 }
1985
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001986 memset(swap_map, 0, maxpages);
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001987 nr_good_pages = maxpages - 1; /* omit header page */
1988
Hugh Dickins81e33972009-01-06 14:39:49 -08001989 for (i = 0; i < swap_header->info.nr_badpages; i++) {
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001990 unsigned int page_nr = swap_header->info.badpages[i];
1991 if (page_nr == 0 || page_nr > swap_header->info.last_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 error = -EINVAL;
1993 goto bad_swap;
1994 }
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001995 if (page_nr < maxpages) {
1996 swap_map[page_nr] = SWAP_MAP_BAD;
1997 nr_good_pages--;
1998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 }
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002000
2001 error = swap_cgroup_swapon(type, maxpages);
2002 if (error)
2003 goto bad_swap;
2004
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002005 if (nr_good_pages) {
Hugh Dickins78ecba02008-07-23 21:28:23 -07002006 swap_map[0] = SWAP_MAP_BAD;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002007 p->max = maxpages;
2008 p->pages = nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07002009 nr_extents = setup_swap_extents(p, &span);
2010 if (nr_extents < 0) {
2011 error = nr_extents;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002012 goto bad_swap;
Hugh Dickins53092a72005-09-03 15:54:34 -07002013 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002014 nr_good_pages = p->pages;
2015 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 if (!nr_good_pages) {
2017 printk(KERN_WARNING "Empty swap-file\n");
2018 error = -EINVAL;
2019 goto bad_swap;
2020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002022 if (p->bdev) {
2023 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2024 p->flags |= SWP_SOLIDSTATE;
2025 p->cluster_next = 1 + (random32() % p->highest_bit);
2026 }
2027 if (discard_swap(p) == 0)
2028 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002029 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002030
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002031 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002032 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07002033 if (swap_flags & SWAP_FLAG_PREFER)
2034 p->prio =
2035 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2036 else
2037 p->prio = --least_priority;
2038 p->swap_map = swap_map;
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08002039 p->flags |= SWP_WRITEOK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 nr_swap_pages += nr_good_pages;
2041 total_swap_pages += nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07002042
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002043 printk(KERN_INFO "Adding %uk swap on %s. "
Hugh Dickins20137a42009-01-06 14:39:54 -08002044 "Priority:%d extents:%d across:%lluk %s%s\n",
Hugh Dickins53092a72005-09-03 15:54:34 -07002045 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002046 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
Hugh Dickins20137a42009-01-06 14:39:54 -08002047 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2048 (p->flags & SWP_DISCARDABLE) ? "D" : "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
2050 /* insert swap space into swap_list: */
2051 prev = -1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002052 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
2053 if (p->prio >= swap_info[i]->prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 prev = i;
2056 }
2057 p->next = i;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002058 if (prev < 0)
2059 swap_list.head = swap_list.next = type;
2060 else
2061 swap_info[prev]->next = type;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002062 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002063 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 error = 0;
2065 goto out;
2066bad_swap:
2067 if (bdev) {
2068 set_blocksize(bdev, p->old_block_size);
2069 bd_release(bdev);
2070 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002071 destroy_swap_extents(p);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002072 swap_cgroup_swapoff(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073bad_swap_2:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002074 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002077 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 vfree(swap_map);
2079 if (swap_file)
2080 filp_close(swap_file, NULL);
2081out:
2082 if (page && !IS_ERR(page)) {
2083 kunmap(page);
2084 page_cache_release(page);
2085 }
2086 if (name)
2087 putname(name);
2088 if (did_down) {
2089 if (!error)
2090 inode->i_flags |= S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002091 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 }
2093 return error;
2094}
2095
2096void si_swapinfo(struct sysinfo *val)
2097{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002098 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 unsigned long nr_to_be_unused = 0;
2100
Hugh Dickins5d337b92005-09-03 15:54:41 -07002101 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002102 for (type = 0; type < nr_swapfiles; type++) {
2103 struct swap_info_struct *si = swap_info[type];
2104
2105 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2106 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 }
2108 val->freeswap = nr_swap_pages + nr_to_be_unused;
2109 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002110 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111}
2112
2113/*
2114 * Verify that a swap entry is valid and increment its swap map count.
