blob: 03aa2d55f1a2f54fc5eb17b099806edbc83a4049 [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 Dickins570a335b2009-12-14 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 Dickins570a335b2009-12-14 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 Dickins570a335b2009-12-14 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,
Dmitry Monakhovfbd9b092010-04-28 17:55:06 +0400142 nr_blocks, GFP_KERNEL,
143 BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800144 if (err)
145 return err;
146 cond_resched();
147 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800148
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800149 list_for_each_entry(se, &si->first_swap_extent.list, list) {
150 start_block = se->start_block << (PAGE_SHIFT - 9);
151 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800152
153 err = blkdev_issue_discard(si->bdev, start_block,
Dmitry Monakhovfbd9b092010-04-28 17:55:06 +0400154 nr_blocks, GFP_KERNEL,
155 BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800156 if (err)
157 break;
158
159 cond_resched();
160 }
161 return err; /* That will often be -EOPNOTSUPP */
162}
163
Hugh Dickins7992fde2009-01-06 14:39:53 -0800164/*
165 * swap allocation tell device that a cluster of swap can now be discarded,
166 * to allow the swap device to optimize its wear-levelling.
167 */
168static void discard_swap_cluster(struct swap_info_struct *si,
169 pgoff_t start_page, pgoff_t nr_pages)
170{
171 struct swap_extent *se = si->curr_swap_extent;
172 int found_extent = 0;
173
174 while (nr_pages) {
175 struct list_head *lh;
176
177 if (se->start_page <= start_page &&
178 start_page < se->start_page + se->nr_pages) {
179 pgoff_t offset = start_page - se->start_page;
180 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800181 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800182
183 if (nr_blocks > nr_pages)
184 nr_blocks = nr_pages;
185 start_page += nr_blocks;
186 nr_pages -= nr_blocks;
187
188 if (!found_extent++)
189 si->curr_swap_extent = se;
190
191 start_block <<= PAGE_SHIFT - 9;
192 nr_blocks <<= PAGE_SHIFT - 9;
193 if (blkdev_issue_discard(si->bdev, start_block,
Dmitry Monakhovfbd9b092010-04-28 17:55:06 +0400194 nr_blocks, GFP_NOIO, BLKDEV_IFL_WAIT |
195 BLKDEV_IFL_BARRIER))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800196 break;
197 }
198
199 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800200 se = list_entry(lh, struct swap_extent, list);
201 }
202}
203
204static int wait_for_discard(void *word)
205{
206 schedule();
207 return 0;
208}
209
Hugh Dickins048c27f2005-09-03 15:54:40 -0700210#define SWAPFILE_CLUSTER 256
211#define LATENCY_LIMIT 256
212
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700213static inline unsigned long scan_swap_map(struct swap_info_struct *si,
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800214 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800216 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800217 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800218 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700219 int latency_ration = LATENCY_LIMIT;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800220 int found_free_cluster = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800222 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700223 * We try to cluster swap pages by allocating them sequentially
224 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
225 * way, however, we resort to first-free allocation, starting
226 * a new cluster. This prevents us from scattering swap pages
227 * all over the entire swap partition, so that we reduce
228 * overall disk seek times between swap pages. -- sct
229 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800230 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700231 */
232
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700233 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800234 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800235
236 if (unlikely(!si->cluster_nr--)) {
237 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
238 si->cluster_nr = SWAPFILE_CLUSTER - 1;
239 goto checks;
240 }
Hugh Dickins7992fde2009-01-06 14:39:53 -0800241 if (si->flags & SWP_DISCARDABLE) {
242 /*
243 * Start range check on racing allocations, in case
244 * they overlap the cluster we eventually decide on
245 * (we scan without swap_lock to allow preemption).
246 * It's hardly conceivable that cluster_nr could be
247 * wrapped during our scan, but don't depend on it.
248 */
249 if (si->lowest_alloc)
250 goto checks;
251 si->lowest_alloc = si->max;
252 si->highest_alloc = 0;
253 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700254 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700255
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800256 /*
257 * If seek is expensive, start searching for new cluster from
258 * start of partition, to minimize the span of allocated swap.
259 * But if seek is cheap, search from our current position, so
260 * that swap is allocated from all over the partition: if the
261 * Flash Translation Layer only remaps within limited zones,
262 * we don't want to wear out the first zone too quickly.
263 */
264 if (!(si->flags & SWP_SOLIDSTATE))
265 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700266 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
267
268 /* Locate the first empty (unaligned) cluster */
269 for (; last_in_cluster <= si->highest_bit; offset++) {
270 if (si->swap_map[offset])
271 last_in_cluster = offset + SWAPFILE_CLUSTER;
272 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700273 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800274 offset -= SWAPFILE_CLUSTER - 1;
275 si->cluster_next = offset;
276 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800277 found_free_cluster = 1;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800278 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700280 if (unlikely(--latency_ration < 0)) {
281 cond_resched();
282 latency_ration = LATENCY_LIMIT;
283 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700284 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800285
286 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800287 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
288
289 /* Locate the first empty (unaligned) cluster */
290 for (; last_in_cluster < scan_base; offset++) {
291 if (si->swap_map[offset])
292 last_in_cluster = offset + SWAPFILE_CLUSTER;
293 else if (offset == last_in_cluster) {
294 spin_lock(&swap_lock);
295 offset -= SWAPFILE_CLUSTER - 1;
296 si->cluster_next = offset;
297 si->cluster_nr = SWAPFILE_CLUSTER - 1;
298 found_free_cluster = 1;
299 goto checks;
300 }
301 if (unlikely(--latency_ration < 0)) {
302 cond_resched();
303 latency_ration = LATENCY_LIMIT;
304 }
305 }
306
307 offset = scan_base;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700308 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800309 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800310 si->lowest_alloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700312
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800313checks:
314 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700315 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700316 if (!si->highest_bit)
317 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800318 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800319 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700320
321 /* reuse swap entry of cache-only swap if not busy. */
322 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
323 int swap_was_freed;
324 spin_unlock(&swap_lock);
325 swap_was_freed = __try_to_reclaim_swap(si, offset);
326 spin_lock(&swap_lock);
327 /* entry was freed successfully, try to use this again */
328 if (swap_was_freed)
329 goto checks;
330 goto scan; /* check next one */
331 }
332
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800333 if (si->swap_map[offset])
334 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700335
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800336 if (offset == si->lowest_bit)
337 si->lowest_bit++;
338 if (offset == si->highest_bit)
339 si->highest_bit--;
340 si->inuse_pages++;
341 if (si->inuse_pages == si->pages) {
342 si->lowest_bit = si->max;
343 si->highest_bit = 0;
344 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800345 si->swap_map[offset] = usage;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800346 si->cluster_next = offset + 1;
347 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800348
349 if (si->lowest_alloc) {
350 /*
351 * Only set when SWP_DISCARDABLE, and there's a scan
352 * for a free cluster in progress or just completed.
353 */
354 if (found_free_cluster) {
355 /*
356 * To optimize wear-levelling, discard the
357 * old data of the cluster, taking care not to
358 * discard any of its pages that have already
359 * been allocated by racing tasks (offset has
360 * already stepped over any at the beginning).
361 */
362 if (offset < si->highest_alloc &&
363 si->lowest_alloc <= last_in_cluster)
364 last_in_cluster = si->lowest_alloc - 1;
365 si->flags |= SWP_DISCARDING;
366 spin_unlock(&swap_lock);
367
368 if (offset < last_in_cluster)
369 discard_swap_cluster(si, offset,
370 last_in_cluster - offset + 1);
371
372 spin_lock(&swap_lock);
373 si->lowest_alloc = 0;
374 si->flags &= ~SWP_DISCARDING;
375
376 smp_mb(); /* wake_up_bit advises this */
377 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
378
379 } else if (si->flags & SWP_DISCARDING) {
380 /*
381 * Delay using pages allocated by racing tasks
382 * until the whole discard has been issued. We
383 * could defer that delay until swap_writepage,
384 * but it's easier to keep this self-contained.