2115 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002116 * Returns error code in following case.
2117 * - success -> 0
2118 * - swp_entry is invalid -> EINVAL
2119 * - swp_entry is migration entry -> EINVAL
2120 * - swap-cache reference is requested but there is already one. -> EEXIST
2121 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a3352009-12-14 17:58:46 -08002122 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002124static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002126 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002128 unsigned char count;
2129 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002130 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Andi Kleena7420aa2009-09-16 11:50:05 +02002132 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002133 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 type = swp_type(entry);
2136 if (type >= nr_swapfiles)
2137 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002138 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 offset = swp_offset(entry);
2140
Hugh Dickins5d337b92005-09-03 15:54:41 -07002141 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002142 if (unlikely(offset >= p->max))
2143 goto unlock_out;
2144
Hugh Dickins253d5532009-12-14 17:58:44 -08002145 count = p->swap_map[offset];
2146 has_cache = count & SWAP_HAS_CACHE;
2147 count &= ~SWAP_HAS_CACHE;
2148 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002149
Hugh Dickins253d5532009-12-14 17:58:44 -08002150 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002151
2152 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002153 if (!has_cache && count)
2154 has_cache = SWAP_HAS_CACHE;
2155 else if (has_cache) /* someone else added cache */
2156 err = -EEXIST;
2157 else /* no users remaining */
2158 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002159
2160 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002161
Hugh Dickins570a3352009-12-14 17:58:46 -08002162 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2163 count += usage;
2164 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002165 err = -EINVAL;
Hugh Dickins570a3352009-12-14 17:58:46 -08002166 else if (swap_count_continued(p, offset, count))
2167 count = COUNT_CONTINUED;
2168 else
2169 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002170 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002171 err = -ENOENT; /* unused swap entry */
2172
2173 p->swap_map[offset] = count | has_cache;
2174
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002175unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002176 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002178 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
2180bad_file:
2181 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2182 goto out;
2183}
Hugh Dickins253d5532009-12-14 17:58:44 -08002184
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002185/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002186 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2187 * (in which case its reference count is never incremented).
2188 */
2189void swap_shmem_alloc(swp_entry_t entry)
2190{
2191 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2192}
2193
2194/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002195 * Increase reference count of swap entry by 1.
2196 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2197 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2198 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2199 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002200 */
Hugh Dickins570a3352009-12-14 17:58:46 -08002201int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002202{
Hugh Dickins570a3352009-12-14 17:58:46 -08002203 int err = 0;
2204
2205 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2206 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2207 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002208}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002210/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002211 * @entry: swap entry for which we allocate swap cache.
2212 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002213 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002214 * This can return error codes. Returns 0 at success.
2215 * -EBUSY means there is a swap cache.
2216 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002217 */
2218int swapcache_prepare(swp_entry_t entry)
2219{
Hugh Dickins253d5532009-12-14 17:58:44 -08002220 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002221}
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07002224 * swap_lock prevents swap_map being freed. Don't grab an extra
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 * reference on the swaphandle, it doesn't matter if it becomes unused.