385 */
386 spin_unlock(&swap_lock);
387 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
388 wait_for_discard, TASK_UNINTERRUPTIBLE);
389 spin_lock(&swap_lock);
390 } else {
391 /*
392 * Note pages allocated by racing tasks while
393 * scan for a free cluster is in progress, so
394 * that its final discard can exclude them.
395 */
396 if (offset < si->lowest_alloc)
397 si->lowest_alloc = offset;
398 if (offset > si->highest_alloc)
399 si->highest_alloc = offset;
400 }
401 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800402 return offset;
403
404scan:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700405 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700406 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700407 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700408 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700409 goto checks;
410 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700411 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
412 spin_lock(&swap_lock);
413 goto checks;
414 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700415 if (unlikely(--latency_ration < 0)) {
416 cond_resched();
417 latency_ration = LATENCY_LIMIT;
418 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700419 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800420 offset = si->lowest_bit;
421 while (++offset < scan_base) {
422 if (!si->swap_map[offset]) {
423 spin_lock(&swap_lock);
424 goto checks;
425 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700426 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
427 spin_lock(&swap_lock);
428 goto checks;
429 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800430 if (unlikely(--latency_ration < 0)) {
431 cond_resched();
432 latency_ration = LATENCY_LIMIT;
433 }
434 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700435 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700436
437no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700438 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 return 0;
440}
441
442swp_entry_t get_swap_page(void)
443{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700444 struct swap_info_struct *si;
445 pgoff_t offset;
446 int type, next;
447 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Hugh Dickins5d337b92005-09-03 15:54:41 -0700449 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700451 goto noswap;
452 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700454 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800455 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700456 next = si->next;
457 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800458 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700459 next = swap_list.head;
460 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700462
463 if (!si->highest_bit)
464 continue;
465 if (!(si->flags & SWP_WRITEOK))
466 continue;
467
468 swap_list.next = next;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700469 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800470 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700471 if (offset) {
472 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700473 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700474 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700475 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700477
478 nr_swap_pages++;
479noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700480 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700481 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700484/* The only caller of this function is now susupend routine */
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800485swp_entry_t get_swap_page_of_type(int type)
486{
487 struct swap_info_struct *si;
488 pgoff_t offset;
489
490 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800491 si = swap_info[type];
492 if (si && (si->flags & SWP_WRITEOK)) {
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800493 nr_swap_pages--;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700494 /* This is called for allocating swap entry, not cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800495 offset = scan_swap_map(si, 1);
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800496 if (offset) {
497 spin_unlock(&swap_lock);
498 return swp_entry(type, offset);
499 }
500 nr_swap_pages++;
501 }
502 spin_unlock(&swap_lock);
503 return (swp_entry_t) {0};
504}
505
Hugh Dickins73c34b62009-12-14 17:58:43 -0800506static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800508 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 unsigned long offset, type;
510
511 if (!entry.val)
512 goto out;
513 type = swp_type(entry);
514 if (type >= nr_swapfiles)
515 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800516 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (!(p->flags & SWP_USED))
518 goto bad_device;
519 offset = swp_offset(entry);
520 if (offset >= p->max)
521 goto bad_offset;
522 if (!p->swap_map[offset])
523 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700524 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return p;
526
527bad_free:
528 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
529 goto out;
530bad_offset:
531 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
532 goto out;
533bad_device:
534 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
535 goto out;
536bad_nofile:
537 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
538out:
539 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800540}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800542static unsigned char swap_entry_free(struct swap_info_struct *p,
543 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544{
Hugh Dickins253d5532009-12-14 17:58:44 -0800545 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800546 unsigned char count;
547 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700549 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800550 has_cache = count & SWAP_HAS_CACHE;
551 count &= ~SWAP_HAS_CACHE;
552
553 if (usage == SWAP_HAS_CACHE) {
554 VM_BUG_ON(!has_cache);
555 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800556 } else if (count == SWAP_MAP_SHMEM) {
557 /*
558 * Or we could insist on shmem.c using a special
559 * swap_shmem_free() and free_shmem_swap_and_cache()...
560 */
561 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800562 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
563 if (count == COUNT_CONTINUED) {
564 if (swap_count_continued(p, offset, count))
565 count = SWAP_MAP_MAX | COUNT_CONTINUED;
566 else
567 count = SWAP_MAP_MAX;
568 } else
569 count--;
570 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800571
572 if (!count)
573 mem_cgroup_uncharge_swap(entry);
574
575 usage = count | has_cache;
576 p->swap_map[offset] = usage;
577
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700578 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800579 if (!usage) {
Nitin Guptab3a27d02010-05-17 11:02:43 +0530580 struct gendisk *disk = p->bdev->bd_disk;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700581 if (offset < p->lowest_bit)
582 p->lowest_bit = offset;
583 if (offset > p->highest_bit)
584 p->highest_bit = offset;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800585 if (swap_list.next >= 0 &&
586 p->prio > swap_info[swap_list.next]->prio)
587 swap_list.next = p->type;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700588 nr_swap_pages++;
589 p->inuse_pages--;
Nitin Guptab3a27d02010-05-17 11:02:43 +0530590 if ((p->flags & SWP_BLKDEV) &&
591 disk->fops->swap_slot_free_notify)
592 disk->fops->swap_slot_free_notify(p->bdev, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800594
595 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
598/*
599 * Caller has made sure that the swapdevice corresponding to entry
600 * is still around or has not been recycled.
601 */
602void swap_free(swp_entry_t entry)
603{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800604 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 p = swap_info_get(entry);
607 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800608 swap_entry_free(p, entry, 1);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700609 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
611}
612
613/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700614 * Called after dropping swapcache to decrease refcnt to swap entries.
615 */
616void swapcache_free(swp_entry_t entry, struct page *page)
617{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700618 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800619 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700620
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700621 p = swap_info_get(entry);
622 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800623 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
624 if (page)
625 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700626 spin_unlock(&swap_lock);
627 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700628}
629
630/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700631 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800632 * This does not give an exact answer when swap count is continued,
633 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700635static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700637 int count = 0;
638 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 swp_entry_t entry;
640
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700641 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 p = swap_info_get(entry);
643 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700644 count = swap_count(p->swap_map[swp_offset(entry)]);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700645 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700647 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
649
650/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800651 * We can write to an anon page without COW if there are no other references
652 * to it. And as a side-effect, free up its swap: because the old content
653 * on disk will never be read, and seeking back there to write new content
654 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800656int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700658 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Hugh Dickins51726b12009-01-06 14:39:25 -0800660 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800661 if (unlikely(PageKsm(page)))
662 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700663 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800664 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700665 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800666 if (count == 1 && !PageWriteback(page)) {
667 delete_from_swap_cache(page);
668 SetPageDirty(page);
669 }
670 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800671 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
674/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800675 * If swap is getting full, or if there are no more mappings of this page,
676 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800678int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Hugh Dickins51726b12009-01-06 14:39:25 -0800680 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 if (!PageSwapCache(page))
683 return 0;
684 if (PageWriteback(page))
685 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800686 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 return 0;
688
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800689 delete_from_swap_cache(page);
690 SetPageDirty(page);
691 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700692}
693
694/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 * Free the swap entry like above, but also try to
696 * free the page cache entry if it is the last user.
697 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800698int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800700 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 struct page *page = NULL;
702
Andi Kleena7420aa2009-09-16 11:50:05 +0200703 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800704 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700705
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 p = swap_info_get(entry);
707 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800708 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Nick Piggin93fac702006-03-31 02:29:56 -0800709 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700710 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800711 page_cache_release(page);
712 page = NULL;
713 }
714 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700715 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 }
717 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800718 /*
719 * Not mapped elsewhere, or swap space full? Free it!
720 * Also recheck PageSwapCache now page is locked (above).
721 */
Nick Piggin93fac702006-03-31 02:29:56 -0800722 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800723 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 delete_from_swap_cache(page);
725 SetPageDirty(page);
726 }
727 unlock_page(page);
728 page_cache_release(page);
729 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800730 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732
Daisuke Nishimura02491442010-03-10 15:22:17 -0800733#ifdef CONFIG_CGROUP_MEM_RES_CTLR
734/**
735 * mem_cgroup_count_swap_user - count the user of a swap entry
736 * @ent: the swap entry to be checked
737 * @pagep: the pointer for the swap cache page of the entry to be stored
738 *
739 * Returns the number of the user of the swap entry. The number is valid only
740 * for swaps of anonymous pages.
741 * If the entry is found on swap cache, the page is stored to pagep with
742 * refcount of it being incremented.
743 */
744int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep)
745{
746 struct page *page;
747 struct swap_info_struct *p;
748 int count = 0;
749
750 page = find_get_page(&swapper_space, ent.val);
751 if (page)
752 count += page_mapcount(page);
753 p = swap_info_get(ent);
754 if (p) {
755 count += swap_count(p->swap_map[swp_offset(ent)]);
756 spin_unlock(&swap_lock);
757 }
758
759 *pagep = page;
760 return count;
761}
762#endif
763
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200764#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800765/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800766 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800767 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800768 * @offset - number of the PAGE_SIZE-sized block of the device, starting
769 * from 0, in which the swap header is expected to be located.
770 *
771 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800772 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800773int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800774{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800775 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800776 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800777
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800778 if (device)
779 bdev = bdget(device);
780
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800781 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800782 for (type = 0; type < nr_swapfiles; type++) {
783 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800784
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800785 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800786 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700787
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800788 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800789 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200790 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800791
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800792 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800793 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800794 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800795 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800796 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800797
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800798 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800799 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200800 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800801
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800802 spin_unlock(&swap_lock);
803 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800804 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800805 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800806 }
807 }
808 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800809 if (bdev)
810 bdput(bdev);
811
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800812 return -ENODEV;
813}
814
815/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800816 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
817 * corresponding to given index in swap_info (swap type).
818 */
819sector_t swapdev_block(int type, pgoff_t offset)
820{
821 struct block_device *bdev;
822
823 if ((unsigned int)type >= nr_swapfiles)
824 return 0;
825 if (!(swap_info[type]->flags & SWP_WRITEOK))
826 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800827 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800828}
829
830/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800831 * Return either the total number of swap pages of given type, or the number
832 * of free pages of that type (depending on @free)
833 *
834 * This is needed for software suspend
835 */
836unsigned int count_swap_pages(int type, int free)
837{
838 unsigned int n = 0;
839
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800840 spin_lock(&swap_lock);
841 if ((unsigned int)type < nr_swapfiles) {
842 struct swap_info_struct *sis = swap_info[type];
843
844 if (sis->flags & SWP_WRITEOK) {
845 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800846 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800847 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800848 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800849 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800850 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800851 return n;
852}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800853#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700856 * No need to decide whether this PTE shares the swap entry with others,
857 * just let do_wp_page work it out if a write is requested later - to
858 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800860static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 unsigned long addr, swp_entry_t entry, struct page *page)
862{
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800863 struct mem_cgroup *ptr = NULL;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800864 spinlock_t *ptl;
865 pte_t *pte;
866 int ret = 1;
867
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800868 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800869 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800870 goto out_nolock;
871 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800872
873 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
874 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
875 if (ret > 0)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800876 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800877 ret = 0;
878 goto out;
879 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800880
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800881 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800882 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 get_page(page);
884 set_pte_at(vma->vm_mm, addr, pte,
885 pte_mkold(mk_pte(page, vma->vm_page_prot)));
886 page_add_anon_rmap(page, vma, addr);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800887 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 swap_free(entry);
889 /*
890 * Move the page to the active list so it is not
891 * immediately swapped out again after swapon.
892 */
893 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800894out:
895 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800896out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800897 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
900static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
901 unsigned long addr, unsigned long end,
902 swp_entry_t entry, struct page *page)
903{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700905 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800906 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Hugh Dickins044d66c2008-02-07 00:14:04 -0800908 /*
909 * We don't actually need pte lock while scanning for swp_pte: since
910 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
911 * page table while we're scanning; though it could get zapped, and on
912 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
913 * of unmatched parts which look like swp_pte, so unuse_pte must
914 * recheck under pte lock. Scanning without pte lock lets it be
915 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
916 */
917 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 do {
919 /*
920 * swapoff spends a _lot_ of time in this loop!
921 * Test inline before going to call unuse_pte.
922 */
923 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800924 pte_unmap(pte);
925 ret = unuse_pte(vma, pmd, addr, entry, page);
926 if (ret)
927 goto out;
928 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 }
930 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800931 pte_unmap(pte - 1);
932out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800933 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934}
935
936static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
937 unsigned long addr, unsigned long end,
938 swp_entry_t entry, struct page *page)
939{
940 pmd_t *pmd;
941 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800942 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 pmd = pmd_offset(pud, addr);
945 do {
946 next = pmd_addr_end(addr, end);
947 if (pmd_none_or_clear_bad(pmd))
948 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800949 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
950 if (ret)
951 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 } while (pmd++, addr = next, addr != end);
953 return 0;
954}
955
956static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
957 unsigned long addr, unsigned long end,
958 swp_entry_t entry, struct page *page)
959{
960 pud_t *pud;
961 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800962 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
964 pud = pud_offset(pgd, addr);
965 do {
966 next = pud_addr_end(addr, end);
967 if (pud_none_or_clear_bad(pud))
968 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800969 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
970 if (ret)
971 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 } while (pud++, addr = next, addr != end);
973 return 0;
974}
975
976static int unuse_vma(struct vm_area_struct *vma,
977 swp_entry_t entry, struct page *page)
978{
979 pgd_t *pgd;
980 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800981 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800983 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 addr = page_address_in_vma(page, vma);
985 if (addr == -EFAULT)
986 return 0;
987 else
988 end = addr + PAGE_SIZE;
989 } else {
990 addr = vma->vm_start;
991 end = vma->vm_end;
992 }
993
994 pgd = pgd_offset(vma->vm_mm, addr);
995 do {
996 next = pgd_addr_end(addr, end);
997 if (pgd_none_or_clear_bad(pgd))
998 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800999 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1000 if (ret)
1001 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 } while (pgd++, addr = next, addr != end);
1003 return 0;
1004}
1005
1006static int unuse_mm(struct mm_struct *mm,
1007 swp_entry_t entry, struct page *page)
1008{
1009 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001010 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
1012 if (!down_read_trylock(&mm->mmap_sem)) {
1013 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001014 * Activate page so shrink_inactive_list is unlikely to unmap
1015 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001017 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 unlock_page(page);
1019 down_read(&mm->mmap_sem);
1020 lock_page(page);
1021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001023 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 break;
1025 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001027 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028}
1029
1030/*
1031 * Scan swap_map from current position to next entry still in use.
1032 * Recycle to start on reaching the end, returning 0 when empty.
1033 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001034static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1035 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001037 unsigned int max = si->max;
1038 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001039 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001042 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 * for whether an entry is in use, not modifying it; false
1044 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001045 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 */
1047 for (;;) {
1048 if (++i >= max) {
1049 if (!prev) {
1050 i = 0;
1051 break;
1052 }
1053 /*
1054 * No entries in use at top of swap_map,
1055 * loop back to start and recheck there.
1056 */
1057 max = prev + 1;
1058 prev = 0;
1059 i = 1;
1060 }
1061 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001062 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 break;
1064 }
1065 return i;
1066}
1067
1068/*
1069 * We completely avoid races by reading each swap page in advance,
1070 * and then search for the process using it. All the necessary
1071 * page table adjustments can then be made atomically.
1072 */
1073static int try_to_unuse(unsigned int type)
1074{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001075 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001077 unsigned char *swap_map;
1078 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 struct page *page;
1080 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001081 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
1084 /*
1085 * When searching mms for an entry, a good strategy is to
1086 * start at the first mm we freed the previous entry from
1087 * (though actually we don't notice whether we or coincidence
1088 * freed the entry). Initialize this start_mm with a hold.
1089 *
1090 * A simpler strategy would be to start at the last mm we
1091 * freed the previous entry from; but that would take less
1092 * advantage of mmlist ordering, which clusters forked mms
1093 * together, child after parent. If we race with dup_mmap(), we
1094 * prefer to resolve parent before child, lest we miss entries
1095 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001096 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 */
1098 start_mm = &init_mm;
1099 atomic_inc(&init_mm.mm_users);
1100
1101 /*
1102 * Keep on scanning until all entries have gone. Usually,
1103 * one pass through swap_map is enough, but not necessarily:
1104 * there are races when an instance of an entry might be missed.