2226 */
2227int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2228{
Hugh Dickins89528982008-02-04 22:28:45 -08002229 struct swap_info_struct *si;
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002230 int our_page_cluster = page_cluster;
Hugh Dickins89528982008-02-04 22:28:45 -08002231 pgoff_t target, toff;
2232 pgoff_t base, end;
2233 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002235 if (!our_page_cluster) /* no readahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 return 0;
Hugh Dickins89528982008-02-04 22:28:45 -08002237
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002238 si = swap_info[swp_type(entry)];
Hugh Dickins89528982008-02-04 22:28:45 -08002239 target = swp_offset(entry);
2240 base = (target >> our_page_cluster) << our_page_cluster;
2241 end = base + (1 << our_page_cluster);
2242 if (!base) /* first page is swap header */
2243 base++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
Hugh Dickins5d337b92005-09-03 15:54:41 -07002245 spin_lock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002246 if (end > si->max) /* don't go beyond end of map */
2247 end = si->max;
2248
2249 /* Count contiguous allocated slots above our target */
2250 for (toff = target; ++toff < end; nr_pages++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 /* Don't read in free or bad pages */
Hugh Dickins89528982008-02-04 22:28:45 -08002252 if (!si->swap_map[toff])
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002254 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 break;
Hugh Dickins89528982008-02-04 22:28:45 -08002256 }
2257 /* Count contiguous allocated slots below our target */
2258 for (toff = target; --toff >= base; nr_pages++) {
2259 /* Don't read in free or bad pages */
2260 if (!si->swap_map[toff])
2261 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002262 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Hugh Dickins89528982008-02-04 22:28:45 -08002263 break;
2264 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07002265 spin_unlock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002266
2267 /*
2268 * Indicate starting offset, and return number of pages to get:
2269 * if only 1, say 0, since there's then no readahead to be done.
2270 */
2271 *offset = ++toff;
2272 return nr_pages? ++nr_pages: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273}
Hugh Dickins570a3352009-12-14 17:58:46 -08002274
2275/*
2276 * add_swap_count_continuation - called when a swap count is duplicated
2277 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2278 * page of the original vmalloc'ed swap_map, to hold the continuation count
2279 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2280 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2281 *
2282 * These continuation pages are seldom referenced: the common paths all work
2283 * on the original swap_map, only referring to a continuation page when the
2284 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2285 *
2286 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2287 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2288 * can be called after dropping locks.
2289 */
2290int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2291{
2292 struct swap_info_struct *si;
2293 struct page *head;
2294 struct page *page;
2295 struct page *list_page;
2296 pgoff_t offset;
2297 unsigned char count;
2298
2299 /*
2300 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2301 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2302 */
2303 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2304
2305 si = swap_info_get(entry);
2306 if (!si) {
2307 /*
2308 * An acceptable race has occurred since the failing
2309 * __swap_duplicate(): the swap entry has been freed,
2310 * perhaps even the whole swap_map cleared for swapoff.
2311 */
2312 goto outer;
2313 }
2314
2315 offset = swp_offset(entry);
2316 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2317
2318 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2319 /*
2320 * The higher the swap count, the more likely it is that tasks
2321 * will race to add swap count continuation: we need to avoid
2322 * over-provisioning.
2323 */
2324 goto out;
2325 }
2326
2327 if (!page) {
2328 spin_unlock(&swap_lock);
2329 return -ENOMEM;
2330 }
2331
2332 /*
2333 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2334 * no architecture is using highmem pages for kernel pagetables: so it
2335 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2336 */
2337 head = vmalloc_to_page(si->swap_map + offset);
2338 offset &= ~PAGE_MASK;
2339
2340 /*
2341 * Page allocation does not initialize the page's lru field,
2342 * but it does always reset its private field.
2343 */
2344 if (!page_private(head)) {
2345 BUG_ON(count & COUNT_CONTINUED);
2346 INIT_LIST_HEAD(&head->lru);
2347 set_page_private(head, SWP_CONTINUED);
2348 si->flags |= SWP_CONTINUED;
2349 }
2350
2351 list_for_each_entry(list_page, &head->lru, lru) {
2352 unsigned char *map;
2353
2354 /*
2355 * If the previous map said no continuation, but we've found
2356 * a continuation page, free our allocation and use this one.
2357 */
2358 if (!(count & COUNT_CONTINUED))
2359 goto out;
2360
2361 map = kmap_atomic(list_page, KM_USER0) + offset;
2362 count = *map;
2363 kunmap_atomic(map, KM_USER0);
2364
2365 /*
2366 * If this continuation count now has some space in it,
2367 * free our allocation and use this one.