1105 */
1106 while ((i = find_next_to_unuse(si, i)) != 0) {
1107 if (signal_pending(current)) {
1108 retval = -EINTR;
1109 break;
1110 }
1111
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001112 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 * Get a page for the entry, using the existing swap
1114 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001115 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 */
1117 swap_map = &si->swap_map[i];
1118 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001119 page = read_swap_cache_async(entry,
1120 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 if (!page) {
1122 /*
1123 * Either swap_duplicate() failed because entry
1124 * has been freed independently, and will not be
1125 * reused since sys_swapoff() already disabled
1126 * allocation from here, or alloc_page() failed.
1127 */
1128 if (!*swap_map)
1129 continue;
1130 retval = -ENOMEM;
1131 break;
1132 }
1133
1134 /*
1135 * Don't hold on to start_mm if it looks like exiting.
1136 */
1137 if (atomic_read(&start_mm->mm_users) == 1) {
1138 mmput(start_mm);
1139 start_mm = &init_mm;
1140 atomic_inc(&init_mm.mm_users);
1141 }
1142
1143 /*
1144 * Wait for and lock page. When do_swap_page races with
1145 * try_to_unuse, do_swap_page can handle the fault much
1146 * faster than try_to_unuse can locate the entry. This
1147 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1148 * defer to do_swap_page in such a case - in some tests,
1149 * do_swap_page and try_to_unuse repeatedly compete.
1150 */
1151 wait_on_page_locked(page);
1152 wait_on_page_writeback(page);
1153 lock_page(page);
1154 wait_on_page_writeback(page);
1155
1156 /*
1157 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001160 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1161 retval = shmem_unuse(entry, page);
1162 /* page has already been unlocked and released */
1163 if (retval < 0)
1164 break;
1165 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001167 if (swap_count(swcount) && start_mm != &init_mm)
1168 retval = unuse_mm(start_mm, entry, page);
1169
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001170 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 int set_start_mm = (*swap_map >= swcount);
1172 struct list_head *p = &start_mm->mmlist;
1173 struct mm_struct *new_start_mm = start_mm;
1174 struct mm_struct *prev_mm = start_mm;
1175 struct mm_struct *mm;
1176
1177 atomic_inc(&new_start_mm->mm_users);
1178 atomic_inc(&prev_mm->mm_users);
1179 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001180 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 (p = p->next) != &start_mm->mmlist) {
1182 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001183 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 spin_unlock(&mmlist_lock);
1186 mmput(prev_mm);
1187 prev_mm = mm;
1188
1189 cond_resched();
1190
1191 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001192 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001194 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001196 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001198
Bo Liu32c5fc12009-11-02 16:50:33 +00001199 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 mmput(new_start_mm);
1201 atomic_inc(&mm->mm_users);
1202 new_start_mm = mm;
1203 set_start_mm = 0;
1204 }
1205 spin_lock(&mmlist_lock);
1206 }
1207 spin_unlock(&mmlist_lock);
1208 mmput(prev_mm);
1209 mmput(start_mm);
1210 start_mm = new_start_mm;
1211 }
1212 if (retval) {
1213 unlock_page(page);
1214 page_cache_release(page);
1215 break;
1216 }
1217
1218 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 * If a reference remains (rare), we would like to leave
1220 * the page in the swap cache; but try_to_unmap could
1221 * then re-duplicate the entry once we drop page lock,
1222 * so we might loop indefinitely; also, that page could
1223 * not be swapped out to other storage meanwhile. So:
1224 * delete from cache even if there's another reference,
1225 * after ensuring that the data has been saved to disk -
1226 * since if the reference remains (rarer), it will be
1227 * read from disk into another page. Splitting into two
1228 * pages would be incorrect if swap supported "shared
1229 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001230 *
1231 * Given how unuse_vma() targets one particular offset
1232 * in an anon_vma, once the anon_vma has been determined,
1233 * this splitting happens to be just what is needed to
1234 * handle where KSM pages have been swapped out: re-reading
1235 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001237 if (swap_count(*swap_map) &&
1238 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 struct writeback_control wbc = {
1240 .sync_mode = WB_SYNC_NONE,
1241 };
1242
1243 swap_writepage(page, &wbc);
1244 lock_page(page);
1245 wait_on_page_writeback(page);
1246 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001247
1248 /*
1249 * It is conceivable that a racing task removed this page from
1250 * swap cache just before we acquired the page lock at the top,
1251 * or while we dropped it in unuse_mm(). The page might even
1252 * be back in swap cache on another swap area: that we must not
1253 * delete, since it may not have been written out to swap yet.
1254 */
1255 if (PageSwapCache(page) &&
1256 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001257 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 /*
1260 * So we could skip searching mms once swap count went
1261 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001262 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 */
1264 SetPageDirty(page);
1265 unlock_page(page);
1266 page_cache_release(page);
1267
1268 /*
1269 * Make sure that we aren't completely killing
1270 * interactive performance.
1271 */
1272 cond_resched();
1273 }
1274
1275 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 return retval;
1277}
1278
1279/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001280 * After a successful try_to_unuse, if no swap is now in use, we know
1281 * we can empty the mmlist. swap_lock must be held on entry and exit.
1282 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 * added to the mmlist just after page_duplicate - before would be racy.
1284 */
1285static void drain_mmlist(void)
1286{
1287 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001288 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001290 for (type = 0; type < nr_swapfiles; type++)
1291 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 return;
1293 spin_lock(&mmlist_lock);
1294 list_for_each_safe(p, next, &init_mm.mmlist)
1295 list_del_init(p);
1296 spin_unlock(&mmlist_lock);
1297}
1298
1299/*
1300 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001301 * corresponds to page offset for the specified swap entry.
1302 * Note that the type of this function is sector_t, but it returns page offset
1303 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001305static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001307 struct swap_info_struct *sis;
1308 struct swap_extent *start_se;
1309 struct swap_extent *se;
1310 pgoff_t offset;
1311
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001312 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001313 *bdev = sis->bdev;
1314
1315 offset = swp_offset(entry);
1316 start_se = sis->curr_swap_extent;
1317 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
1319 for ( ; ; ) {
1320 struct list_head *lh;
1321
1322 if (se->start_page <= offset &&
1323 offset < (se->start_page + se->nr_pages)) {
1324 return se->start_block + (offset - se->start_page);
1325 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001326 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 se = list_entry(lh, struct swap_extent, list);
1328 sis->curr_swap_extent = se;
1329 BUG_ON(se == start_se); /* It *must* be present */
1330 }
1331}
1332
1333/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001334 * Returns the page offset into bdev for the specified page's swap entry.
1335 */
1336sector_t map_swap_page(struct page *page, struct block_device **bdev)
1337{
1338 swp_entry_t entry;
1339 entry.val = page_private(page);
1340 return map_swap_entry(entry, bdev);
1341}
1342
1343/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 * Free all of a swapdev's extent information
1345 */
1346static void destroy_swap_extents(struct swap_info_struct *sis)
1347{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001348 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 struct swap_extent *se;
1350
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001351 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 struct swap_extent, list);
1353 list_del(&se->list);
1354 kfree(se);
1355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356}
1357
1358/*
1359 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001360 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001362 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 */
1364static int
1365add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1366 unsigned long nr_pages, sector_t start_block)
1367{
1368 struct swap_extent *se;
1369 struct swap_extent *new_se;
1370 struct list_head *lh;
1371
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001372 if (start_page == 0) {
1373 se = &sis->first_swap_extent;
1374 sis->curr_swap_extent = se;
1375 se->start_page = 0;
1376 se->nr_pages = nr_pages;
1377 se->start_block = start_block;
1378 return 1;
1379 } else {
1380 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001382 BUG_ON(se->start_page + se->nr_pages != start_page);
1383 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 /* Merge it */
1385 se->nr_pages += nr_pages;
1386 return 0;
1387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 }
1389
1390 /*
1391 * No merge. Insert a new extent, preserving ordering.
1392 */
1393 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1394 if (new_se == NULL)
1395 return -ENOMEM;
1396 new_se->start_page = start_page;
1397 new_se->nr_pages = nr_pages;
1398 new_se->start_block = start_block;
1399
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001400 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001401 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402}
1403
1404/*
1405 * A `swap extent' is a simple thing which maps a contiguous range of pages
1406 * onto a contiguous range of disk blocks. An ordered list of swap extents
1407 * is built at swapon time and is then used at swap_writepage/swap_readpage
1408 * time for locating where on disk a page belongs.
1409 *
1410 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1411 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1412 * swap files identically.
1413 *
1414 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1415 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1416 * swapfiles are handled *identically* after swapon time.
1417 *
1418 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1419 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1420 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1421 * requirements, they are simply tossed out - we will never use those blocks
1422 * for swapping.
1423 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001424 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1426 * which will scribble on the fs.
1427 *
1428 * The amount of disk space which a single swap extent represents varies.
1429 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1430 * extents in the list. To avoid much list walking, we cache the previous
1431 * search location in `curr_swap_extent', and start new searches from there.
1432 * This is extremely effective. The average number of iterations in
1433 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1434 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001435static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
1437 struct inode *inode;
1438 unsigned blocks_per_page;
1439 unsigned long page_no;
1440 unsigned blkbits;
1441 sector_t probe_block;
1442 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001443 sector_t lowest_block = -1;
1444 sector_t highest_block = 0;
1445 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 int ret;
1447
1448 inode = sis->swap_file->f_mapping->host;
1449 if (S_ISBLK(inode->i_mode)) {
1450 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001451 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001452 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 }
1454
1455 blkbits = inode->i_blkbits;
1456 blocks_per_page = PAGE_SIZE >> blkbits;
1457
1458 /*
1459 * Map all the blocks into the extent list. This code doesn't try
1460 * to be very smart.
1461 */
1462 probe_block = 0;
1463 page_no = 0;
1464 last_block = i_size_read(inode) >> blkbits;
1465 while ((probe_block + blocks_per_page) <= last_block &&
1466 page_no < sis->max) {
1467 unsigned block_in_page;
1468 sector_t first_block;
1469
1470 first_block = bmap(inode, probe_block);
1471 if (first_block == 0)
1472 goto bad_bmap;
1473
1474 /*
1475 * It must be PAGE_SIZE aligned on-disk
1476 */
1477 if (first_block & (blocks_per_page - 1)) {
1478 probe_block++;
1479 goto reprobe;
1480 }
1481
1482 for (block_in_page = 1; block_in_page < blocks_per_page;
1483 block_in_page++) {
1484 sector_t block;
1485
1486 block = bmap(inode, probe_block + block_in_page);
1487 if (block == 0)
1488 goto bad_bmap;
1489 if (block != first_block + block_in_page) {
1490 /* Discontiguity */
1491 probe_block++;
1492 goto reprobe;
1493 }
1494 }
1495
Hugh Dickins53092a72005-09-03 15:54:34 -07001496 first_block >>= (PAGE_SHIFT - blkbits);
1497 if (page_no) { /* exclude the header page */
1498 if (first_block < lowest_block)
1499 lowest_block = first_block;
1500 if (first_block > highest_block)
1501 highest_block = first_block;
1502 }
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 /*
1505 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1506 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001507 ret = add_swap_extent(sis, page_no, 1, first_block);
1508 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001510 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 page_no++;
1512 probe_block += blocks_per_page;
1513reprobe:
1514 continue;
1515 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001516 ret = nr_extents;
1517 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001519 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001521 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001523out:
1524 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525bad_bmap:
1526 printk(KERN_ERR "swapon: swapfile has holes\n");
1527 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001528 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529}
1530
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001531SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001533 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001534 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 struct file *swap_file, *victim;
1536 struct address_space *mapping;
1537 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001538 char *pathname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 int i, type, prev;
1540 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 if (!capable(CAP_SYS_ADMIN))
1543 return -EPERM;
1544
1545 pathname = getname(specialfile);
1546 err = PTR_ERR(pathname);
1547 if (IS_ERR(pathname))
1548 goto out;
1549
1550 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1551 putname(pathname);
1552 err = PTR_ERR(victim);
1553 if (IS_ERR(victim))
1554 goto out;
1555
1556 mapping = victim->f_mapping;
1557 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001558 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001559 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1560 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001561 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if (p->swap_file->f_mapping == mapping)
1563 break;
1564 }
1565 prev = type;
1566 }
1567 if (type < 0) {
1568 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001569 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 goto out_dput;
1571 }
1572 if (!security_vm_enough_memory(p->pages))
1573 vm_unacct_memory(p->pages);
1574 else {
1575 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001576 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 goto out_dput;
1578 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001579 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001581 else
1582 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 if (type == swap_list.next) {
1584 /* just pick something that's safe... */
1585 swap_list.next = swap_list.head;
1586 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001587 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001588 for (i = p->next; i >= 0; i = swap_info[i]->next)
1589 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001590 least_priority++;
1591 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 nr_swap_pages -= p->pages;
1593 total_swap_pages -= p->pages;
1594 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001595 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001596
Hugh Dickins35451be2009-09-21 17:02:27 -07001597 current->flags |= PF_OOM_ORIGIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 err = try_to_unuse(type);
Hugh Dickins35451be2009-09-21 17:02:27 -07001599 current->flags &= ~PF_OOM_ORIGIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 if (err) {
1602 /* re-insert swap space back into swap_list */
Hugh Dickins5d337b92005-09-03 15:54:41 -07001603 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001604 if (p->prio < 0)
1605 p->prio = --least_priority;
1606 prev = -1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001607 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1608 if (p->prio >= swap_info[i]->prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 break;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001610 prev = i;
1611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 p->next = i;
1613 if (prev < 0)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001614 swap_list.head = swap_list.next = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 else
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001616 swap_info[prev]->next = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 nr_swap_pages += p->pages;
1618 total_swap_pages += p->pages;
1619 p->flags |= SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001620 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 goto out_dput;
1622 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001623
1624 /* wait for any unplug function to finish */
1625 down_write(&swap_unplug_sem);
1626 up_write(&swap_unplug_sem);
1627
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001628 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001629 if (p->flags & SWP_CONTINUED)
1630 free_swap_count_continuations(p);
1631
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001632 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001633 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001635
1636 /* wait for anyone still in scan_swap_map */
1637 p->highest_bit = 0; /* cuts scans short */
1638 while (p->flags >= SWP_SCANNING) {
1639 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001640 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001641 spin_lock(&swap_lock);
1642 }
1643
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 swap_file = p->swap_file;
1645 p->swap_file = NULL;
1646 p->max = 0;
1647 swap_map = p->swap_map;
1648 p->swap_map = NULL;
1649 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001650 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001651 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 vfree(swap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001653 /* Destroy swap account informatin */
1654 swap_cgroup_swapoff(type);
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 inode = mapping->host;
1657 if (S_ISBLK(inode->i_mode)) {
1658 struct block_device *bdev = I_BDEV(inode);
1659 set_blocksize(bdev, p->old_block_size);
1660 bd_release(bdev);
1661 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001662 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001664 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 }