2368 */
2369 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2370 goto out;
2371 }
2372
2373 list_add_tail(&page->lru, &head->lru);
2374 page = NULL; /* now it's attached, don't free it */
2375out:
2376 spin_unlock(&swap_lock);
2377outer:
2378 if (page)
2379 __free_page(page);
2380 return 0;
2381}
2382
2383/*
2384 * swap_count_continued - when the original swap_map count is incremented
2385 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2386 * into, carry if so, or else fail until a new continuation page is allocated;
2387 * when the original swap_map count is decremented from 0 with continuation,
2388 * borrow from the continuation and report whether it still holds more.
2389 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2390 */
2391static bool swap_count_continued(struct swap_info_struct *si,
2392 pgoff_t offset, unsigned char count)
2393{
2394 struct page *head;
2395 struct page *page;
2396 unsigned char *map;
2397
2398 head = vmalloc_to_page(si->swap_map + offset);
2399 if (page_private(head) != SWP_CONTINUED) {
2400 BUG_ON(count & COUNT_CONTINUED);
2401 return false; /* need to add count continuation */
2402 }
2403
2404 offset &= ~PAGE_MASK;
2405 page = list_entry(head->lru.next, struct page, lru);
2406 map = kmap_atomic(page, KM_USER0) + offset;
2407
2408 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2409 goto init_map; /* jump over SWAP_CONT_MAX checks */
2410
2411 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2412 /*
2413 * Think of how you add 1 to 999
2414 */
2415 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2416 kunmap_atomic(map, KM_USER0);
2417 page = list_entry(page->lru.next, struct page, lru);
2418 BUG_ON(page == head);
2419 map = kmap_atomic(page, KM_USER0) + offset;
2420 }
2421 if (*map == SWAP_CONT_MAX) {
2422 kunmap_atomic(map, KM_USER0);
2423 page = list_entry(page->lru.next, struct page, lru);
2424 if (page == head)
2425 return false; /* add count continuation */
2426 map = kmap_atomic(page, KM_USER0) + offset;
2427init_map: *map = 0; /* we didn't zero the page */
2428 }
2429 *map += 1;
2430 kunmap_atomic(map, KM_USER0);
2431 page = list_entry(page->lru.prev, struct page, lru);
2432 while (page != head) {
2433 map = kmap_atomic(page, KM_USER0) + offset;
2434 *map = COUNT_CONTINUED;
2435 kunmap_atomic(map, KM_USER0);
2436 page = list_entry(page->lru.prev, struct page, lru);
2437 }
2438 return true; /* incremented */
2439
2440 } else { /* decrementing */
2441 /*
2442 * Think of how you subtract 1 from 1000
2443 */
2444 BUG_ON(count != COUNT_CONTINUED);
2445 while (*map == COUNT_CONTINUED) {
2446 kunmap_atomic(map, KM_USER0);
2447 page = list_entry(page->lru.next, struct page, lru);
2448 BUG_ON(page == head);
2449 map = kmap_atomic(page, KM_USER0) + offset;
2450 }
2451 BUG_ON(*map == 0);
2452 *map -= 1;
2453 if (*map == 0)
2454 count = 0;
2455 kunmap_atomic(map, KM_USER0);
2456 page = list_entry(page->lru.prev, struct page, lru);
2457 while (page != head) {
2458 map = kmap_atomic(page, KM_USER0) + offset;
2459 *map = SWAP_CONT_MAX | count;
2460 count = COUNT_CONTINUED;
2461 kunmap_atomic(map, KM_USER0);
2462 page = list_entry(page->lru.prev, struct page, lru);
2463 }
2464 return count == COUNT_CONTINUED;
2465 }
2466}
2467
2468/*
2469 * free_swap_count_continuations - swapoff free all the continuation pages
2470 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2471 */
2472static void free_swap_count_continuations(struct swap_info_struct *si)
2473{
2474 pgoff_t offset;
2475
2476 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2477 struct page *head;
2478 head = vmalloc_to_page(si->swap_map + offset);
2479 if (page_private(head)) {
2480 struct list_head *this, *next;
2481 list_for_each_safe(this, next, &head->lru) {
2482 struct page *page;
2483 page = list_entry(this, struct page, lru);
2484 list_del(this);
2485 __free_page(page);
2486 }
2487 }
2488 }
2489}