1666 filp_close(swap_file, NULL);
1667 err = 0;
1668
1669out_dput:
1670 filp_close(victim, NULL);
1671out:
1672 return err;
1673}
1674
1675#ifdef CONFIG_PROC_FS
1676/* iterator */
1677static void *swap_start(struct seq_file *swap, loff_t *pos)
1678{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001679 struct swap_info_struct *si;
1680 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 loff_t l = *pos;
1682
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001683 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001685 if (!l)
1686 return SEQ_START_TOKEN;
1687
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001688 for (type = 0; type < nr_swapfiles; type++) {
1689 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1690 si = swap_info[type];
1691 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001693 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001694 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 }
1696
1697 return NULL;
1698}
1699
1700static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1701{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001702 struct swap_info_struct *si = v;
1703 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001705 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001706 type = 0;
1707 else
1708 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001709
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001710 for (; type < nr_swapfiles; type++) {
1711 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1712 si = swap_info[type];
1713 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 continue;
1715 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001716 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 }
1718
1719 return NULL;
1720}
1721
1722static void swap_stop(struct seq_file *swap, void *v)
1723{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001724 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725}
1726
1727static int swap_show(struct seq_file *swap, void *v)
1728{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001729 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 struct file *file;
1731 int len;
1732
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001733 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001734 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1735 return 0;
1736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001738 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001739 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001740 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001741 len < 40 ? 40 - len : 1, " ",
1742 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001744 si->pages << (PAGE_SHIFT - 10),
1745 si->inuse_pages << (PAGE_SHIFT - 10),
1746 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 return 0;
1748}
1749
Helge Deller15ad7cd2006-12-06 20:40:36 -08001750static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 .start = swap_start,
1752 .next = swap_next,
1753 .stop = swap_stop,
1754 .show = swap_show
1755};
1756
1757static int swaps_open(struct inode *inode, struct file *file)
1758{
1759 return seq_open(file, &swaps_op);
1760}
1761
Helge Deller15ad7cd2006-12-06 20:40:36 -08001762static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 .open = swaps_open,
1764 .read = seq_read,
1765 .llseek = seq_lseek,
1766 .release = seq_release,
1767};
1768
1769static int __init procswaps_init(void)
1770{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001771 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 return 0;
1773}
1774__initcall(procswaps_init);
1775#endif /* CONFIG_PROC_FS */
1776
Jan Beulich17963162008-12-16 11:35:24 +00001777#ifdef MAX_SWAPFILES_CHECK
1778static int __init max_swapfiles_check(void)
1779{
1780 MAX_SWAPFILES_CHECK();
1781 return 0;
1782}
1783late_initcall(max_swapfiles_check);
1784#endif
1785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786/*
1787 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1788 *
1789 * The swapon system call
1790 */
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001791SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001793 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 char *name = NULL;
1795 struct block_device *bdev = NULL;
1796 struct file *swap_file = NULL;
1797 struct address_space *mapping;
1798 unsigned int type;
1799 int i, prev;
1800 int error;
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001801 union swap_header *swap_header;
1802 unsigned int nr_good_pages;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001803 int nr_extents = 0;
Hugh Dickins53092a72005-09-03 15:54:34 -07001804 sector_t span;
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001805 unsigned long maxpages;
Hugh Dickins73fd8742009-01-06 14:39:47 -08001806 unsigned long swapfilepages;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001807 unsigned char *swap_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 struct page *page = NULL;
1809 struct inode *inode = NULL;
1810 int did_down = 0;
1811
1812 if (!capable(CAP_SYS_ADMIN))
1813 return -EPERM;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001814
1815 p = kzalloc(sizeof(*p), GFP_KERNEL);
1816 if (!p)
1817 return -ENOMEM;
1818
Hugh Dickins5d337b92005-09-03 15:54:41 -07001819 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001820 for (type = 0; type < nr_swapfiles; type++) {
1821 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 error = -EPERM;
Christoph Lameter06972122006-06-23 02:03:35 -07001825 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001826 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001827 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 goto out;
1829 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001830 if (type >= nr_swapfiles) {
1831 p->type = type;
1832 swap_info[type] = p;
1833 /*
1834 * Write swap_info[type] before nr_swapfiles, in case a
1835 * racing procfs swap_start() or swap_next() is reading them.
1836 * (We never shrink nr_swapfiles, we never free this entry.)
1837 */
1838 smp_wmb();
1839 nr_swapfiles++;
1840 } else {
1841 kfree(p);
1842 p = swap_info[type];
1843 /*
1844 * Do not memset this entry: a racing procfs swap_next()
1845 * would be relying on p->type to remain valid.
1846 */
1847 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001848 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001851 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 name = getname(specialfile);
1854 error = PTR_ERR(name);
1855 if (IS_ERR(name)) {
1856 name = NULL;
1857 goto bad_swap_2;
1858 }
1859 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1860 error = PTR_ERR(swap_file);
1861 if (IS_ERR(swap_file)) {
1862 swap_file = NULL;
1863 goto bad_swap_2;
1864 }
1865
1866 p->swap_file = swap_file;
1867 mapping = swap_file->f_mapping;
1868 inode = mapping->host;
1869
1870 error = -EBUSY;
1871 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001872 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874 if (i == type || !q->swap_file)
1875 continue;
1876 if (mapping == q->swap_file->f_mapping)
1877 goto bad_swap;
1878 }
1879
1880 error = -EINVAL;
1881 if (S_ISBLK(inode->i_mode)) {
1882 bdev = I_BDEV(inode);
1883 error = bd_claim(bdev, sys_swapon);
1884 if (error < 0) {
1885 bdev = NULL;
Rob Landleyf7b3a432005-09-22 21:44:27 -07001886 error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 goto bad_swap;
1888 }
1889 p->old_block_size = block_size(bdev);
1890 error = set_blocksize(bdev, PAGE_SIZE);
1891 if (error < 0)
1892 goto bad_swap;
1893 p->bdev = bdev;
Nitin Guptab2725642010-05-17 11:02:42 +05301894 p->flags |= SWP_BLKDEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 } else if (S_ISREG(inode->i_mode)) {
1896 p->bdev = inode->i_sb->s_bdev;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001897 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 did_down = 1;
1899 if (IS_SWAPFILE(inode)) {
1900 error = -EBUSY;
1901 goto bad_swap;
1902 }
1903 } else {
1904 goto bad_swap;
1905 }
1906
Hugh Dickins73fd8742009-01-06 14:39:47 -08001907 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
1909 /*
1910 * Read the swap header.
1911 */
1912 if (!mapping->a_ops->readpage) {
1913 error = -EINVAL;
1914 goto bad_swap;
1915 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07001916 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 if (IS_ERR(page)) {
1918 error = PTR_ERR(page);
1919 goto bad_swap;
1920 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001921 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
Hugh Dickins81e33972009-01-06 14:39:49 -08001923 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Jesper Juhle97a3112006-01-11 01:50:28 +01001924 printk(KERN_ERR "Unable to find swap-space signature\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 error = -EINVAL;
1926 goto bad_swap;
1927 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001928
Hugh Dickins81e33972009-01-06 14:39:49 -08001929 /* swap partition endianess hack... */
1930 if (swab32(swap_header->info.version) == 1) {
1931 swab32s(&swap_header->info.version);
1932 swab32s(&swap_header->info.last_page);
1933 swab32s(&swap_header->info.nr_badpages);
1934 for (i = 0; i < swap_header->info.nr_badpages; i++)
1935 swab32s(&swap_header->info.badpages[i]);
1936 }
1937 /* Check the swap header's sub-version */
1938 if (swap_header->info.version != 1) {
1939 printk(KERN_WARNING
1940 "Unable to handle swap header version %d\n",
1941 swap_header->info.version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 error = -EINVAL;
1943 goto bad_swap;
Hugh Dickins81e33972009-01-06 14:39:49 -08001944 }
1945
1946 p->lowest_bit = 1;
1947 p->cluster_next = 1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001948 p->cluster_nr = 0;
Hugh Dickins81e33972009-01-06 14:39:49 -08001949
1950 /*
1951 * Find out how many pages are allowed for a single swap
1952 * device. There are two limiting factors: 1) the number of
1953 * bits for the swap offset in the swp_entry_t type and
1954 * 2) the number of bits in the a swap pte as defined by
1955 * the different architectures. In order to find the
1956 * largest possible bit mask a swap entry with swap type 0
1957 * and swap offset ~0UL is created, encoded to a swap pte,
1958 * decoded to a swp_entry_t again and finally the swap
1959 * offset is extracted. This will mask all the bits from
1960 * the initial ~0UL mask that can't be encoded in either
1961 * the swp_entry_t or the architecture definition of a
1962 * swap pte.
1963 */
1964 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001965 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
1966 if (maxpages > swap_header->info.last_page) {
1967 maxpages = swap_header->info.last_page + 1;
1968 /* p->max is an unsigned int: don't overflow it */
1969 if ((unsigned int)maxpages == 0)
1970 maxpages = UINT_MAX;
1971 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001972 p->highest_bit = maxpages - 1;
1973
1974 error = -EINVAL;
1975 if (!maxpages)
1976 goto bad_swap;
1977 if (swapfilepages && maxpages > swapfilepages) {
1978 printk(KERN_WARNING
1979 "Swap area shorter than signature indicates\n");
1980 goto bad_swap;
1981 }
1982 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1983 goto bad_swap;
1984 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1985 goto bad_swap;
1986
1987 /* OK, set up the swap map and apply the bad block list */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001988 swap_map = vmalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08001989 if (!swap_map) {
1990 error = -ENOMEM;
1991 goto bad_swap;
1992 }
1993
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001994 memset(swap_map, 0, maxpages);
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001995 nr_good_pages = maxpages - 1; /* omit header page */
1996
Hugh Dickins81e33972009-01-06 14:39:49 -08001997 for (i = 0; i < swap_header->info.nr_badpages; i++) {
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08001998 unsigned int page_nr = swap_header->info.badpages[i];
1999 if (page_nr == 0 || page_nr > swap_header->info.last_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 error = -EINVAL;
2001 goto bad_swap;
2002 }
Hugh Dickinsad2bd7e2010-03-05 13:42:12 -08002003 if (page_nr < maxpages) {
2004 swap_map[page_nr] = SWAP_MAP_BAD;
2005 nr_good_pages--;
2006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 }
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002008
2009 error = swap_cgroup_swapon(type, maxpages);
2010 if (error)
2011 goto bad_swap;
2012
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002013 if (nr_good_pages) {
Hugh Dickins78ecba02008-07-23 21:28:23 -07002014 swap_map[0] = SWAP_MAP_BAD;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002015 p->max = maxpages;
2016 p->pages = nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07002017 nr_extents = setup_swap_extents(p, &span);
2018 if (nr_extents < 0) {
2019 error = nr_extents;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002020 goto bad_swap;
Hugh Dickins53092a72005-09-03 15:54:34 -07002021 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07002022 nr_good_pages = p->pages;
2023 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 if (!nr_good_pages) {
2025 printk(KERN_WARNING "Empty swap-file\n");
2026 error = -EINVAL;
2027 goto bad_swap;
2028 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002030 if (p->bdev) {
2031 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2032 p->flags |= SWP_SOLIDSTATE;
2033 p->cluster_next = 1 + (random32() % p->highest_bit);
2034 }
2035 if (discard_swap(p) == 0)
2036 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002037 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002038
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002039 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002040 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07002041 if (swap_flags & SWAP_FLAG_PREFER)
2042 p->prio =
2043 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2044 else
2045 p->prio = --least_priority;
2046 p->swap_map = swap_map;
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08002047 p->flags |= SWP_WRITEOK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 nr_swap_pages += nr_good_pages;
2049 total_swap_pages += nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07002050
Hugh Dickins6eb396d2005-09-03 15:54:35 -07002051 printk(KERN_INFO "Adding %uk swap on %s. "
Hugh Dickins20137a42009-01-06 14:39:54 -08002052 "Priority:%d extents:%d across:%lluk %s%s\n",
Hugh Dickins53092a72005-09-03 15:54:34 -07002053 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002054 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
Hugh Dickins20137a42009-01-06 14:39:54 -08002055 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2056 (p->flags & SWP_DISCARDABLE) ? "D" : "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
2058 /* insert swap space into swap_list: */
2059 prev = -1;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002060 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
2061 if (p->prio >= swap_info[i]->prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 prev = i;
2064 }
2065 p->next = i;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002066 if (prev < 0)
2067 swap_list.head = swap_list.next = type;
2068 else
2069 swap_info[prev]->next = type;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002070 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002071 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 error = 0;
2073 goto out;
2074bad_swap:
2075 if (bdev) {
2076 set_blocksize(bdev, p->old_block_size);
2077 bd_release(bdev);
2078 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002079 destroy_swap_extents(p);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08002080 swap_cgroup_swapoff(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081bad_swap_2:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002082 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002085 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 vfree(swap_map);
2087 if (swap_file)
2088 filp_close(swap_file, NULL);
2089out:
2090 if (page && !IS_ERR(page)) {
2091 kunmap(page);
2092 page_cache_release(page);
2093 }
2094 if (name)
2095 putname(name);
2096 if (did_down) {
2097 if (!error)
2098 inode->i_flags |= S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002099 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 }
2101 return error;
2102}
2103
2104void si_swapinfo(struct sysinfo *val)
2105{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002106 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 unsigned long nr_to_be_unused = 0;
2108
Hugh Dickins5d337b92005-09-03 15:54:41 -07002109 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002110 for (type = 0; type < nr_swapfiles; type++) {
2111 struct swap_info_struct *si = swap_info[type];
2112
2113 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2114 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 }
2116 val->freeswap = nr_swap_pages + nr_to_be_unused;
2117 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002118 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119}
2120
2121/*
2122 * Verify that a swap entry is valid and increment its swap map count.
2123 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002124 * Returns error code in following case.
2125 * - success -> 0
2126 * - swp_entry is invalid -> EINVAL
2127 * - swp_entry is migration entry -> EINVAL
2128 * - swap-cache reference is requested but there is already one. -> EEXIST
2129 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002130 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002132static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002134 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002136 unsigned char count;
2137 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002138 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
Andi Kleena7420aa2009-09-16 11:50:05 +02002140 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002141 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002142
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 type = swp_type(entry);
2144 if (type >= nr_swapfiles)
2145 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002146 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 offset = swp_offset(entry);
2148
Hugh Dickins5d337b92005-09-03 15:54:41 -07002149 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002150 if (unlikely(offset >= p->max))
2151 goto unlock_out;
2152
Hugh Dickins253d5532009-12-14 17:58:44 -08002153 count = p->swap_map[offset];
2154 has_cache = count & SWAP_HAS_CACHE;
2155 count &= ~SWAP_HAS_CACHE;
2156 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002157
Hugh Dickins253d5532009-12-14 17:58:44 -08002158 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002159
2160 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002161 if (!has_cache && count)
2162 has_cache = SWAP_HAS_CACHE;
2163 else if (has_cache) /* someone else added cache */
2164 err = -EEXIST;
2165 else /* no users remaining */
2166 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002167
2168 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002169
Hugh Dickins570a335b2009-12-14 17:58:46 -08002170 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2171 count += usage;
2172 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002173 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002174 else if (swap_count_continued(p, offset, count))
2175 count = COUNT_CONTINUED;
2176 else
2177 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002178 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002179 err = -ENOENT; /* unused swap entry */
2180
2181 p->swap_map[offset] = count | has_cache;
2182
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002183unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002184 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002186 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
2188bad_file:
2189 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2190 goto out;
2191}
Hugh Dickins253d5532009-12-14 17:58:44 -08002192
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002193/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002194 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2195 * (in which case its reference count is never incremented).
2196 */
2197void swap_shmem_alloc(swp_entry_t entry)
2198{
2199 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2200}
2201
2202/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002203 * Increase reference count of swap entry by 1.
2204 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2205 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2206 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2207 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002208 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002209int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002210{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002211 int err = 0;
2212
2213 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2214 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2215 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002216}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002218/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002219 * @entry: swap entry for which we allocate swap cache.
2220 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002221 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002222 * This can return error codes. Returns 0 at success.
2223 * -EBUSY means there is a swap cache.
2224 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002225 */
2226int swapcache_prepare(swp_entry_t entry)
2227{
Hugh Dickins253d5532009-12-14 17:58:44 -08002228 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002229}
2230
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07002232 * swap_lock prevents swap_map being freed. Don't grab an extra
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 * reference on the swaphandle, it doesn't matter if it becomes unused.
2234 */
2235int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2236{
Hugh Dickins89528982008-02-04 22:28:45 -08002237 struct swap_info_struct *si;
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002238 int our_page_cluster = page_cluster;
Hugh Dickins89528982008-02-04 22:28:45 -08002239 pgoff_t target, toff;
2240 pgoff_t base, end;
2241 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002243 if (!our_page_cluster) /* no readahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 return 0;
Hugh Dickins89528982008-02-04 22:28:45 -08002245
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002246 si = swap_info[swp_type(entry)];
Hugh Dickins89528982008-02-04 22:28:45 -08002247 target = swp_offset(entry);
2248 base = (target >> our_page_cluster) << our_page_cluster;
2249 end = base + (1 << our_page_cluster);
2250 if (!base) /* first page is swap header */
2251 base++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
Hugh Dickins5d337b92005-09-03 15:54:41 -07002253 spin_lock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002254 if (end > si->max) /* don't go beyond end of map */
2255 end = si->max;
2256
2257 /* Count contiguous allocated slots above our target */
2258 for (toff = target; ++toff < end; nr_pages++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 /* Don't read in free or bad pages */
Hugh Dickins89528982008-02-04 22:28:45 -08002260 if (!si->swap_map[toff])
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002262 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 break;
Hugh Dickins89528982008-02-04 22:28:45 -08002264 }
2265 /* Count contiguous allocated slots below our target */
2266 for (toff = target; --toff >= base; nr_pages++) {
2267 /* Don't read in free or bad pages */
2268 if (!si->swap_map[toff])
2269 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002270 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Hugh Dickins89528982008-02-04 22:28:45 -08002271 break;
2272 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07002273 spin_unlock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002274
2275 /*
2276 * Indicate starting offset, and return number of pages to get:
2277 * if only 1, say 0, since there's then no readahead to be done.
2278 */
2279 *offset = ++toff;
2280 return nr_pages? ++nr_pages: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281}
Hugh Dickins570a335b2009-12-14 17:58:46 -08002282
2283/*
2284 * add_swap_count_continuation - called when a swap count is duplicated
2285 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2286 * page of the original vmalloc'ed swap_map, to hold the continuation count
2287 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2288 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2289 *
2290 * These continuation pages are seldom referenced: the common paths all work
2291 * on the original swap_map, only referring to a continuation page when the
2292 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2293 *
2294 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2295 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2296 * can be called after dropping locks.
2297 */
2298int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2299{
2300 struct swap_info_struct *si;
2301 struct page *head;
2302 struct page *page;
2303 struct page *list_page;
2304 pgoff_t offset;
2305 unsigned char count;
2306
2307 /*
2308 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2309 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2310 */
2311 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2312
2313 si = swap_info_get(entry);
2314 if (!si) {
2315 /*
2316 * An acceptable race has occurred since the failing
2317 * __swap_duplicate(): the swap entry has been freed,
2318 * perhaps even the whole swap_map cleared for swapoff.
2319 */
2320 goto outer;
2321 }
2322
2323 offset = swp_offset(entry);
2324 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2325
2326 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2327 /*
2328 * The higher the swap count, the more likely it is that tasks
2329 * will race to add swap count continuation: we need to avoid
2330 * over-provisioning.
2331 */
2332 goto out;
2333 }
2334
2335 if (!page) {
2336 spin_unlock(&swap_lock);
2337 return -ENOMEM;
2338 }
2339
2340 /*
2341 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2342 * no architecture is using highmem pages for kernel pagetables: so it
2343 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2344 */
2345 head = vmalloc_to_page(si->swap_map + offset);
2346 offset &= ~PAGE_MASK;
2347
2348 /*
2349 * Page allocation does not initialize the page's lru field,
2350 * but it does always reset its private field.
2351 */
2352 if (!page_private(head)) {
2353 BUG_ON(count & COUNT_CONTINUED);
2354 INIT_LIST_HEAD(&head->lru);
2355 set_page_private(head, SWP_CONTINUED);
2356 si->flags |= SWP_CONTINUED;
2357 }
2358
2359 list_for_each_entry(list_page, &head->lru, lru) {
2360 unsigned char *map;
2361
2362 /*
2363 * If the previous map said no continuation, but we've found
2364 * a continuation page, free our allocation and use this one.
2365 */
2366 if (!(count & COUNT_CONTINUED))
2367 goto out;
2368
2369 map = kmap_atomic(list_page, KM_USER0) + offset;
2370 count = *map;
2371 kunmap_atomic(map, KM_USER0);
2372
2373 /*
2374 * If this continuation count now has some space in it,
2375 * free our allocation and use this one.
2376 */
2377 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2378 goto out;
2379 }
2380
2381 list_add_tail(&page->lru, &head->lru);
2382 page = NULL; /* now it's attached, don't free it */
2383out:
2384 spin_unlock(&swap_lock);
2385outer:
2386 if (page)
2387 __free_page(page);
2388 return 0;
2389}
2390
2391/*
2392 * swap_count_continued - when the original swap_map count is incremented
2393 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2394 * into, carry if so, or else fail until a new continuation page is allocated;
2395 * when the original swap_map count is decremented from 0 with continuation,
2396 * borrow from the continuation and report whether it still holds more.
2397 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2398 */
2399static bool swap_count_continued(struct swap_info_struct *si,
2400 pgoff_t offset, unsigned char count)
2401{
2402 struct page *head;
2403 struct page *page;
2404 unsigned char *map;
2405
2406 head = vmalloc_to_page(si->swap_map + offset);
2407 if (page_private(head) != SWP_CONTINUED) {
2408 BUG_ON(count & COUNT_CONTINUED);
2409 return false; /* need to add count continuation */
2410 }
2411
2412 offset &= ~PAGE_MASK;
2413 page = list_entry(head->lru.next, struct page, lru);
2414 map = kmap_atomic(page, KM_USER0) + offset;
2415
2416 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2417 goto init_map; /* jump over SWAP_CONT_MAX checks */
2418
2419 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2420 /*
2421 * Think of how you add 1 to 999
2422 */
2423 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2424 kunmap_atomic(map, KM_USER0);
2425 page = list_entry(page->lru.next, struct page, lru);
2426 BUG_ON(page == head);
2427 map = kmap_atomic(page, KM_USER0) + offset;
2428 }
2429 if (*map == SWAP_CONT_MAX) {
2430 kunmap_atomic(map, KM_USER0);
2431 page = list_entry(page->lru.next, struct page, lru);
2432 if (page == head)
2433 return false; /* add count continuation */
2434 map = kmap_atomic(page, KM_USER0) + offset;
2435init_map: *map = 0; /* we didn't zero the page */
2436 }
2437 *map += 1;
2438 kunmap_atomic(map, KM_USER0);
2439 page = list_entry(page->lru.prev, struct page, lru);
2440 while (page != head) {
2441 map = kmap_atomic(page, KM_USER0) + offset;
2442 *map = COUNT_CONTINUED;
2443 kunmap_atomic(map, KM_USER0);
2444 page = list_entry(page->lru.prev, struct page, lru);
2445 }
2446 return true; /* incremented */
2447
2448 } else { /* decrementing */
2449 /*
2450 * Think of how you subtract 1 from 1000
2451 */
2452 BUG_ON(count != COUNT_CONTINUED);
2453 while (*map == COUNT_CONTINUED) {
2454 kunmap_atomic(map, KM_USER0);
2455 page = list_entry(page->lru.next, struct page, lru);
2456 BUG_ON(page == head);
2457 map = kmap_atomic(page, KM_USER0) + offset;
2458 }
2459 BUG_ON(*map == 0);
2460 *map -= 1;
2461 if (*map == 0)
2462 count = 0;
2463 kunmap_atomic(map, KM_USER0);
2464 page = list_entry(page->lru.prev, struct page, lru);
2465 while (page != head) {
2466 map = kmap_atomic(page, KM_USER0) + offset;
2467 *map = SWAP_CONT_MAX | count;
2468 count = COUNT_CONTINUED;
2469 kunmap_atomic(map, KM_USER0);
2470 page = list_entry(page->lru.prev, struct page, lru);
2471 }
2472 return count == COUNT_CONTINUED;
2473 }
2474}
2475
2476/*
2477 * free_swap_count_continuations - swapoff free all the continuation pages
2478 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2479 */
2480static void free_swap_count_continuations(struct swap_info_struct *si)
2481{
2482 pgoff_t offset;
2483
2484 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2485 struct page *head;
2486 head = vmalloc_to_page(si->swap_map + offset);
2487 if (page_private(head)) {
2488 struct list_head *this, *next;
2489 list_for_each_safe(this, next, &head->lru) {
2490 struct page *page;
2491 page = list_entry(this, struct page, lru);
2492 list_del(this);
2493 __free_page(page);
2494 }
2495 }
2496 }
2497}